* [RFCv2 1/6] ring: common functions for 'move head' ops
2024-09-06 13:13 ` [RFCv2 0/6] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
@ 2024-09-06 13:13 ` Konstantin Ananyev
2024-09-06 13:13 ` [RFCv2 2/6] ring: make copying functions generic Konstantin Ananyev
` (5 subsequent siblings)
6 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-06 13:13 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, Konstantin Ananyev
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
Note upfront: that change doesn't introduce any functional or
performance changes.
It is just a code-reordering for:
- code deduplication
- ability in future to re-use the same code to introduce new functionality
For each sync mode corresponding move_prod_head() and
move_cons_head() are nearly identical to each other,
the only differences are:
- do we need to use a @capacity to calculate number of entries or not.
- what we need to update (prod/cons) and what is used as
read-only counterpart.
So instead of having 2 copies of nearly identical functions,
introduce a new common one that could be used by both functions:
move_prod_head() and move_cons_head().
As another positive thing - we can get rid of referencing whole rte_ring
structure in that new common sub-function.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
---
lib/ring/rte_ring_c11_pvt.h | 134 +++++--------------------------
lib/ring/rte_ring_elem_pvt.h | 66 +++++++++++++++
lib/ring/rte_ring_generic_pvt.h | 121 ++++------------------------
lib/ring/rte_ring_hts_elem_pvt.h | 85 ++++++--------------
lib/ring/rte_ring_rts_elem_pvt.h | 85 ++++++--------------
5 files changed, 149 insertions(+), 342 deletions(-)
diff --git a/lib/ring/rte_ring_c11_pvt.h b/lib/ring/rte_ring_c11_pvt.h
index 629b2d9288..048933ddc6 100644
--- a/lib/ring/rte_ring_c11_pvt.h
+++ b/lib/ring/rte_ring_c11_pvt.h
@@ -28,41 +28,19 @@ __rte_ring_update_tail(struct rte_ring_headtail *ht, uint32_t old_val,
rte_atomic_store_explicit(&ht->tail, new_val, rte_memory_order_release);
}
-/**
- * @internal This function updates the producer head for enqueue
- *
- * @param r
- * A pointer to the ring structure
- * @param is_sp
- * Indicates whether multi-producer path is needed or not
- * @param n
- * The number of elements we will want to enqueue, i.e. how far should the
- * head be moved
- * @param behavior
- * RTE_RING_QUEUE_FIXED: Enqueue a fixed number of items from a ring
- * RTE_RING_QUEUE_VARIABLE: Enqueue as many items as possible from ring
- * @param old_head
- * Returns head value as it was before the move, i.e. where enqueue starts
- * @param new_head
- * Returns the current/new head value i.e. where enqueue finishes
- * @param free_entries
- * Returns the amount of free space in the ring BEFORE head was moved
- * @return
- * Actual number of objects enqueued.
- * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
- */
static __rte_always_inline unsigned int
-__rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
- unsigned int n, enum rte_ring_queue_behavior behavior,
- uint32_t *old_head, uint32_t *new_head,
- uint32_t *free_entries)
+__rte_ring_headtail_move_head(struct rte_ring_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity,
+ unsigned int is_st, unsigned int n,
+ enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head, uint32_t *entries)
{
- const uint32_t capacity = r->capacity;
- uint32_t cons_tail;
- unsigned int max = n;
+ uint32_t stail;
int success;
+ unsigned int max = n;
- *old_head = rte_atomic_load_explicit(&r->prod.head, rte_memory_order_relaxed);
+ *old_head = rte_atomic_load_explicit(&d->head,
+ rte_memory_order_relaxed);
do {
/* Reset n to the initial burst count */
n = max;
@@ -73,112 +51,36 @@ __rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
/* load-acquire synchronize with store-release of ht->tail
* in update_tail.
*/
- cons_tail = rte_atomic_load_explicit(&r->cons.tail,
+ stail = rte_atomic_load_explicit(&s->tail,
rte_memory_order_acquire);
/* The subtraction is done between two unsigned 32bits value
* (the result is always modulo 32 bits even if we have
- * *old_head > cons_tail). So 'free_entries' is always between 0
+ * *old_head > s->tail). So 'free_entries' is always between 0
* and capacity (which is < size).
*/
- *free_entries = (capacity + cons_tail - *old_head);
+ *entries = (capacity + stail - *old_head);
/* check that we have enough room in ring */
- if (unlikely(n > *free_entries))
+ if (unlikely(n > *entries))
n = (behavior == RTE_RING_QUEUE_FIXED) ?
- 0 : *free_entries;
+ 0 : *entries;
if (n == 0)
return 0;
*new_head = *old_head + n;
- if (is_sp) {
- r->prod.head = *new_head;
+ if (is_st) {
+ d->head = *new_head;
success = 1;
} else
/* on failure, *old_head is updated */
- success = rte_atomic_compare_exchange_strong_explicit(&r->prod.head,
- old_head, *new_head,
+ success = rte_atomic_compare_exchange_strong_explicit(
+ &d->head, old_head, *new_head,
rte_memory_order_relaxed,
rte_memory_order_relaxed);
} while (unlikely(success == 0));
return n;
}
-/**
- * @internal This function updates the consumer head for dequeue
- *
- * @param r
- * A pointer to the ring structure
- * @param is_sc
- * Indicates whether multi-consumer path is needed or not
- * @param n
- * The number of elements we will want to dequeue, i.e. how far should the
- * head be moved
- * @param behavior
- * RTE_RING_QUEUE_FIXED: Dequeue a fixed number of items from a ring
- * RTE_RING_QUEUE_VARIABLE: Dequeue as many items as possible from ring
- * @param old_head
- * Returns head value as it was before the move, i.e. where dequeue starts
- * @param new_head
- * Returns the current/new head value i.e. where dequeue finishes
- * @param entries
- * Returns the number of entries in the ring BEFORE head was moved
- * @return
- * - Actual number of objects dequeued.
- * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
- */
-static __rte_always_inline unsigned int
-__rte_ring_move_cons_head(struct rte_ring *r, int is_sc,
- unsigned int n, enum rte_ring_queue_behavior behavior,
- uint32_t *old_head, uint32_t *new_head,
- uint32_t *entries)
-{
- unsigned int max = n;
- uint32_t prod_tail;
- int success;
-
- /* move cons.head atomically */
- *old_head = rte_atomic_load_explicit(&r->cons.head, rte_memory_order_relaxed);
- do {
- /* Restore n as it may change every loop */
- n = max;
-
- /* Ensure the head is read before tail */
- rte_atomic_thread_fence(rte_memory_order_acquire);
-
- /* this load-acquire synchronize with store-release of ht->tail
- * in update_tail.
- */
- prod_tail = rte_atomic_load_explicit(&r->prod.tail,
- rte_memory_order_acquire);
-
- /* The subtraction is done between two unsigned 32bits value
- * (the result is always modulo 32 bits even if we have
- * cons_head > prod_tail). So 'entries' is always between 0
- * and size(ring)-1.
- */
- *entries = (prod_tail - *old_head);
-
- /* Set the actual entries for dequeue */
- if (n > *entries)
- n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries;
-
- if (unlikely(n == 0))
- return 0;
-
- *new_head = *old_head + n;
- if (is_sc) {
- r->cons.head = *new_head;
- success = 1;
- } else
- /* on failure, *old_head will be updated */
- success = rte_atomic_compare_exchange_strong_explicit(&r->cons.head,
- old_head, *new_head,
- rte_memory_order_relaxed,
- rte_memory_order_relaxed);
- } while (unlikely(success == 0));
- return n;
-}
-
#endif /* _RTE_RING_C11_PVT_H_ */
diff --git a/lib/ring/rte_ring_elem_pvt.h b/lib/ring/rte_ring_elem_pvt.h
index 4b80f58980..3a83668a08 100644
--- a/lib/ring/rte_ring_elem_pvt.h
+++ b/lib/ring/rte_ring_elem_pvt.h
@@ -293,6 +293,72 @@ __rte_ring_dequeue_elems(struct rte_ring *r, uint32_t cons_head,
#include "rte_ring_generic_pvt.h"
#endif
+/**
+ * @internal This function updates the producer head for enqueue
+ *
+ * @param r
+ * A pointer to the ring structure
+ * @param is_sp
+ * Indicates whether multi-producer path is needed or not
+ * @param n
+ * The number of elements we will want to enqueue, i.e. how far should the
+ * head be moved
+ * @param behavior
+ * RTE_RING_QUEUE_FIXED: Enqueue a fixed number of items from a ring
+ * RTE_RING_QUEUE_VARIABLE: Enqueue as many items as possible from ring
+ * @param old_head
+ * Returns head value as it was before the move, i.e. where enqueue starts
+ * @param new_head
+ * Returns the current/new head value i.e. where enqueue finishes
+ * @param free_entries
+ * Returns the amount of free space in the ring BEFORE head was moved
+ * @return
+ * Actual number of objects enqueued.
+ * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
+ */
+static __rte_always_inline unsigned int
+__rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
+ unsigned int n, enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head,
+ uint32_t *free_entries)
+{
+ return __rte_ring_headtail_move_head(&r->prod, &r->cons, r->capacity,
+ is_sp, n, behavior, old_head, new_head, free_entries);
+}
+
+/**
+ * @internal This function updates the consumer head for dequeue
+ *
+ * @param r
+ * A pointer to the ring structure
+ * @param is_sc
+ * Indicates whether multi-consumer path is needed or not
+ * @param n
+ * The number of elements we will want to dequeue, i.e. how far should the
+ * head be moved
+ * @param behavior
+ * RTE_RING_QUEUE_FIXED: Dequeue a fixed number of items from a ring
+ * RTE_RING_QUEUE_VARIABLE: Dequeue as many items as possible from ring
+ * @param old_head
+ * Returns head value as it was before the move, i.e. where dequeue starts
+ * @param new_head
+ * Returns the current/new head value i.e. where dequeue finishes
+ * @param entries
+ * Returns the number of entries in the ring BEFORE head was moved
+ * @return
+ * - Actual number of objects dequeued.
+ * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
+ */
+static __rte_always_inline unsigned int
+__rte_ring_move_cons_head(struct rte_ring *r, unsigned int is_sc,
+ unsigned int n, enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head,
+ uint32_t *entries)
+{
+ return __rte_ring_headtail_move_head(&r->cons, &r->prod, 0,
+ is_sc, n, behavior, old_head, new_head, entries);
+}
+
/**
* @internal Enqueue several objects on the ring
*
diff --git a/lib/ring/rte_ring_generic_pvt.h b/lib/ring/rte_ring_generic_pvt.h
index 457f41dab3..12f3595926 100644
--- a/lib/ring/rte_ring_generic_pvt.h
+++ b/lib/ring/rte_ring_generic_pvt.h
@@ -29,36 +29,13 @@ __rte_ring_update_tail(struct rte_ring_headtail *ht, uint32_t old_val,
ht->tail = new_val;
}
-/**
- * @internal This function updates the producer head for enqueue
- *
- * @param r
- * A pointer to the ring structure
- * @param is_sp
- * Indicates whether multi-producer path is needed or not
- * @param n
- * The number of elements we will want to enqueue, i.e. how far should the
- * head be moved
- * @param behavior
- * RTE_RING_QUEUE_FIXED: Enqueue a fixed number of items from a ring
- * RTE_RING_QUEUE_VARIABLE: Enqueue as many items as possible from ring
- * @param old_head
- * Returns head value as it was before the move, i.e. where enqueue starts
- * @param new_head
- * Returns the current/new head value i.e. where enqueue finishes
- * @param free_entries
- * Returns the amount of free space in the ring BEFORE head was moved
- * @return
- * Actual number of objects enqueued.
- * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
- */
static __rte_always_inline unsigned int
-__rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
- unsigned int n, enum rte_ring_queue_behavior behavior,
- uint32_t *old_head, uint32_t *new_head,
- uint32_t *free_entries)
+__rte_ring_headtail_move_head(struct rte_ring_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity,
+ unsigned int is_st, unsigned int n,
+ enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head, uint32_t *entries)
{
- const uint32_t capacity = r->capacity;
unsigned int max = n;
int success;
@@ -66,7 +43,7 @@ __rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
/* Reset n to the initial burst count */
n = max;
- *old_head = r->prod.head;
+ *old_head = d->head;
/* add rmb barrier to avoid load/load reorder in weak
* memory model. It is noop on x86
@@ -76,97 +53,27 @@ __rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
/*
* The subtraction is done between two unsigned 32bits value
* (the result is always modulo 32 bits even if we have
- * *old_head > cons_tail). So 'free_entries' is always between 0
+ * *old_head > s->tail). So 'free_entries' is always between 0
* and capacity (which is < size).
*/
- *free_entries = (capacity + r->cons.tail - *old_head);
+ *entries = (capacity + s->tail - *old_head);
/* check that we have enough room in ring */
- if (unlikely(n > *free_entries))
+ if (unlikely(n > *entries))
n = (behavior == RTE_RING_QUEUE_FIXED) ?
- 0 : *free_entries;
+ 0 : *entries;
if (n == 0)
return 0;
*new_head = *old_head + n;
- if (is_sp) {
- r->prod.head = *new_head;
+ if (is_st) {
+ d->head = *new_head;
success = 1;
} else
- success = rte_atomic32_cmpset((uint32_t *)(uintptr_t)&r->prod.head,
- *old_head, *new_head);
- } while (unlikely(success == 0));
- return n;
-}
-
-/**
- * @internal This function updates the consumer head for dequeue
- *
- * @param r
- * A pointer to the ring structure
- * @param is_sc
- * Indicates whether multi-consumer path is needed or not
- * @param n
- * The number of elements we will want to dequeue, i.e. how far should the
- * head be moved
- * @param behavior
- * RTE_RING_QUEUE_FIXED: Dequeue a fixed number of items from a ring
- * RTE_RING_QUEUE_VARIABLE: Dequeue as many items as possible from ring
- * @param old_head
- * Returns head value as it was before the move, i.e. where dequeue starts
- * @param new_head
- * Returns the current/new head value i.e. where dequeue finishes
- * @param entries
- * Returns the number of entries in the ring BEFORE head was moved
- * @return
- * - Actual number of objects dequeued.
- * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
- */
-static __rte_always_inline unsigned int
-__rte_ring_move_cons_head(struct rte_ring *r, unsigned int is_sc,
- unsigned int n, enum rte_ring_queue_behavior behavior,
- uint32_t *old_head, uint32_t *new_head,
- uint32_t *entries)
-{
- unsigned int max = n;
- int success;
-
- /* move cons.head atomically */
- do {
- /* Restore n as it may change every loop */
- n = max;
-
- *old_head = r->cons.head;
-
- /* add rmb barrier to avoid load/load reorder in weak
- * memory model. It is noop on x86
- */
- rte_smp_rmb();
-
- /* The subtraction is done between two unsigned 32bits value
- * (the result is always modulo 32 bits even if we have
- * cons_head > prod_tail). So 'entries' is always between 0
- * and size(ring)-1.
- */
- *entries = (r->prod.tail - *old_head);
-
- /* Set the actual entries for dequeue */
- if (n > *entries)
- n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries;
-
- if (unlikely(n == 0))
- return 0;
-
- *new_head = *old_head + n;
- if (is_sc) {
- r->cons.head = *new_head;
- rte_smp_rmb();
- success = 1;
- } else {
- success = rte_atomic32_cmpset((uint32_t *)(uintptr_t)&r->cons.head,
+ success = rte_atomic32_cmpset(
+ (uint32_t *)(uintptr_t)&d->head,
*old_head, *new_head);
- }
} while (unlikely(success == 0));
return n;
}
diff --git a/lib/ring/rte_ring_hts_elem_pvt.h b/lib/ring/rte_ring_hts_elem_pvt.h
index 91f5eeccb9..ed5f16879f 100644
--- a/lib/ring/rte_ring_hts_elem_pvt.h
+++ b/lib/ring/rte_ring_hts_elem_pvt.h
@@ -50,20 +50,16 @@ __rte_ring_hts_head_wait(const struct rte_ring_hts_headtail *ht,
}
}
-/**
- * @internal This function updates the producer head for enqueue
- */
-static __rte_always_inline unsigned int
-__rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
+static __rte_always_inline uint32_t
+__rte_ring_hts_move_head(struct rte_ring_hts_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity, unsigned int num,
enum rte_ring_queue_behavior behavior, uint32_t *old_head,
- uint32_t *free_entries)
+ uint32_t *entries)
{
uint32_t n;
union __rte_ring_hts_pos np, op;
- const uint32_t capacity = r->capacity;
-
- op.raw = rte_atomic_load_explicit(&r->hts_prod.ht.raw, rte_memory_order_acquire);
+ op.raw = rte_atomic_load_explicit(&d->ht.raw, rte_memory_order_acquire);
do {
/* Reset n to the initial burst count */
@@ -74,7 +70,7 @@ __rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
* make sure that we read prod head/tail *before*
* reading cons tail.
*/
- __rte_ring_hts_head_wait(&r->hts_prod, &op);
+ __rte_ring_hts_head_wait(d, &op);
/*
* The subtraction is done between two unsigned 32bits value
@@ -82,12 +78,12 @@ __rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
* *old_head > cons_tail). So 'free_entries' is always between 0
* and capacity (which is < size).
*/
- *free_entries = capacity + r->cons.tail - op.pos.head;
+ *entries = capacity + s->tail - op.pos.head;
/* check that we have enough room in ring */
- if (unlikely(n > *free_entries))
+ if (unlikely(n > *entries))
n = (behavior == RTE_RING_QUEUE_FIXED) ?
- 0 : *free_entries;
+ 0 : *entries;
if (n == 0)
break;
@@ -100,13 +96,25 @@ __rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
* - OOO reads of cons tail value
* - OOO copy of elems from the ring
*/
- } while (rte_atomic_compare_exchange_strong_explicit(&r->hts_prod.ht.raw,
+ } while (rte_atomic_compare_exchange_strong_explicit(&d->ht.raw,
(uint64_t *)(uintptr_t)&op.raw, np.raw,
- rte_memory_order_acquire, rte_memory_order_acquire) == 0);
+ rte_memory_order_acquire,
+ rte_memory_order_acquire) == 0);
*old_head = op.pos.head;
return n;
}
+/**
+ * @internal This function updates the producer head for enqueue
+ */
+static __rte_always_inline unsigned int
+__rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
+ enum rte_ring_queue_behavior behavior, uint32_t *old_head,
+ uint32_t *free_entries)
+{
+ return __rte_ring_hts_move_head(&r->hts_prod, &r->cons,
+ r->capacity, num, behavior, old_head, free_entries);
+}
/**
* @internal This function updates the consumer head for dequeue
@@ -116,51 +124,8 @@ __rte_ring_hts_move_cons_head(struct rte_ring *r, unsigned int num,
enum rte_ring_queue_behavior behavior, uint32_t *old_head,
uint32_t *entries)
{
- uint32_t n;
- union __rte_ring_hts_pos np, op;
-
- op.raw = rte_atomic_load_explicit(&r->hts_cons.ht.raw, rte_memory_order_acquire);
-
- /* move cons.head atomically */
- do {
- /* Restore n as it may change every loop */
- n = num;
-
- /*
- * wait for tail to be equal to head,
- * make sure that we read cons head/tail *before*
- * reading prod tail.
- */
- __rte_ring_hts_head_wait(&r->hts_cons, &op);
-
- /* The subtraction is done between two unsigned 32bits value
- * (the result is always modulo 32 bits even if we have
- * cons_head > prod_tail). So 'entries' is always between 0
- * and size(ring)-1.
- */
- *entries = r->prod.tail - op.pos.head;
-
- /* Set the actual entries for dequeue */
- if (n > *entries)
- n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries;
-
- if (unlikely(n == 0))
- break;
-
- np.pos.tail = op.pos.tail;
- np.pos.head = op.pos.head + n;
-
- /*
- * this CAS(ACQUIRE, ACQUIRE) serves as a hoist barrier to prevent:
- * - OOO reads of prod tail value
- * - OOO copy of elems from the ring
- */
- } while (rte_atomic_compare_exchange_strong_explicit(&r->hts_cons.ht.raw,
- (uint64_t *)(uintptr_t)&op.raw, np.raw,
- rte_memory_order_acquire, rte_memory_order_acquire) == 0);
-
- *old_head = op.pos.head;
- return n;
+ return __rte_ring_hts_move_head(&r->hts_cons, &r->prod,
+ 0, num, behavior, old_head, entries);
}
/**
diff --git a/lib/ring/rte_ring_rts_elem_pvt.h b/lib/ring/rte_ring_rts_elem_pvt.h
index 122650346b..027409a3fa 100644
--- a/lib/ring/rte_ring_rts_elem_pvt.h
+++ b/lib/ring/rte_ring_rts_elem_pvt.h
@@ -64,20 +64,17 @@ __rte_ring_rts_head_wait(const struct rte_ring_rts_headtail *ht,
}
}
-/**
- * @internal This function updates the producer head for enqueue.
- */
static __rte_always_inline uint32_t
-__rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
+__rte_ring_rts_move_head(struct rte_ring_rts_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity, uint32_t num,
enum rte_ring_queue_behavior behavior, uint32_t *old_head,
- uint32_t *free_entries)
+ uint32_t *entries)
{
uint32_t n;
union __rte_ring_rts_poscnt nh, oh;
- const uint32_t capacity = r->capacity;
-
- oh.raw = rte_atomic_load_explicit(&r->rts_prod.head.raw, rte_memory_order_acquire);
+ oh.raw = rte_atomic_load_explicit(&d->head.raw,
+ rte_memory_order_acquire);
do {
/* Reset n to the initial burst count */
@@ -88,7 +85,7 @@ __rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
* make sure that we read prod head *before*
* reading cons tail.
*/
- __rte_ring_rts_head_wait(&r->rts_prod, &oh);
+ __rte_ring_rts_head_wait(d, &oh);
/*
* The subtraction is done between two unsigned 32bits value
@@ -96,12 +93,12 @@ __rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
* *old_head > cons_tail). So 'free_entries' is always between 0
* and capacity (which is < size).
*/
- *free_entries = capacity + r->cons.tail - oh.val.pos;
+ *entries = capacity + s->tail - oh.val.pos;
/* check that we have enough room in ring */
- if (unlikely(n > *free_entries))
+ if (unlikely(n > *entries))
n = (behavior == RTE_RING_QUEUE_FIXED) ?
- 0 : *free_entries;
+ 0 : *entries;
if (n == 0)
break;
@@ -114,14 +111,27 @@ __rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
* - OOO reads of cons tail value
* - OOO copy of elems to the ring
*/
- } while (rte_atomic_compare_exchange_strong_explicit(&r->rts_prod.head.raw,
+ } while (rte_atomic_compare_exchange_strong_explicit(&d->head.raw,
(uint64_t *)(uintptr_t)&oh.raw, nh.raw,
- rte_memory_order_acquire, rte_memory_order_acquire) == 0);
+ rte_memory_order_acquire,
+ rte_memory_order_acquire) == 0);
*old_head = oh.val.pos;
return n;
}
+/**
+ * @internal This function updates the producer head for enqueue.
+ */
+static __rte_always_inline uint32_t
+__rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
+ enum rte_ring_queue_behavior behavior, uint32_t *old_head,
+ uint32_t *free_entries)
+{
+ return __rte_ring_rts_move_head(&r->rts_prod, &r->cons,
+ r->capacity, num, behavior, old_head, free_entries);
+}
+
/**
* @internal This function updates the consumer head for dequeue
*/
@@ -130,51 +140,8 @@ __rte_ring_rts_move_cons_head(struct rte_ring *r, uint32_t num,
enum rte_ring_queue_behavior behavior, uint32_t *old_head,
uint32_t *entries)
{
- uint32_t n;
- union __rte_ring_rts_poscnt nh, oh;
-
- oh.raw = rte_atomic_load_explicit(&r->rts_cons.head.raw, rte_memory_order_acquire);
-
- /* move cons.head atomically */
- do {
- /* Restore n as it may change every loop */
- n = num;
-
- /*
- * wait for cons head/tail distance,
- * make sure that we read cons head *before*
- * reading prod tail.
- */
- __rte_ring_rts_head_wait(&r->rts_cons, &oh);
-
- /* The subtraction is done between two unsigned 32bits value
- * (the result is always modulo 32 bits even if we have
- * cons_head > prod_tail). So 'entries' is always between 0
- * and size(ring)-1.
- */
- *entries = r->prod.tail - oh.val.pos;
-
- /* Set the actual entries for dequeue */
- if (n > *entries)
- n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries;
-
- if (unlikely(n == 0))
- break;
-
- nh.val.pos = oh.val.pos + n;
- nh.val.cnt = oh.val.cnt + 1;
-
- /*
- * this CAS(ACQUIRE, ACQUIRE) serves as a hoist barrier to prevent:
- * - OOO reads of prod tail value
- * - OOO copy of elems from the ring
- */
- } while (rte_atomic_compare_exchange_strong_explicit(&r->rts_cons.head.raw,
- (uint64_t *)(uintptr_t)&oh.raw, nh.raw,
- rte_memory_order_acquire, rte_memory_order_acquire) == 0);
-
- *old_head = oh.val.pos;
- return n;
+ return __rte_ring_rts_move_head(&r->rts_cons, &r->prod,
+ 0, num, behavior, old_head, entries);
}
/**
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [RFCv2 2/6] ring: make copying functions generic
2024-09-06 13:13 ` [RFCv2 0/6] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
2024-09-06 13:13 ` [RFCv2 1/6] ring: common functions for 'move head' ops Konstantin Ananyev
@ 2024-09-06 13:13 ` Konstantin Ananyev
2024-09-06 13:13 ` [RFCv2 3/6] ring: make dump function more verbose Konstantin Ananyev
` (4 subsequent siblings)
6 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-06 13:13 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, Konstantin Ananyev
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
Note upfront: that change doesn't introduce any functional
or performance changes.
It is just a code-reordering for:
- improve code modularity and re-usability
- ability in future to re-use the same code to introduce new functionality
There is no real need for enqueue_elems()/dequeue_elems()
to get pointer to actual rte_ring structure, instead it is enough to pass
a pointer to actual elements buffer inside the ring.
In return, we'll get a copying functions that could be used for other
queueing abstractions that do have circular ring buffer inside.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
---
lib/ring/rte_ring_elem_pvt.h | 117 ++++++++++++++++++++---------------
1 file changed, 68 insertions(+), 49 deletions(-)
diff --git a/lib/ring/rte_ring_elem_pvt.h b/lib/ring/rte_ring_elem_pvt.h
index 3a83668a08..216cb6089f 100644
--- a/lib/ring/rte_ring_elem_pvt.h
+++ b/lib/ring/rte_ring_elem_pvt.h
@@ -17,12 +17,14 @@
#endif
static __rte_always_inline void
-__rte_ring_enqueue_elems_32(struct rte_ring *r, const uint32_t size,
- uint32_t idx, const void *obj_table, uint32_t n)
+__rte_ring_enqueue_elems_32(void *ring_table, const void *obj_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- uint32_t *ring = (uint32_t *)&r[1];
+
+ uint32_t *ring = ring_table;
const uint32_t *obj = (const uint32_t *)obj_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x7); i += 8, idx += 8) {
ring[idx] = obj[i];
@@ -60,14 +62,14 @@ __rte_ring_enqueue_elems_32(struct rte_ring *r, const uint32_t size,
}
static __rte_always_inline void
-__rte_ring_enqueue_elems_64(struct rte_ring *r, uint32_t prod_head,
- const void *obj_table, uint32_t n)
+__rte_ring_enqueue_elems_64(void *ring_table, const void *obj_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- const uint32_t size = r->size;
- uint32_t idx = prod_head & r->mask;
- uint64_t *ring = (uint64_t *)&r[1];
+
+ uint64_t *ring = ring_table;
const unaligned_uint64_t *obj = (const unaligned_uint64_t *)obj_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x3); i += 4, idx += 4) {
ring[idx] = obj[i];
@@ -93,14 +95,14 @@ __rte_ring_enqueue_elems_64(struct rte_ring *r, uint32_t prod_head,
}
static __rte_always_inline void
-__rte_ring_enqueue_elems_128(struct rte_ring *r, uint32_t prod_head,
- const void *obj_table, uint32_t n)
+__rte_ring_enqueue_elems_128(void *ring_table, const void *obj_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- const uint32_t size = r->size;
- uint32_t idx = prod_head & r->mask;
- rte_int128_t *ring = (rte_int128_t *)&r[1];
+
+ rte_int128_t *ring = ring_table;
const rte_int128_t *obj = (const rte_int128_t *)obj_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x1); i += 2, idx += 2)
memcpy((void *)(ring + idx),
@@ -126,37 +128,47 @@ __rte_ring_enqueue_elems_128(struct rte_ring *r, uint32_t prod_head,
* single and multi producer enqueue functions.
*/
static __rte_always_inline void
-__rte_ring_enqueue_elems(struct rte_ring *r, uint32_t prod_head,
- const void *obj_table, uint32_t esize, uint32_t num)
+__rte_ring_do_enqueue_elems(void *ring_table, const void *obj_table,
+ uint32_t size, uint32_t idx, uint32_t esize, uint32_t num)
{
/* 8B and 16B copies implemented individually to retain
* the current performance.
*/
if (esize == 8)
- __rte_ring_enqueue_elems_64(r, prod_head, obj_table, num);
+ __rte_ring_enqueue_elems_64(ring_table, obj_table, size,
+ idx, num);
else if (esize == 16)
- __rte_ring_enqueue_elems_128(r, prod_head, obj_table, num);
+ __rte_ring_enqueue_elems_128(ring_table, obj_table, size,
+ idx, num);
else {
- uint32_t idx, scale, nr_idx, nr_num, nr_size;
+ uint32_t scale, nr_idx, nr_num, nr_size;
/* Normalize to uint32_t */
scale = esize / sizeof(uint32_t);
nr_num = num * scale;
- idx = prod_head & r->mask;
nr_idx = idx * scale;
- nr_size = r->size * scale;
- __rte_ring_enqueue_elems_32(r, nr_size, nr_idx,
- obj_table, nr_num);
+ nr_size = size * scale;
+ __rte_ring_enqueue_elems_32(ring_table, obj_table, nr_size,
+ nr_idx, nr_num);
}
}
static __rte_always_inline void
-__rte_ring_dequeue_elems_32(struct rte_ring *r, const uint32_t size,
- uint32_t idx, void *obj_table, uint32_t n)
+__rte_ring_enqueue_elems(struct rte_ring *r, uint32_t prod_head,
+ const void *obj_table, uint32_t esize, uint32_t num)
+{
+ __rte_ring_do_enqueue_elems(&r[1], obj_table, r->size,
+ prod_head & r->mask, esize, num);
+}
+
+static __rte_always_inline void
+__rte_ring_dequeue_elems_32(void *obj_table, const void *ring_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- uint32_t *ring = (uint32_t *)&r[1];
- uint32_t *obj = (uint32_t *)obj_table;
+ uint32_t *obj = obj_table;
+ const uint32_t *ring = (const uint32_t *)ring_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x7); i += 8, idx += 8) {
obj[i] = ring[idx];
@@ -194,14 +206,13 @@ __rte_ring_dequeue_elems_32(struct rte_ring *r, const uint32_t size,
}
static __rte_always_inline void
-__rte_ring_dequeue_elems_64(struct rte_ring *r, uint32_t cons_head,
- void *obj_table, uint32_t n)
+__rte_ring_dequeue_elems_64(void *obj_table, const void *ring_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- const uint32_t size = r->size;
- uint32_t idx = cons_head & r->mask;
- uint64_t *ring = (uint64_t *)&r[1];
unaligned_uint64_t *obj = (unaligned_uint64_t *)obj_table;
+ const uint64_t *ring = (const uint64_t *)ring_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x3); i += 4, idx += 4) {
obj[i] = ring[idx];
@@ -227,27 +238,26 @@ __rte_ring_dequeue_elems_64(struct rte_ring *r, uint32_t cons_head,
}
static __rte_always_inline void
-__rte_ring_dequeue_elems_128(struct rte_ring *r, uint32_t cons_head,
- void *obj_table, uint32_t n)
+__rte_ring_dequeue_elems_128(void *obj_table, const void *ring_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- const uint32_t size = r->size;
- uint32_t idx = cons_head & r->mask;
- rte_int128_t *ring = (rte_int128_t *)&r[1];
rte_int128_t *obj = (rte_int128_t *)obj_table;
+ const rte_int128_t *ring = (const rte_int128_t *)ring_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x1); i += 2, idx += 2)
- memcpy((void *)(obj + i), (void *)(ring + idx), 32);
+ memcpy((obj + i), (const void *)(ring + idx), 32);
switch (n & 0x1) {
case 1:
- memcpy((void *)(obj + i), (void *)(ring + idx), 16);
+ memcpy((obj + i), (const void *)(ring + idx), 16);
}
} else {
for (i = 0; idx < size; i++, idx++)
- memcpy((void *)(obj + i), (void *)(ring + idx), 16);
+ memcpy((obj + i), (const void *)(ring + idx), 16);
/* Start at the beginning */
for (idx = 0; i < n; i++, idx++)
- memcpy((void *)(obj + i), (void *)(ring + idx), 16);
+ memcpy((obj + i), (const void *)(ring + idx), 16);
}
}
@@ -256,30 +266,39 @@ __rte_ring_dequeue_elems_128(struct rte_ring *r, uint32_t cons_head,
* single and multi producer enqueue functions.
*/
static __rte_always_inline void
-__rte_ring_dequeue_elems(struct rte_ring *r, uint32_t cons_head,
- void *obj_table, uint32_t esize, uint32_t num)
+__rte_ring_do_dequeue_elems(void *obj_table, const void *ring_table,
+ uint32_t size, uint32_t idx, uint32_t esize, uint32_t num)
{
/* 8B and 16B copies implemented individually to retain
* the current performance.
*/
if (esize == 8)
- __rte_ring_dequeue_elems_64(r, cons_head, obj_table, num);
+ __rte_ring_dequeue_elems_64(obj_table, ring_table, size,
+ idx, num);
else if (esize == 16)
- __rte_ring_dequeue_elems_128(r, cons_head, obj_table, num);
+ __rte_ring_dequeue_elems_128(obj_table, ring_table, size,
+ idx, num);
else {
- uint32_t idx, scale, nr_idx, nr_num, nr_size;
+ uint32_t scale, nr_idx, nr_num, nr_size;
/* Normalize to uint32_t */
scale = esize / sizeof(uint32_t);
nr_num = num * scale;
- idx = cons_head & r->mask;
nr_idx = idx * scale;
- nr_size = r->size * scale;
- __rte_ring_dequeue_elems_32(r, nr_size, nr_idx,
- obj_table, nr_num);
+ nr_size = size * scale;
+ __rte_ring_dequeue_elems_32(obj_table, ring_table, nr_size,
+ nr_idx, nr_num);
}
}
+static __rte_always_inline void
+__rte_ring_dequeue_elems(struct rte_ring *r, uint32_t cons_head,
+ void *obj_table, uint32_t esize, uint32_t num)
+{
+ __rte_ring_do_dequeue_elems(obj_table, &r[1], r->size,
+ cons_head & r->mask, esize, num);
+}
+
/* Between load and load. there might be cpu reorder in weak model
* (powerpc/arm).
* There are 2 choices for the users
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [RFCv2 3/6] ring: make dump function more verbose
2024-09-06 13:13 ` [RFCv2 0/6] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
2024-09-06 13:13 ` [RFCv2 1/6] ring: common functions for 'move head' ops Konstantin Ananyev
2024-09-06 13:13 ` [RFCv2 2/6] ring: make copying functions generic Konstantin Ananyev
@ 2024-09-06 13:13 ` Konstantin Ananyev
2024-09-06 13:13 ` [RFCv2 4/6] ring/soring: introduce Staged Ordered Ring Konstantin Ananyev
` (3 subsequent siblings)
6 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-06 13:13 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, Eimear Morrissey
From: Eimear Morrissey <eimear.morrissey@huawei.com>
The current rte_ring_dump function uses the generic rte_ring_headtail
structure to access head/tail positions. This is incorrect for the RTS
case where the head is stored in a different offset in the union of
structs. Switching to a separate function for each sync type allows
to dump correct head/tail values and extra metadata.
Signed-off-by: Eimear Morrissey <eimear.morrissey@huawei.com>
---
.mailmap | 1 +
app/test/test_ring_stress_impl.h | 1 +
lib/ring/rte_ring.c | 87 ++++++++++++++++++++++++++++++--
lib/ring/rte_ring.h | 15 ++++++
lib/ring/version.map | 7 +++
5 files changed, 107 insertions(+), 4 deletions(-)
diff --git a/.mailmap b/.mailmap
index 4a508bafad..3da86393c0 100644
--- a/.mailmap
+++ b/.mailmap
@@ -379,6 +379,7 @@ Eduard Serra <eserra@vmware.com>
Edward Makarov <makarov@kraftway.ru>
Edwin Brossette <edwin.brossette@6wind.com>
Eelco Chaudron <echaudro@redhat.com>
+Eimear Morrissey <eimear.morrissey@huawei.com>
Elad Nachman <eladv6@gmail.com>
Elad Persiko <eladpe@mellanox.com>
Elena Agostini <eagostini@nvidia.com>
diff --git a/app/test/test_ring_stress_impl.h b/app/test/test_ring_stress_impl.h
index 8b0bfb11fe..8449cb4b15 100644
--- a/app/test/test_ring_stress_impl.h
+++ b/app/test/test_ring_stress_impl.h
@@ -380,6 +380,7 @@ test_mt1(int (*test)(void *))
}
lcore_stat_dump(stdout, UINT32_MAX, &arg[mc].stats);
+ rte_ring_dump(stdout, r);
mt1_fini(r, data);
return rc;
}
diff --git a/lib/ring/rte_ring.c b/lib/ring/rte_ring.c
index aebb6d6728..261f2a06db 100644
--- a/lib/ring/rte_ring.c
+++ b/lib/ring/rte_ring.c
@@ -364,20 +364,99 @@ rte_ring_free(struct rte_ring *r)
rte_free(te);
}
+static const char *
+ring_get_sync_type(const enum rte_ring_sync_type st)
+{
+ switch (st) {
+ case RTE_RING_SYNC_ST:
+ return "single thread";
+ case RTE_RING_SYNC_MT:
+ return "multi thread";
+ case RTE_RING_SYNC_MT_RTS:
+ return "multi thread - RTS";
+ case RTE_RING_SYNC_MT_HTS:
+ return "multi thread - HTS";
+ default:
+ return "unknown";
+ }
+}
+
+static void
+ring_dump_ht_headtail(FILE *f, const char *prefix,
+ const struct rte_ring_headtail *ht)
+{
+ fprintf(f, "%ssync_type=%s\n", prefix,
+ ring_get_sync_type(ht->sync_type));
+ fprintf(f, "%shead=%"PRIu32"\n", prefix, ht->head);
+ fprintf(f, "%stail=%"PRIu32"\n", prefix, ht->tail);
+}
+
+static void
+ring_dump_rts_headtail(FILE *f, const char *prefix,
+ const struct rte_ring_rts_headtail *rts)
+{
+ fprintf(f, "%ssync_type=%s\n", prefix,
+ ring_get_sync_type(rts->sync_type));
+ fprintf(f, "%shead.pos=%"PRIu32"\n", prefix, rts->head.val.pos);
+ fprintf(f, "%shead.cnt=%"PRIu32"\n", prefix, rts->head.val.cnt);
+ fprintf(f, "%stail.pos=%"PRIu32"\n", prefix, rts->tail.val.pos);
+ fprintf(f, "%stail.cnt=%"PRIu32"\n", prefix, rts->tail.val.cnt);
+ fprintf(f, "%shtd_max=%"PRIu32"\n", prefix, rts->htd_max);
+}
+
+static void
+ring_dump_hts_headtail(FILE *f, const char *prefix,
+ const struct rte_ring_hts_headtail *hts)
+{
+ fprintf(f, "%ssync_type=%s\n", prefix,
+ ring_get_sync_type(hts->sync_type));
+ fprintf(f, "%shead=%"PRIu32"\n", prefix, hts->ht.pos.head);
+ fprintf(f, "%stail=%"PRIu32"\n", prefix, hts->ht.pos.tail);
+}
+
+void
+rte_ring_headtail_dump(FILE *f, const char *prefix,
+ const struct rte_ring_headtail *r)
+{
+ if (f == NULL || r == NULL)
+ return;
+
+ prefix = (prefix != NULL) ? prefix : "";
+
+ switch (r->sync_type) {
+ case RTE_RING_SYNC_ST:
+ case RTE_RING_SYNC_MT:
+ ring_dump_ht_headtail(f, prefix, r);
+ break;
+ case RTE_RING_SYNC_MT_RTS:
+ ring_dump_rts_headtail(f, prefix,
+ (const struct rte_ring_rts_headtail *)r);
+ break;
+ case RTE_RING_SYNC_MT_HTS:
+ ring_dump_hts_headtail(f, prefix,
+ (const struct rte_ring_hts_headtail *)r);
+ break;
+ default:
+ RING_LOG(ERR, "Invalid ring sync type detected");
+ }
+}
+
/* dump the status of the ring on the console */
void
rte_ring_dump(FILE *f, const struct rte_ring *r)
{
+ if (f == NULL || r == NULL)
+ return;
+
fprintf(f, "ring <%s>@%p\n", r->name, r);
fprintf(f, " flags=%x\n", r->flags);
fprintf(f, " size=%"PRIu32"\n", r->size);
fprintf(f, " capacity=%"PRIu32"\n", r->capacity);
- fprintf(f, " ct=%"PRIu32"\n", r->cons.tail);
- fprintf(f, " ch=%"PRIu32"\n", r->cons.head);
- fprintf(f, " pt=%"PRIu32"\n", r->prod.tail);
- fprintf(f, " ph=%"PRIu32"\n", r->prod.head);
fprintf(f, " used=%u\n", rte_ring_count(r));
fprintf(f, " avail=%u\n", rte_ring_free_count(r));
+
+ rte_ring_headtail_dump(f, " cons.", &(r->cons));
+ rte_ring_headtail_dump(f, " prod.", &(r->prod));
}
/* dump the status of all rings on the console */
diff --git a/lib/ring/rte_ring.h b/lib/ring/rte_ring.h
index c709f30497..d6f9e85c64 100644
--- a/lib/ring/rte_ring.h
+++ b/lib/ring/rte_ring.h
@@ -204,6 +204,21 @@ void rte_ring_free(struct rte_ring *r);
*/
void rte_ring_dump(FILE *f, const struct rte_ring *r);
+/**
+ * Dump the status of a headtail to a file.
+ *
+ * @param f
+ * A pointer to a file for output
+ * @param prefix
+ * A string to prefix each output line with
+ * @param r
+ * A pointer to a ring headtail structure.
+ */
+__rte_experimental
+void
+rte_ring_headtail_dump(FILE *f, const char *prefix,
+ const struct rte_ring_headtail *r);
+
/**
* Enqueue several objects on the ring (multi-producers safe).
*
diff --git a/lib/ring/version.map b/lib/ring/version.map
index 8da094a69a..61f7464f5a 100644
--- a/lib/ring/version.map
+++ b/lib/ring/version.map
@@ -14,3 +14,10 @@ DPDK_25 {
local: *;
};
+
+EXPERIMENTAL {
+ global:
+
+ # added in 24.11
+ rte_ring_headtail_dump;
+};
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [RFCv2 4/6] ring/soring: introduce Staged Ordered Ring
2024-09-06 13:13 ` [RFCv2 0/6] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
` (2 preceding siblings ...)
2024-09-06 13:13 ` [RFCv2 3/6] ring: make dump function more verbose Konstantin Ananyev
@ 2024-09-06 13:13 ` Konstantin Ananyev
2024-09-06 13:13 ` [RFCv2 5/6] app/test: add unit tests for soring API Konstantin Ananyev
` (2 subsequent siblings)
6 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-06 13:13 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, Konstantin Ananyev, Eimear Morrissey
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
Staged-Ordered-Ring (SORING) provides a SW abstraction for 'ordered' queues
with multiple processing 'stages'.
It is based on conventional DPDK rte_ring, re-uses many of its concepts,
and even substantial part of its code.
It can be viewed as an 'extension' of rte_ring functionality.
In particular, main SORING properties:
- circular ring buffer with fixed size objects
- producer, consumer plus multiple processing stages in the middle.
- allows to split objects processing into multiple stages.
- objects remain in the same ring while moving from one stage to the other,
initial order is preserved, no extra copying needed.
- preserves the ingress order of objects within the queue across multiple
stages, i.e.:
at the same stage multiple threads can process objects from the ring in
any order, but for the next stage objects will always appear in the
original order.
- each stage (and producer/consumer) can be served by single and/or
multiple threads.
- number of stages, size and number of objects in the ring are
configurable at ring initialization time.
Data-path API provides four main operations:
- enqueue/dequeue works in the same manner as for conventional rte_ring,
all rte_ring synchronization types are supported.
- acquire/release - for each stage there is an acquire (start) and
release (finish) operation.
after some objects are 'acquired' - given thread can safely assume that
it has exclusive possession of these objects till 'release' for them is
invoked.
Note that right now user has to release exactly the same number of
objects that was acquired before.
After 'release', objects can be 'acquired' by next stage and/or dequeued
by the consumer (in case of last stage).
Expected use-case: applications that uses pipeline model
(probably with multiple stages) for packet processing, when preserving
incoming packet order is important. I.E.: IPsec processing, etc.
Signed-off-by: Eimear Morrissey <eimear.morrissey@huawei.com>
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
---
lib/ring/meson.build | 4 +-
lib/ring/rte_soring.c | 182 ++++++++++++++
lib/ring/rte_soring.h | 547 +++++++++++++++++++++++++++++++++++++++++
lib/ring/soring.c | 548 ++++++++++++++++++++++++++++++++++++++++++
lib/ring/soring.h | 124 ++++++++++
lib/ring/version.map | 19 ++
6 files changed, 1422 insertions(+), 2 deletions(-)
create mode 100644 lib/ring/rte_soring.c
create mode 100644 lib/ring/rte_soring.h
create mode 100644 lib/ring/soring.c
create mode 100644 lib/ring/soring.h
diff --git a/lib/ring/meson.build b/lib/ring/meson.build
index 7fca958ed7..21f2c12989 100644
--- a/lib/ring/meson.build
+++ b/lib/ring/meson.build
@@ -1,8 +1,8 @@
# SPDX-License-Identifier: BSD-3-Clause
# Copyright(c) 2017 Intel Corporation
-sources = files('rte_ring.c')
-headers = files('rte_ring.h')
+sources = files('rte_ring.c', 'rte_soring.c', 'soring.c')
+headers = files('rte_ring.h', 'rte_soring.h')
# most sub-headers are not for direct inclusion
indirect_headers += files (
'rte_ring_core.h',
diff --git a/lib/ring/rte_soring.c b/lib/ring/rte_soring.c
new file mode 100644
index 0000000000..b6bc71b8c9
--- /dev/null
+++ b/lib/ring/rte_soring.c
@@ -0,0 +1,182 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#include <inttypes.h>
+
+#include "soring.h"
+#include <rte_string_fns.h>
+
+RTE_LOG_REGISTER_DEFAULT(soring_logtype, INFO);
+#define RTE_LOGTYPE_SORING soring_logtype
+#define SORING_LOG(level, ...) \
+ RTE_LOG_LINE(level, SORING, "" __VA_ARGS__)
+
+static uint32_t
+soring_calc_elem_num(uint32_t count)
+{
+ return rte_align32pow2(count + 1);
+}
+
+static int
+soring_check_param(uint32_t esize, uint32_t stsize, uint32_t count,
+ uint32_t stages)
+{
+ if (stages == 0) {
+ SORING_LOG(ERR, "invalid number of stages: %u", stages);
+ return -EINVAL;
+ }
+
+ /* Check if element size is a multiple of 4B */
+ if (esize == 0 || esize % 4 != 0) {
+ SORING_LOG(ERR, "invalid element size: %u", esize);
+ return -EINVAL;
+ }
+
+ /* Check if ret-code size is a multiple of 4B */
+ if (stsize % 4 != 0) {
+ SORING_LOG(ERR, "invalid retcode size: %u", stsize);
+ return -EINVAL;
+ }
+
+ /* count must be a power of 2 */
+ if (rte_is_power_of_2(count) == 0 ||
+ (count > RTE_SORING_ELEM_MAX + 1)) {
+ SORING_LOG(ERR, "invalid number of elements: %u", count);
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+/*
+ * Calculate size offsets for SORING internal data layout.
+ */
+static size_t
+soring_get_szofs(uint32_t esize, uint32_t stsize, uint32_t count,
+ uint32_t stages, size_t *elst_ofs, size_t *state_ofs,
+ size_t *stage_ofs)
+{
+ size_t sz;
+ const struct rte_soring * const r = NULL;
+
+ sz = sizeof(r[0]) + (size_t)count * esize;
+ sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE);
+
+ if (elst_ofs != NULL)
+ *elst_ofs = sz;
+
+ sz = sz + (size_t)count * stsize;
+ sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE);
+
+ if (state_ofs != NULL)
+ *state_ofs = sz;
+
+ sz += sizeof(r->state[0]) * count;
+ sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE);
+
+ if (stage_ofs != NULL)
+ *stage_ofs = sz;
+
+ sz += sizeof(r->stage[0]) * stages;
+ sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE);
+
+ return sz;
+}
+
+static void
+soring_dump_stage_headtail(FILE *f, const char *prefix,
+ struct soring_stage *st)
+{
+ fprintf(f, "%stail.pos=%"PRIu32"\n", prefix, st->sht.tail.pos);
+ fprintf(f, "%stail.sync=%"PRIu32"\n", prefix, st->sht.tail.sync);
+ fprintf(f, "%shead=%"PRIu32"\n", prefix, st->sht.head);
+}
+
+void
+rte_soring_dump(FILE *f, const struct rte_soring *r)
+{
+ uint32_t i;
+ char buf[32];
+
+ if (f == NULL || r == NULL)
+ return;
+
+ fprintf(f, "soring <%s>@%p\n", r->name, r);
+ fprintf(f, " size=%"PRIu32"\n", r->size);
+ fprintf(f, " capacity=%"PRIu32"\n", r->capacity);
+ fprintf(f, " esize=%"PRIu32"\n", r->esize);
+ fprintf(f, " msize=%"PRIu32"\n", r->msize);
+ fprintf(f, " used=%u\n", rte_soring_count(r));
+ fprintf(f, " avail=%u\n", rte_soring_free_count(r));
+
+ rte_ring_headtail_dump(f, " cons.", &(r->cons.ht));
+ rte_ring_headtail_dump(f, " prod.", &(r->prod.ht));
+
+ fprintf(f, " nb_stage=%"PRIu32"\n", r->nb_stage);
+ for (i = 0; i < r->nb_stage; i++) {
+ snprintf(buf, sizeof(buf), " stage[%u].", i);
+ soring_dump_stage_headtail(f, buf, r->stage + i);
+ }
+}
+
+ssize_t
+rte_soring_get_memsize(const struct rte_soring_param *prm)
+{
+ int32_t rc;
+ uint32_t count;
+
+ count = soring_calc_elem_num(prm->elems);
+ rc = soring_check_param(prm->elem_size, prm->meta_size, count,
+ prm->stages);
+ if (rc != 0)
+ return rc;
+
+ return soring_get_szofs(prm->elem_size, prm->meta_size, count,
+ prm->stages, NULL, NULL, NULL);
+}
+
+int
+rte_soring_init(struct rte_soring *r, const struct rte_soring_param *prm)
+{
+ int32_t rc;
+ uint32_t n;
+ size_t meta_ofs, stage_ofs, state_ofs;
+
+ if (r == NULL || prm == NULL)
+ return -EINVAL;
+
+ n = soring_calc_elem_num(prm->elems);
+ rc = soring_check_param(prm->elem_size, prm->meta_size, n, prm->stages);
+ if (rc != 0)
+ return rc;
+
+ soring_get_szofs(prm->elem_size, prm->meta_size, n, prm->stages,
+ &meta_ofs, &state_ofs, &stage_ofs);
+
+ memset(r, 0, sizeof(*r));
+ rc = strlcpy(r->name, prm->name, sizeof(r->name));
+ if (rc < 0 || rc >= (int)sizeof(r->name))
+ return -ENAMETOOLONG;
+
+ r->size = n;
+ r->mask = r->size - 1;
+ r->capacity = prm->elems;
+ r->esize = prm->elem_size;
+ r->msize = prm->meta_size;
+
+ r->prod.ht.sync_type = prm->prod_synt;
+ r->cons.ht.sync_type = prm->cons_synt;
+
+ r->state = (union soring_state *)((uintptr_t)r + state_ofs);
+ memset(r->state, 0, sizeof(r->state[0]) * r->size);
+
+ r->stage = (struct soring_stage *)((uintptr_t)r + stage_ofs);
+ r->nb_stage = prm->stages;
+ memset(r->stage, 0, r->nb_stage * sizeof(r->stage[0]));
+
+ if (r->msize != 0)
+ r->meta = (void *)((uintptr_t)r + meta_ofs);
+
+ return 0;
+}
diff --git a/lib/ring/rte_soring.h b/lib/ring/rte_soring.h
new file mode 100644
index 0000000000..dd52a507e3
--- /dev/null
+++ b/lib/ring/rte_soring.h
@@ -0,0 +1,547 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#ifndef _RTE_SORING_H_
+#define _RTE_SORING_H_
+
+/**
+ * @file
+ * This file contains definition of RTE soring (Staged Ordered Ring) public API.
+ * Brief description:
+ * enqueue/dequeue works the same as for conventional rte_ring:
+ * any rte_ring sync types can be used, etc.
+ * Plus there could be multiple 'stages'.
+ * For each stage there is an acquire (start) and release (finish) operation.
+ * after some elems are 'acquired' - user can safely assume that he has
+ * exclusive possession of these elems till 'release' for them is done.
+ * Note that right now user has to release exactly the same number of elems
+ * he acquired before.
+ * After 'release', elems can be 'acquired' by next stage and/or dequeued
+ * (in case of last stage).
+ */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <rte_ring.h>
+
+/* upper 2 bits are used for status */
+#define RTE_SORING_ST_BIT 30
+
+/* max possible number of elements in the soring */
+#define RTE_SORING_ELEM_MAX (RTE_BIT32(RTE_SORING_ST_BIT) - 1)
+
+struct rte_soring_param {
+ /** expected name of the ring */
+ const char *name;
+ /** number of elemnts in the ring */
+ uint32_t elems;
+ /** size of elements in the ring, must be a multiple of 4 */
+ uint32_t elem_size;
+ /**
+ * size of metadata for each elem, must be a multiple of 4.
+ * This parameter defines a size of supplementary and optional
+ * array of metadata associated with each object in the soring.
+ * While element size is configurable (see @elem_size parameter above),
+ * so user can specify it big enough to hold both object and its
+ * metadata together, for performance reasons it might be plausible
+ * to access them as separate arrays.
+ * Common usage scenario when such separation helps:
+ * enqueue() - writes to objects array
+ * acquire() - reads from objects array
+ * release() - writes to metadata array (as an example: return code)
+ * dequeue() - reads both objects and metadata array
+ */
+ uint32_t meta_size;
+ /** number of stages in the ring */
+ uint32_t stages;
+ /** sync type for producer */
+ enum rte_ring_sync_type prod_synt;
+ /** sync type for consumer */
+ enum rte_ring_sync_type cons_synt;
+};
+
+struct rte_soring;
+
+/**
+ * Calculate the memory size needed for a soring
+ *
+ * This function returns the number of bytes needed for a ring, given
+ * the expected parameters for it. This value is the sum of the size of
+ * the internal metadata and the size of the memory needed by the
+ * actual ring elements and theri rec-codes. The value is aligned to a cache
+ * line size.
+ *
+ * @param prm
+ * Pointer to the structure that contains soring creation paramers.
+ * @return
+ * - The memory size needed for the soring on success.
+ * - -EINVAL if provided paramer values are invalid.
+ */
+__rte_experimental
+ssize_t
+rte_soring_get_memsize(const struct rte_soring_param *prm);
+
+/**
+ * Initialize a soring structure.
+ *
+ * Initialize a soring structure in memory pointed by "r".
+ * The size of the memory area must be large enough to store the soring
+ * internal structures plus the objects and ret-code tables.
+ * It is strongly advised to use rte_ring_get_memsize() to get the
+ * appropriate size.
+ *
+ * @param r
+ * Pointer to the soring structure.
+ * @param prm
+ * Pointer to the structure that contains soring creation paramers.
+ * @return
+ * - 0 on success, or a negative error code.
+ */
+__rte_experimental
+int
+rte_soring_init(struct rte_soring *r, const struct rte_soring_param *prm);
+
+/**
+ * Return the total number of filled entries in a ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @return
+ * The number of entries in the ring.
+ */
+__rte_experimental
+unsigned int
+rte_soring_count(const struct rte_soring *r);
+
+/**
+ * Return the total number of unfilled entries in a ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @return
+ * The number of free entries in the ring.
+ */
+__rte_experimental
+unsigned int
+rte_soring_free_count(const struct rte_soring *r);
+
+/**
+ * Dump the status of the soring
+ *
+ * @param f
+ * A pointer to a file for output
+ * @param r
+ * Pointer to the soring structure.
+ */
+__rte_experimental
+void
+rte_soring_dump(FILE *f, const struct rte_soring *r);
+
+/**
+ * Enqueue several objects on the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to enqueue.
+ * Size of objects to enqueue must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param n
+ * The number of objects to add in the ring from the @objs.
+ * @param free_space
+ * if non-NULL, returns the amount of space in the ring after the
+ * enqueue operation has finished.
+ * @return
+ * - Actual number of objects enqueued, either 0 or @n.
+ */
+__rte_experimental
+uint32_t
+rte_soring_enqueue_bulk(struct rte_soring *r, const void *objs,
+ uint32_t n, uint32_t *free_space);
+
+/**
+ * Enqueue several objects plus metadata on the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to enqueue.
+ * Size of objects to enqueue must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each object to enqueue.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as @meta_size
+ * parameter used while creating the ring. If user created the soring with
+ * @meta_size value equals zero, then @meta parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param n
+ * The number of objects to add in the ring from the @objs.
+ * @param free_space
+ * if non-NULL, returns the amount of space in the ring after the
+ * enqueue operation has finished.
+ * @return
+ * - Actual number of objects enqueued, either 0 or @n.
+ */
+__rte_experimental
+uint32_t
+rte_soring_enqueux_bulk(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, uint32_t *free_space);
+
+/**
+ * Enqueue several objects on the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to enqueue.
+ * Size of objects to enqueue must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param n
+ * The number of objects to add in the ring from the @objs.
+ * @param free_space
+ * if non-NULL, returns the amount of space in the ring after the
+ * enqueue operation has finished.
+ * @return
+ * - Actual number of objects enqueued.
+ */
+__rte_experimental
+uint32_t
+rte_soring_enqueue_burst(struct rte_soring *r, const void *objs,
+ uint32_t n, uint32_t *free_space);
+
+/**
+ * Enqueue several objects plus metadata on the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to enqueue.
+ * Size of objects to enqueue must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each object to enqueue.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as @meta_size
+ * parameter used while creating the ring. If user created the soring with
+ * @meta_size value equals zero, then @meta parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param n
+ * The number of objects to add in the ring from the @objs.
+ * @param free_space
+ * if non-NULL, returns the amount of space in the ring after the
+ * enqueue operation has finished.
+ * @return
+ * - Actual number of objects enqueued.
+ */
+__rte_experimental
+uint32_t
+rte_soring_enqueux_burst(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, uint32_t *free_space);
+
+/**
+ * Dequeue several objects from the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to dequeue.
+ * Size of objects to enqueue must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param num
+ * The number of objects to dequeue from the ring into the objs.
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries after the
+ * dequeue has finished.
+ * @return
+ * - Actual number of objects dequeued, either 0 or @num.
+ */
+__rte_experimental
+uint32_t
+rte_soring_dequeue_bulk(struct rte_soring *r, void *objs,
+ uint32_t num, uint32_t *available);
+
+/**
+ * Dequeue several objects plus metadata from the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to dequeue.
+ * Size of objects to enqueue must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to array of metadata values for each object to dequeue.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as @meta_size
+ * parameter used while creating the ring. If user created the soring with
+ * @meta_size value equals zero, then @meta parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param num
+ * The number of objects to dequeue from the ring into the objs.
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries after the
+ * dequeue has finished.
+ * @return
+ * - Actual number of objects dequeued, either 0 or @num.
+ */
+__rte_experimental
+uint32_t
+rte_soring_dequeux_bulk(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, uint32_t *available);
+
+/**
+ * Dequeue several objects from the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to dequeue.
+ * Size of objects to enqueue must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param num
+ * The number of objects to dequeue from the ring into the objs.
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries after the
+ * dequeue has finished.
+ * @return
+ * - Actual number of objects dequeued.
+ */
+__rte_experimental
+uint32_t
+rte_soring_dequeue_burst(struct rte_soring *r, void *objs,
+ uint32_t num, uint32_t *available);
+
+/**
+ * Dequeue several objects plus metadata from the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to dequeue.
+ * Size of objects to enqueue must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to array of metadata values for each object to dequeue.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as @meta_size
+ * parameter used while creating the ring. If user created the soring with
+ * @meta_size value equals zero, then @meta parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param num
+ * The number of objects to dequeue from the ring into the objs.
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries after the
+ * dequeue has finished.
+ * @return
+ * - Actual number of objects dequeued.
+ */
+__rte_experimental
+uint32_t
+rte_soring_dequeux_burst(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, uint32_t *available);
+
+/**
+ * Acquire several objects from the ring for given stage.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to acquire.
+ * Size of objects must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param stage
+ * Stage to acquire objects for.
+ * @param num
+ * The number of objects to acquire.
+ * @param ftoken
+ * Pointer to the opaque 'token' value used by release() op.
+ * User has to store this value somewhere, and later provide to the
+ * release().
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries for given stage
+ * after the acquire has finished.
+ * @return
+ * - Actual number of objects acquired, either 0 or @num.
+ */
+__rte_experimental
+uint32_t
+rte_soring_acquire_bulk(struct rte_soring *r, void *objs,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available);
+
+/**
+ * Acquire several objects plus metadata from the ring for given stage.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to acquire.
+ * Size of objects must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each for each acquired object.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as @meta_size
+ * parameter used while creating the ring. If user created the soring with
+ * @meta_size value equals zero, then @meta parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param stage
+ * Stage to acquire objects for.
+ * @param num
+ * The number of objects to acquire.
+ * @param ftoken
+ * Pointer to the opaque 'token' value used by release() op.
+ * User has to store this value somewhere, and later provide to the
+ * release().
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries for given stage
+ * after the acquire has finished.
+ * @return
+ * - Actual number of objects acquired, either 0 or @num.
+ */
+__rte_experimental
+uint32_t
+rte_soring_acquirx_bulk(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available);
+
+/**
+ * Acquire several objects from the ring for given stage.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to acquire.
+ * Size of objects must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param stage
+ * Stage to acquire objects for.
+ * @param num
+ * The number of objects to acquire.
+ * @param ftoken
+ * Pointer to the opaque 'token' value used by release() op.
+ * User has to store this value somewhere, and later provide to the
+ * release().
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries for given stage
+ * after the acquire has finished.
+ * @return
+ * - Actual number of objects acquired.
+ */
+__rte_experimental
+uint32_t
+rte_soring_acquire_burst(struct rte_soring *r, void *objs,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available);
+
+/**
+ * Acquire several objects plus metadata from the ring for given stage.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to acquire.
+ * Size of objects must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each for each acquired object.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as @meta_size
+ * parameter used while creating the ring. If user created the soring with
+ * @meta_size value equals zero, then @meta parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param stage
+ * Stage to acquire objects for.
+ * @param num
+ * The number of objects to acquire.
+ * @param ftoken
+ * Pointer to the opaque 'token' value used by release() op.
+ * User has to store this value somewhere, and later provide to the
+ * release().
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries for given stage
+ * after the acquire has finished.
+ * @return
+ * - Actual number of objects acquired.
+ */
+__rte_experimental
+uint32_t
+rte_soring_acquirx_burst(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available);
+
+/**
+ * Release several objects for given stage back to the ring.
+ * Note that it means these objects become avaialble for next stage or
+ * dequeue.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to relase.
+ * Note that unless user needs to overwrite ring objects this parameter
+ * can be NULL.
+ * Size of objects must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param stage
+ * Current stage.
+ * @param n
+ * The number of objects to release.
+ * Has to be the same value as returned by acquire() op.
+ * @param ftoken
+ * Opaque 'token' value obtained from acquire() op.
+ * @return
+ * - None.
+ */
+__rte_experimental
+void
+rte_soring_release(struct rte_soring *r, const void *objs,
+ uint32_t stage, uint32_t n, uint32_t ftoken);
+
+/**
+ * Release several objects plus metadata for given stage back to the ring.
+ * Note that it means these objects become avaialble for next stage or
+ * dequeue.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to relase.
+ * Note that unless user needs to overwrite ring objects this parameter
+ * can be NULL.
+ * Size of objects must be the same value as @elem_size parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each object to release.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as @meta_size
+ * parameter used while creating the ring. If user created the soring with
+ * @meta_size value equals zero, then meta parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param stage
+ * Current stage.
+ * @param n
+ * The number of objects to release.
+ * Has to be the same value as returned by acquire() op.
+ * @param ftoken
+ * Opaque 'token' value obtained from acquire() op.
+ * @return
+ * - None.
+ */
+__rte_experimental
+void
+rte_soring_releasx(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t stage, uint32_t n, uint32_t ftoken);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _RTE_SORING_H_ */
diff --git a/lib/ring/soring.c b/lib/ring/soring.c
new file mode 100644
index 0000000000..97fa38df66
--- /dev/null
+++ b/lib/ring/soring.c
@@ -0,0 +1,548 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+/**
+ * @file
+ * This file contains implementation of SORING 'datapath' functions.
+ * Brief description:
+ * ==================
+ * enqueue/dequeue works the same as for conventional rte_ring:
+ * any rte_ring sync types can be used, etc.
+ * Plus there could be multiple 'stages'.
+ * For each stage there is an acquire (start) and release (finish) operation.
+ * After some elems are 'acquired' - user can safely assume that he has
+ * exclusive possession of these elems till 'release' for them is done.
+ * Note that right now user has to release exactly the same number of elems
+ * he acquired before.
+ * After 'release', elems can be 'acquired' by next stage and/or dequeued
+ * (in case of last stage).
+ * Internal structure:
+ * ===================
+ * In addition to 'normal' ring of elems, it also has a ring of states of the
+ * same size. Each state[] corresponds to exactly one elem[].
+ * state[] will be used by acquire/release/dequeue functions to store internal
+ * information and should not be accessed by the user directly.
+ * How it works:
+ * =============
+ * 'acquire()' just moves stage's head (same as rte_ring move_head does),
+ * plus it saves in state[stage.cur_head] information about how many elems
+ * were acquired, current head position and special flag value to indicate
+ * that elems are acquired (RTE_SORING_ST_START).
+ * Note that 'acquire()' returns to the user a special 'ftoken' that user has
+ * to provide for 'release()' (in fact it is just a position for current head).
+ * 'release()' extracts old head value from provided ftoken and checks that
+ * corresponding 'state[]' contains expected values(mostly for sanity
+ * purposes).
+ * * Then it marks this state[] with 'RTE_SORING_ST_FINISH' flag to indicate
+ * that given subset of objects was released.
+ * After that, it checks does old head value equals to current tail value?
+ * If yes, then it performs 'finalize()' operation, otherwise 'release()'
+ * just returns (without spinning on stage tail value).
+ * As updated state[] is shared by all threads, some other thread can do
+ * 'finalize()' for given stage.
+ * That allows 'release()' to avoid excessive waits on the tail value.
+ * Main purpose of 'finalize()' operation is to walk through 'state[]'
+ * from current stage tail up to its head, check state[] and move stage tail
+ * through elements that already are in RTE_SORING_ST_FINISH state.
+ * Along with that, corresponding state[] values are reset to zero.
+ * Note that 'finalize()' for given stage can be done from multiple places:
+ * 'release()' for that stage or from 'acquire()' for next stage
+ * even from consumer's 'dequeue()' - in case given stage is the last one.
+ * So 'finalize()' has to be MT-safe and inside it we have to
+ * guarantee that only one thread will update state[] and stage's tail values.
+ */
+
+#include "soring.h"
+
+/*
+ * Inline functions (fastpath) start here.
+ */
+static __rte_always_inline uint32_t
+__rte_soring_stage_finalize(struct soring_stage_headtail *sht,
+ union soring_state *rstate, uint32_t rmask, uint32_t maxn)
+{
+ int32_t rc;
+ uint32_t head, i, idx, k, n, tail;
+ union soring_stage_tail nt, ot;
+ union soring_state st;
+
+ /* try to grab exclusive right to update tail value */
+ ot.raw = rte_atomic_load_explicit(&sht->tail.raw,
+ rte_memory_order_acquire);
+
+ /* other thread already finalizing it for us */
+ if (ot.sync != 0)
+ return 0;
+
+ nt.pos = ot.pos;
+ nt.sync = 1;
+ rc = rte_atomic_compare_exchange_strong_explicit(&sht->tail.raw,
+ &ot.raw, nt.raw, rte_memory_order_acquire,
+ rte_memory_order_relaxed);
+
+ /* other thread won the race */
+ if (rc == 0)
+ return 0;
+
+ /*
+ * start with current tail and walk through states that are
+ * already finished.
+ */
+
+ head = rte_atomic_load_explicit(&sht->head, rte_memory_order_relaxed);
+ n = RTE_MIN(head - ot.pos, maxn);
+ for (i = 0, tail = ot.pos; i < n; i += k, tail += k) {
+
+ idx = tail & rmask;
+ st.raw = rte_atomic_load_explicit(&rstate[idx].raw,
+ rte_memory_order_relaxed);
+ if ((st.stnum & RTE_SORING_ST_MASK) != RTE_SORING_ST_FINISH ||
+ st.ftoken != tail)
+ break;
+
+ k = st.stnum & ~RTE_SORING_ST_MASK;
+ rte_atomic_store_explicit(&rstate[idx].raw, 0,
+ rte_memory_order_relaxed);
+ }
+
+
+ /* release exclusive right to update along with new tail value */
+ ot.pos = tail;
+ rte_atomic_store_explicit(&sht->tail.raw, ot.raw,
+ rte_memory_order_release);
+
+ return i;
+}
+
+static __rte_always_inline uint32_t
+__rte_soring_move_prod_head(struct rte_soring *r, uint32_t num,
+ enum rte_ring_queue_behavior behavior, enum rte_ring_sync_type st,
+ uint32_t *head, uint32_t *next, uint32_t *free)
+{
+ uint32_t n;
+
+ switch (st) {
+ case RTE_RING_SYNC_ST:
+ case RTE_RING_SYNC_MT:
+ n = __rte_ring_headtail_move_head(&r->prod.ht, &r->cons.ht,
+ r->capacity, st, num, behavior, head, next, free);
+ break;
+ case RTE_RING_SYNC_MT_RTS:
+ n = __rte_ring_rts_move_head(&r->prod.rts, &r->cons.ht,
+ r->capacity, num, behavior, head, free);
+ *next = *head + n;
+ break;
+ case RTE_RING_SYNC_MT_HTS:
+ n = __rte_ring_hts_move_head(&r->prod.hts, &r->cons.ht,
+ r->capacity, num, behavior, head, free);
+ *next = *head + n;
+ break;
+ default:
+ /* unsupported mode, shouldn't be here */
+ RTE_ASSERT(0);
+ *free = 0;
+ n = 0;
+ }
+
+ return n;
+}
+
+static __rte_always_inline uint32_t
+__rte_soring_move_cons_head(struct rte_soring *r, uint32_t stage, uint32_t num,
+ enum rte_ring_queue_behavior behavior, enum rte_ring_sync_type st,
+ uint32_t *head, uint32_t *next, uint32_t *avail)
+{
+ uint32_t n;
+
+ switch (st) {
+ case RTE_RING_SYNC_ST:
+ case RTE_RING_SYNC_MT:
+ n = __rte_ring_headtail_move_head(&r->cons.ht,
+ &r->stage[stage].ht, 0, st, num, behavior,
+ head, next, avail);
+ break;
+ case RTE_RING_SYNC_MT_RTS:
+ n = __rte_ring_rts_move_head(&r->cons.rts, &r->stage[stage].ht,
+ 0, num, behavior, head, avail);
+ *next = *head + n;
+ break;
+ case RTE_RING_SYNC_MT_HTS:
+ n = __rte_ring_hts_move_head(&r->cons.hts, &r->stage[stage].ht,
+ 0, num, behavior, head, avail);
+ *next = *head + n;
+ break;
+ default:
+ /* unsupported mode, shouldn't be here */
+ RTE_ASSERT(0);
+ *avail = 0;
+ n = 0;
+ }
+
+ return n;
+}
+
+static __rte_always_inline void
+__rte_soring_update_tail(struct __rte_ring_headtail *rht,
+ enum rte_ring_sync_type st, uint32_t head, uint32_t next, uint32_t enq)
+{
+ uint32_t n;
+
+ switch (st) {
+ case RTE_RING_SYNC_ST:
+ case RTE_RING_SYNC_MT:
+ __rte_ring_update_tail(&rht->ht, head, next, st, enq);
+ break;
+ case RTE_RING_SYNC_MT_RTS:
+ __rte_ring_rts_update_tail(&rht->rts);
+ break;
+ case RTE_RING_SYNC_MT_HTS:
+ n = next - head;
+ __rte_ring_hts_update_tail(&rht->hts, head, n, enq);
+ break;
+ default:
+ /* unsupported mode, shouldn't be here */
+ RTE_ASSERT(0);
+ }
+}
+
+static __rte_always_inline uint32_t
+__rte_soring_stage_move_head(struct soring_stage_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity, uint32_t num,
+ enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head, uint32_t *avail)
+{
+ uint32_t n, tail;
+
+ *old_head = rte_atomic_load_explicit(&d->head,
+ rte_memory_order_acquire);
+
+ do {
+ n = num;
+ tail = rte_atomic_load_explicit(&s->tail,
+ rte_memory_order_relaxed);
+ *avail = capacity + tail - *old_head;
+ if (n > *avail)
+ n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *avail;
+ if (n == 0)
+ return 0;
+ *new_head = *old_head + n;
+ } while (rte_atomic_compare_exchange_strong_explicit(&d->head,
+ old_head, *new_head, rte_memory_order_acquire,
+ rte_memory_order_acquire) == 0);
+
+ return n;
+}
+
+static __rte_always_inline uint32_t
+soring_enqueue(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, enum rte_ring_queue_behavior behavior,
+ uint32_t *free_space)
+{
+ enum rte_ring_sync_type st;
+ uint32_t nb_free, prod_head, prod_next;
+
+ RTE_ASSERT(r != NULL && r->nb_stage > 0);
+ RTE_ASSERT(meta == NULL || r->meta != NULL);
+
+ st = r->prod.ht.sync_type;
+
+ n = __rte_soring_move_prod_head(r, n, behavior, st,
+ &prod_head, &prod_next, &nb_free);
+ if (n != 0) {
+ __rte_ring_do_enqueue_elems(&r[1], objs, r->size,
+ prod_head & r->mask, r->esize, n);
+ if (meta != NULL)
+ __rte_ring_do_enqueue_elems(r->meta, meta, r->size,
+ prod_head & r->mask, r->msize, n);
+ __rte_soring_update_tail(&r->prod, st, prod_head, prod_next, 1);
+ }
+
+ if (free_space != NULL)
+ *free_space = nb_free - n;
+ return n;
+}
+
+static __rte_always_inline uint32_t
+soring_dequeue(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, enum rte_ring_queue_behavior behavior,
+ uint32_t *available)
+{
+ enum rte_ring_sync_type st;
+ uint32_t entries, cons_head, cons_next, n, ns, reqn;
+
+ RTE_ASSERT(r != NULL && r->nb_stage > 0);
+ RTE_ASSERT(meta == NULL || r->meta != NULL);
+
+ ns = r->nb_stage - 1;
+ st = r->cons.ht.sync_type;
+
+ /* try to grab exactly @num elems first */
+ n = __rte_soring_move_cons_head(r, ns, num, RTE_RING_QUEUE_FIXED, st,
+ &cons_head, &cons_next, &entries);
+ if (n == 0) {
+ /* try to finalize some elems from previous stage */
+ n = __rte_soring_stage_finalize(&r->stage[ns].sht,
+ r->state, r->mask, 2 * num);
+ entries += n;
+
+ /* repeat attempt to grab elems */
+ reqn = (behavior == RTE_RING_QUEUE_FIXED) ? num : 0;
+ if (entries >= reqn)
+ n = __rte_soring_move_cons_head(r, ns, num, behavior,
+ st, &cons_head, &cons_next, &entries);
+ else
+ n = 0;
+ }
+
+ /* we have some elems to consume */
+ if (n != 0) {
+ __rte_ring_do_dequeue_elems(objs, &r[1], r->size,
+ cons_head & r->mask, r->esize, n);
+ if (meta != NULL)
+ __rte_ring_do_dequeue_elems(meta, r->meta, r->size,
+ cons_head & r->mask, r->msize, n);
+ __rte_soring_update_tail(&r->cons, st, cons_head, cons_next, 0);
+ }
+
+ if (available != NULL)
+ *available = entries - n;
+ return n;
+}
+
+static __rte_always_inline uint32_t
+soring_acquire(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, enum rte_ring_queue_behavior behavior,
+ uint32_t *ftoken, uint32_t *available)
+{
+ uint32_t avail, head, idx, n, next, reqn;
+ union soring_state st;
+ struct soring_stage *pstg;
+ struct soring_stage_headtail *cons;
+
+ RTE_ASSERT(r != NULL && stage < r->nb_stage);
+ RTE_ASSERT(meta == NULL || r->meta != NULL);
+
+ cons = &r->stage[stage].sht;
+
+ if (stage == 0)
+ n = __rte_soring_stage_move_head(cons, &r->prod.ht, 0, num,
+ behavior, &head, &next, &avail);
+ else {
+ pstg = r->stage + stage - 1;
+
+ /* try to grab exactly @num elems */
+ n = __rte_soring_stage_move_head(cons, &pstg->ht, 0, num,
+ RTE_RING_QUEUE_FIXED, &head, &next, &avail);
+ if (n == 0) {
+ /* try to finalize some elems from previous stage */
+ n = __rte_soring_stage_finalize(&pstg->sht, r->state,
+ r->mask, 2 * num);
+ avail += n;
+
+ /* repeat attempt to grab elems */
+ reqn = (behavior == RTE_RING_QUEUE_FIXED) ? num : 0;
+ if (avail >= reqn)
+ n = __rte_soring_stage_move_head(cons,
+ &pstg->ht, 0, num, behavior, &head,
+ &next, &avail);
+ else
+ n = 0;
+ }
+ }
+
+ if (n != 0) {
+
+ idx = head & r->mask;
+
+ /* copy elems that are ready for given stage */
+ __rte_ring_do_dequeue_elems(objs, &r[1], r->size, idx,
+ r->esize, n);
+ if (meta != NULL)
+ __rte_ring_do_dequeue_elems(meta, r->meta,
+ r->size, idx, r->msize, n);
+
+ /* update status ring */
+ st.ftoken = head;
+ st.stnum = (RTE_SORING_ST_START | n);
+
+ rte_atomic_store_explicit(&r->state[idx].raw, st.raw,
+ rte_memory_order_relaxed);
+ *ftoken = head;
+ }
+
+ if (available != NULL)
+ *available = avail - n;
+ return n;
+}
+
+static __rte_always_inline void
+soring_release(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t stage, uint32_t n, uint32_t ftoken)
+{
+ uint32_t idx, tail;
+ struct soring_stage *stg;
+ union soring_state st;
+
+ RTE_ASSERT(r != NULL && stage < r->nb_stage);
+ RTE_ASSERT(meta == NULL || r->meta != NULL);
+
+ stg = r->stage + stage;
+ tail = rte_atomic_load_explicit(&stg->sht.tail.pos,
+ rte_memory_order_relaxed);
+
+ idx = ftoken & r->mask;
+ st.raw = rte_atomic_load_explicit(&r->state[idx].raw,
+ rte_memory_order_acquire);
+
+ /* check state ring contents */
+ RTE_VERIFY(st.stnum == (RTE_SORING_ST_START | n) &&
+ st.ftoken == ftoken);
+
+ /* update contents of the ring, if necessary */
+ if (objs != NULL)
+ __rte_ring_do_enqueue_elems(&r[1], objs, r->size, idx,
+ r->esize, n);
+ if (meta != NULL)
+ __rte_ring_do_enqueue_elems(r->meta, meta, r->size, idx,
+ r->msize, n);
+
+ /* set state to FINISH, make sure it is not reordered */
+ st.stnum = RTE_SORING_ST_FINISH | n;
+ rte_atomic_store_explicit(&r->state[idx].raw, st.raw,
+ rte_memory_order_release);
+
+ if (tail == ftoken)
+ __rte_soring_stage_finalize(&stg->sht, r->state, r->mask,
+ r->capacity);
+}
+
+/*
+ * Public functions (data-path) start here.
+ */
+
+void
+rte_soring_release(struct rte_soring *r, const void *objs,
+ uint32_t stage, uint32_t n, uint32_t ftoken)
+{
+ return soring_release(r, objs, NULL, stage, n, ftoken);
+}
+
+
+void
+rte_soring_releasx(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t stage, uint32_t n, uint32_t ftoken)
+{
+ return soring_release(r, objs, meta, stage, n, ftoken);
+}
+
+uint32_t
+rte_soring_enqueue_bulk(struct rte_soring *r, const void *objs, uint32_t n,
+ uint32_t *free_space)
+{
+ return soring_enqueue(r, objs, NULL, n, RTE_RING_QUEUE_FIXED,
+ free_space);
+}
+
+uint32_t
+rte_soring_enqueux_bulk(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, uint32_t *free_space)
+{
+ return soring_enqueue(r, objs, meta, n, RTE_RING_QUEUE_FIXED,
+ free_space);
+}
+
+uint32_t
+rte_soring_enqueue_burst(struct rte_soring *r, const void *objs, uint32_t n,
+ uint32_t *free_space)
+{
+ return soring_enqueue(r, objs, NULL, n, RTE_RING_QUEUE_VARIABLE,
+ free_space);
+}
+
+uint32_t
+rte_soring_enqueux_burst(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, uint32_t *free_space)
+{
+ return soring_enqueue(r, objs, meta, n, RTE_RING_QUEUE_VARIABLE,
+ free_space);
+}
+
+uint32_t
+rte_soring_dequeue_bulk(struct rte_soring *r, void *objs, uint32_t num,
+ uint32_t *available)
+{
+ return soring_dequeue(r, objs, NULL, num, RTE_RING_QUEUE_FIXED,
+ available);
+}
+
+uint32_t
+rte_soring_dequeux_bulk(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, uint32_t *available)
+{
+ return soring_dequeue(r, objs, meta, num, RTE_RING_QUEUE_FIXED,
+ available);
+}
+
+uint32_t
+rte_soring_dequeue_burst(struct rte_soring *r, void *objs, uint32_t num,
+ uint32_t *available)
+{
+ return soring_dequeue(r, objs, NULL, num, RTE_RING_QUEUE_VARIABLE,
+ available);
+}
+
+uint32_t
+rte_soring_dequeux_burst(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, uint32_t *available)
+{
+ return soring_dequeue(r, objs, meta, num, RTE_RING_QUEUE_VARIABLE,
+ available);
+}
+
+uint32_t
+rte_soring_acquire_bulk(struct rte_soring *r, void *objs,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available)
+{
+ return soring_acquire(r, objs, NULL, stage, num,
+ RTE_RING_QUEUE_FIXED, ftoken, available);
+}
+
+uint32_t
+rte_soring_acquirx_bulk(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available)
+{
+ return soring_acquire(r, objs, meta, stage, num,
+ RTE_RING_QUEUE_FIXED, ftoken, available);
+}
+
+uint32_t
+rte_soring_acquire_burst(struct rte_soring *r, void *objs,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available)
+{
+ return soring_acquire(r, objs, NULL, stage, num,
+ RTE_RING_QUEUE_VARIABLE, ftoken, available);
+}
+
+uint32_t
+rte_soring_acquirx_burst(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available)
+{
+ return soring_acquire(r, objs, meta, stage, num,
+ RTE_RING_QUEUE_VARIABLE, ftoken, available);
+}
+
+unsigned int
+rte_soring_count(const struct rte_soring *r)
+{
+ uint32_t prod_tail = r->prod.ht.tail;
+ uint32_t cons_tail = r->cons.ht.tail;
+ uint32_t count = (prod_tail - cons_tail) & r->mask;
+ return (count > r->capacity) ? r->capacity : count;
+}
+
+unsigned int
+rte_soring_free_count(const struct rte_soring *r)
+{
+ return r->capacity - rte_soring_count(r);
+}
diff --git a/lib/ring/soring.h b/lib/ring/soring.h
new file mode 100644
index 0000000000..c44d8912d9
--- /dev/null
+++ b/lib/ring/soring.h
@@ -0,0 +1,124 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#ifndef _SORING_H_
+#define _SORING_H_
+
+/**
+ * @file
+ * This file contains internal strctures of RTE soring: Staged Ordered Ring.
+ * Sort of extension of conventional RTE ring.
+ * Internal structure:
+ * In addition to 'normal' ring of elems, it also has a ring of states of the
+ * same size. Each state[] corresponds to exactly one elem[].
+ * state[] will be used by acquire/release/dequeue functions to store internal
+ * information and should not be accessed by the user directly.
+ * For actual implementation details, please refer to soring.c
+ */
+
+#include <rte_soring.h>
+
+union soring_state {
+ alignas(sizeof(uint64_t)) RTE_ATOMIC(uint64_t) raw;
+ struct {
+ RTE_ATOMIC(uint32_t) ftoken;
+ RTE_ATOMIC(uint32_t) stnum;
+ };
+};
+
+/* upper 2 bits are used for status */
+#define RTE_SORING_ST_START RTE_SHIFT_VAL32(1, RTE_SORING_ST_BIT)
+#define RTE_SORING_ST_FINISH RTE_SHIFT_VAL32(2, RTE_SORING_ST_BIT)
+
+#define RTE_SORING_ST_MASK \
+ RTE_GENMASK32(sizeof(uint32_t) * CHAR_BIT - 1, RTE_SORING_ST_BIT)
+
+/**
+ * SORING re-uses rte_ring internal structures and implementation
+ * for enqueue/dequeue operations.
+ */
+struct __rte_ring_headtail {
+
+ union __rte_cache_aligned {
+ struct rte_ring_headtail ht;
+ struct rte_ring_hts_headtail hts;
+ struct rte_ring_rts_headtail rts;
+ };
+
+ RTE_CACHE_GUARD;
+};
+
+union soring_stage_tail {
+ /** raw 8B value to read/write *sync* and *pos* as one atomic op */
+ alignas(sizeof(uint64_t)) RTE_ATOMIC(uint64_t) raw;
+ struct {
+ RTE_ATOMIC(uint32_t) sync;
+ RTE_ATOMIC(uint32_t) pos;
+ };
+};
+
+struct soring_stage_headtail {
+ volatile union soring_stage_tail tail;
+ enum rte_ring_sync_type unused; /**< unused */
+ volatile RTE_ATOMIC(uint32_t) head;
+};
+
+/**
+ * SORING stage head_tail structure is 'compatible' with rte_ring ones.
+ * rte_ring internal functions for moving producer/consumer head
+ * can work transparently with stage head_tail and visa-versa
+ * (no modifications required).
+ */
+struct soring_stage {
+
+ union __rte_cache_aligned {
+ struct rte_ring_headtail ht;
+ struct soring_stage_headtail sht;
+ };
+
+ RTE_CACHE_GUARD;
+};
+
+/**
+ * RTE soring internal structure.
+ * As with rte_ring actual elements array supposed to be located direclty
+ * after the rte_soring structure.
+ */
+struct __rte_cache_aligned rte_soring {
+ uint32_t size; /**< Size of ring. */
+ uint32_t mask; /**< Mask (size-1) of ring. */
+ uint32_t capacity; /**< Usable size of ring */
+ uint32_t esize;
+ /**< size of elements in the ring, must be a multiple of 4*/
+ uint32_t msize;
+ /**< size of metadata value for each elem, must be a multiple of 4 */
+
+ /** Ring stages */
+ struct soring_stage *stage;
+ uint32_t nb_stage;
+
+ /** Ring of states (one per element) */
+ union soring_state *state;
+
+ /** Pointer to the buffer where metadata values for each elements
+ * are stored. This is supplementary and optional information that
+ * user can attach to each element of the ring.
+ * While it is possible to incorporate this information inside
+ * user-defined element, in many cases it is plausible to maintain it
+ * as a separate array (mainly for performance reasons).
+ */
+ void *meta;
+
+ RTE_CACHE_GUARD;
+
+ /** Ring producer status. */
+ struct __rte_ring_headtail prod;
+
+ /** Ring consumer status. */
+ struct __rte_ring_headtail cons;
+
+ char name[RTE_RING_NAMESIZE]; /**< Name of the ring. */
+};
+
+#endif /* _SORING_H_ */
diff --git a/lib/ring/version.map b/lib/ring/version.map
index 61f7464f5a..081034c30e 100644
--- a/lib/ring/version.map
+++ b/lib/ring/version.map
@@ -20,4 +20,23 @@ EXPERIMENTAL {
# added in 24.11
rte_ring_headtail_dump;
+ rte_soring_acquire_bulk;
+ rte_soring_acquire_burst;
+ rte_soring_acquirx_bulk;
+ rte_soring_acquirx_burst;
+ rte_soring_count;
+ rte_soring_dequeue_bulk;
+ rte_soring_dequeue_burst;
+ rte_soring_dequeux_bulk;
+ rte_soring_dequeux_burst;
+ rte_soring_enqueue_bulk;
+ rte_soring_enqueue_burst;
+ rte_soring_enqueux_bulk;
+ rte_soring_enqueux_burst;
+ rte_soring_dump;
+ rte_soring_free_count;
+ rte_soring_get_memsize;
+ rte_soring_init;
+ rte_soring_release;
+ rte_soring_releasx;
};
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [RFCv2 5/6] app/test: add unit tests for soring API
2024-09-06 13:13 ` [RFCv2 0/6] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
` (3 preceding siblings ...)
2024-09-06 13:13 ` [RFCv2 4/6] ring/soring: introduce Staged Ordered Ring Konstantin Ananyev
@ 2024-09-06 13:13 ` Konstantin Ananyev
2024-09-06 13:13 ` [RFCv2 6/6] examples/l3fwd: make ACL work in pipeline and eventdev modes Konstantin Ananyev
2024-09-16 12:37 ` [PATCH v3 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
6 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-06 13:13 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, Konstantin Ananyev, Eimear Morrissey
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
Add both functional and stess test-cases for soring API.
Stress test serves as both functional and performance test of soring
enqueue/dequeue/acquire/release operations under high contention
(for both over committed and non-over committed scenarios).
Signed-off-by: Eimear Morrissey <eimear.morrissey@huawei.com>
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
---
app/test/meson.build | 3 +
app/test/test_soring.c | 442 +++++++++++++++
app/test/test_soring_mt_stress.c | 40 ++
app/test/test_soring_stress.c | 48 ++
app/test/test_soring_stress.h | 35 ++
app/test/test_soring_stress_impl.h | 827 +++++++++++++++++++++++++++++
6 files changed, 1395 insertions(+)
create mode 100644 app/test/test_soring.c
create mode 100644 app/test/test_soring_mt_stress.c
create mode 100644 app/test/test_soring_stress.c
create mode 100644 app/test/test_soring_stress.h
create mode 100644 app/test/test_soring_stress_impl.h
diff --git a/app/test/meson.build b/app/test/meson.build
index e29258e6ec..c290162e43 100644
--- a/app/test/meson.build
+++ b/app/test/meson.build
@@ -175,6 +175,9 @@ source_file_deps = {
'test_security_proto.c' : ['cryptodev', 'security'],
'test_seqlock.c': [],
'test_service_cores.c': [],
+ 'test_soring.c': [],
+ 'test_soring_mt_stress.c': [],
+ 'test_soring_stress.c': [],
'test_spinlock.c': [],
'test_stack.c': ['stack'],
'test_stack_perf.c': ['stack'],
diff --git a/app/test/test_soring.c b/app/test/test_soring.c
new file mode 100644
index 0000000000..b2110305a7
--- /dev/null
+++ b/app/test/test_soring.c
@@ -0,0 +1,442 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#include <string.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <inttypes.h>
+#include <errno.h>
+#include <sys/queue.h>
+
+#include <rte_common.h>
+#include <rte_log.h>
+#include <rte_memory.h>
+#include <rte_launch.h>
+#include <rte_cycles.h>
+#include <rte_eal.h>
+#include <rte_per_lcore.h>
+#include <rte_lcore.h>
+#include <rte_branch_prediction.h>
+#include <rte_malloc.h>
+#include <rte_random.h>
+#include <rte_errno.h>
+#include <rte_hexdump.h>
+
+#include <rte_soring.h>
+
+#include "test.h"
+
+#define MAX_ACQUIRED 20
+
+#define SORING_TEST_ASSERT(val, expected) do { \
+ RTE_TEST_ASSERT(expected == val, \
+ "%s: expected %u got %u\n", #val, expected, val); \
+} while (0)
+
+static void
+set_soring_init_param(struct rte_soring_param *prm,
+ const char *name, uint32_t esize, uint32_t elems,
+ uint32_t stages, uint32_t stsize,
+ enum rte_ring_sync_type rst_prod,
+ enum rte_ring_sync_type rst_cons)
+{
+ prm->name = name;
+ prm->elem_size = esize;
+ prm->elems = elems;
+ prm->stages = stages;
+ prm->meta_size = stsize;
+ prm->prod_synt = rst_prod;
+ prm->cons_synt = rst_cons;
+}
+
+static int
+move_forward_stage(struct rte_soring *sor,
+ uint32_t num_packets, uint32_t stage)
+{
+ uint32_t acquired;
+ uint32_t ftoken;
+ uint32_t *acquired_objs[MAX_ACQUIRED];
+
+ acquired = rte_soring_acquire_bulk(sor, acquired_objs, stage,
+ num_packets, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, num_packets);
+ rte_soring_release(sor, NULL, stage, num_packets,
+ ftoken);
+
+ return 0;
+}
+
+/*
+ * struct rte_soring_param param checking.
+ */
+static int
+test_soring_init(void)
+{
+ struct rte_soring *sor = NULL;
+ struct rte_soring_param prm;
+ int rc;
+ size_t sz;
+ memset(&prm, 0, sizeof(prm));
+
+/*init memory*/
+ set_soring_init_param(&prm, "alloc_memory", sizeof(uintptr_t),
+ 4, 1, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ sor = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ RTE_TEST_ASSERT_NOT_NULL(sor, "could not allocate memory for soring");
+
+ set_soring_init_param(&prm, "test_invalid_stages", sizeof(uintptr_t),
+ 4, 0, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with invalid num stages");
+
+ set_soring_init_param(&prm, "test_invalid_esize", 0,
+ 4, 1, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with 0 esize");
+
+ set_soring_init_param(&prm, "test_invalid_esize", 9,
+ 4, 1, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with esize not multiple of 4");
+
+ set_soring_init_param(&prm, "test_invalid_rsize", sizeof(uintptr_t),
+ 4, 1, 3, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with rcsize not multiple of 4");
+
+ set_soring_init_param(&prm, "test_invalid_elems", sizeof(uintptr_t),
+ RTE_SORING_ELEM_MAX + 1, 1, 4, RTE_RING_SYNC_MT,
+ RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with invalid num elements");
+
+ rte_free(sor);
+ return 0;
+}
+
+static int
+test_soring_get_memsize(void)
+{
+
+ struct rte_soring_param prm;
+ ssize_t sz;
+
+ set_soring_init_param(&prm, "memsize", sizeof(uint32_t *),
+ 10, 1, 0, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ RTE_TEST_ASSERT(sz > 0, "failed to calculate size");
+
+ set_soring_init_param(&prm, "memsize", sizeof(uint8_t),
+ 16, UINT32_MAX, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ RTE_TEST_ASSERT_EQUAL(sz, -EINVAL, "calcuated size incorrect");
+
+ set_soring_init_param(&prm, "memsize", 0,
+ 16, UINT32_MAX, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ RTE_TEST_ASSERT_EQUAL(sz, -EINVAL, "calculated size incorrect");
+
+ return 0;
+
+}
+
+static int
+test_soring_stages(void)
+{
+ struct rte_soring *sor = NULL;
+ struct rte_soring_param prm;
+ uint32_t objs[32];
+ uint32_t rcs[32];
+ uint32_t acquired_objs[32];
+ uint32_t acquired_rcs[32];
+ uint32_t dequeued_rcs[32];
+ uint32_t dequeued_objs[32];
+ size_t ssz;
+ uint32_t stage, enqueued, dequeued, acquired;
+ uint32_t i, ftoken;
+
+ memset(&prm, 0, sizeof(prm));
+ set_soring_init_param(&prm, "stages", sizeof(uint32_t), 32,
+ 10000, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ ssz = rte_soring_get_memsize(&prm);
+ RTE_TEST_ASSERT(ssz > 0, "parameter error calculating ring size");
+ sor = rte_zmalloc(NULL, ssz, RTE_CACHE_LINE_SIZE);
+ RTE_TEST_ASSERT_NOT_NULL(sor, "couldn't allocate memory for soring");
+ rte_soring_init(sor, &prm);
+
+ for (i = 0; i < 32; i++) {
+ rcs[i] = i;
+ objs[i] = i + 32;
+ }
+
+ enqueued = rte_soring_enqueux_burst(sor, objs, rcs, 32, NULL);
+ SORING_TEST_ASSERT(enqueued, 32);
+
+ for (stage = 0; stage < 10000; stage++) {
+ int j;
+ dequeued = rte_soring_dequeue_bulk(sor, dequeued_objs,
+ 32, NULL);
+ /* check none at end stage */
+ SORING_TEST_ASSERT(dequeued, 0);
+
+ acquired = rte_soring_acquirx_bulk(sor, acquired_objs,
+ acquired_rcs, stage, 32, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, 32);
+
+ for (j = 0; j < 32; j++) {
+ SORING_TEST_ASSERT(acquired_rcs[j], j + stage);
+ SORING_TEST_ASSERT(acquired_objs[j], j + stage + 32);
+ /* modify both queue object and rc */
+ acquired_objs[j]++;
+ acquired_rcs[j]++;
+ }
+
+ rte_soring_releasx(sor, acquired_objs,
+ acquired_rcs, stage, 32,
+ ftoken);
+ }
+
+ dequeued = rte_soring_dequeux_bulk(sor, dequeued_objs, dequeued_rcs,
+ 32, NULL);
+ SORING_TEST_ASSERT(dequeued, 32);
+ for (i = 0; i < 32; i++) {
+ /* ensure we ended up with the expected values in order*/
+ SORING_TEST_ASSERT(dequeued_rcs[i], i + 10000);
+ SORING_TEST_ASSERT(dequeued_objs[i], i + 32 + 10000);
+ }
+ rte_free(sor);
+ return 0;
+}
+
+static int
+test_soring_enqueue_dequeue(void)
+{
+ struct rte_soring *sor = NULL;
+ struct rte_soring_param prm;
+ int rc;
+ uint32_t i;
+ size_t sz;
+ uint32_t queue_objs[10];
+ uint32_t *queue_objs_p[10];
+ uint32_t free_space;
+ uint32_t enqueued, dequeued;
+ uint32_t *dequeued_objs[10];
+
+ memset(&prm, 0, sizeof(prm));
+ for (i = 0; i < 10; i++) {
+ queue_objs[i] = i + 1;
+ queue_objs_p[i] = &queue_objs[i];
+ }
+
+/*init memory*/
+ set_soring_init_param(&prm, "enqueue/dequeue", sizeof(uint32_t *),
+ 10, 1, 0, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ sor = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ RTE_TEST_ASSERT_NOT_NULL(sor, "alloc failed for soring");
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_SUCCESS(rc, "Failed to init soring");
+
+ free_space = 0;
+
+ enqueued = rte_soring_enqueue_burst(sor, queue_objs_p, 5, &free_space);
+
+ SORING_TEST_ASSERT(free_space, 5);
+ SORING_TEST_ASSERT(enqueued, 5);
+
+ /* fixed amount enqueue */
+ enqueued = rte_soring_enqueue_bulk(sor, queue_objs_p, 7, &free_space);
+
+ SORING_TEST_ASSERT(free_space, 5);
+ SORING_TEST_ASSERT(enqueued, 0);
+
+ /* variable amount enqueue */
+ enqueued = rte_soring_enqueue_burst(sor, queue_objs_p + 5, 7,
+ &free_space);
+ SORING_TEST_ASSERT(free_space, 0);
+ SORING_TEST_ASSERT(enqueued, 5);
+
+ /* test no dequeue while stage 0 has not completed */
+ dequeued = rte_soring_dequeue_bulk(sor, dequeued_objs, 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 0);
+ dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 0);
+
+ move_forward_stage(sor, 8, 0);
+
+ /* can only dequeue as many as have completed stage*/
+ dequeued = rte_soring_dequeue_bulk(sor, dequeued_objs, 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 0);
+ dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 8);
+ /* packets remain in order */
+ for (i = 0; i < dequeued; i++) {
+ RTE_TEST_ASSERT_EQUAL(dequeued_objs[i],
+ queue_objs_p[i], "dequeued != enqueued");
+ }
+
+ dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 1, NULL);
+ SORING_TEST_ASSERT(dequeued, 0);
+
+ move_forward_stage(sor, 2, 0);
+ dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 2, NULL);
+ SORING_TEST_ASSERT(dequeued, 2);
+ /* packets remain in order */
+ for (i = 0; i < dequeued; i++) {
+ RTE_TEST_ASSERT_EQUAL(dequeued_objs[i],
+ queue_objs_p[i + 8], "dequeued != enqueued");
+ }
+
+ rte_soring_dump(stdout, sor);
+ rte_free(sor);
+ return 0;
+}
+
+static int
+test_soring_acquire_release(void)
+{
+
+ struct rte_soring *sor = NULL;
+ struct rte_soring_param prm;
+ int rc, i;
+ size_t sz;
+
+ memset(&prm, 0, sizeof(prm));
+ uint32_t queue_objs[10];
+ uint32_t rc_objs[10];
+ uint32_t acquired_objs[10];
+ uint32_t dequeued_objs[10];
+ uint32_t dequeued_rcs[10];
+ uint32_t s1_acquired_rcs[10];
+ uint32_t free_space, enqueued, ftoken, acquired, dequeued;
+
+ for (i = 0; i < 10; i++) {
+ queue_objs[i] = i + 5;
+ rc_objs[i] = i + 10;
+ }
+
+/*init memory*/
+ set_soring_init_param(&prm, "test_acquire_release", sizeof(uint32_t),
+ 20, 2, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ sor = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ if (sor == NULL) {
+ printf("%s: alloc(%zu) for FIFO with %u elems failed",
+ __func__, sz, prm.elems);
+ return -ENOMEM;
+ }
+
+ /* init ring */
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_SUCCESS(rc, "failed to init soring");
+
+ /* enqueue with associated rc */
+ enqueued = rte_soring_enqueux_burst(sor, queue_objs, rc_objs, 5,
+ &free_space);
+ SORING_TEST_ASSERT(enqueued, 5);
+ /* enqueue without associated rc */
+ enqueued = rte_soring_enqueue_burst(sor, queue_objs + 5, 5,
+ &free_space);
+ SORING_TEST_ASSERT(enqueued, 5);
+
+ /* acquire the objects with rc's and ensure they are as expected */
+ acquired = rte_soring_acquirx_burst(sor, acquired_objs,
+ s1_acquired_rcs, 0, 5, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, 5);
+ for (i = 0; i < 5; i++) {
+ RTE_TEST_ASSERT_EQUAL(s1_acquired_rcs[i], rc_objs[i],
+ "acquired rc[%d]: %u != enqueued rc: %u",
+ i, s1_acquired_rcs[i], rc_objs[i]);
+ RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i],
+ "acquired obj[%d]: %u != enqueued obj %u",
+ i, acquired_objs[i], queue_objs[i]);
+ }
+ rte_soring_release(sor, NULL, 0, 5, ftoken);
+
+ /* acquire the objects without rc's and ensure they are as expected */
+ acquired = rte_soring_acquirx_burst(sor, acquired_objs,
+ s1_acquired_rcs, 0, 5, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, 5);
+ for (i = 0; i < 5; i++) {
+ /* as the rc area of memory is zero'd at init this is true
+ * but this is a detail of implementation rather than
+ * a guarantee.
+ */
+ RTE_TEST_ASSERT_EQUAL(s1_acquired_rcs[i], 0,
+ "acquired rc not empty");
+ RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i + 5],
+ "acquired obj[%d]: %u != enqueued obj %u",
+ i, acquired_objs[i], queue_objs[i + 5]);
+ }
+ /*release the objects, adding rc's */
+ rte_soring_releasx(sor, NULL, rc_objs + 5, 0, 5,
+ ftoken);
+
+ acquired = rte_soring_acquirx_burst(sor, acquired_objs,
+ s1_acquired_rcs, 1, 10, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, 10);
+ for (i = 0; i < 10; i++) {
+ /* ensure the associated rc's are the ones added at relase */
+ RTE_TEST_ASSERT_EQUAL(s1_acquired_rcs[i], rc_objs[i],
+ "acquired rc[%d]: %u != enqueued rc: %u",
+ i, s1_acquired_rcs[i], rc_objs[i]);
+ RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i],
+ "acquired obj[%d]: %u != enqueued obj %u",
+ i, acquired_objs[i], queue_objs[i]);
+ }
+ /*release the objects, with rc's set to NULL */
+ rte_soring_release(sor, NULL, 1, 10, ftoken);
+
+ dequeued = rte_soring_dequeux_burst(sor, dequeued_objs, dequeued_rcs,
+ 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 10);
+ for (i = 0; i < 10; i++) {
+ /* ensure the associated rc's are the ones added at relase */
+ RTE_TEST_ASSERT_EQUAL(dequeued_rcs[i], rc_objs[i],
+ "dequeued rc[%d]: %u != enqueued rc: %u",
+ i, dequeued_rcs[i], rc_objs[i]);
+ RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i],
+ "acquired obj[%d]: %u != enqueued obj %u",
+ i, acquired_objs[i], queue_objs[i]);
+ }
+ rte_soring_dump(stdout, sor);
+ rte_free(sor);
+ return 0;
+}
+
+static int
+test_soring(void)
+{
+
+ /* Negative test cases */
+ if (test_soring_init() < 0)
+ goto test_fail;
+
+ /* Memory calculations */
+ if (test_soring_get_memsize() < 0)
+ goto test_fail;
+
+ /* Basic enqueue/dequeue operations */
+ if (test_soring_enqueue_dequeue() < 0)
+ goto test_fail;
+
+ /* Acquire/release */
+ if (test_soring_acquire_release() < 0)
+ goto test_fail;
+
+ /* Test large number of stages */
+ if (test_soring_stages() < 0)
+ goto test_fail;
+
+ return 0;
+
+test_fail:
+ return -1;
+}
+
+REGISTER_FAST_TEST(soring_autotest, true, true, test_soring);
+
diff --git a/app/test/test_soring_mt_stress.c b/app/test/test_soring_mt_stress.c
new file mode 100644
index 0000000000..2f90bb4598
--- /dev/null
+++ b/app/test/test_soring_mt_stress.c
@@ -0,0 +1,40 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#include "test_soring_stress_impl.h"
+
+static inline uint32_t
+_st_ring_dequeue_burst(struct rte_soring *r, void **obj, uint32_t n,
+ uint32_t *avail)
+{
+ return rte_soring_dequeue_burst(r, obj, n, avail);
+}
+
+static inline uint32_t
+_st_ring_enqueue_bulk(struct rte_soring *r, void * const *obj, uint32_t n,
+ uint32_t *free)
+{
+ return rte_soring_enqueue_bulk(r, obj, n, free);
+}
+
+static inline uint32_t
+_st_ring_acquire_burst(struct rte_soring *r, uint32_t stage, void **obj,
+ uint32_t num, uint32_t *token, uint32_t *avail)
+{
+ return rte_soring_acquire_burst(r, obj, stage, num, token, avail);
+}
+
+static inline void
+_st_ring_release(struct rte_soring *r, uint32_t stage, uint32_t token,
+ void * const *obj, uint32_t num)
+{
+ RTE_SET_USED(obj);
+ rte_soring_release(r, NULL, stage, num, token);
+}
+
+const struct test test_soring_mt_stress = {
+ .name = "MT",
+ .nb_case = RTE_DIM(tests),
+ .cases = tests,
+};
diff --git a/app/test/test_soring_stress.c b/app/test/test_soring_stress.c
new file mode 100644
index 0000000000..334af6a29c
--- /dev/null
+++ b/app/test/test_soring_stress.c
@@ -0,0 +1,48 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#include "test_soring_stress.h"
+
+static int
+run_test(const struct test *test)
+{
+ int32_t rc;
+ uint32_t i, k;
+
+ for (i = 0, k = 0; i != test->nb_case; i++) {
+
+ printf("TEST-CASE %s %s START\n",
+ test->name, test->cases[i].name);
+
+ rc = test->cases[i].func(test->cases[i].wfunc);
+ k += (rc == 0);
+
+ if (rc != 0)
+ printf("TEST-CASE %s %s FAILED\n",
+ test->name, test->cases[i].name);
+ else
+ printf("TEST-CASE %s %s OK\n",
+ test->name, test->cases[i].name);
+ }
+
+ return k;
+}
+
+static int
+test_ring_stress(void)
+{
+ uint32_t n, k;
+
+ n = 0;
+ k = 0;
+
+ n += test_soring_mt_stress.nb_case;
+ k += run_test(&test_soring_mt_stress);
+
+ printf("Number of tests:\t%u\nSuccess:\t%u\nFailed:\t%u\n",
+ n, k, n - k);
+ return (k != n);
+}
+
+REGISTER_TEST_COMMAND(soring_stress_autotest, test_ring_stress);
diff --git a/app/test/test_soring_stress.h b/app/test/test_soring_stress.h
new file mode 100644
index 0000000000..6190e96117
--- /dev/null
+++ b/app/test/test_soring_stress.h
@@ -0,0 +1,35 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+
+#include <inttypes.h>
+#include <stddef.h>
+#include <stdalign.h>
+#include <string.h>
+#include <stdio.h>
+#include <unistd.h>
+
+#include <rte_soring.h>
+#include <rte_cycles.h>
+#include <rte_launch.h>
+#include <rte_pause.h>
+#include <rte_random.h>
+#include <rte_malloc.h>
+#include <rte_spinlock.h>
+
+#include "test.h"
+
+struct test_case {
+ const char *name;
+ int (*func)(int (*)(void *));
+ int (*wfunc)(void *arg);
+};
+
+struct test {
+ const char *name;
+ uint32_t nb_case;
+ const struct test_case *cases;
+};
+
+extern const struct test test_soring_mt_stress;
diff --git a/app/test/test_soring_stress_impl.h b/app/test/test_soring_stress_impl.h
new file mode 100644
index 0000000000..1ba78a31f5
--- /dev/null
+++ b/app/test/test_soring_stress_impl.h
@@ -0,0 +1,827 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#include <stdalign.h>
+
+#include "test_soring_stress.h"
+
+/**
+ * Stress test for soring enqueue/dequeue/acquire/release operations.
+ * Depending on the role, performs at least one of the following patterns
+ * on each worker:
+ * - dequeue/read-write data from/to the dequeued objects/enqueue.
+ * - acquire/read-write data from/to the acquired objects/release.
+ * Serves as both functional and performance test of soring
+ * data-path API under high contention
+ * (for both over committed and non-over committed scenarios).
+ */
+
+#define RING_NAME "SORING_STRESS"
+#define BULK_NUM 32
+#define RING_SIZE (2 * BULK_NUM * RTE_MAX_LCORE)
+
+#define MAX_STAGES 16
+
+enum {
+ WRK_CMD_STOP,
+ WRK_CMD_RUN,
+};
+
+static uint32_t wrk_cmd __rte_cache_aligned = WRK_CMD_STOP;
+
+/* test run-time in seconds */
+static const uint32_t run_time = 60;
+static const uint32_t verbose;
+
+struct lcore_op_stat {
+ uint64_t nb_lcore;
+ uint64_t nb_call;
+ uint64_t nb_obj;
+ uint64_t nb_cycle;
+ uint64_t max_cycle;
+ uint64_t min_cycle;
+};
+
+struct lcore_stat {
+ uint64_t nb_cycle;
+ struct lcore_op_stat deqenq;
+ uint32_t role_mask;
+ uint32_t nb_stage;
+ struct lcore_op_stat stage[MAX_STAGES];
+};
+
+#define ROLE_DEQENQ RTE_BIT32(0)
+#define ROLE_STAGE(n) RTE_BIT32(n + 1)
+
+struct lcore_arg {
+ struct rte_soring *rng;
+ struct lcore_stat stats;
+} __rte_cache_aligned;
+
+struct ring_elem {
+ uint32_t cnt[RTE_CACHE_LINE_SIZE / sizeof(uint32_t)];
+} __rte_cache_aligned;
+
+/*
+ * redefinable functions
+ */
+
+static uint32_t
+_st_ring_dequeue_burst(struct rte_soring *r, void **obj, uint32_t n,
+ uint32_t *avail);
+
+static uint32_t
+_st_ring_enqueue_bulk(struct rte_soring *r, void * const *obj, uint32_t n,
+ uint32_t *free);
+
+static uint32_t
+_st_ring_acquire_burst(struct rte_soring *r, uint32_t stage, void **obj,
+ uint32_t num, uint32_t *token, uint32_t *avail);
+
+static void
+_st_ring_release(struct rte_soring *r, uint32_t stage, uint32_t token,
+ void * const *obj, uint32_t num);
+
+static void
+lcore_op_stat_update(struct lcore_op_stat *ls, uint64_t call, uint64_t obj,
+ uint64_t tm, int32_t prcs)
+{
+ ls->nb_call += call;
+ ls->nb_obj += obj;
+ ls->nb_cycle += tm;
+ if (prcs) {
+ ls->max_cycle = RTE_MAX(ls->max_cycle, tm);
+ ls->min_cycle = RTE_MIN(ls->min_cycle, tm);
+ }
+}
+
+static void
+lcore_stat_update(struct lcore_stat *ls, uint64_t call, uint64_t obj,
+ uint64_t tm, int32_t prcs)
+{
+ uint32_t i;
+
+ ls->nb_cycle += tm;
+ lcore_op_stat_update(&ls->deqenq, call, obj, tm, prcs);
+ for (i = 0; i != ls->nb_stage; i++)
+ lcore_op_stat_update(ls->stage + i, call, obj, tm, prcs);
+}
+
+static void
+lcore_op_stat_aggr(struct lcore_op_stat *ms, const struct lcore_op_stat *ls)
+{
+ ms->nb_call += ls->nb_call;
+ ms->nb_obj += ls->nb_obj;
+ ms->nb_cycle += ls->nb_cycle;
+ ms->max_cycle = RTE_MAX(ms->max_cycle, ls->max_cycle);
+ ms->min_cycle = RTE_MIN(ms->min_cycle, ls->min_cycle);
+}
+
+static void
+lcore_stat_aggr(struct lcore_stat *ms, const struct lcore_stat *ls)
+{
+ uint32_t i;
+
+ ms->nb_cycle = RTE_MAX(ms->nb_cycle, ls->nb_cycle);
+ lcore_op_stat_aggr(&ms->deqenq, &ls->deqenq);
+ ms->deqenq.nb_lcore += ((ls->role_mask & ROLE_DEQENQ) != 0);
+ for (i = 0; i != ms->nb_stage; i++) {
+ lcore_op_stat_aggr(ms->stage + i, ls->stage + i);
+ ms->stage[i].nb_lcore += ((ls->role_mask & ROLE_STAGE(i)) != 0);
+ }
+}
+
+static void
+lcore_op_stat_dump(FILE *f, const struct lcore_op_stat *ls, const char *cap,
+ long double st)
+{
+ fprintf(f, "\t%s={\n", cap);
+
+ fprintf(f, "\t\tnb_lcore=%" PRIu64 ",\n", ls->nb_lcore);
+ fprintf(f, "\t\tnb_call=%" PRIu64 ",\n", ls->nb_call);
+ fprintf(f, "\t\tnb_obj=%" PRIu64 ",\n", ls->nb_obj);
+ fprintf(f, "\t\tnb_cycle=%" PRIu64 ",\n", ls->nb_cycle);
+ fprintf(f, "\t\tobj/call(avg): %.2Lf\n",
+ (long double)ls->nb_obj / ls->nb_call);
+ fprintf(f, "\t\tcycles/obj(avg): %.2Lf\n",
+ (long double)ls->nb_cycle / ls->nb_obj);
+ fprintf(f, "\t\tcycles/call(avg): %.2Lf\n",
+ (long double)ls->nb_cycle / ls->nb_call);
+
+ /* if min/max cycles per call stats was collected */
+ if (ls->min_cycle != UINT64_MAX) {
+ fprintf(f, "\t\tmax cycles/call=%" PRIu64 "(%.2Lf usec),\n",
+ ls->max_cycle,
+ (long double)ls->max_cycle / st);
+ fprintf(f, "\t\tmin cycles/call=%" PRIu64 "(%.2Lf usec),\n",
+ ls->min_cycle,
+ (long double)ls->min_cycle / st);
+ }
+
+ fprintf(f, "\t},\n");
+}
+
+static void
+lcore_stat_dump(FILE *f, uint32_t lc, const struct lcore_stat *ls)
+{
+ uint32_t i;
+ long double st;
+ char cap[64];
+
+ st = (long double)rte_get_timer_hz() / US_PER_S;
+
+ if (lc == UINT32_MAX)
+ fprintf(f, "%s(AGGREGATE)={\n", __func__);
+ else
+ fprintf(f, "%s(lcore=%u)={\n", __func__, lc);
+
+ fprintf(f, "\tnb_cycle=%" PRIu64 "(%.2Lf usec),\n",
+ ls->nb_cycle, (long double)ls->nb_cycle / st);
+
+ lcore_op_stat_dump(f, &ls->deqenq, "DEQ+ENQ", st);
+ for (i = 0; i != ls->nb_stage; i++) {
+ snprintf(cap, sizeof(cap), "%s#%u", "STAGE", i);
+ lcore_op_stat_dump(f, ls->stage + i, cap, st);
+ }
+
+ fprintf(f, "};\n");
+}
+
+static void
+fill_ring_elm(struct ring_elem *elm, uint32_t fill)
+{
+ uint32_t i;
+
+ for (i = 0; i != RTE_DIM(elm->cnt); i++)
+ elm->cnt[i] = fill;
+}
+
+static int32_t
+check_updt_elem(struct ring_elem *elm[], uint32_t num,
+ const struct ring_elem *check, const struct ring_elem *fill)
+{
+ uint32_t i;
+
+ static rte_spinlock_t dump_lock;
+
+ for (i = 0; i != num; i++) {
+ if (memcmp(check, elm[i], sizeof(*check)) != 0) {
+ rte_spinlock_lock(&dump_lock);
+ printf("%s(lc=%u, num=%u) failed at %u-th iter, "
+ "offending object: %p\n",
+ __func__, rte_lcore_id(), num, i, elm[i]);
+ rte_memdump(stdout, "expected", check, sizeof(*check));
+ rte_memdump(stdout, "result", elm[i], sizeof(*elm[i]));
+ rte_spinlock_unlock(&dump_lock);
+ return -EINVAL;
+ }
+ memcpy(elm[i], fill, sizeof(*elm[i]));
+ }
+
+ return 0;
+}
+
+static int
+check_ring_op(uint32_t exp, uint32_t res, uint32_t lc,
+ enum rte_ring_queue_behavior bhv, const char *fname, const char *opname)
+{
+ if (bhv == RTE_RING_QUEUE_FIXED && exp != res) {
+ printf("%s(lc=%u) failure: %s expected: %u, returned %u\n",
+ fname, lc, opname, exp, res);
+ return -ENOSPC;
+ }
+ if (bhv == RTE_RING_QUEUE_VARIABLE && exp < res) {
+ printf("%s(lc=%u) failure: %s expected: %u, returned %u\n",
+ fname, lc, opname, exp, res);
+ return -ENOSPC;
+ }
+ return 0;
+}
+
+/* num in interval [7/8, 11/8] of BULK_NUM */
+static inline uint32_t
+rand_elem_num(void)
+{
+ uint32_t num;
+
+ num = 7 * BULK_NUM / 8 + rte_rand() % (BULK_NUM / 2);
+ return num;
+}
+
+/*
+ * for each enabled stage do:
+ * acquire burst of objects
+ * read and check their contents
+ * update and check their contents
+ * release burst of objects
+ * done
+ */
+static int32_t
+test_worker_stages(struct lcore_arg *la, uint32_t lc, const char *fname,
+ struct ring_elem *obj[2 * BULK_NUM],
+ const struct ring_elem *def_elm, const struct ring_elem *loc_elm,
+ const struct ring_elem stg_elm[MAX_STAGES], int32_t prcs)
+{
+ int32_t rc;
+ uint32_t i, n, num, tkn;
+ uint64_t tm0, tm1;
+ const struct ring_elem *celm, *pelm;
+
+ num = rand_elem_num();
+
+ rc = 0;
+ tkn = 0;
+ for (i = 0, pelm = def_elm; i != la->stats.nb_stage; pelm = celm, i++) {
+
+ celm = stg_elm + i;
+
+ /* given stage is not enabled on that lcore */
+ if ((la->stats.role_mask & ROLE_STAGE(i)) == 0)
+ continue;
+
+ /* reset all pointer values */
+ memset(obj, 0, sizeof(*obj) * num);
+
+ /* acquire num elems */
+ tm0 = (prcs != 0) ? rte_rdtsc_precise() : 0;
+ n = _st_ring_acquire_burst(la->rng, i, (void **)obj, num,
+ &tkn, NULL);
+ tm0 = (prcs != 0) ? rte_rdtsc_precise() - tm0 : 0;
+
+ /* check return value and objects */
+ rc = check_ring_op(num, n, lc, RTE_RING_QUEUE_VARIABLE, fname,
+ RTE_STR(_st_ring_stage_acquire));
+ if (rc == 0)
+ rc = check_updt_elem(obj, n, pelm, loc_elm);
+ if (rc != 0)
+ break;
+
+ /* release num elems */
+ rte_compiler_barrier();
+ rc = check_updt_elem(obj, n, loc_elm, celm);
+ if (rc != 0)
+ break;
+
+ if (n == 0)
+ tm1 = 0;
+ else {
+ tm1 = (prcs != 0) ? rte_rdtsc_precise() : 0;
+ _st_ring_release(la->rng, i, tkn,
+ (void **)obj, n);
+ tm1 = (prcs != 0) ? rte_rdtsc_precise() - tm1 : 0;
+ }
+ lcore_op_stat_update(la->stats.stage + i, 1, n, tm0 + tm1,
+ prcs);
+ }
+
+ return rc;
+}
+
+static int32_t
+test_worker_deqenq(struct lcore_arg *la, uint32_t lc, const char *fname,
+ struct ring_elem *obj[2 * BULK_NUM],
+ const struct ring_elem *def_elm, const struct ring_elem *loc_elm,
+ const struct ring_elem *pelm, int32_t prcs)
+{
+ int32_t rc;
+ uint32_t k, n, num;
+ uint64_t tm0, tm1;
+
+ num = rand_elem_num();
+
+ /* reset all pointer values */
+ memset(obj, 0, sizeof(*obj) * num);
+
+ /* dequeue num elems */
+ tm0 = (prcs != 0) ? rte_rdtsc_precise() : 0;
+ n = _st_ring_dequeue_burst(la->rng, (void **)obj, num, NULL);
+
+ tm0 = (prcs != 0) ? rte_rdtsc_precise() - tm0 : 0;
+
+ /* check return value and objects */
+ rc = check_ring_op(num, n, lc, RTE_RING_QUEUE_VARIABLE, fname,
+ RTE_STR(_st_ring_dequeue_bulk));
+ if (rc == 0)
+ rc = check_updt_elem(obj, n, pelm, loc_elm);
+ if (rc != 0)
+ return rc;
+
+ /* enqueue n elems */
+ rte_compiler_barrier();
+ rc = check_updt_elem(obj, n, loc_elm, def_elm);
+ if (rc != 0)
+ return rc;
+
+ tm1 = (prcs != 0) ? rte_rdtsc_precise() : 0;
+ k = _st_ring_enqueue_bulk(la->rng, (void **)obj, n, NULL);
+ tm1 = (prcs != 0) ? rte_rdtsc_precise() - tm1 : 0;
+
+ /* check return value */
+ rc = check_ring_op(n, k, lc, RTE_RING_QUEUE_FIXED, fname,
+ RTE_STR(_st_ring_enqueue_bulk));
+ if (rc != 0)
+ return rc;
+
+ lcore_op_stat_update(&la->stats.deqenq, 1, n, tm0 + tm1, prcs);
+ return 0;
+}
+
+static int
+test_worker(void *arg, const char *fname, int32_t prcs)
+{
+ int32_t rc;
+ uint32_t i, lc;
+ uint64_t cl;
+ struct lcore_arg *la;
+ struct ring_elem *obj[2 * BULK_NUM];
+ struct ring_elem *pelm, def_elm, loc_elm, stg_elm[MAX_STAGES];
+
+ la = arg;
+ lc = rte_lcore_id();
+
+ fill_ring_elm(&def_elm, UINT32_MAX);
+ fill_ring_elm(&loc_elm, lc);
+
+ for (i = 0; i != RTE_DIM(stg_elm); i++)
+ fill_ring_elm(stg_elm + i, (i + 1) << 24);
+
+ pelm = stg_elm + la->stats.nb_stage - 1;
+
+ /* Acquire ordering is not required as the main is not
+ * really releasing any data through 'wrk_cmd' to
+ * the worker.
+ */
+ while (__atomic_load_n(&wrk_cmd, __ATOMIC_RELAXED) != WRK_CMD_RUN)
+ rte_pause();
+
+ cl = rte_rdtsc_precise();
+
+ do {
+ if ((la->stats.role_mask & ~ROLE_DEQENQ) != 0) {
+ rc = test_worker_stages(la, lc, fname, obj,
+ &def_elm, &loc_elm, stg_elm, prcs);
+ if (rc != 0)
+ break;
+ }
+
+ if ((la->stats.role_mask & ROLE_DEQENQ) != 0) {
+ rc = test_worker_deqenq(la, lc, fname, obj,
+ &def_elm, &loc_elm, pelm, prcs);
+ if (rc != 0)
+ break;
+ }
+
+ } while (__atomic_load_n(&wrk_cmd, __ATOMIC_RELAXED) == WRK_CMD_RUN);
+
+ cl = rte_rdtsc_precise() - cl;
+ if (prcs == 0)
+ lcore_stat_update(&la->stats, 0, 0, cl, 0);
+ la->stats.nb_cycle = cl;
+ return rc;
+}
+static int
+test_worker_prcs(void *arg)
+{
+ return test_worker(arg, __func__, 1);
+}
+
+static int
+test_worker_avg(void *arg)
+{
+ return test_worker(arg, __func__, 0);
+}
+
+static void
+mt1_fini(struct rte_soring *rng, void *data)
+{
+ rte_free(rng);
+ rte_free(data);
+}
+
+static int
+mt1_init(struct rte_soring **rng, void **data, uint32_t num,
+ enum rte_ring_sync_type prod_synt, enum rte_ring_sync_type cons_synt,
+ uint32_t nb_stages)
+{
+ int32_t rc;
+ size_t sz;
+ uint32_t i;
+ struct rte_soring *r;
+ struct ring_elem *elm;
+ void *p;
+ struct rte_soring_param prm;
+
+ *rng = NULL;
+ *data = NULL;
+
+ sz = num * sizeof(*elm);
+ elm = rte_zmalloc(NULL, sz, alignof(typeof(*elm)));
+ if (elm == NULL) {
+ printf("%s: alloc(%zu) for %u elems data failed",
+ __func__, sz, num);
+ return -ENOMEM;
+ }
+
+ *data = elm;
+
+ /* alloc soring */
+ memset(&prm, 0, sizeof(prm));
+
+ prm.name = __func__;
+ prm.elems = num;
+ prm.elem_size = sizeof(uintptr_t);
+ prm.stages = nb_stages;
+ prm.prod_synt = prod_synt;
+ prm.cons_synt = cons_synt;
+
+ sz = rte_soring_get_memsize(&prm);
+ r = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ if (r == NULL) {
+ printf("%s: alloc(%zu) for FIFO with %u elems failed",
+ __func__, sz, prm.elems);
+ return -ENOMEM;
+ }
+
+ *rng = r;
+
+ rc = rte_soring_init(r, &prm);
+ if (rc != 0) {
+ printf("%s: rte_soring_init(r=%p,elems=%u,stages=%u) failed, "
+ "error: %d(%s)\n",
+ __func__, r, prm.elems, prm.stages, rc, strerror(-rc));
+ return rc;
+ }
+
+ for (i = 0; i != num; i++) {
+ fill_ring_elm(elm + i, UINT32_MAX);
+ p = elm + i;
+ if (_st_ring_enqueue_bulk(r, &p, 1, NULL) != 1)
+ break;
+ }
+
+ if (i != num) {
+ printf("%s: _st_ring_enqueue(%p, %u) returned %u\n",
+ __func__, r, num, i);
+ return -ENOSPC;
+ }
+
+ return 0;
+}
+
+static int
+test_mt(int (*test)(void *), enum rte_ring_sync_type prod_synt,
+ enum rte_ring_sync_type cons_synt, uint32_t nb_stage,
+ const uint32_t role_mask[RTE_MAX_LCORE])
+{
+ int32_t rc;
+ uint32_t i, lc, mc;
+ struct rte_soring *r;
+ void *data;
+ struct lcore_arg arg[RTE_MAX_LCORE];
+
+ static const struct lcore_op_stat init_stat = {
+ .min_cycle = UINT64_MAX,
+ };
+
+ rc = mt1_init(&r, &data, RING_SIZE, prod_synt, cons_synt, nb_stage);
+
+ if (rc != 0) {
+ mt1_fini(r, data);
+ return rc;
+ }
+
+ memset(arg, 0, sizeof(arg));
+
+ /* launch on all workers */
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ arg[lc].rng = r;
+ arg[lc].stats.deqenq = init_stat;
+ arg[lc].stats.nb_stage = nb_stage;
+ arg[lc].stats.role_mask = role_mask[lc];
+ for (i = 0; i != arg[lc].stats.nb_stage; i++)
+ arg[lc].stats.stage[i] = init_stat;
+ rte_eal_remote_launch(test, &arg[lc], lc);
+ }
+
+ /* signal workers to start test */
+ __atomic_store_n(&wrk_cmd, WRK_CMD_RUN, __ATOMIC_RELEASE);
+
+ rte_delay_us(run_time * US_PER_S);
+
+ /* signal workers to stop test */
+ __atomic_store_n(&wrk_cmd, WRK_CMD_STOP, __ATOMIC_RELEASE);
+
+ /* wait for workers and collect stats. */
+ mc = rte_lcore_id();
+ arg[mc].stats.deqenq = init_stat;
+ arg[mc].stats.nb_stage = nb_stage;
+ for (i = 0; i != arg[mc].stats.nb_stage; i++)
+ arg[mc].stats.stage[i] = init_stat;
+
+ rc = 0;
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ rc |= rte_eal_wait_lcore(lc);
+ lcore_stat_aggr(&arg[mc].stats, &arg[lc].stats);
+ if (verbose != 0)
+ lcore_stat_dump(stdout, lc, &arg[lc].stats);
+ }
+
+ lcore_stat_dump(stdout, UINT32_MAX, &arg[mc].stats);
+ rte_soring_dump(stdout, r);
+ mt1_fini(r, data);
+ return rc;
+}
+
+/*
+ * launch all stages and deq+enq on all worker lcores
+ */
+static void
+role_mask_sym(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE])
+{
+ uint32_t lc;
+ const uint32_t msk = RTE_BIT32(nb_stage + 2) - 1;
+
+ memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE);
+ RTE_LCORE_FOREACH_WORKER(lc)
+ role_mask[lc] = msk;
+}
+
+/*
+ * Divide all workers in two (nearly) equal groups:
+ * - workers from 'even' group do deque+enque
+ * - workers from 'odd' group do acquire/release (for all stages)
+ */
+static void
+role_mask_even_odd(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE])
+{
+ uint32_t i, lc;
+ const uint32_t msk[2] = {
+ [0] = ROLE_DEQENQ,
+ [1] = RTE_GENMASK32(nb_stage + 1, 1),
+ };
+
+ memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE);
+
+ i = 0;
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ role_mask[lc] = msk[i & 1];
+ i++;
+ }
+ if (i == 1) {
+ lc = rte_get_next_lcore(-1, 1, 0);
+ role_mask[lc] |= msk[i & 1];
+ }
+}
+
+/*
+ * Divide all workers (nearly) evenly among all possible stages
+ */
+static void
+role_mask_div(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE])
+{
+ uint32_t i, lc;
+ uint32_t msk[nb_stage + 1];
+
+ memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE);
+
+ for (i = 0; i != RTE_DIM(msk); i++) {
+ msk[i] = RTE_BIT32(i);
+ };
+
+ i = 0;
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ role_mask[lc] = msk[i % RTE_DIM(msk)];
+ i++;
+ }
+ if (i < RTE_DIM(msk)) {
+ lc = rte_get_next_lcore(-1, 1, 0);
+ for (; i != RTE_DIM(msk); i++)
+ role_mask[lc] |= msk[i % RTE_DIM(msk)];
+ }
+}
+
+/*
+ * one worker does ST enqueue+dequeue, while all others - stages processing.
+ */
+static void
+role_mask_denq_st(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE])
+{
+ uint32_t i, lc;
+ const uint32_t msk[2] = {
+ [0] = ROLE_DEQENQ,
+ [1] = RTE_GENMASK32(nb_stage + 1, 1),
+ };
+
+ memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE);
+
+ i = 0;
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ if (i == 0)
+ role_mask[lc] = msk[0];
+ else
+ role_mask[lc] = msk[1];
+ i++;
+ }
+ if (i == 1) {
+ lc = rte_get_next_lcore(-1, 1, 0);
+ role_mask[lc] |= msk[1];
+ }
+}
+
+
+static int
+test_sym_mt1(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+ const uint32_t nb_stage = 1;
+
+ role_mask_sym(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT,
+ nb_stage, role_mask);
+}
+
+static int
+test_sym_mt4(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 4;
+
+ role_mask_sym(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT,
+ nb_stage, role_mask);
+}
+
+static int
+test_sym_mt_rts4(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 4;
+
+ role_mask_sym(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT_RTS, RTE_RING_SYNC_MT_RTS,
+ nb_stage, role_mask);
+}
+
+static int
+test_sym_mt_hts4(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 4;
+
+ role_mask_sym(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT_HTS, RTE_RING_SYNC_MT_HTS,
+ nb_stage, role_mask);
+}
+
+static int
+test_stdenq_stage4(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 4;
+
+ role_mask_denq_st(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_ST, RTE_RING_SYNC_ST,
+ nb_stage, role_mask);
+}
+
+
+static int
+test_even_odd_mt5(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 5;
+
+ role_mask_even_odd(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT,
+ nb_stage, role_mask);
+}
+
+static int
+test_div_mt3(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 3;
+
+ role_mask_div(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT,
+ nb_stage, role_mask);
+}
+
+static const struct test_case tests[] = {
+ {
+ .name = "MT_DEQENQ-MT_STG1-PRCS",
+ .func = test_sym_mt1,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG1-AVG",
+ .func = test_sym_mt1,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "ST_DEQENQ-MT_STG4-PRCS",
+ .func = test_stdenq_stage4,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "ST_DEQENQ-MT_STG4-AVG",
+ .func = test_stdenq_stage4,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG4-PRCS",
+ .func = test_sym_mt4,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG4-AVG",
+ .func = test_sym_mt4,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MTRTS_DEQENQ-MT_STG4-PRCS",
+ .func = test_sym_mt_rts4,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MTRTS_DEQENQ-MT_STG4-AVG",
+ .func = test_sym_mt_rts4,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MTHTS_DEQENQ-MT_STG4-PRCS",
+ .func = test_sym_mt_hts4,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MTHTS_DEQENQ-MT_STG4-AVG",
+ .func = test_sym_mt_hts4,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG5-1:1-PRCS",
+ .func = test_even_odd_mt5,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG5-1:1-AVG",
+ .func = test_even_odd_mt5,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG3-1:3-PRCS",
+ .func = test_div_mt3,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MT_DEQENQ_MT_STG3-1:3-AVG",
+ .func = test_div_mt3,
+ .wfunc = test_worker_avg,
+ },
+};
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [RFCv2 6/6] examples/l3fwd: make ACL work in pipeline and eventdev modes
2024-09-06 13:13 ` [RFCv2 0/6] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
` (4 preceding siblings ...)
2024-09-06 13:13 ` [RFCv2 5/6] app/test: add unit tests for soring API Konstantin Ananyev
@ 2024-09-06 13:13 ` Konstantin Ananyev
2024-09-16 12:37 ` [PATCH v3 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
6 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-06 13:13 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, Konstantin Ananyev
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
Note upfront:
This is a huge commit that is combined from several ones.
For now, I submit it just for reference and demonstration purposes and
will probably remove it in future versions.
If will decide to go ahead with it, then it needs to be reworked and split
into several proper commits.
It adds for l3fwd:
- eventdev mode for ACL lookup-mode
- Introduce a worker-pool-mode
(right now implemented for ACL lookup-mode only).
Worker-Pool mode is a simple pipeline model, with the following stages:
1) I/O thread receives packets from NIC RX HW queues and enqueues them
into the work queue
2) Worker thread reads packets from the work queue(s),
process them and then puts processed packets back into the
work queue along with the processing status (routing info/error code).
3) I/O thread dequeues packets and their status from the work queue,
and based on it either TX packet or drops it.
Very similar to l3fwd-eventdev working model.
Note that it could be several I/O threads, each can serve one or multiple
HW RX queues. Also there could be several Worker threads, each of them can
process packets from multiple work queues in round-robin fashion.
Work queue can be one of the following types:
- wqorder: allows Worker threads to process packets in any order,
but guarantees that on dequeue stage the ingress order of packets
will be preserved. I.E. at stage #3, I/O thread will get packets
exactly in the same order as they were enqueued at stage #1.
- wqunorder: doesn't provide any ordered guarantees.
'wqunroder' mode is implemented using 2 rte_ring structures per queue.
'wqorder' mode is implemtened using rte_soring structure per queue.
To facilitate this new functionality, command line parameters were
extended:
--mode:
Possible values one of: poll/eventdev/wqorder/wqorderS/wqunorder/wqunorderS
Default value: poll
- wqorder: Worker-Pool ordered mode with a separate work queue for each
HW RX queue.
- wqorderS: Worker-Pool ordered mode with one work queue per I/O thread.
- wqunorder: Worker-Pool un-ordered mode with a separate work queue for each
HW RX queue.
- wqunorderS: Worker-Pool un-ordered mode with oen work queue per I/O thread.
--wqsize: number of elements for each worker queue.
--lookup-iter: forces to perform ACL lookup several times over the same
packet. This is artificial parameter and is added temporally for
benchmarking purposes. Will be removed in latest versions (if any).
Note that in Worker-Pool mode all free lcores that were not assigned as
I/O threads will be used as Worker threads.
As an example:
dpdk-l3fwd --lcores=53,55,57,59,61 ... -- \
-P -p f --config '(0,0,53)(1,0,53)(2,0,53)(3,0,53)' --lookup acl \
--parse-ptype --mode=wqorder ...
In that case lcore 53 will be used as I/O thread (stages #1,3)
to serve 4 HW RX queues,
while lcores 55,57,59,61 will serve as Worker threads (stage #2).
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
---
examples/l3fwd/l3fwd.h | 55 +++++++
examples/l3fwd/l3fwd_acl.c | 125 +++++++++++---
examples/l3fwd/l3fwd_acl_event.h | 258 +++++++++++++++++++++++++++++
examples/l3fwd/l3fwd_event.c | 14 ++
examples/l3fwd/l3fwd_event.h | 1 +
examples/l3fwd/l3fwd_sse.h | 49 +++++-
examples/l3fwd/l3fwd_wqp.c | 274 +++++++++++++++++++++++++++++++
examples/l3fwd/l3fwd_wqp.h | 130 +++++++++++++++
examples/l3fwd/main.c | 75 ++++++++-
examples/l3fwd/meson.build | 1 +
10 files changed, 954 insertions(+), 28 deletions(-)
create mode 100644 examples/l3fwd/l3fwd_acl_event.h
create mode 100644 examples/l3fwd/l3fwd_wqp.c
create mode 100644 examples/l3fwd/l3fwd_wqp.h
diff --git a/examples/l3fwd/l3fwd.h b/examples/l3fwd/l3fwd.h
index 93ce652d02..218f363764 100644
--- a/examples/l3fwd/l3fwd.h
+++ b/examples/l3fwd/l3fwd.h
@@ -77,6 +77,42 @@ struct __rte_cache_aligned lcore_rx_queue {
uint16_t queue_id;
};
+enum L3FWD_WORKER_MODE {
+ L3FWD_WORKER_POLL,
+ L3FWD_WORKER_UNQUE,
+ L3FWD_WORKER_ORQUE,
+};
+
+struct l3fwd_wqp_param {
+ enum L3FWD_WORKER_MODE mode;
+ uint32_t qsize; /**< Number of elems in worker queue */
+ int32_t single; /**< use single queue per I/O (poll) thread */
+};
+
+extern struct l3fwd_wqp_param l3fwd_wqp_param;
+
+enum {
+ LCORE_WQ_IN,
+ LCORE_WQ_OUT,
+ LCORE_WQ_NUM,
+};
+
+union lcore_wq {
+ struct rte_ring *r[LCORE_WQ_NUM];
+ struct {
+ struct rte_soring *sor;
+ /* used by WQ, sort of thred-local var */
+ uint32_t ftoken;
+ };
+};
+
+struct lcore_wq_pool {
+ uint32_t nb_queue;
+ uint32_t qmask;
+ union lcore_wq queue[MAX_RX_QUEUE_PER_LCORE];
+ struct l3fwd_wqp_param prm;
+};
+
struct __rte_cache_aligned lcore_conf {
uint16_t n_rx_queue;
struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
@@ -86,6 +122,7 @@ struct __rte_cache_aligned lcore_conf {
struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
void *ipv4_lookup_struct;
void *ipv6_lookup_struct;
+ struct lcore_wq_pool wqpool;
};
extern volatile bool force_quit;
@@ -115,6 +152,8 @@ extern struct acl_algorithms acl_alg[];
extern uint32_t max_pkt_len;
+extern uint32_t l3fwd_lookup_iter_num;
+
/* Send burst of packets on an output interface */
static inline int
send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
@@ -308,6 +347,22 @@ fib_event_main_loop_tx_q_vector(__rte_unused void *dummy);
int
fib_event_main_loop_tx_q_burst_vector(__rte_unused void *dummy);
+int
+acl_event_main_loop_tx_d(__rte_unused void *dummy);
+int
+acl_event_main_loop_tx_d_burst(__rte_unused void *dummy);
+int
+acl_event_main_loop_tx_q(__rte_unused void *dummy);
+int
+acl_event_main_loop_tx_q_burst(__rte_unused void *dummy);
+int
+acl_event_main_loop_tx_d_vector(__rte_unused void *dummy);
+int
+acl_event_main_loop_tx_d_burst_vector(__rte_unused void *dummy);
+int
+acl_event_main_loop_tx_q_vector(__rte_unused void *dummy);
+int
+acl_event_main_loop_tx_q_burst_vector(__rte_unused void *dummy);
/* Return ipv4/ipv6 fwd lookup struct for ACL, LPM, EM or FIB. */
void *
diff --git a/examples/l3fwd/l3fwd_acl.c b/examples/l3fwd/l3fwd_acl.c
index b635011ef7..8b550e495a 100644
--- a/examples/l3fwd/l3fwd_acl.c
+++ b/examples/l3fwd/l3fwd_acl.c
@@ -4,6 +4,7 @@
#include "l3fwd.h"
#include "l3fwd_route.h"
+#include "l3fwd_wqp.h"
/*
* Rule and trace formats definitions.
@@ -1003,19 +1004,21 @@ acl_process_pkts(struct rte_mbuf *pkts[MAX_PKT_BURST],
/* split packets burst depending on packet type (IPv4/IPv6) */
l3fwd_acl_prepare_acl_parameter(pkts, &acl_search, num);
- if (acl_search.num_ipv4)
- rte_acl_classify(acl_config.acx_ipv4[socketid],
+ for (i = l3fwd_lookup_iter_num; i != 0; i--) {
+ if (acl_search.num_ipv4)
+ rte_acl_classify(acl_config.acx_ipv4[socketid],
acl_search.data_ipv4,
acl_search.res_ipv4,
acl_search.num_ipv4,
DEFAULT_MAX_CATEGORIES);
- if (acl_search.num_ipv6)
- rte_acl_classify(acl_config.acx_ipv6[socketid],
+ if (acl_search.num_ipv6)
+ rte_acl_classify(acl_config.acx_ipv6[socketid],
acl_search.data_ipv6,
acl_search.res_ipv6,
acl_search.num_ipv6,
DEFAULT_MAX_CATEGORIES);
+ }
/* combine lookup results back, into one array of next hops */
n4 = 0;
@@ -1042,34 +1045,36 @@ acl_process_pkts(struct rte_mbuf *pkts[MAX_PKT_BURST],
static inline void
acl_send_packets(struct lcore_conf *qconf, struct rte_mbuf *pkts[],
- uint16_t hops[], uint32_t num)
+ uint16_t hops[], uint32_t num, int step3)
{
#if defined ACL_SEND_MULTI
- send_packets_multi(qconf, pkts, hops, num);
+ __send_packets_multi(qconf, pkts, hops, num, step3);
#else
- send_packets_single(qconf, pkts, hops, num);
+ if (step3 != 0)
+ send_packets_single(qconf, pkts, hops, num);
+ else {
+ uint32_t i;
+ for (i = 0; i != num; i++)
+ send_single_packet(qconf, pkts[i], hops[i]);
+ }
#endif
}
/* main processing loop */
-int
-acl_main_loop(__rte_unused void *dummy)
+static int
+acl_poll_loop(struct lcore_conf *qconf, uint32_t lcore_id)
{
struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
uint16_t hops[MAX_PKT_BURST];
- unsigned int lcore_id;
uint64_t prev_tsc, diff_tsc, cur_tsc;
- int i, nb_rx;
+ uint32_t i, n, nb_rx;
uint16_t portid;
uint16_t queueid;
- struct lcore_conf *qconf;
int socketid;
const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
/ US_PER_S * BURST_TX_DRAIN_US;
prev_tsc = 0;
- lcore_id = rte_lcore_id();
- qconf = &lcore_conf[lcore_id];
socketid = rte_lcore_to_socket_id(lcore_id);
if (qconf->n_rx_queue == 0) {
@@ -1121,17 +1126,99 @@ acl_main_loop(__rte_unused void *dummy)
nb_rx = rte_eth_rx_burst(portid, queueid,
pkts_burst, MAX_PKT_BURST);
- if (nb_rx > 0) {
- acl_process_pkts(pkts_burst, hops, nb_rx,
- socketid);
- acl_send_packets(qconf, pkts_burst, hops,
- nb_rx);
+ if (nb_rx != 0) {
+ if (l3fwd_wqp_param.mode == L3FWD_WORKER_POLL) {
+ acl_process_pkts(pkts_burst, hops,
+ nb_rx, socketid);
+ acl_send_packets(qconf, pkts_burst,
+ hops, nb_rx, 1);
+ } else {
+ n = lcore_wq_submit(&qconf->wqpool, i,
+ pkts_burst, nb_rx);
+ if (n != nb_rx) {
+ /* update stats counter */
+ rte_pktmbuf_free_bulk(
+ pkts_burst + n,
+ nb_rx - n);
+ }
+ }
+ }
+ if (l3fwd_wqp_param.mode != L3FWD_WORKER_POLL) {
+ nb_rx = lcore_wq_receive(&qconf->wqpool, i,
+ pkts_burst, hops, MAX_PKT_BURST);
+ if (nb_rx != 0)
+ acl_send_packets(qconf, pkts_burst,
+ hops, nb_rx, 0);
+ }
+ }
+ }
+ return 0;
+}
+
+/* WT processing loop */
+static int
+acl_wqp_loop(struct lcore_conf *qconf, uint32_t lcore_id)
+{
+ int32_t socketid;
+ uint32_t i, k, n;
+ struct rte_mbuf *pkts[MAX_PKT_BURST];
+ uint16_t hops[MAX_PKT_BURST];
+
+ socketid = rte_lcore_to_socket_id(lcore_id);
+
+ if (qconf->wqpool.nb_queue == 0) {
+ RTE_LOG(INFO, L3FWD, "%s: lcore %u has nothing to do\n",
+ __func__, lcore_id);
+ return 0;
+ }
+
+ RTE_LOG(INFO, L3FWD, "%s: entering loop on lcore %u\n",
+ __func__, lcore_id);
+
+ while (!force_quit) {
+
+ /*
+ * Read packet from internal queues and process them
+ */
+ for (i = 0; i < qconf->wqpool.nb_queue; ++i) {
+
+ n = lcore_wq_pull(&qconf->wqpool, i, pkts,
+ RTE_DIM(pkts));
+ if (n == 0)
+ continue;
+
+ acl_process_pkts(pkts, hops, n, socketid);
+ process_step3_burst(pkts, hops, n);
+ k = lcore_wq_push(&qconf->wqpool, i, pkts, hops, n);
+ if (n != k) {
+ /* stats update */
+ rte_pktmbuf_free_bulk(pkts + k, n - k);
}
}
}
return 0;
}
+/* main processing loop */
+int
+acl_main_loop(__rte_unused void *dummy)
+{
+ uint32_t lcore_id;
+ struct lcore_conf *qconf;
+
+ lcore_id = rte_lcore_id();
+ qconf = &lcore_conf[lcore_id];
+
+ if (qconf->n_rx_queue != 0)
+ return acl_poll_loop(qconf, lcore_id);
+ else
+ return acl_wqp_loop(qconf, lcore_id);
+}
+
+#ifdef RTE_LIB_EVENTDEV
+#include "l3fwd_acl_event.h"
+#endif
+
/* Not used by L3fwd ACL. */
void *
acl_get_ipv4_l3fwd_lookup_struct(__rte_unused const int socketid)
diff --git a/examples/l3fwd/l3fwd_acl_event.h b/examples/l3fwd/l3fwd_acl_event.h
new file mode 100644
index 0000000000..240dd3fb03
--- /dev/null
+++ b/examples/l3fwd/l3fwd_acl_event.h
@@ -0,0 +1,258 @@
+#include "l3fwd_event.h"
+
+/* One eventdev loop for single and burst using acl. */
+static __rte_always_inline void
+acl_event_loop(struct l3fwd_event_resources *evt_rsrc,
+ const uint8_t flags)
+{
+ uint32_t i, lcore_id, nb_deq, nb_enq;
+ int32_t socketid;
+ uint16_t hops[MAX_PKT_BURST];
+ struct rte_mbuf *pkts[MAX_PKT_BURST];
+ struct rte_event events[MAX_PKT_BURST];
+
+ const int event_p_id = l3fwd_get_free_event_port(evt_rsrc);
+ const uint8_t tx_q_id = evt_rsrc->evq.event_q_id[
+ evt_rsrc->evq.nb_queues - 1];
+ const uint8_t event_d_id = evt_rsrc->event_d_id;
+ const uint16_t deq_len = RTE_MIN(evt_rsrc->deq_depth, MAX_PKT_BURST);
+
+ if (event_p_id < 0)
+ return;
+
+ lcore_id = rte_lcore_id();
+ socketid = rte_lcore_to_socket_id(lcore_id);
+
+ RTE_LOG(INFO, L3FWD, "entering %s on lcore %u\n", __func__, lcore_id);
+
+ nb_deq = 0;
+ nb_enq = 0;
+
+ while (!force_quit) {
+ /* Read events from RX queues. */
+ nb_deq = rte_event_dequeue_burst(event_d_id, event_p_id,
+ events, deq_len, 0);
+ if (nb_deq == 0) {
+ rte_pause();
+ continue;
+ }
+
+ for (i = 0; i != nb_deq; i++) {
+ pkts[i] = events[i].mbuf;
+ if (flags & L3FWD_EVENT_TX_ENQ) {
+ events[i].queue_id = tx_q_id;
+ events[i].op = RTE_EVENT_OP_FORWARD;
+ }
+ rte_event_eth_tx_adapter_txq_set(pkts[i], 0);
+ }
+
+ acl_process_pkts(pkts, hops, nb_deq, socketid);
+
+ for (i = 0; i != nb_deq; i++) {
+ process_packet(pkts[i], &hops[i]);
+ pkts[i]->port = (hops[i] != BAD_PORT) ?
+ hops[i] : pkts[i]->port;
+ }
+
+ if (flags & L3FWD_EVENT_TX_ENQ) {
+ nb_enq = rte_event_enqueue_burst(event_d_id, event_p_id,
+ events, nb_deq);
+ while (nb_enq < nb_deq && !force_quit)
+ nb_enq += rte_event_enqueue_burst(event_d_id,
+ event_p_id, events + nb_enq,
+ nb_deq - nb_enq);
+ }
+
+ if (flags & L3FWD_EVENT_TX_DIRECT) {
+ nb_enq = rte_event_eth_tx_adapter_enqueue(event_d_id,
+ event_p_id, events, nb_deq, 0);
+ while (nb_enq < nb_deq && !force_quit)
+ nb_enq += rte_event_eth_tx_adapter_enqueue(
+ event_d_id, event_p_id,
+ events + nb_enq,
+ nb_deq - nb_enq, 0);
+ }
+ }
+
+ l3fwd_event_worker_cleanup(event_d_id, event_p_id, events, nb_enq,
+ nb_deq, 0);
+}
+
+int __rte_noinline
+acl_event_main_loop_tx_d(__rte_unused void *dummy)
+{
+ struct l3fwd_event_resources *evt_rsrc =
+ l3fwd_get_eventdev_rsrc();
+
+ acl_event_loop(evt_rsrc, L3FWD_EVENT_TX_DIRECT);
+ return 0;
+}
+
+int __rte_noinline
+acl_event_main_loop_tx_d_burst(__rte_unused void *dummy)
+{
+ struct l3fwd_event_resources *evt_rsrc =
+ l3fwd_get_eventdev_rsrc();
+
+ acl_event_loop(evt_rsrc, L3FWD_EVENT_TX_DIRECT);
+ return 0;
+}
+
+int __rte_noinline
+acl_event_main_loop_tx_q(__rte_unused void *dummy)
+{
+ struct l3fwd_event_resources *evt_rsrc =
+ l3fwd_get_eventdev_rsrc();
+
+ acl_event_loop(evt_rsrc, L3FWD_EVENT_TX_ENQ);
+ return 0;
+}
+
+int __rte_noinline
+acl_event_main_loop_tx_q_burst(__rte_unused void *dummy)
+{
+ struct l3fwd_event_resources *evt_rsrc =
+ l3fwd_get_eventdev_rsrc();
+
+ acl_event_loop(evt_rsrc, L3FWD_EVENT_TX_ENQ);
+ return 0;
+}
+
+static __rte_always_inline void
+acl_process_event_vector(struct rte_event_vector *vec, uint16_t *hops,
+ int32_t socketid)
+{
+ uint32_t i, k;
+
+ for (i = 0; i != vec->nb_elem; i += k) {
+ k = RTE_MIN(vec->nb_elem - i, (uint32_t)MAX_PKT_BURST);
+ acl_process_pkts(vec->mbufs + i, hops + i, k, socketid);
+ }
+
+#if defined ACL_SEND_MULTI
+ k = RTE_ALIGN_FLOOR(vec->nb_elem, FWDSTEP);
+
+ for (i = 0; i != k; i += FWDSTEP)
+ processx4_step3(&vec->mbufs[i], &hops[i]);
+ for (; i < vec->nb_elem; i++)
+ process_packet(vec->mbufs[i], &hops[i]);
+#else
+ for (i = 0; i < vec->nb_elem; i++)
+ process_packet(vec->mbufs[i], &hops[i]);
+#endif
+
+ process_event_vector(vec, hops);
+}
+
+static __rte_always_inline void
+acl_event_loop_vector(struct l3fwd_event_resources *evt_rsrc,
+ const uint8_t flags)
+{
+ uint16_t *hops;
+ int32_t socketid;
+ uint32_t i, lcore_id, nb_deq, nb_enq;
+ struct rte_event events[MAX_PKT_BURST];
+
+ const int event_p_id = l3fwd_get_free_event_port(evt_rsrc);
+ const uint8_t tx_q_id =
+ evt_rsrc->evq.event_q_id[evt_rsrc->evq.nb_queues - 1];
+ const uint8_t event_d_id = evt_rsrc->event_d_id;
+ const uint16_t deq_len = evt_rsrc->deq_depth;
+
+ if (event_p_id < 0)
+ return;
+
+ lcore_id = rte_lcore_id();
+ socketid = rte_lcore_to_socket_id(lcore_id);
+
+ hops = rte_zmalloc_socket(NULL, sizeof(hops[0]) * evt_rsrc->vector_size,
+ RTE_CACHE_LINE_SIZE, socketid);
+ if (hops == NULL) {
+ RTE_LOG(ERR, L3FWD,
+ "%s: failed to alloc internal buffers on lcore %u\n",
+ __func__, lcore_id);
+ return;
+ }
+
+ RTE_LOG(INFO, L3FWD, "entering %s on lcore %u\n", __func__, lcore_id);
+
+ nb_deq = 0;
+ nb_enq = 0;
+
+ while (!force_quit) {
+ /* Read events from RX queues. */
+ nb_deq = rte_event_dequeue_burst(event_d_id, event_p_id, events,
+ deq_len, 0);
+ if (nb_deq == 0) {
+ rte_pause();
+ continue;
+ }
+
+ for (i = 0; i < nb_deq; i++) {
+ if (flags & L3FWD_EVENT_TX_ENQ) {
+ events[i].queue_id = tx_q_id;
+ events[i].op = RTE_EVENT_OP_FORWARD;
+ }
+
+ acl_process_event_vector(events[i].vec, hops, socketid);
+ }
+
+ if (flags & L3FWD_EVENT_TX_ENQ) {
+ nb_enq = rte_event_enqueue_burst(event_d_id, event_p_id,
+ events, nb_deq);
+ while (nb_enq < nb_deq && !force_quit)
+ nb_enq += rte_event_enqueue_burst(
+ event_d_id, event_p_id, events + nb_enq,
+ nb_deq - nb_enq);
+ }
+
+ if (flags & L3FWD_EVENT_TX_DIRECT) {
+ nb_enq = rte_event_eth_tx_adapter_enqueue(
+ event_d_id, event_p_id, events, nb_deq, 0);
+ while (nb_enq < nb_deq && !force_quit)
+ nb_enq += rte_event_eth_tx_adapter_enqueue(
+ event_d_id, event_p_id, events + nb_enq,
+ nb_deq - nb_enq, 0);
+ }
+ }
+
+ l3fwd_event_worker_cleanup(event_d_id, event_p_id, events, nb_enq,
+ nb_deq, 1);
+ rte_free(hops);
+}
+
+int __rte_noinline
+acl_event_main_loop_tx_d_vector(__rte_unused void *dummy)
+{
+ struct l3fwd_event_resources *evt_rsrc = l3fwd_get_eventdev_rsrc();
+
+ acl_event_loop_vector(evt_rsrc, L3FWD_EVENT_TX_DIRECT);
+ return 0;
+}
+
+int __rte_noinline
+acl_event_main_loop_tx_d_burst_vector(__rte_unused void *dummy)
+{
+ struct l3fwd_event_resources *evt_rsrc = l3fwd_get_eventdev_rsrc();
+
+ acl_event_loop_vector(evt_rsrc, L3FWD_EVENT_TX_DIRECT);
+ return 0;
+}
+
+int __rte_noinline
+acl_event_main_loop_tx_q_vector(__rte_unused void *dummy)
+{
+ struct l3fwd_event_resources *evt_rsrc = l3fwd_get_eventdev_rsrc();
+
+ acl_event_loop_vector(evt_rsrc, L3FWD_EVENT_TX_ENQ);
+ return 0;
+}
+
+int __rte_noinline
+acl_event_main_loop_tx_q_burst_vector(__rte_unused void *dummy)
+{
+ struct l3fwd_event_resources *evt_rsrc = l3fwd_get_eventdev_rsrc();
+
+ acl_event_loop_vector(evt_rsrc, L3FWD_EVENT_TX_ENQ);
+ return 0;
+}
diff --git a/examples/l3fwd/l3fwd_event.c b/examples/l3fwd/l3fwd_event.c
index 20be22c6db..18ddd4b8b8 100644
--- a/examples/l3fwd/l3fwd_event.c
+++ b/examples/l3fwd/l3fwd_event.c
@@ -250,6 +250,16 @@ l3fwd_event_resource_setup(struct rte_eth_conf *port_conf)
[1][1][0] = fib_event_main_loop_tx_q_vector,
[1][1][1] = fib_event_main_loop_tx_q_burst_vector,
};
+ const event_loop_cb acl_event_loop[2][2][2] = {
+ [0][0][0] = acl_event_main_loop_tx_d,
+ [0][0][1] = acl_event_main_loop_tx_d_burst,
+ [0][1][0] = acl_event_main_loop_tx_q,
+ [0][1][1] = acl_event_main_loop_tx_q_burst,
+ [1][0][0] = acl_event_main_loop_tx_d_vector,
+ [1][0][1] = acl_event_main_loop_tx_d_burst_vector,
+ [1][1][0] = acl_event_main_loop_tx_q_vector,
+ [1][1][1] = acl_event_main_loop_tx_q_burst_vector,
+ };
uint32_t event_queue_cfg;
int ret;
@@ -293,6 +303,10 @@ l3fwd_event_resource_setup(struct rte_eth_conf *port_conf)
evt_rsrc->ops.fib_event_loop =
fib_event_loop[evt_rsrc->vector_enabled][evt_rsrc->tx_mode_q]
[evt_rsrc->has_burst];
+
+ evt_rsrc->ops.acl_event_loop =
+ acl_event_loop[evt_rsrc->vector_enabled][evt_rsrc->tx_mode_q]
+ [evt_rsrc->has_burst];
}
static void
diff --git a/examples/l3fwd/l3fwd_event.h b/examples/l3fwd/l3fwd_event.h
index c6a4a89127..21aa24c396 100644
--- a/examples/l3fwd/l3fwd_event.h
+++ b/examples/l3fwd/l3fwd_event.h
@@ -58,6 +58,7 @@ struct l3fwd_event_setup_ops {
event_loop_cb lpm_event_loop;
event_loop_cb em_event_loop;
event_loop_cb fib_event_loop;
+ event_loop_cb acl_event_loop;
};
struct l3fwd_event_resources {
diff --git a/examples/l3fwd/l3fwd_sse.h b/examples/l3fwd/l3fwd_sse.h
index 083729cdef..29c5c7c014 100644
--- a/examples/l3fwd/l3fwd_sse.h
+++ b/examples/l3fwd/l3fwd_sse.h
@@ -86,12 +86,35 @@ process_packet(struct rte_mbuf *pkt, uint16_t *dst_port)
_mm_storeu_si128((__m128i *)eth_hdr, te);
}
+static inline void
+process_step3_burst(struct rte_mbuf *pkt[], uint16_t dst_port[], uint32_t num)
+{
+ uint32_t i, k;
+
+ k = RTE_ALIGN_FLOOR(num, FWDSTEP);
+
+ for (i = 0; i != k; i += FWDSTEP)
+ processx4_step3(pkt + i, dst_port + i);
+
+ /* Process up to last 3 packets one by one. */
+ switch (num % FWDSTEP) {
+ case 3:
+ process_packet(pkt[i + 2], dst_port + i + 2);
+ /* fall-through */
+ case 2:
+ process_packet(pkt[i + 1], dst_port + i + 1);
+ /* fall-through */
+ case 1:
+ process_packet(pkt[i], dst_port + i);
+ }
+}
+
/**
* Send packets burst from pkts_burst to the ports in dst_port array
*/
static __rte_always_inline void
-send_packets_multi(struct lcore_conf *qconf, struct rte_mbuf **pkts_burst,
- uint16_t dst_port[MAX_PKT_BURST], int nb_rx)
+__send_packets_multi(struct lcore_conf *qconf, struct rte_mbuf **pkts_burst,
+ uint16_t dst_port[MAX_PKT_BURST], int nb_rx, int step3)
{
int32_t k;
int j = 0;
@@ -110,13 +133,15 @@ send_packets_multi(struct lcore_conf *qconf, struct rte_mbuf **pkts_burst,
lp = pnum;
lp[0] = 1;
- processx4_step3(pkts_burst, dst_port);
+ if (step3 != 0)
+ processx4_step3(pkts_burst, dst_port);
/* dp1: <d[0], d[1], d[2], d[3], ... > */
dp1 = _mm_loadu_si128((__m128i *)dst_port);
for (j = FWDSTEP; j != k; j += FWDSTEP) {
- processx4_step3(&pkts_burst[j], &dst_port[j]);
+ if (step3 != 0)
+ processx4_step3(&pkts_burst[j], &dst_port[j]);
/*
* dp2:
@@ -155,17 +180,20 @@ send_packets_multi(struct lcore_conf *qconf, struct rte_mbuf **pkts_burst,
/* Process up to last 3 packets one by one. */
switch (nb_rx % FWDSTEP) {
case 3:
- process_packet(pkts_burst[j], dst_port + j);
+ if (step3 != 0)
+ process_packet(pkts_burst[j], dst_port + j);
GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
j++;
/* fall-through */
case 2:
- process_packet(pkts_burst[j], dst_port + j);
+ if (step3 != 0)
+ process_packet(pkts_burst[j], dst_port + j);
GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
j++;
/* fall-through */
case 1:
- process_packet(pkts_burst[j], dst_port + j);
+ if (step3 != 0)
+ process_packet(pkts_burst[j], dst_port + j);
GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
j++;
}
@@ -194,6 +222,13 @@ send_packets_multi(struct lcore_conf *qconf, struct rte_mbuf **pkts_burst,
}
}
+static __rte_always_inline void
+send_packets_multi(struct lcore_conf *qconf, struct rte_mbuf **pkts_burst,
+ uint16_t dst_port[MAX_PKT_BURST], int nb_rx)
+{
+ __send_packets_multi(qconf, pkts_burst, dst_port, nb_rx, 1);
+}
+
static __rte_always_inline uint16_t
process_dst_port(uint16_t *dst_ports, uint16_t nb_elem)
{
diff --git a/examples/l3fwd/l3fwd_wqp.c b/examples/l3fwd/l3fwd_wqp.c
new file mode 100644
index 0000000000..74f745d1af
--- /dev/null
+++ b/examples/l3fwd/l3fwd_wqp.c
@@ -0,0 +1,274 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#include "l3fwd.h"
+#include "l3fwd_wqp.h"
+
+static int
+wqp_ring_init(struct rte_ring **r, uint32_t num, size_t sz, int32_t sid,
+ uint32_t flags)
+{
+ char name[RTE_RING_NAMESIZE];
+
+ *r = rte_zmalloc_socket(NULL, sz, RTE_CACHE_LINE_SIZE, sid);
+ if (*r == NULL)
+ return -ENOMEM;
+
+ snprintf(name, sizeof(name), "%p", *r);
+ return rte_ring_init(*r, name, num, flags);
+}
+
+static int
+wqp_soring_init(struct rte_soring **r, struct rte_soring_param *prm, size_t sz,
+ int32_t sid)
+{
+ char name[RTE_RING_NAMESIZE];
+
+ *r = rte_zmalloc_socket(NULL, sz, RTE_CACHE_LINE_SIZE, sid);
+ if (*r == NULL)
+ return -ENOMEM;
+
+ snprintf(name, sizeof(name), "%p", *r);
+ prm->name = name;
+ return rte_soring_init(*r, prm);
+}
+
+static void
+wqp_fini(struct lcore_conf *lc)
+{
+ uint32_t i, j;
+
+ if (lc->n_rx_queue == 0)
+ return;
+
+ for (i = 0; i != lc->wqpool.nb_queue; i++) {
+ for (j = 0; j != RTE_DIM(lc->wqpool.queue[i].r); j++)
+ rte_free(lc->wqpool.queue[i].r[j]);
+ }
+
+ memset(&lc->wqpool, 0, sizeof(lc->wqpool));
+}
+
+static int
+l3fwd_wqp_unque_init(struct lcore_conf *lc, const struct l3fwd_wqp_param *prm,
+ uint32_t lcid)
+{
+ int32_t rc, sid;
+ uint32_t i, n, nq;
+ size_t szi, szo;
+
+ sid = rte_lcore_to_socket_id(lcid);
+
+ n = rte_align32pow2(prm->qsize);
+
+ szi = rte_ring_get_memsize(n);
+ szo = rte_ring_get_memsize_elem(sizeof(struct wqelm), n);
+
+ nq = (prm->single == 0) ? lc->n_rx_queue : 1;
+ lc->wqpool.nb_queue = nq;
+ lc->wqpool.qmask = (prm->single != 0) ? 0 : UINT32_MAX;
+
+ rc = 0;
+ for (i = 0; i != nq; i++) {
+
+ rc = wqp_ring_init(&lc->wqpool.queue[i].r[LCORE_WQ_IN], n, szi,
+ sid, RING_F_SP_ENQ);
+ if (rc != 0)
+ break;
+
+ rc = wqp_ring_init(&lc->wqpool.queue[i].r[LCORE_WQ_OUT], n, szo,
+ sid, RING_F_SC_DEQ);
+ if (rc != 0)
+ break;
+ }
+
+ if (i != nq) {
+ printf("error: %s failed at %u-th queue, error code: %d\n",
+ __func__, i, rc);
+ wqp_fini(lc);
+ }
+
+ lc->wqpool.prm = *prm;
+ return rc;
+}
+
+static int
+l3fwd_wqp_orque_init(struct lcore_conf *lc, const struct l3fwd_wqp_param *qprm,
+ uint32_t lcid)
+{
+ int32_t rc, sid;
+ uint32_t i, n, nq;
+ ssize_t sz;
+ struct rte_soring_param prm;
+
+ sid = rte_lcore_to_socket_id(lcid);
+
+ memset(&prm, 0, sizeof(prm));
+
+ n = 2 * qprm->qsize;
+ prm.elems = n;
+ prm.elem_size = sizeof(uintptr_t);
+ prm.meta_size = sizeof(uint32_t);
+ prm.stages = 1;
+ prm.prod_synt = RTE_RING_SYNC_ST;
+ prm.cons_synt = RTE_RING_SYNC_ST;
+
+ sz = rte_soring_get_memsize(&prm);
+ if (sz < 0)
+ return sz;
+
+ nq = (qprm->single == 0) ? lc->n_rx_queue : 1;
+ lc->wqpool.nb_queue = nq;
+ lc->wqpool.qmask = (qprm->single != 0) ? 0 : UINT32_MAX;
+
+ rc = 0;
+ for (i = 0; i != nq; i++) {
+
+ rc = wqp_soring_init(&lc->wqpool.queue[i].sor, &prm, sz, sid);
+ if (rc != 0)
+ break;
+ }
+
+ if (i != nq) {
+ printf("error: %s failed at %u-th queue, error code: %d\n",
+ __func__, i, rc);
+ wqp_fini(lc);
+ }
+
+ lc->wqpool.prm = *qprm;
+ return rc;
+}
+
+static int
+wqp_init(struct lcore_conf *lc, const struct l3fwd_wqp_param *prm,
+ uint32_t lcid)
+{
+ /* this is I/O poll lcore */
+ if (lc->n_rx_queue != 0) {
+ if (prm->mode == L3FWD_WORKER_UNQUE)
+ return l3fwd_wqp_unque_init(lc, prm, lcid);
+ else if (prm->mode == L3FWD_WORKER_ORQUE)
+ return l3fwd_wqp_orque_init(lc, prm, lcid);
+ else
+ return -ENOTSUP;
+ }
+
+ return -ENOTSUP;
+}
+
+void
+l3fwd_wqp_fini(struct lcore_conf lc[RTE_MAX_LCORE])
+{
+ uint32_t lcid;
+
+ for (lcid = 0; lcid != RTE_MAX_LCORE; lcid++)
+ wqp_fini(lc + lcid);
+}
+
+static int
+check_set_wqp_param(struct l3fwd_wqp_param *prm)
+{
+ uint32_t n;
+
+ if (prm->qsize == 0) {
+ n = RTE_MAX(nb_rxd, nb_txd);
+ n = n + n / 2;
+ prm->qsize = n;
+ }
+
+ return 0;
+}
+
+static void print_wqp_param(const struct l3fwd_wqp_param *prm)
+{
+ printf("%s(%p): mode=%d, qsize=%u, single=%d\n",
+ __func__, prm, prm->mode, prm->qsize, prm->single);
+}
+
+int
+l3fwd_wqp_init(struct lcore_conf lc[RTE_MAX_LCORE],
+ const struct l3fwd_wqp_param *qprm)
+{
+ int32_t rc;
+ uint32_t i, j, k, lcid, m, n, nrxq, nwqt;
+ union lcore_wq *wqp;
+ struct l3fwd_wqp_param prm;
+
+ if (qprm->mode == L3FWD_WORKER_POLL)
+ return 0;
+
+ prm = *qprm;
+ rc = check_set_wqp_param(&prm);
+ print_wqp_param(&prm);
+ if (rc < 0) {
+ printf("error: %s invalid paramer values\n", __func__);
+ return rc;
+ }
+
+ nrxq = 0;
+ nwqt = 0;
+ for (lcid = 0; lcid != RTE_MAX_LCORE; lcid++) {
+ if (rte_lcore_is_enabled(lcid) == 0)
+ continue;
+ if (lc[lcid].n_rx_queue != 0)
+ nrxq += (prm.single != 0) ? 1 : lc[lcid].n_rx_queue;
+ nwqt += (lc[lcid].n_rx_queue == 0);
+ }
+
+ printf("%s: total worker queues: %u, total WQ threads: %u\n",
+ __func__, nrxq, nwqt);
+ if (nrxq == 0)
+ return 0;
+
+ if (nrxq > nwqt * MAX_RX_QUEUE_PER_LCORE) {
+ printf("error: %s not enough WQ threads to handle all RXQs\n",
+ __func__);
+ return -EINVAL;
+ }
+
+ for (lcid = 0; lcid != RTE_MAX_LCORE; lcid++) {
+ if (rte_lcore_is_enabled(lcid) == 0 || lc[lcid].n_rx_queue == 0)
+ continue;
+ rc = wqp_init(lc + lcid, &prm, lcid);
+ if (rc != 0)
+ break;
+ }
+ if (rc != 0)
+ return rc;
+
+ /* create a temp pool of all RX queues */
+ wqp = malloc(sizeof(wqp[0]) * nrxq);
+ if (wqp == NULL) {
+ l3fwd_wqp_fini(lc);
+ return -ENOMEM;
+ }
+
+ n = 0;
+ for (lcid = 0; lcid != RTE_MAX_LCORE; lcid++) {
+ memcpy(wqp + n, lc[lcid].wqpool.queue,
+ lc[lcid].wqpool.nb_queue * sizeof(wqp[0]));
+ n += lc[lcid].wqpool.nb_queue;
+ }
+
+ /* distribute them across all worker threads */
+ k = 0;
+ m = RTE_MIN(RTE_DIM(lc[lcid].wqpool.queue), n);
+ for (lcid = 0; lcid != RTE_MAX_LCORE; lcid++) {
+ if (rte_lcore_is_enabled(lcid) == 0 ||
+ lc[lcid].wqpool.nb_queue != 0)
+ continue;
+ j = k;
+ for (i = 0; i != m; i++) {
+ lc[lcid].wqpool.queue[i] = wqp[j];
+ j = (j + 1) % n;
+ }
+ lc[lcid].wqpool.nb_queue = i;
+ lc[lcid].wqpool.qmask = UINT32_MAX;
+ lc[lcid].wqpool.prm = prm;
+ k = j;
+ }
+
+ free(wqp);
+ return rc;
+}
diff --git a/examples/l3fwd/l3fwd_wqp.h b/examples/l3fwd/l3fwd_wqp.h
new file mode 100644
index 0000000000..bd6ce88657
--- /dev/null
+++ b/examples/l3fwd/l3fwd_wqp.h
@@ -0,0 +1,130 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#ifndef L3FWD_WQP_H
+#define L3FWD_WQP_H
+
+#include <rte_soring.h>
+
+struct wqelm {
+ struct rte_mbuf *mb;
+ uint32_t rc;
+} __rte_packed;
+
+
+void l3fwd_wqp_fini(struct lcore_conf lc[RTE_MAX_LCORE]);
+int l3fwd_wqp_init(struct lcore_conf lc[RTE_MAX_LCORE],
+ const struct l3fwd_wqp_param *prm);
+
+static inline uint32_t
+lcore_wq_submit(const struct lcore_wq_pool *wqp, uint32_t idx,
+ struct rte_mbuf * const pkts[MAX_PKT_BURST], uint32_t num)
+{
+ idx &= wqp->qmask;
+
+ if (wqp->prm.mode == L3FWD_WORKER_UNQUE) {
+ struct rte_ring *r = wqp->queue[idx].r[LCORE_WQ_IN];
+ return rte_ring_enqueue_burst(r, (void * const *)pkts, num,
+ NULL);
+ } else if (wqp->prm.mode == L3FWD_WORKER_ORQUE) {
+ struct rte_soring *sor = wqp->queue[idx].sor;
+ return rte_soring_enqueue_burst(sor, pkts, num, NULL);
+ }
+
+ rte_errno = ENOTSUP;
+ return 0;
+}
+
+static inline uint32_t
+lcore_wq_receive(const struct lcore_wq_pool *wqp, uint32_t idx,
+ struct rte_mbuf *pkts[MAX_PKT_BURST], uint16_t hops[MAX_PKT_BURST],
+ uint32_t num)
+{
+ uint32_t i, n;
+ uint32_t rcs[MAX_PKT_BURST];
+ struct wqelm elm[MAX_PKT_BURST];
+
+ idx &= wqp->qmask;
+
+ if (wqp->prm.mode == L3FWD_WORKER_UNQUE) {
+ struct rte_ring *r = wqp->queue[idx].r[LCORE_WQ_OUT];
+
+ n = rte_ring_dequeue_burst_elem(r, elm, sizeof(elm[0]), num,
+ NULL);
+ for (i = 0; i != n; i++) {
+ pkts[i] = elm[i].mb;
+ hops[i] = elm[i].rc;
+ }
+
+ return n;
+
+ } else if (wqp->prm.mode == L3FWD_WORKER_ORQUE) {
+ struct rte_soring *sor = wqp->queue[idx].sor;
+
+ n = rte_soring_dequeux_burst(sor, pkts, rcs, num, NULL);
+ for (i = 0; i != n; i++)
+ hops[i] = rcs[i];
+
+ return n;
+ }
+
+ rte_errno = ENOTSUP;
+ return 0;
+}
+
+static inline uint32_t
+lcore_wq_pull(struct lcore_wq_pool *wqp, uint32_t idx,
+ struct rte_mbuf *pkts[MAX_PKT_BURST], uint32_t num)
+{
+ idx &= wqp->qmask;
+
+ if (wqp->prm.mode == L3FWD_WORKER_UNQUE) {
+ struct rte_ring *r = wqp->queue[idx].r[LCORE_WQ_IN];
+ return rte_ring_dequeue_burst(r, (void **)pkts, num, NULL);
+
+ } else if (wqp->prm.mode == L3FWD_WORKER_ORQUE) {
+ struct rte_soring *sor = wqp->queue[idx].sor;
+ return rte_soring_acquire_burst(sor, pkts, 0, num,
+ &wqp->queue[idx].ftoken, NULL);
+ }
+
+ rte_errno = ENOTSUP;
+ return 0;
+}
+
+static inline uint32_t
+lcore_wq_push(const struct lcore_wq_pool *wqp, uint32_t idx,
+ struct rte_mbuf * const pkts[MAX_PKT_BURST],
+ const uint16_t hops[MAX_PKT_BURST], uint32_t num)
+{
+ uint32_t i;
+ uint32_t rcs[MAX_PKT_BURST];
+ struct wqelm elm[MAX_PKT_BURST];
+
+ idx &= wqp->qmask;
+
+ if (wqp->prm.mode == L3FWD_WORKER_UNQUE) {
+ struct rte_ring *r = wqp->queue[idx].r[LCORE_WQ_OUT];
+
+ for (i = 0; i != num; i++) {
+ elm[i].mb = pkts[i];
+ elm[i].rc = hops[i];
+ }
+ return rte_ring_enqueue_burst_elem(r, elm, sizeof(elm[0]), num,
+ NULL);
+ } else if (wqp->prm.mode == L3FWD_WORKER_ORQUE) {
+ struct rte_soring *sor = wqp->queue[idx].sor;
+
+ for (i = 0; i != num; i++)
+ rcs[i] = hops[i];
+ rte_soring_releasx(sor, NULL, rcs, 0, num,
+ wqp->queue[idx].ftoken);
+ return num;
+ }
+
+ rte_errno = ENOTSUP;
+ return 0;
+}
+
+#endif /* L3FWD_WQP_H */
diff --git a/examples/l3fwd/main.c b/examples/l3fwd/main.c
index 01b763e5ba..5d2bc5c611 100644
--- a/examples/l3fwd/main.c
+++ b/examples/l3fwd/main.c
@@ -47,6 +47,7 @@
#include "l3fwd.h"
#include "l3fwd_event.h"
#include "l3fwd_route.h"
+#include "l3fwd_wqp.h"
#define MAX_TX_QUEUE_PER_PORT RTE_MAX_LCORE
#define MAX_RX_QUEUE_PER_PORT 128
@@ -69,6 +70,10 @@ enum L3FWD_LOOKUP_MODE {
};
static enum L3FWD_LOOKUP_MODE lookup_mode;
+struct l3fwd_wqp_param l3fwd_wqp_param = {
+ .mode = L3FWD_WORKER_POLL,
+};
+
/* Global variables. */
static int numa_on = 1; /**< NUMA is enabled by default. */
static int parse_ptype; /**< Parse packet type using rx callback, and */
@@ -246,6 +251,8 @@ const struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
{{32, 1, 2, 0, 0, 0, 0, 15, 0, 0, 0, 0, 0, 0, 0, 0}, 64, 15},
};
+uint32_t l3fwd_lookup_iter_num = 1;
+
/*
* API's called during initialization to setup ACL/EM/LPM rules.
*/
@@ -453,6 +460,23 @@ print_usage(const char *prgname)
ACL_LEAD_CHAR, ROUTE_LEAD_CHAR, alg);
}
+static int
+parse_uint_val(const char *str, uint32_t *val, uint32_t min, uint32_t max)
+{
+ char *end = NULL;
+ unsigned long v;
+
+ errno = 0;
+ v = strtoul(str, &end, 0);
+ if (errno != 0 || end == NULL || *end != '\0')
+ return -1;
+ if (v < min || v > max)
+ return -1;
+
+ *val = v;
+ return 0;
+}
+
static int
parse_max_pkt_len(const char *pktlen)
{
@@ -572,16 +596,35 @@ parse_eth_dest(const char *optarg)
}
static void
-parse_mode(const char *optarg __rte_unused)
+parse_mode(const char *optarg)
{
+ l3fwd_wqp_param.mode = L3FWD_WORKER_POLL;
+
#ifdef RTE_LIB_EVENTDEV
struct l3fwd_event_resources *evt_rsrc = l3fwd_get_eventdev_rsrc();
+ evt_rsrc->enabled = false;
+
if (!strcmp(optarg, "poll"))
evt_rsrc->enabled = false;
else if (!strcmp(optarg, "eventdev"))
evt_rsrc->enabled = true;
+ else
#endif
+ if (strcmp(optarg, "wqorder") == 0) {
+ l3fwd_wqp_param.mode = L3FWD_WORKER_ORQUE;
+ l3fwd_wqp_param.single = 0;
+ } else if (strcmp(optarg, "wqunorder") == 0) {
+ l3fwd_wqp_param.mode = L3FWD_WORKER_UNQUE;
+ l3fwd_wqp_param.single = 0;
+ } else if (strcmp(optarg, "wqorderS") == 0) {
+ l3fwd_wqp_param.mode = L3FWD_WORKER_ORQUE;
+ l3fwd_wqp_param.single = 1;
+ } else if (strcmp(optarg, "wqunorderS") == 0) {
+ l3fwd_wqp_param.mode = L3FWD_WORKER_UNQUE;
+ l3fwd_wqp_param.single = 1;
+ } else
+ rte_exit(EXIT_FAILURE, "unknown mode: %s\n", optarg);
}
static void
@@ -698,6 +741,8 @@ static const char short_options[] =
#define CMD_LINE_OPT_RULE_IPV4 "rule_ipv4"
#define CMD_LINE_OPT_RULE_IPV6 "rule_ipv6"
#define CMD_LINE_OPT_ALG "alg"
+#define CMD_LINE_OPT_WQSIZE "wqsize"
+#define CMD_LINE_OPT_LOOKUP_ITER "lookup-iter"
enum {
/* long options mapped to a short option */
@@ -726,7 +771,9 @@ enum {
CMD_LINE_OPT_LOOKUP_NUM,
CMD_LINE_OPT_ENABLE_VECTOR_NUM,
CMD_LINE_OPT_VECTOR_SIZE_NUM,
- CMD_LINE_OPT_VECTOR_TMO_NS_NUM
+ CMD_LINE_OPT_VECTOR_TMO_NS_NUM,
+ CMD_LINE_OPT_WQSIZE_NUM,
+ CMD_LINE_OPT_LOOKUP_ITER_NUM,
};
static const struct option lgopts[] = {
@@ -753,6 +800,8 @@ static const struct option lgopts[] = {
{CMD_LINE_OPT_RULE_IPV4, 1, 0, CMD_LINE_OPT_RULE_IPV4_NUM},
{CMD_LINE_OPT_RULE_IPV6, 1, 0, CMD_LINE_OPT_RULE_IPV6_NUM},
{CMD_LINE_OPT_ALG, 1, 0, CMD_LINE_OPT_ALG_NUM},
+ {CMD_LINE_OPT_WQSIZE, 1, 0, CMD_LINE_OPT_WQSIZE_NUM},
+ {CMD_LINE_OPT_LOOKUP_ITER, 1, 0, CMD_LINE_OPT_LOOKUP_ITER_NUM},
{NULL, 0, 0, 0}
};
@@ -934,6 +983,18 @@ parse_args(int argc, char **argv)
case CMD_LINE_OPT_ALG_NUM:
l3fwd_set_alg(optarg);
break;
+ case CMD_LINE_OPT_WQSIZE_NUM:
+ ret = parse_uint_val(optarg, &l3fwd_wqp_param.qsize,
+ RX_DESC_DEFAULT, UINT16_MAX);
+ if (ret < 0)
+ return ret;
+ break;
+ case CMD_LINE_OPT_LOOKUP_ITER_NUM:
+ ret = parse_uint_val(optarg, &l3fwd_lookup_iter_num,
+ 1, UINT16_MAX);
+ if (ret < 0)
+ return ret;
+ break;
default:
print_usage(prgname);
return -1;
@@ -1588,6 +1649,8 @@ main(int argc, char **argv)
l3fwd_lkp.main_loop = evt_rsrc->ops.em_event_loop;
else if (lookup_mode == L3FWD_LOOKUP_FIB)
l3fwd_lkp.main_loop = evt_rsrc->ops.fib_event_loop;
+ else if (lookup_mode == L3FWD_LOOKUP_ACL)
+ l3fwd_lkp.main_loop = evt_rsrc->ops.acl_event_loop;
else
l3fwd_lkp.main_loop = evt_rsrc->ops.lpm_event_loop;
} else
@@ -1640,6 +1703,12 @@ main(int argc, char **argv)
}
}
+ /* init worker queues for lcores (if any) */
+ ret = l3fwd_wqp_init(lcore_conf, &l3fwd_wqp_param);
+ if (ret != 0)
+ rte_exit(EXIT_FAILURE, "l3fwd_wqp_init: err=%d, lcore=%u\n",
+ ret, lcore_id);
+
check_all_ports_link_status(enabled_port_mask);
ret = 0;
@@ -1695,6 +1764,8 @@ main(int argc, char **argv)
/* clean up config file routes */
l3fwd_lkp.free_routes();
+ l3fwd_wqp_fini(lcore_conf);
+
/* clean up the EAL */
rte_eal_cleanup();
diff --git a/examples/l3fwd/meson.build b/examples/l3fwd/meson.build
index c25de77bba..a024492fb1 100644
--- a/examples/l3fwd/meson.build
+++ b/examples/l3fwd/meson.build
@@ -16,6 +16,7 @@ sources = files(
'l3fwd_event_generic.c',
'l3fwd_fib.c',
'l3fwd_lpm.c',
+ 'l3fwd_wqp.c',
'main.c',
)
if dpdk_conf.has('RTE_LIB_EVENTDEV')
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [PATCH v3 0/5] Stage-Ordered API and other extensions for ring library
2024-09-06 13:13 ` [RFCv2 0/6] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
` (5 preceding siblings ...)
2024-09-06 13:13 ` [RFCv2 6/6] examples/l3fwd: make ACL work in pipeline and eventdev modes Konstantin Ananyev
@ 2024-09-16 12:37 ` Konstantin Ananyev
2024-09-16 12:37 ` [PATCH v3 1/5] ring: common functions for 'move head' ops Konstantin Ananyev
` (5 more replies)
6 siblings, 6 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-16 12:37 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, Konstantin Ananyev
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
v2 -> v3:
- fix compilation/doxygen complains
- dropped patch:
"examples/l3fwd: make ACL work in pipeline and eventdev modes": [2]
As was mentioned in the patch desctiption it was way too big,
controversial and incomplete. If the community is ok to introduce
pipeline model into the l3fwd, then it is propbably worth to be
a separate patch series.
v1 -> v2:
- rename 'elmst/objst' to 'meta' (Morten)
- introduce new data-path APIs set: one with both meta{} and objs[],
second with just objs[] (Morten)
- split data-path APIs into burst/bulk flavours (same as rte_ring)
- added dump function for te_soring and improved dump() for rte_ring.
- dropped patch:
" ring: minimize reads of the counterpart cache-line"
- no performance gain observed
- actually it does change behavior of conventional rte_ring
enqueue/dequeue APIs -
it could return available/free less then actually exist in the ring.
As in some other libs we reliy on that information - it will
introduce problems.
The main aim of these series is to extend ring library with
new API that allows user to create/use Staged-Ordered-Ring (SORING)
abstraction. In addition to that there are few other patches that serve
different purposes:
- first two patches are just code reordering to de-duplicate
and generalize existing rte_ring code.
- patch #3 extends rte_ring_dump() to correctly print head/tail metadata
for different sync modes.
- next two patches introduce SORING API into the ring library and
provide UT for it.
SORING overview
==============
Staged-Ordered-Ring (SORING) provides a SW abstraction for 'ordered' queues
with multiple processing 'stages'. It is based on conventional DPDK
rte_ring, re-uses many of its concepts, and even substantial part of
its code.
It can be viewed as an 'extension' of rte_ring functionality.
In particular, main SORING properties:
- circular ring buffer with fixed size objects
- producer, consumer plus multiple processing stages in between.
- allows to split objects processing into multiple stages.
- objects remain in the same ring while moving from one stage to the other,
initial order is preserved, no extra copying needed.
- preserves the ingress order of objects within the queue across multiple
stages
- each stage (and producer/consumer) can be served by single and/or
multiple threads.
- number of stages, size and number of objects in the ring are
configurable at ring initialization time.
Data-path API provides four main operations:
- enqueue/dequeue works in the same manner as for conventional rte_ring,
all rte_ring synchronization types are supported.
- acquire/release - for each stage there is an acquire (start) and
release (finish) operation. After some objects are 'acquired' -
given thread can safely assume that it has exclusive ownership of
these objects till it will invoke 'release' for them.
After 'release', objects can be 'acquired' by next stage and/or dequeued
by the consumer (in case of last stage).
Expected use-case: applications that uses pipeline model
(probably with multiple stages) for packet processing, when preserving
incoming packet order is important.
The concept of ‘ring with stages’ is similar to DPDK OPDL eventdev PMD [1],
but the internals are different.
In particular, SORING maintains internal array of 'states' for each element
in the ring that is shared by all threads/processes that access the ring.
That allows 'release' to avoid excessive waits on the tail value and helps
to improve performancei and scalability.
In terms of performance, with our measurements rte_soring and
conventional rte_ring provide nearly identical numbers.
As an example, on our SUT: Intel ICX CPU @ 2.00GHz,
l3fwd (--lookup=acl) in pipeline mode [2] both
rte_ring and rte_soring reach ~20Mpps for single I/O lcore and same
number of worker lcores.
[1] https://www.dpdk.org/wp-content/uploads/sites/35/2018/06/DPDK-China2017-Ma-OPDL.pdf
[2] https://patchwork.dpdk.org/project/dpdk/patch/20240906131348.804-7-konstantin.v.ananyev@yandex.ru/
Eimear Morrissey (1):
ring: make dump function more verbose
Konstantin Ananyev (4):
ring: common functions for 'move head' ops
ring: make copying functions generic
ring/soring: introduce Staged Ordered Ring
app/test: add unit tests for soring API
.mailmap | 1 +
app/test/meson.build | 3 +
app/test/test_ring_stress_impl.h | 1 +
app/test/test_soring.c | 442 +++++++++++++++
app/test/test_soring_mt_stress.c | 40 ++
app/test/test_soring_stress.c | 48 ++
app/test/test_soring_stress.h | 35 ++
app/test/test_soring_stress_impl.h | 827 +++++++++++++++++++++++++++++
lib/ring/meson.build | 4 +-
lib/ring/rte_ring.c | 87 ++-
lib/ring/rte_ring.h | 15 +
lib/ring/rte_ring_c11_pvt.h | 134 +----
lib/ring/rte_ring_elem_pvt.h | 181 +++++--
lib/ring/rte_ring_generic_pvt.h | 121 +----
lib/ring/rte_ring_hts_elem_pvt.h | 85 +--
lib/ring/rte_ring_rts_elem_pvt.h | 85 +--
lib/ring/rte_soring.c | 182 +++++++
lib/ring/rte_soring.h | 547 +++++++++++++++++++
lib/ring/soring.c | 548 +++++++++++++++++++
lib/ring/soring.h | 124 +++++
lib/ring/version.map | 26 +
21 files changed, 3140 insertions(+), 396 deletions(-)
create mode 100644 app/test/test_soring.c
create mode 100644 app/test/test_soring_mt_stress.c
create mode 100644 app/test/test_soring_stress.c
create mode 100644 app/test/test_soring_stress.h
create mode 100644 app/test/test_soring_stress_impl.h
create mode 100644 lib/ring/rte_soring.c
create mode 100644 lib/ring/rte_soring.h
create mode 100644 lib/ring/soring.c
create mode 100644 lib/ring/soring.h
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [PATCH v3 1/5] ring: common functions for 'move head' ops
2024-09-16 12:37 ` [PATCH v3 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
@ 2024-09-16 12:37 ` Konstantin Ananyev
2024-09-16 12:37 ` [PATCH v3 2/5] ring: make copying functions generic Konstantin Ananyev
` (4 subsequent siblings)
5 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-16 12:37 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, Konstantin Ananyev
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
Note upfront: that change doesn't introduce any functional or
performance changes.
It is just a code-reordering for:
- code deduplication
- ability in future to re-use the same code to introduce new functionality
For each sync mode corresponding move_prod_head() and
move_cons_head() are nearly identical to each other,
the only differences are:
- do we need to use a @capacity to calculate number of entries or not.
- what we need to update (prod/cons) and what is used as
read-only counterpart.
So instead of having 2 copies of nearly identical functions,
introduce a new common one that could be used by both functions:
move_prod_head() and move_cons_head().
As another positive thing - we can get rid of referencing whole rte_ring
structure in that new common sub-function.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
---
lib/ring/rte_ring_c11_pvt.h | 134 +++++--------------------------
lib/ring/rte_ring_elem_pvt.h | 66 +++++++++++++++
lib/ring/rte_ring_generic_pvt.h | 121 ++++------------------------
lib/ring/rte_ring_hts_elem_pvt.h | 85 ++++++--------------
lib/ring/rte_ring_rts_elem_pvt.h | 85 ++++++--------------
5 files changed, 149 insertions(+), 342 deletions(-)
diff --git a/lib/ring/rte_ring_c11_pvt.h b/lib/ring/rte_ring_c11_pvt.h
index 629b2d9288..048933ddc6 100644
--- a/lib/ring/rte_ring_c11_pvt.h
+++ b/lib/ring/rte_ring_c11_pvt.h
@@ -28,41 +28,19 @@ __rte_ring_update_tail(struct rte_ring_headtail *ht, uint32_t old_val,
rte_atomic_store_explicit(&ht->tail, new_val, rte_memory_order_release);
}
-/**
- * @internal This function updates the producer head for enqueue
- *
- * @param r
- * A pointer to the ring structure
- * @param is_sp
- * Indicates whether multi-producer path is needed or not
- * @param n
- * The number of elements we will want to enqueue, i.e. how far should the
- * head be moved
- * @param behavior
- * RTE_RING_QUEUE_FIXED: Enqueue a fixed number of items from a ring
- * RTE_RING_QUEUE_VARIABLE: Enqueue as many items as possible from ring
- * @param old_head
- * Returns head value as it was before the move, i.e. where enqueue starts
- * @param new_head
- * Returns the current/new head value i.e. where enqueue finishes
- * @param free_entries
- * Returns the amount of free space in the ring BEFORE head was moved
- * @return
- * Actual number of objects enqueued.
- * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
- */
static __rte_always_inline unsigned int
-__rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
- unsigned int n, enum rte_ring_queue_behavior behavior,
- uint32_t *old_head, uint32_t *new_head,
- uint32_t *free_entries)
+__rte_ring_headtail_move_head(struct rte_ring_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity,
+ unsigned int is_st, unsigned int n,
+ enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head, uint32_t *entries)
{
- const uint32_t capacity = r->capacity;
- uint32_t cons_tail;
- unsigned int max = n;
+ uint32_t stail;
int success;
+ unsigned int max = n;
- *old_head = rte_atomic_load_explicit(&r->prod.head, rte_memory_order_relaxed);
+ *old_head = rte_atomic_load_explicit(&d->head,
+ rte_memory_order_relaxed);
do {
/* Reset n to the initial burst count */
n = max;
@@ -73,112 +51,36 @@ __rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
/* load-acquire synchronize with store-release of ht->tail
* in update_tail.
*/
- cons_tail = rte_atomic_load_explicit(&r->cons.tail,
+ stail = rte_atomic_load_explicit(&s->tail,
rte_memory_order_acquire);
/* The subtraction is done between two unsigned 32bits value
* (the result is always modulo 32 bits even if we have
- * *old_head > cons_tail). So 'free_entries' is always between 0
+ * *old_head > s->tail). So 'free_entries' is always between 0
* and capacity (which is < size).
*/
- *free_entries = (capacity + cons_tail - *old_head);
+ *entries = (capacity + stail - *old_head);
/* check that we have enough room in ring */
- if (unlikely(n > *free_entries))
+ if (unlikely(n > *entries))
n = (behavior == RTE_RING_QUEUE_FIXED) ?
- 0 : *free_entries;
+ 0 : *entries;
if (n == 0)
return 0;
*new_head = *old_head + n;
- if (is_sp) {
- r->prod.head = *new_head;
+ if (is_st) {
+ d->head = *new_head;
success = 1;
} else
/* on failure, *old_head is updated */
- success = rte_atomic_compare_exchange_strong_explicit(&r->prod.head,
- old_head, *new_head,
+ success = rte_atomic_compare_exchange_strong_explicit(
+ &d->head, old_head, *new_head,
rte_memory_order_relaxed,
rte_memory_order_relaxed);
} while (unlikely(success == 0));
return n;
}
-/**
- * @internal This function updates the consumer head for dequeue
- *
- * @param r
- * A pointer to the ring structure
- * @param is_sc
- * Indicates whether multi-consumer path is needed or not
- * @param n
- * The number of elements we will want to dequeue, i.e. how far should the
- * head be moved
- * @param behavior
- * RTE_RING_QUEUE_FIXED: Dequeue a fixed number of items from a ring
- * RTE_RING_QUEUE_VARIABLE: Dequeue as many items as possible from ring
- * @param old_head
- * Returns head value as it was before the move, i.e. where dequeue starts
- * @param new_head
- * Returns the current/new head value i.e. where dequeue finishes
- * @param entries
- * Returns the number of entries in the ring BEFORE head was moved
- * @return
- * - Actual number of objects dequeued.
- * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
- */
-static __rte_always_inline unsigned int
-__rte_ring_move_cons_head(struct rte_ring *r, int is_sc,
- unsigned int n, enum rte_ring_queue_behavior behavior,
- uint32_t *old_head, uint32_t *new_head,
- uint32_t *entries)
-{
- unsigned int max = n;
- uint32_t prod_tail;
- int success;
-
- /* move cons.head atomically */
- *old_head = rte_atomic_load_explicit(&r->cons.head, rte_memory_order_relaxed);
- do {
- /* Restore n as it may change every loop */
- n = max;
-
- /* Ensure the head is read before tail */
- rte_atomic_thread_fence(rte_memory_order_acquire);
-
- /* this load-acquire synchronize with store-release of ht->tail
- * in update_tail.
- */
- prod_tail = rte_atomic_load_explicit(&r->prod.tail,
- rte_memory_order_acquire);
-
- /* The subtraction is done between two unsigned 32bits value
- * (the result is always modulo 32 bits even if we have
- * cons_head > prod_tail). So 'entries' is always between 0
- * and size(ring)-1.
- */
- *entries = (prod_tail - *old_head);
-
- /* Set the actual entries for dequeue */
- if (n > *entries)
- n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries;
-
- if (unlikely(n == 0))
- return 0;
-
- *new_head = *old_head + n;
- if (is_sc) {
- r->cons.head = *new_head;
- success = 1;
- } else
- /* on failure, *old_head will be updated */
- success = rte_atomic_compare_exchange_strong_explicit(&r->cons.head,
- old_head, *new_head,
- rte_memory_order_relaxed,
- rte_memory_order_relaxed);
- } while (unlikely(success == 0));
- return n;
-}
-
#endif /* _RTE_RING_C11_PVT_H_ */
diff --git a/lib/ring/rte_ring_elem_pvt.h b/lib/ring/rte_ring_elem_pvt.h
index 4b80f58980..3a83668a08 100644
--- a/lib/ring/rte_ring_elem_pvt.h
+++ b/lib/ring/rte_ring_elem_pvt.h
@@ -293,6 +293,72 @@ __rte_ring_dequeue_elems(struct rte_ring *r, uint32_t cons_head,
#include "rte_ring_generic_pvt.h"
#endif
+/**
+ * @internal This function updates the producer head for enqueue
+ *
+ * @param r
+ * A pointer to the ring structure
+ * @param is_sp
+ * Indicates whether multi-producer path is needed or not
+ * @param n
+ * The number of elements we will want to enqueue, i.e. how far should the
+ * head be moved
+ * @param behavior
+ * RTE_RING_QUEUE_FIXED: Enqueue a fixed number of items from a ring
+ * RTE_RING_QUEUE_VARIABLE: Enqueue as many items as possible from ring
+ * @param old_head
+ * Returns head value as it was before the move, i.e. where enqueue starts
+ * @param new_head
+ * Returns the current/new head value i.e. where enqueue finishes
+ * @param free_entries
+ * Returns the amount of free space in the ring BEFORE head was moved
+ * @return
+ * Actual number of objects enqueued.
+ * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
+ */
+static __rte_always_inline unsigned int
+__rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
+ unsigned int n, enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head,
+ uint32_t *free_entries)
+{
+ return __rte_ring_headtail_move_head(&r->prod, &r->cons, r->capacity,
+ is_sp, n, behavior, old_head, new_head, free_entries);
+}
+
+/**
+ * @internal This function updates the consumer head for dequeue
+ *
+ * @param r
+ * A pointer to the ring structure
+ * @param is_sc
+ * Indicates whether multi-consumer path is needed or not
+ * @param n
+ * The number of elements we will want to dequeue, i.e. how far should the
+ * head be moved
+ * @param behavior
+ * RTE_RING_QUEUE_FIXED: Dequeue a fixed number of items from a ring
+ * RTE_RING_QUEUE_VARIABLE: Dequeue as many items as possible from ring
+ * @param old_head
+ * Returns head value as it was before the move, i.e. where dequeue starts
+ * @param new_head
+ * Returns the current/new head value i.e. where dequeue finishes
+ * @param entries
+ * Returns the number of entries in the ring BEFORE head was moved
+ * @return
+ * - Actual number of objects dequeued.
+ * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
+ */
+static __rte_always_inline unsigned int
+__rte_ring_move_cons_head(struct rte_ring *r, unsigned int is_sc,
+ unsigned int n, enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head,
+ uint32_t *entries)
+{
+ return __rte_ring_headtail_move_head(&r->cons, &r->prod, 0,
+ is_sc, n, behavior, old_head, new_head, entries);
+}
+
/**
* @internal Enqueue several objects on the ring
*
diff --git a/lib/ring/rte_ring_generic_pvt.h b/lib/ring/rte_ring_generic_pvt.h
index 457f41dab3..12f3595926 100644
--- a/lib/ring/rte_ring_generic_pvt.h
+++ b/lib/ring/rte_ring_generic_pvt.h
@@ -29,36 +29,13 @@ __rte_ring_update_tail(struct rte_ring_headtail *ht, uint32_t old_val,
ht->tail = new_val;
}
-/**
- * @internal This function updates the producer head for enqueue
- *
- * @param r
- * A pointer to the ring structure
- * @param is_sp
- * Indicates whether multi-producer path is needed or not
- * @param n
- * The number of elements we will want to enqueue, i.e. how far should the
- * head be moved
- * @param behavior
- * RTE_RING_QUEUE_FIXED: Enqueue a fixed number of items from a ring
- * RTE_RING_QUEUE_VARIABLE: Enqueue as many items as possible from ring
- * @param old_head
- * Returns head value as it was before the move, i.e. where enqueue starts
- * @param new_head
- * Returns the current/new head value i.e. where enqueue finishes
- * @param free_entries
- * Returns the amount of free space in the ring BEFORE head was moved
- * @return
- * Actual number of objects enqueued.
- * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
- */
static __rte_always_inline unsigned int
-__rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
- unsigned int n, enum rte_ring_queue_behavior behavior,
- uint32_t *old_head, uint32_t *new_head,
- uint32_t *free_entries)
+__rte_ring_headtail_move_head(struct rte_ring_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity,
+ unsigned int is_st, unsigned int n,
+ enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head, uint32_t *entries)
{
- const uint32_t capacity = r->capacity;
unsigned int max = n;
int success;
@@ -66,7 +43,7 @@ __rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
/* Reset n to the initial burst count */
n = max;
- *old_head = r->prod.head;
+ *old_head = d->head;
/* add rmb barrier to avoid load/load reorder in weak
* memory model. It is noop on x86
@@ -76,97 +53,27 @@ __rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
/*
* The subtraction is done between two unsigned 32bits value
* (the result is always modulo 32 bits even if we have
- * *old_head > cons_tail). So 'free_entries' is always between 0
+ * *old_head > s->tail). So 'free_entries' is always between 0
* and capacity (which is < size).
*/
- *free_entries = (capacity + r->cons.tail - *old_head);
+ *entries = (capacity + s->tail - *old_head);
/* check that we have enough room in ring */
- if (unlikely(n > *free_entries))
+ if (unlikely(n > *entries))
n = (behavior == RTE_RING_QUEUE_FIXED) ?
- 0 : *free_entries;
+ 0 : *entries;
if (n == 0)
return 0;
*new_head = *old_head + n;
- if (is_sp) {
- r->prod.head = *new_head;
+ if (is_st) {
+ d->head = *new_head;
success = 1;
} else
- success = rte_atomic32_cmpset((uint32_t *)(uintptr_t)&r->prod.head,
- *old_head, *new_head);
- } while (unlikely(success == 0));
- return n;
-}
-
-/**
- * @internal This function updates the consumer head for dequeue
- *
- * @param r
- * A pointer to the ring structure
- * @param is_sc
- * Indicates whether multi-consumer path is needed or not
- * @param n
- * The number of elements we will want to dequeue, i.e. how far should the
- * head be moved
- * @param behavior
- * RTE_RING_QUEUE_FIXED: Dequeue a fixed number of items from a ring
- * RTE_RING_QUEUE_VARIABLE: Dequeue as many items as possible from ring
- * @param old_head
- * Returns head value as it was before the move, i.e. where dequeue starts
- * @param new_head
- * Returns the current/new head value i.e. where dequeue finishes
- * @param entries
- * Returns the number of entries in the ring BEFORE head was moved
- * @return
- * - Actual number of objects dequeued.
- * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
- */
-static __rte_always_inline unsigned int
-__rte_ring_move_cons_head(struct rte_ring *r, unsigned int is_sc,
- unsigned int n, enum rte_ring_queue_behavior behavior,
- uint32_t *old_head, uint32_t *new_head,
- uint32_t *entries)
-{
- unsigned int max = n;
- int success;
-
- /* move cons.head atomically */
- do {
- /* Restore n as it may change every loop */
- n = max;
-
- *old_head = r->cons.head;
-
- /* add rmb barrier to avoid load/load reorder in weak
- * memory model. It is noop on x86
- */
- rte_smp_rmb();
-
- /* The subtraction is done between two unsigned 32bits value
- * (the result is always modulo 32 bits even if we have
- * cons_head > prod_tail). So 'entries' is always between 0
- * and size(ring)-1.
- */
- *entries = (r->prod.tail - *old_head);
-
- /* Set the actual entries for dequeue */
- if (n > *entries)
- n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries;
-
- if (unlikely(n == 0))
- return 0;
-
- *new_head = *old_head + n;
- if (is_sc) {
- r->cons.head = *new_head;
- rte_smp_rmb();
- success = 1;
- } else {
- success = rte_atomic32_cmpset((uint32_t *)(uintptr_t)&r->cons.head,
+ success = rte_atomic32_cmpset(
+ (uint32_t *)(uintptr_t)&d->head,
*old_head, *new_head);
- }
} while (unlikely(success == 0));
return n;
}
diff --git a/lib/ring/rte_ring_hts_elem_pvt.h b/lib/ring/rte_ring_hts_elem_pvt.h
index 91f5eeccb9..ed5f16879f 100644
--- a/lib/ring/rte_ring_hts_elem_pvt.h
+++ b/lib/ring/rte_ring_hts_elem_pvt.h
@@ -50,20 +50,16 @@ __rte_ring_hts_head_wait(const struct rte_ring_hts_headtail *ht,
}
}
-/**
- * @internal This function updates the producer head for enqueue
- */
-static __rte_always_inline unsigned int
-__rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
+static __rte_always_inline uint32_t
+__rte_ring_hts_move_head(struct rte_ring_hts_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity, unsigned int num,
enum rte_ring_queue_behavior behavior, uint32_t *old_head,
- uint32_t *free_entries)
+ uint32_t *entries)
{
uint32_t n;
union __rte_ring_hts_pos np, op;
- const uint32_t capacity = r->capacity;
-
- op.raw = rte_atomic_load_explicit(&r->hts_prod.ht.raw, rte_memory_order_acquire);
+ op.raw = rte_atomic_load_explicit(&d->ht.raw, rte_memory_order_acquire);
do {
/* Reset n to the initial burst count */
@@ -74,7 +70,7 @@ __rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
* make sure that we read prod head/tail *before*
* reading cons tail.
*/
- __rte_ring_hts_head_wait(&r->hts_prod, &op);
+ __rte_ring_hts_head_wait(d, &op);
/*
* The subtraction is done between two unsigned 32bits value
@@ -82,12 +78,12 @@ __rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
* *old_head > cons_tail). So 'free_entries' is always between 0
* and capacity (which is < size).
*/
- *free_entries = capacity + r->cons.tail - op.pos.head;
+ *entries = capacity + s->tail - op.pos.head;
/* check that we have enough room in ring */
- if (unlikely(n > *free_entries))
+ if (unlikely(n > *entries))
n = (behavior == RTE_RING_QUEUE_FIXED) ?
- 0 : *free_entries;
+ 0 : *entries;
if (n == 0)
break;
@@ -100,13 +96,25 @@ __rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
* - OOO reads of cons tail value
* - OOO copy of elems from the ring
*/
- } while (rte_atomic_compare_exchange_strong_explicit(&r->hts_prod.ht.raw,
+ } while (rte_atomic_compare_exchange_strong_explicit(&d->ht.raw,
(uint64_t *)(uintptr_t)&op.raw, np.raw,
- rte_memory_order_acquire, rte_memory_order_acquire) == 0);
+ rte_memory_order_acquire,
+ rte_memory_order_acquire) == 0);
*old_head = op.pos.head;
return n;
}
+/**
+ * @internal This function updates the producer head for enqueue
+ */
+static __rte_always_inline unsigned int
+__rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
+ enum rte_ring_queue_behavior behavior, uint32_t *old_head,
+ uint32_t *free_entries)
+{
+ return __rte_ring_hts_move_head(&r->hts_prod, &r->cons,
+ r->capacity, num, behavior, old_head, free_entries);
+}
/**
* @internal This function updates the consumer head for dequeue
@@ -116,51 +124,8 @@ __rte_ring_hts_move_cons_head(struct rte_ring *r, unsigned int num,
enum rte_ring_queue_behavior behavior, uint32_t *old_head,
uint32_t *entries)
{
- uint32_t n;
- union __rte_ring_hts_pos np, op;
-
- op.raw = rte_atomic_load_explicit(&r->hts_cons.ht.raw, rte_memory_order_acquire);
-
- /* move cons.head atomically */
- do {
- /* Restore n as it may change every loop */
- n = num;
-
- /*
- * wait for tail to be equal to head,
- * make sure that we read cons head/tail *before*
- * reading prod tail.
- */
- __rte_ring_hts_head_wait(&r->hts_cons, &op);
-
- /* The subtraction is done between two unsigned 32bits value
- * (the result is always modulo 32 bits even if we have
- * cons_head > prod_tail). So 'entries' is always between 0
- * and size(ring)-1.
- */
- *entries = r->prod.tail - op.pos.head;
-
- /* Set the actual entries for dequeue */
- if (n > *entries)
- n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries;
-
- if (unlikely(n == 0))
- break;
-
- np.pos.tail = op.pos.tail;
- np.pos.head = op.pos.head + n;
-
- /*
- * this CAS(ACQUIRE, ACQUIRE) serves as a hoist barrier to prevent:
- * - OOO reads of prod tail value
- * - OOO copy of elems from the ring
- */
- } while (rte_atomic_compare_exchange_strong_explicit(&r->hts_cons.ht.raw,
- (uint64_t *)(uintptr_t)&op.raw, np.raw,
- rte_memory_order_acquire, rte_memory_order_acquire) == 0);
-
- *old_head = op.pos.head;
- return n;
+ return __rte_ring_hts_move_head(&r->hts_cons, &r->prod,
+ 0, num, behavior, old_head, entries);
}
/**
diff --git a/lib/ring/rte_ring_rts_elem_pvt.h b/lib/ring/rte_ring_rts_elem_pvt.h
index 122650346b..027409a3fa 100644
--- a/lib/ring/rte_ring_rts_elem_pvt.h
+++ b/lib/ring/rte_ring_rts_elem_pvt.h
@@ -64,20 +64,17 @@ __rte_ring_rts_head_wait(const struct rte_ring_rts_headtail *ht,
}
}
-/**
- * @internal This function updates the producer head for enqueue.
- */
static __rte_always_inline uint32_t
-__rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
+__rte_ring_rts_move_head(struct rte_ring_rts_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity, uint32_t num,
enum rte_ring_queue_behavior behavior, uint32_t *old_head,
- uint32_t *free_entries)
+ uint32_t *entries)
{
uint32_t n;
union __rte_ring_rts_poscnt nh, oh;
- const uint32_t capacity = r->capacity;
-
- oh.raw = rte_atomic_load_explicit(&r->rts_prod.head.raw, rte_memory_order_acquire);
+ oh.raw = rte_atomic_load_explicit(&d->head.raw,
+ rte_memory_order_acquire);
do {
/* Reset n to the initial burst count */
@@ -88,7 +85,7 @@ __rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
* make sure that we read prod head *before*
* reading cons tail.
*/
- __rte_ring_rts_head_wait(&r->rts_prod, &oh);
+ __rte_ring_rts_head_wait(d, &oh);
/*
* The subtraction is done between two unsigned 32bits value
@@ -96,12 +93,12 @@ __rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
* *old_head > cons_tail). So 'free_entries' is always between 0
* and capacity (which is < size).
*/
- *free_entries = capacity + r->cons.tail - oh.val.pos;
+ *entries = capacity + s->tail - oh.val.pos;
/* check that we have enough room in ring */
- if (unlikely(n > *free_entries))
+ if (unlikely(n > *entries))
n = (behavior == RTE_RING_QUEUE_FIXED) ?
- 0 : *free_entries;
+ 0 : *entries;
if (n == 0)
break;
@@ -114,14 +111,27 @@ __rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
* - OOO reads of cons tail value
* - OOO copy of elems to the ring
*/
- } while (rte_atomic_compare_exchange_strong_explicit(&r->rts_prod.head.raw,
+ } while (rte_atomic_compare_exchange_strong_explicit(&d->head.raw,
(uint64_t *)(uintptr_t)&oh.raw, nh.raw,
- rte_memory_order_acquire, rte_memory_order_acquire) == 0);
+ rte_memory_order_acquire,
+ rte_memory_order_acquire) == 0);
*old_head = oh.val.pos;
return n;
}
+/**
+ * @internal This function updates the producer head for enqueue.
+ */
+static __rte_always_inline uint32_t
+__rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
+ enum rte_ring_queue_behavior behavior, uint32_t *old_head,
+ uint32_t *free_entries)
+{
+ return __rte_ring_rts_move_head(&r->rts_prod, &r->cons,
+ r->capacity, num, behavior, old_head, free_entries);
+}
+
/**
* @internal This function updates the consumer head for dequeue
*/
@@ -130,51 +140,8 @@ __rte_ring_rts_move_cons_head(struct rte_ring *r, uint32_t num,
enum rte_ring_queue_behavior behavior, uint32_t *old_head,
uint32_t *entries)
{
- uint32_t n;
- union __rte_ring_rts_poscnt nh, oh;
-
- oh.raw = rte_atomic_load_explicit(&r->rts_cons.head.raw, rte_memory_order_acquire);
-
- /* move cons.head atomically */
- do {
- /* Restore n as it may change every loop */
- n = num;
-
- /*
- * wait for cons head/tail distance,
- * make sure that we read cons head *before*
- * reading prod tail.
- */
- __rte_ring_rts_head_wait(&r->rts_cons, &oh);
-
- /* The subtraction is done between two unsigned 32bits value
- * (the result is always modulo 32 bits even if we have
- * cons_head > prod_tail). So 'entries' is always between 0
- * and size(ring)-1.
- */
- *entries = r->prod.tail - oh.val.pos;
-
- /* Set the actual entries for dequeue */
- if (n > *entries)
- n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries;
-
- if (unlikely(n == 0))
- break;
-
- nh.val.pos = oh.val.pos + n;
- nh.val.cnt = oh.val.cnt + 1;
-
- /*
- * this CAS(ACQUIRE, ACQUIRE) serves as a hoist barrier to prevent:
- * - OOO reads of prod tail value
- * - OOO copy of elems from the ring
- */
- } while (rte_atomic_compare_exchange_strong_explicit(&r->rts_cons.head.raw,
- (uint64_t *)(uintptr_t)&oh.raw, nh.raw,
- rte_memory_order_acquire, rte_memory_order_acquire) == 0);
-
- *old_head = oh.val.pos;
- return n;
+ return __rte_ring_rts_move_head(&r->rts_cons, &r->prod,
+ 0, num, behavior, old_head, entries);
}
/**
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [PATCH v3 2/5] ring: make copying functions generic
2024-09-16 12:37 ` [PATCH v3 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
2024-09-16 12:37 ` [PATCH v3 1/5] ring: common functions for 'move head' ops Konstantin Ananyev
@ 2024-09-16 12:37 ` Konstantin Ananyev
2024-09-16 12:37 ` [PATCH v3 3/5] ring: make dump function more verbose Konstantin Ananyev
` (3 subsequent siblings)
5 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-16 12:37 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, Konstantin Ananyev
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
Note upfront: that change doesn't introduce any functional
or performance changes.
It is just a code-reordering for:
- improve code modularity and re-usability
- ability in future to re-use the same code to introduce new functionality
There is no real need for enqueue_elems()/dequeue_elems()
to get pointer to actual rte_ring structure, instead it is enough to pass
a pointer to actual elements buffer inside the ring.
In return, we'll get a copying functions that could be used for other
queueing abstractions that do have circular ring buffer inside.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
---
lib/ring/rte_ring_elem_pvt.h | 115 ++++++++++++++++++++---------------
1 file changed, 67 insertions(+), 48 deletions(-)
diff --git a/lib/ring/rte_ring_elem_pvt.h b/lib/ring/rte_ring_elem_pvt.h
index 3a83668a08..6eafae121f 100644
--- a/lib/ring/rte_ring_elem_pvt.h
+++ b/lib/ring/rte_ring_elem_pvt.h
@@ -17,12 +17,14 @@
#endif
static __rte_always_inline void
-__rte_ring_enqueue_elems_32(struct rte_ring *r, const uint32_t size,
- uint32_t idx, const void *obj_table, uint32_t n)
+__rte_ring_enqueue_elems_32(void *ring_table, const void *obj_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- uint32_t *ring = (uint32_t *)&r[1];
+
+ uint32_t *ring = (uint32_t *)ring_table;
const uint32_t *obj = (const uint32_t *)obj_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x7); i += 8, idx += 8) {
ring[idx] = obj[i];
@@ -60,14 +62,14 @@ __rte_ring_enqueue_elems_32(struct rte_ring *r, const uint32_t size,
}
static __rte_always_inline void
-__rte_ring_enqueue_elems_64(struct rte_ring *r, uint32_t prod_head,
- const void *obj_table, uint32_t n)
+__rte_ring_enqueue_elems_64(void *ring_table, const void *obj_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- const uint32_t size = r->size;
- uint32_t idx = prod_head & r->mask;
- uint64_t *ring = (uint64_t *)&r[1];
+
+ uint64_t *ring = (uint64_t *)ring_table;
const unaligned_uint64_t *obj = (const unaligned_uint64_t *)obj_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x3); i += 4, idx += 4) {
ring[idx] = obj[i];
@@ -93,14 +95,14 @@ __rte_ring_enqueue_elems_64(struct rte_ring *r, uint32_t prod_head,
}
static __rte_always_inline void
-__rte_ring_enqueue_elems_128(struct rte_ring *r, uint32_t prod_head,
- const void *obj_table, uint32_t n)
+__rte_ring_enqueue_elems_128(void *ring_table, const void *obj_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- const uint32_t size = r->size;
- uint32_t idx = prod_head & r->mask;
- rte_int128_t *ring = (rte_int128_t *)&r[1];
+
+ rte_int128_t *ring = (rte_int128_t *)ring_table;
const rte_int128_t *obj = (const rte_int128_t *)obj_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x1); i += 2, idx += 2)
memcpy((void *)(ring + idx),
@@ -126,37 +128,47 @@ __rte_ring_enqueue_elems_128(struct rte_ring *r, uint32_t prod_head,
* single and multi producer enqueue functions.
*/
static __rte_always_inline void
-__rte_ring_enqueue_elems(struct rte_ring *r, uint32_t prod_head,
- const void *obj_table, uint32_t esize, uint32_t num)
+__rte_ring_do_enqueue_elems(void *ring_table, const void *obj_table,
+ uint32_t size, uint32_t idx, uint32_t esize, uint32_t num)
{
/* 8B and 16B copies implemented individually to retain
* the current performance.
*/
if (esize == 8)
- __rte_ring_enqueue_elems_64(r, prod_head, obj_table, num);
+ __rte_ring_enqueue_elems_64(ring_table, obj_table, size,
+ idx, num);
else if (esize == 16)
- __rte_ring_enqueue_elems_128(r, prod_head, obj_table, num);
+ __rte_ring_enqueue_elems_128(ring_table, obj_table, size,
+ idx, num);
else {
- uint32_t idx, scale, nr_idx, nr_num, nr_size;
+ uint32_t scale, nr_idx, nr_num, nr_size;
/* Normalize to uint32_t */
scale = esize / sizeof(uint32_t);
nr_num = num * scale;
- idx = prod_head & r->mask;
nr_idx = idx * scale;
- nr_size = r->size * scale;
- __rte_ring_enqueue_elems_32(r, nr_size, nr_idx,
- obj_table, nr_num);
+ nr_size = size * scale;
+ __rte_ring_enqueue_elems_32(ring_table, obj_table, nr_size,
+ nr_idx, nr_num);
}
}
static __rte_always_inline void
-__rte_ring_dequeue_elems_32(struct rte_ring *r, const uint32_t size,
- uint32_t idx, void *obj_table, uint32_t n)
+__rte_ring_enqueue_elems(struct rte_ring *r, uint32_t prod_head,
+ const void *obj_table, uint32_t esize, uint32_t num)
+{
+ __rte_ring_do_enqueue_elems(&r[1], obj_table, r->size,
+ prod_head & r->mask, esize, num);
+}
+
+static __rte_always_inline void
+__rte_ring_dequeue_elems_32(void *obj_table, const void *ring_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- uint32_t *ring = (uint32_t *)&r[1];
uint32_t *obj = (uint32_t *)obj_table;
+ const uint32_t *ring = (const uint32_t *)ring_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x7); i += 8, idx += 8) {
obj[i] = ring[idx];
@@ -194,14 +206,13 @@ __rte_ring_dequeue_elems_32(struct rte_ring *r, const uint32_t size,
}
static __rte_always_inline void
-__rte_ring_dequeue_elems_64(struct rte_ring *r, uint32_t cons_head,
- void *obj_table, uint32_t n)
+__rte_ring_dequeue_elems_64(void *obj_table, const void *ring_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- const uint32_t size = r->size;
- uint32_t idx = cons_head & r->mask;
- uint64_t *ring = (uint64_t *)&r[1];
unaligned_uint64_t *obj = (unaligned_uint64_t *)obj_table;
+ const uint64_t *ring = (const uint64_t *)ring_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x3); i += 4, idx += 4) {
obj[i] = ring[idx];
@@ -227,27 +238,26 @@ __rte_ring_dequeue_elems_64(struct rte_ring *r, uint32_t cons_head,
}
static __rte_always_inline void
-__rte_ring_dequeue_elems_128(struct rte_ring *r, uint32_t cons_head,
- void *obj_table, uint32_t n)
+__rte_ring_dequeue_elems_128(void *obj_table, const void *ring_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- const uint32_t size = r->size;
- uint32_t idx = cons_head & r->mask;
- rte_int128_t *ring = (rte_int128_t *)&r[1];
rte_int128_t *obj = (rte_int128_t *)obj_table;
+ const rte_int128_t *ring = (const rte_int128_t *)ring_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x1); i += 2, idx += 2)
- memcpy((void *)(obj + i), (void *)(ring + idx), 32);
+ memcpy((obj + i), (const void *)(ring + idx), 32);
switch (n & 0x1) {
case 1:
- memcpy((void *)(obj + i), (void *)(ring + idx), 16);
+ memcpy((obj + i), (const void *)(ring + idx), 16);
}
} else {
for (i = 0; idx < size; i++, idx++)
- memcpy((void *)(obj + i), (void *)(ring + idx), 16);
+ memcpy((obj + i), (const void *)(ring + idx), 16);
/* Start at the beginning */
for (idx = 0; i < n; i++, idx++)
- memcpy((void *)(obj + i), (void *)(ring + idx), 16);
+ memcpy((obj + i), (const void *)(ring + idx), 16);
}
}
@@ -256,30 +266,39 @@ __rte_ring_dequeue_elems_128(struct rte_ring *r, uint32_t cons_head,
* single and multi producer enqueue functions.
*/
static __rte_always_inline void
-__rte_ring_dequeue_elems(struct rte_ring *r, uint32_t cons_head,
- void *obj_table, uint32_t esize, uint32_t num)
+__rte_ring_do_dequeue_elems(void *obj_table, const void *ring_table,
+ uint32_t size, uint32_t idx, uint32_t esize, uint32_t num)
{
/* 8B and 16B copies implemented individually to retain
* the current performance.
*/
if (esize == 8)
- __rte_ring_dequeue_elems_64(r, cons_head, obj_table, num);
+ __rte_ring_dequeue_elems_64(obj_table, ring_table, size,
+ idx, num);
else if (esize == 16)
- __rte_ring_dequeue_elems_128(r, cons_head, obj_table, num);
+ __rte_ring_dequeue_elems_128(obj_table, ring_table, size,
+ idx, num);
else {
- uint32_t idx, scale, nr_idx, nr_num, nr_size;
+ uint32_t scale, nr_idx, nr_num, nr_size;
/* Normalize to uint32_t */
scale = esize / sizeof(uint32_t);
nr_num = num * scale;
- idx = cons_head & r->mask;
nr_idx = idx * scale;
- nr_size = r->size * scale;
- __rte_ring_dequeue_elems_32(r, nr_size, nr_idx,
- obj_table, nr_num);
+ nr_size = size * scale;
+ __rte_ring_dequeue_elems_32(obj_table, ring_table, nr_size,
+ nr_idx, nr_num);
}
}
+static __rte_always_inline void
+__rte_ring_dequeue_elems(struct rte_ring *r, uint32_t cons_head,
+ void *obj_table, uint32_t esize, uint32_t num)
+{
+ __rte_ring_do_dequeue_elems(obj_table, &r[1], r->size,
+ cons_head & r->mask, esize, num);
+}
+
/* Between load and load. there might be cpu reorder in weak model
* (powerpc/arm).
* There are 2 choices for the users
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [PATCH v3 3/5] ring: make dump function more verbose
2024-09-16 12:37 ` [PATCH v3 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
2024-09-16 12:37 ` [PATCH v3 1/5] ring: common functions for 'move head' ops Konstantin Ananyev
2024-09-16 12:37 ` [PATCH v3 2/5] ring: make copying functions generic Konstantin Ananyev
@ 2024-09-16 12:37 ` Konstantin Ananyev
2024-09-16 12:37 ` [PATCH v3 4/5] ring/soring: introduce Staged Ordered Ring Konstantin Ananyev
` (2 subsequent siblings)
5 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-16 12:37 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, Eimear Morrissey
From: Eimear Morrissey <eimear.morrissey@huawei.com>
The current rte_ring_dump function uses the generic rte_ring_headtail
structure to access head/tail positions. This is incorrect for the RTS
case where the head is stored in a different offset in the union of
structs. Switching to a separate function for each sync type allows
to dump correct head/tail values and extra metadata.
Signed-off-by: Eimear Morrissey <eimear.morrissey@huawei.com>
---
.mailmap | 1 +
app/test/test_ring_stress_impl.h | 1 +
lib/ring/rte_ring.c | 87 ++++++++++++++++++++++++++++++--
lib/ring/rte_ring.h | 15 ++++++
lib/ring/version.map | 7 +++
5 files changed, 107 insertions(+), 4 deletions(-)
diff --git a/.mailmap b/.mailmap
index 4a508bafad..3da86393c0 100644
--- a/.mailmap
+++ b/.mailmap
@@ -379,6 +379,7 @@ Eduard Serra <eserra@vmware.com>
Edward Makarov <makarov@kraftway.ru>
Edwin Brossette <edwin.brossette@6wind.com>
Eelco Chaudron <echaudro@redhat.com>
+Eimear Morrissey <eimear.morrissey@huawei.com>
Elad Nachman <eladv6@gmail.com>
Elad Persiko <eladpe@mellanox.com>
Elena Agostini <eagostini@nvidia.com>
diff --git a/app/test/test_ring_stress_impl.h b/app/test/test_ring_stress_impl.h
index 8b0bfb11fe..8449cb4b15 100644
--- a/app/test/test_ring_stress_impl.h
+++ b/app/test/test_ring_stress_impl.h
@@ -380,6 +380,7 @@ test_mt1(int (*test)(void *))
}
lcore_stat_dump(stdout, UINT32_MAX, &arg[mc].stats);
+ rte_ring_dump(stdout, r);
mt1_fini(r, data);
return rc;
}
diff --git a/lib/ring/rte_ring.c b/lib/ring/rte_ring.c
index aebb6d6728..261f2a06db 100644
--- a/lib/ring/rte_ring.c
+++ b/lib/ring/rte_ring.c
@@ -364,20 +364,99 @@ rte_ring_free(struct rte_ring *r)
rte_free(te);
}
+static const char *
+ring_get_sync_type(const enum rte_ring_sync_type st)
+{
+ switch (st) {
+ case RTE_RING_SYNC_ST:
+ return "single thread";
+ case RTE_RING_SYNC_MT:
+ return "multi thread";
+ case RTE_RING_SYNC_MT_RTS:
+ return "multi thread - RTS";
+ case RTE_RING_SYNC_MT_HTS:
+ return "multi thread - HTS";
+ default:
+ return "unknown";
+ }
+}
+
+static void
+ring_dump_ht_headtail(FILE *f, const char *prefix,
+ const struct rte_ring_headtail *ht)
+{
+ fprintf(f, "%ssync_type=%s\n", prefix,
+ ring_get_sync_type(ht->sync_type));
+ fprintf(f, "%shead=%"PRIu32"\n", prefix, ht->head);
+ fprintf(f, "%stail=%"PRIu32"\n", prefix, ht->tail);
+}
+
+static void
+ring_dump_rts_headtail(FILE *f, const char *prefix,
+ const struct rte_ring_rts_headtail *rts)
+{
+ fprintf(f, "%ssync_type=%s\n", prefix,
+ ring_get_sync_type(rts->sync_type));
+ fprintf(f, "%shead.pos=%"PRIu32"\n", prefix, rts->head.val.pos);
+ fprintf(f, "%shead.cnt=%"PRIu32"\n", prefix, rts->head.val.cnt);
+ fprintf(f, "%stail.pos=%"PRIu32"\n", prefix, rts->tail.val.pos);
+ fprintf(f, "%stail.cnt=%"PRIu32"\n", prefix, rts->tail.val.cnt);
+ fprintf(f, "%shtd_max=%"PRIu32"\n", prefix, rts->htd_max);
+}
+
+static void
+ring_dump_hts_headtail(FILE *f, const char *prefix,
+ const struct rte_ring_hts_headtail *hts)
+{
+ fprintf(f, "%ssync_type=%s\n", prefix,
+ ring_get_sync_type(hts->sync_type));
+ fprintf(f, "%shead=%"PRIu32"\n", prefix, hts->ht.pos.head);
+ fprintf(f, "%stail=%"PRIu32"\n", prefix, hts->ht.pos.tail);
+}
+
+void
+rte_ring_headtail_dump(FILE *f, const char *prefix,
+ const struct rte_ring_headtail *r)
+{
+ if (f == NULL || r == NULL)
+ return;
+
+ prefix = (prefix != NULL) ? prefix : "";
+
+ switch (r->sync_type) {
+ case RTE_RING_SYNC_ST:
+ case RTE_RING_SYNC_MT:
+ ring_dump_ht_headtail(f, prefix, r);
+ break;
+ case RTE_RING_SYNC_MT_RTS:
+ ring_dump_rts_headtail(f, prefix,
+ (const struct rte_ring_rts_headtail *)r);
+ break;
+ case RTE_RING_SYNC_MT_HTS:
+ ring_dump_hts_headtail(f, prefix,
+ (const struct rte_ring_hts_headtail *)r);
+ break;
+ default:
+ RING_LOG(ERR, "Invalid ring sync type detected");
+ }
+}
+
/* dump the status of the ring on the console */
void
rte_ring_dump(FILE *f, const struct rte_ring *r)
{
+ if (f == NULL || r == NULL)
+ return;
+
fprintf(f, "ring <%s>@%p\n", r->name, r);
fprintf(f, " flags=%x\n", r->flags);
fprintf(f, " size=%"PRIu32"\n", r->size);
fprintf(f, " capacity=%"PRIu32"\n", r->capacity);
- fprintf(f, " ct=%"PRIu32"\n", r->cons.tail);
- fprintf(f, " ch=%"PRIu32"\n", r->cons.head);
- fprintf(f, " pt=%"PRIu32"\n", r->prod.tail);
- fprintf(f, " ph=%"PRIu32"\n", r->prod.head);
fprintf(f, " used=%u\n", rte_ring_count(r));
fprintf(f, " avail=%u\n", rte_ring_free_count(r));
+
+ rte_ring_headtail_dump(f, " cons.", &(r->cons));
+ rte_ring_headtail_dump(f, " prod.", &(r->prod));
}
/* dump the status of all rings on the console */
diff --git a/lib/ring/rte_ring.h b/lib/ring/rte_ring.h
index c709f30497..d6f9e85c64 100644
--- a/lib/ring/rte_ring.h
+++ b/lib/ring/rte_ring.h
@@ -204,6 +204,21 @@ void rte_ring_free(struct rte_ring *r);
*/
void rte_ring_dump(FILE *f, const struct rte_ring *r);
+/**
+ * Dump the status of a headtail to a file.
+ *
+ * @param f
+ * A pointer to a file for output
+ * @param prefix
+ * A string to prefix each output line with
+ * @param r
+ * A pointer to a ring headtail structure.
+ */
+__rte_experimental
+void
+rte_ring_headtail_dump(FILE *f, const char *prefix,
+ const struct rte_ring_headtail *r);
+
/**
* Enqueue several objects on the ring (multi-producers safe).
*
diff --git a/lib/ring/version.map b/lib/ring/version.map
index 8da094a69a..61f7464f5a 100644
--- a/lib/ring/version.map
+++ b/lib/ring/version.map
@@ -14,3 +14,10 @@ DPDK_25 {
local: *;
};
+
+EXPERIMENTAL {
+ global:
+
+ # added in 24.11
+ rte_ring_headtail_dump;
+};
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [PATCH v3 4/5] ring/soring: introduce Staged Ordered Ring
2024-09-16 12:37 ` [PATCH v3 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
` (2 preceding siblings ...)
2024-09-16 12:37 ` [PATCH v3 3/5] ring: make dump function more verbose Konstantin Ananyev
@ 2024-09-16 12:37 ` Konstantin Ananyev
2024-09-16 12:37 ` [PATCH v3 5/5] app/test: add unit tests for soring API Konstantin Ananyev
2024-09-17 12:09 ` [PATCH v4 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
5 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-16 12:37 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, Konstantin Ananyev, Eimear Morrissey
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
Staged-Ordered-Ring (SORING) provides a SW abstraction for 'ordered' queues
with multiple processing 'stages'.
It is based on conventional DPDK rte_ring, re-uses many of its concepts,
and even substantial part of its code.
It can be viewed as an 'extension' of rte_ring functionality.
In particular, main SORING properties:
- circular ring buffer with fixed size objects
- producer, consumer plus multiple processing stages in the middle.
- allows to split objects processing into multiple stages.
- objects remain in the same ring while moving from one stage to the other,
initial order is preserved, no extra copying needed.
- preserves the ingress order of objects within the queue across multiple
stages, i.e.:
at the same stage multiple threads can process objects from the ring in
any order, but for the next stage objects will always appear in the
original order.
- each stage (and producer/consumer) can be served by single and/or
multiple threads.
- number of stages, size and number of objects in the ring are
configurable at ring initialization time.
Data-path API provides four main operations:
- enqueue/dequeue works in the same manner as for conventional rte_ring,
all rte_ring synchronization types are supported.
- acquire/release - for each stage there is an acquire (start) and
release (finish) operation.
after some objects are 'acquired' - given thread can safely assume that
it has exclusive possession of these objects till 'release' for them is
invoked.
Note that right now user has to release exactly the same number of
objects that was acquired before.
After 'release', objects can be 'acquired' by next stage and/or dequeued
by the consumer (in case of last stage).
Expected use-case: applications that uses pipeline model
(probably with multiple stages) for packet processing, when preserving
incoming packet order is important. I.E.: IPsec processing, etc.
Signed-off-by: Eimear Morrissey <eimear.morrissey@huawei.com>
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
---
lib/ring/meson.build | 4 +-
lib/ring/rte_soring.c | 182 ++++++++++++++
lib/ring/rte_soring.h | 547 +++++++++++++++++++++++++++++++++++++++++
lib/ring/soring.c | 548 ++++++++++++++++++++++++++++++++++++++++++
lib/ring/soring.h | 124 ++++++++++
lib/ring/version.map | 19 ++
6 files changed, 1422 insertions(+), 2 deletions(-)
create mode 100644 lib/ring/rte_soring.c
create mode 100644 lib/ring/rte_soring.h
create mode 100644 lib/ring/soring.c
create mode 100644 lib/ring/soring.h
diff --git a/lib/ring/meson.build b/lib/ring/meson.build
index 7fca958ed7..21f2c12989 100644
--- a/lib/ring/meson.build
+++ b/lib/ring/meson.build
@@ -1,8 +1,8 @@
# SPDX-License-Identifier: BSD-3-Clause
# Copyright(c) 2017 Intel Corporation
-sources = files('rte_ring.c')
-headers = files('rte_ring.h')
+sources = files('rte_ring.c', 'rte_soring.c', 'soring.c')
+headers = files('rte_ring.h', 'rte_soring.h')
# most sub-headers are not for direct inclusion
indirect_headers += files (
'rte_ring_core.h',
diff --git a/lib/ring/rte_soring.c b/lib/ring/rte_soring.c
new file mode 100644
index 0000000000..b6bc71b8c9
--- /dev/null
+++ b/lib/ring/rte_soring.c
@@ -0,0 +1,182 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#include <inttypes.h>
+
+#include "soring.h"
+#include <rte_string_fns.h>
+
+RTE_LOG_REGISTER_DEFAULT(soring_logtype, INFO);
+#define RTE_LOGTYPE_SORING soring_logtype
+#define SORING_LOG(level, ...) \
+ RTE_LOG_LINE(level, SORING, "" __VA_ARGS__)
+
+static uint32_t
+soring_calc_elem_num(uint32_t count)
+{
+ return rte_align32pow2(count + 1);
+}
+
+static int
+soring_check_param(uint32_t esize, uint32_t stsize, uint32_t count,
+ uint32_t stages)
+{
+ if (stages == 0) {
+ SORING_LOG(ERR, "invalid number of stages: %u", stages);
+ return -EINVAL;
+ }
+
+ /* Check if element size is a multiple of 4B */
+ if (esize == 0 || esize % 4 != 0) {
+ SORING_LOG(ERR, "invalid element size: %u", esize);
+ return -EINVAL;
+ }
+
+ /* Check if ret-code size is a multiple of 4B */
+ if (stsize % 4 != 0) {
+ SORING_LOG(ERR, "invalid retcode size: %u", stsize);
+ return -EINVAL;
+ }
+
+ /* count must be a power of 2 */
+ if (rte_is_power_of_2(count) == 0 ||
+ (count > RTE_SORING_ELEM_MAX + 1)) {
+ SORING_LOG(ERR, "invalid number of elements: %u", count);
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+/*
+ * Calculate size offsets for SORING internal data layout.
+ */
+static size_t
+soring_get_szofs(uint32_t esize, uint32_t stsize, uint32_t count,
+ uint32_t stages, size_t *elst_ofs, size_t *state_ofs,
+ size_t *stage_ofs)
+{
+ size_t sz;
+ const struct rte_soring * const r = NULL;
+
+ sz = sizeof(r[0]) + (size_t)count * esize;
+ sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE);
+
+ if (elst_ofs != NULL)
+ *elst_ofs = sz;
+
+ sz = sz + (size_t)count * stsize;
+ sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE);
+
+ if (state_ofs != NULL)
+ *state_ofs = sz;
+
+ sz += sizeof(r->state[0]) * count;
+ sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE);
+
+ if (stage_ofs != NULL)
+ *stage_ofs = sz;
+
+ sz += sizeof(r->stage[0]) * stages;
+ sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE);
+
+ return sz;
+}
+
+static void
+soring_dump_stage_headtail(FILE *f, const char *prefix,
+ struct soring_stage *st)
+{
+ fprintf(f, "%stail.pos=%"PRIu32"\n", prefix, st->sht.tail.pos);
+ fprintf(f, "%stail.sync=%"PRIu32"\n", prefix, st->sht.tail.sync);
+ fprintf(f, "%shead=%"PRIu32"\n", prefix, st->sht.head);
+}
+
+void
+rte_soring_dump(FILE *f, const struct rte_soring *r)
+{
+ uint32_t i;
+ char buf[32];
+
+ if (f == NULL || r == NULL)
+ return;
+
+ fprintf(f, "soring <%s>@%p\n", r->name, r);
+ fprintf(f, " size=%"PRIu32"\n", r->size);
+ fprintf(f, " capacity=%"PRIu32"\n", r->capacity);
+ fprintf(f, " esize=%"PRIu32"\n", r->esize);
+ fprintf(f, " msize=%"PRIu32"\n", r->msize);
+ fprintf(f, " used=%u\n", rte_soring_count(r));
+ fprintf(f, " avail=%u\n", rte_soring_free_count(r));
+
+ rte_ring_headtail_dump(f, " cons.", &(r->cons.ht));
+ rte_ring_headtail_dump(f, " prod.", &(r->prod.ht));
+
+ fprintf(f, " nb_stage=%"PRIu32"\n", r->nb_stage);
+ for (i = 0; i < r->nb_stage; i++) {
+ snprintf(buf, sizeof(buf), " stage[%u].", i);
+ soring_dump_stage_headtail(f, buf, r->stage + i);
+ }
+}
+
+ssize_t
+rte_soring_get_memsize(const struct rte_soring_param *prm)
+{
+ int32_t rc;
+ uint32_t count;
+
+ count = soring_calc_elem_num(prm->elems);
+ rc = soring_check_param(prm->elem_size, prm->meta_size, count,
+ prm->stages);
+ if (rc != 0)
+ return rc;
+
+ return soring_get_szofs(prm->elem_size, prm->meta_size, count,
+ prm->stages, NULL, NULL, NULL);
+}
+
+int
+rte_soring_init(struct rte_soring *r, const struct rte_soring_param *prm)
+{
+ int32_t rc;
+ uint32_t n;
+ size_t meta_ofs, stage_ofs, state_ofs;
+
+ if (r == NULL || prm == NULL)
+ return -EINVAL;
+
+ n = soring_calc_elem_num(prm->elems);
+ rc = soring_check_param(prm->elem_size, prm->meta_size, n, prm->stages);
+ if (rc != 0)
+ return rc;
+
+ soring_get_szofs(prm->elem_size, prm->meta_size, n, prm->stages,
+ &meta_ofs, &state_ofs, &stage_ofs);
+
+ memset(r, 0, sizeof(*r));
+ rc = strlcpy(r->name, prm->name, sizeof(r->name));
+ if (rc < 0 || rc >= (int)sizeof(r->name))
+ return -ENAMETOOLONG;
+
+ r->size = n;
+ r->mask = r->size - 1;
+ r->capacity = prm->elems;
+ r->esize = prm->elem_size;
+ r->msize = prm->meta_size;
+
+ r->prod.ht.sync_type = prm->prod_synt;
+ r->cons.ht.sync_type = prm->cons_synt;
+
+ r->state = (union soring_state *)((uintptr_t)r + state_ofs);
+ memset(r->state, 0, sizeof(r->state[0]) * r->size);
+
+ r->stage = (struct soring_stage *)((uintptr_t)r + stage_ofs);
+ r->nb_stage = prm->stages;
+ memset(r->stage, 0, r->nb_stage * sizeof(r->stage[0]));
+
+ if (r->msize != 0)
+ r->meta = (void *)((uintptr_t)r + meta_ofs);
+
+ return 0;
+}
diff --git a/lib/ring/rte_soring.h b/lib/ring/rte_soring.h
new file mode 100644
index 0000000000..81e25db86b
--- /dev/null
+++ b/lib/ring/rte_soring.h
@@ -0,0 +1,547 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#ifndef _RTE_SORING_H_
+#define _RTE_SORING_H_
+
+/**
+ * @file
+ * This file contains definition of RTE soring (Staged Ordered Ring) public API.
+ * Brief description:
+ * enqueue/dequeue works the same as for conventional rte_ring:
+ * any rte_ring sync types can be used, etc.
+ * Plus there could be multiple 'stages'.
+ * For each stage there is an acquire (start) and release (finish) operation.
+ * after some elems are 'acquired' - user can safely assume that he has
+ * exclusive possession of these elems till 'release' for them is done.
+ * Note that right now user has to release exactly the same number of elems
+ * he acquired before.
+ * After 'release', elems can be 'acquired' by next stage and/or dequeued
+ * (in case of last stage).
+ */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <rte_ring.h>
+
+/* upper 2 bits are used for status */
+#define RTE_SORING_ST_BIT 30
+
+/* max possible number of elements in the soring */
+#define RTE_SORING_ELEM_MAX (RTE_BIT32(RTE_SORING_ST_BIT) - 1)
+
+struct rte_soring_param {
+ /** expected name of the ring */
+ const char *name;
+ /** number of elemnts in the ring */
+ uint32_t elems;
+ /** size of elements in the ring, must be a multiple of 4 */
+ uint32_t elem_size;
+ /**
+ * size of metadata for each elem, must be a multiple of 4.
+ * This parameter defines a size of supplementary and optional
+ * array of metadata associated with each object in the soring.
+ * While element size is configurable (see 'elem_size' parameter above),
+ * so user can specify it big enough to hold both object and its
+ * metadata together, for performance reasons it might be plausible
+ * to access them as separate arrays.
+ * Common usage scenario when such separation helps:
+ * enqueue() - writes to objects array
+ * acquire() - reads from objects array
+ * release() - writes to metadata array (as an example: return code)
+ * dequeue() - reads both objects and metadata array
+ */
+ uint32_t meta_size;
+ /** number of stages in the ring */
+ uint32_t stages;
+ /** sync type for producer */
+ enum rte_ring_sync_type prod_synt;
+ /** sync type for consumer */
+ enum rte_ring_sync_type cons_synt;
+};
+
+struct rte_soring;
+
+/**
+ * Calculate the memory size needed for a soring
+ *
+ * This function returns the number of bytes needed for a ring, given
+ * the expected parameters for it. This value is the sum of the size of
+ * the internal metadata and the size of the memory needed by the
+ * actual ring elements and theri rec-codes. The value is aligned to a cache
+ * line size.
+ *
+ * @param prm
+ * Pointer to the structure that contains soring creation paramers.
+ * @return
+ * - The memory size needed for the soring on success.
+ * - -EINVAL if provided paramer values are invalid.
+ */
+__rte_experimental
+ssize_t
+rte_soring_get_memsize(const struct rte_soring_param *prm);
+
+/**
+ * Initialize a soring structure.
+ *
+ * Initialize a soring structure in memory pointed by "r".
+ * The size of the memory area must be large enough to store the soring
+ * internal structures plus the objects and ret-code tables.
+ * It is strongly advised to use rte_ring_get_memsize() to get the
+ * appropriate size.
+ *
+ * @param r
+ * Pointer to the soring structure.
+ * @param prm
+ * Pointer to the structure that contains soring creation paramers.
+ * @return
+ * - 0 on success, or a negative error code.
+ */
+__rte_experimental
+int
+rte_soring_init(struct rte_soring *r, const struct rte_soring_param *prm);
+
+/**
+ * Return the total number of filled entries in a ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @return
+ * The number of entries in the ring.
+ */
+__rte_experimental
+unsigned int
+rte_soring_count(const struct rte_soring *r);
+
+/**
+ * Return the total number of unfilled entries in a ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @return
+ * The number of free entries in the ring.
+ */
+__rte_experimental
+unsigned int
+rte_soring_free_count(const struct rte_soring *r);
+
+/**
+ * Dump the status of the soring
+ *
+ * @param f
+ * A pointer to a file for output
+ * @param r
+ * Pointer to the soring structure.
+ */
+__rte_experimental
+void
+rte_soring_dump(FILE *f, const struct rte_soring *r);
+
+/**
+ * Enqueue several objects on the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to enqueue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param n
+ * The number of objects to add in the ring from the 'objs'.
+ * @param free_space
+ * if non-NULL, returns the amount of space in the ring after the
+ * enqueue operation has finished.
+ * @return
+ * - Actual number of objects enqueued, either 0 or @n.
+ */
+__rte_experimental
+uint32_t
+rte_soring_enqueue_bulk(struct rte_soring *r, const void *objs,
+ uint32_t n, uint32_t *free_space);
+
+/**
+ * Enqueue several objects plus metadata on the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to enqueue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each object to enqueue.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then 'meta' parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param n
+ * The number of objects to add in the ring from the 'objs'.
+ * @param free_space
+ * if non-NULL, returns the amount of space in the ring after the
+ * enqueue operation has finished.
+ * @return
+ * - Actual number of objects enqueued, either 0 or @n.
+ */
+__rte_experimental
+uint32_t
+rte_soring_enqueux_bulk(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, uint32_t *free_space);
+
+/**
+ * Enqueue several objects on the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to enqueue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param n
+ * The number of objects to add in the ring from the 'objs'.
+ * @param free_space
+ * if non-NULL, returns the amount of space in the ring after the
+ * enqueue operation has finished.
+ * @return
+ * - Actual number of objects enqueued.
+ */
+__rte_experimental
+uint32_t
+rte_soring_enqueue_burst(struct rte_soring *r, const void *objs,
+ uint32_t n, uint32_t *free_space);
+
+/**
+ * Enqueue several objects plus metadata on the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to enqueue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each object to enqueue.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then 'meta' parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param n
+ * The number of objects to add in the ring from the 'objs'.
+ * @param free_space
+ * if non-NULL, returns the amount of space in the ring after the
+ * enqueue operation has finished.
+ * @return
+ * - Actual number of objects enqueued.
+ */
+__rte_experimental
+uint32_t
+rte_soring_enqueux_burst(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, uint32_t *free_space);
+
+/**
+ * Dequeue several objects from the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to dequeue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param num
+ * The number of objects to dequeue from the ring into the objs.
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries after the
+ * dequeue has finished.
+ * @return
+ * - Actual number of objects dequeued, either 0 or 'num'.
+ */
+__rte_experimental
+uint32_t
+rte_soring_dequeue_bulk(struct rte_soring *r, void *objs,
+ uint32_t num, uint32_t *available);
+
+/**
+ * Dequeue several objects plus metadata from the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to dequeue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to array of metadata values for each object to dequeue.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then 'meta' parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param num
+ * The number of objects to dequeue from the ring into the objs.
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries after the
+ * dequeue has finished.
+ * @return
+ * - Actual number of objects dequeued, either 0 or 'num'.
+ */
+__rte_experimental
+uint32_t
+rte_soring_dequeux_bulk(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, uint32_t *available);
+
+/**
+ * Dequeue several objects from the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to dequeue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param num
+ * The number of objects to dequeue from the ring into the objs.
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries after the
+ * dequeue has finished.
+ * @return
+ * - Actual number of objects dequeued.
+ */
+__rte_experimental
+uint32_t
+rte_soring_dequeue_burst(struct rte_soring *r, void *objs,
+ uint32_t num, uint32_t *available);
+
+/**
+ * Dequeue several objects plus metadata from the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to dequeue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to array of metadata values for each object to dequeue.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then 'meta' parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param num
+ * The number of objects to dequeue from the ring into the objs.
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries after the
+ * dequeue has finished.
+ * @return
+ * - Actual number of objects dequeued.
+ */
+__rte_experimental
+uint32_t
+rte_soring_dequeux_burst(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, uint32_t *available);
+
+/**
+ * Acquire several objects from the ring for given stage.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to acquire.
+ * Size of objects must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param stage
+ * Stage to acquire objects for.
+ * @param num
+ * The number of objects to acquire.
+ * @param ftoken
+ * Pointer to the opaque 'token' value used by release() op.
+ * User has to store this value somewhere, and later provide to the
+ * release().
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries for given stage
+ * after the acquire has finished.
+ * @return
+ * - Actual number of objects acquired, either 0 or 'num'.
+ */
+__rte_experimental
+uint32_t
+rte_soring_acquire_bulk(struct rte_soring *r, void *objs,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available);
+
+/**
+ * Acquire several objects plus metadata from the ring for given stage.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to acquire.
+ * Size of objects must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each for each acquired object.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then 'meta' parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param stage
+ * Stage to acquire objects for.
+ * @param num
+ * The number of objects to acquire.
+ * @param ftoken
+ * Pointer to the opaque 'token' value used by release() op.
+ * User has to store this value somewhere, and later provide to the
+ * release().
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries for given stage
+ * after the acquire has finished.
+ * @return
+ * - Actual number of objects acquired, either 0 or 'num'.
+ */
+__rte_experimental
+uint32_t
+rte_soring_acquirx_bulk(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available);
+
+/**
+ * Acquire several objects from the ring for given stage.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to acquire.
+ * Size of objects must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param stage
+ * Stage to acquire objects for.
+ * @param num
+ * The number of objects to acquire.
+ * @param ftoken
+ * Pointer to the opaque 'token' value used by release() op.
+ * User has to store this value somewhere, and later provide to the
+ * release().
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries for given stage
+ * after the acquire has finished.
+ * @return
+ * - Actual number of objects acquired.
+ */
+__rte_experimental
+uint32_t
+rte_soring_acquire_burst(struct rte_soring *r, void *objs,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available);
+
+/**
+ * Acquire several objects plus metadata from the ring for given stage.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to acquire.
+ * Size of objects must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each for each acquired object.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then 'meta' parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param stage
+ * Stage to acquire objects for.
+ * @param num
+ * The number of objects to acquire.
+ * @param ftoken
+ * Pointer to the opaque 'token' value used by release() op.
+ * User has to store this value somewhere, and later provide to the
+ * release().
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries for given stage
+ * after the acquire has finished.
+ * @return
+ * - Actual number of objects acquired.
+ */
+__rte_experimental
+uint32_t
+rte_soring_acquirx_burst(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available);
+
+/**
+ * Release several objects for given stage back to the ring.
+ * Note that it means these objects become avaialble for next stage or
+ * dequeue.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to relase.
+ * Note that unless user needs to overwrite ring objects this parameter
+ * can be NULL.
+ * Size of objects must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param stage
+ * Current stage.
+ * @param n
+ * The number of objects to release.
+ * Has to be the same value as returned by acquire() op.
+ * @param ftoken
+ * Opaque 'token' value obtained from acquire() op.
+ * @return
+ * - None.
+ */
+__rte_experimental
+void
+rte_soring_release(struct rte_soring *r, const void *objs,
+ uint32_t stage, uint32_t n, uint32_t ftoken);
+
+/**
+ * Release several objects plus metadata for given stage back to the ring.
+ * Note that it means these objects become avaialble for next stage or
+ * dequeue.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to relase.
+ * Note that unless user needs to overwrite ring objects this parameter
+ * can be NULL.
+ * Size of objects must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each object to release.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then meta parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param stage
+ * Current stage.
+ * @param n
+ * The number of objects to release.
+ * Has to be the same value as returned by acquire() op.
+ * @param ftoken
+ * Opaque 'token' value obtained from acquire() op.
+ * @return
+ * - None.
+ */
+__rte_experimental
+void
+rte_soring_releasx(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t stage, uint32_t n, uint32_t ftoken);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _RTE_SORING_H_ */
diff --git a/lib/ring/soring.c b/lib/ring/soring.c
new file mode 100644
index 0000000000..97fa38df66
--- /dev/null
+++ b/lib/ring/soring.c
@@ -0,0 +1,548 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+/**
+ * @file
+ * This file contains implementation of SORING 'datapath' functions.
+ * Brief description:
+ * ==================
+ * enqueue/dequeue works the same as for conventional rte_ring:
+ * any rte_ring sync types can be used, etc.
+ * Plus there could be multiple 'stages'.
+ * For each stage there is an acquire (start) and release (finish) operation.
+ * After some elems are 'acquired' - user can safely assume that he has
+ * exclusive possession of these elems till 'release' for them is done.
+ * Note that right now user has to release exactly the same number of elems
+ * he acquired before.
+ * After 'release', elems can be 'acquired' by next stage and/or dequeued
+ * (in case of last stage).
+ * Internal structure:
+ * ===================
+ * In addition to 'normal' ring of elems, it also has a ring of states of the
+ * same size. Each state[] corresponds to exactly one elem[].
+ * state[] will be used by acquire/release/dequeue functions to store internal
+ * information and should not be accessed by the user directly.
+ * How it works:
+ * =============
+ * 'acquire()' just moves stage's head (same as rte_ring move_head does),
+ * plus it saves in state[stage.cur_head] information about how many elems
+ * were acquired, current head position and special flag value to indicate
+ * that elems are acquired (RTE_SORING_ST_START).
+ * Note that 'acquire()' returns to the user a special 'ftoken' that user has
+ * to provide for 'release()' (in fact it is just a position for current head).
+ * 'release()' extracts old head value from provided ftoken and checks that
+ * corresponding 'state[]' contains expected values(mostly for sanity
+ * purposes).
+ * * Then it marks this state[] with 'RTE_SORING_ST_FINISH' flag to indicate
+ * that given subset of objects was released.
+ * After that, it checks does old head value equals to current tail value?
+ * If yes, then it performs 'finalize()' operation, otherwise 'release()'
+ * just returns (without spinning on stage tail value).
+ * As updated state[] is shared by all threads, some other thread can do
+ * 'finalize()' for given stage.
+ * That allows 'release()' to avoid excessive waits on the tail value.
+ * Main purpose of 'finalize()' operation is to walk through 'state[]'
+ * from current stage tail up to its head, check state[] and move stage tail
+ * through elements that already are in RTE_SORING_ST_FINISH state.
+ * Along with that, corresponding state[] values are reset to zero.
+ * Note that 'finalize()' for given stage can be done from multiple places:
+ * 'release()' for that stage or from 'acquire()' for next stage
+ * even from consumer's 'dequeue()' - in case given stage is the last one.
+ * So 'finalize()' has to be MT-safe and inside it we have to
+ * guarantee that only one thread will update state[] and stage's tail values.
+ */
+
+#include "soring.h"
+
+/*
+ * Inline functions (fastpath) start here.
+ */
+static __rte_always_inline uint32_t
+__rte_soring_stage_finalize(struct soring_stage_headtail *sht,
+ union soring_state *rstate, uint32_t rmask, uint32_t maxn)
+{
+ int32_t rc;
+ uint32_t head, i, idx, k, n, tail;
+ union soring_stage_tail nt, ot;
+ union soring_state st;
+
+ /* try to grab exclusive right to update tail value */
+ ot.raw = rte_atomic_load_explicit(&sht->tail.raw,
+ rte_memory_order_acquire);
+
+ /* other thread already finalizing it for us */
+ if (ot.sync != 0)
+ return 0;
+
+ nt.pos = ot.pos;
+ nt.sync = 1;
+ rc = rte_atomic_compare_exchange_strong_explicit(&sht->tail.raw,
+ &ot.raw, nt.raw, rte_memory_order_acquire,
+ rte_memory_order_relaxed);
+
+ /* other thread won the race */
+ if (rc == 0)
+ return 0;
+
+ /*
+ * start with current tail and walk through states that are
+ * already finished.
+ */
+
+ head = rte_atomic_load_explicit(&sht->head, rte_memory_order_relaxed);
+ n = RTE_MIN(head - ot.pos, maxn);
+ for (i = 0, tail = ot.pos; i < n; i += k, tail += k) {
+
+ idx = tail & rmask;
+ st.raw = rte_atomic_load_explicit(&rstate[idx].raw,
+ rte_memory_order_relaxed);
+ if ((st.stnum & RTE_SORING_ST_MASK) != RTE_SORING_ST_FINISH ||
+ st.ftoken != tail)
+ break;
+
+ k = st.stnum & ~RTE_SORING_ST_MASK;
+ rte_atomic_store_explicit(&rstate[idx].raw, 0,
+ rte_memory_order_relaxed);
+ }
+
+
+ /* release exclusive right to update along with new tail value */
+ ot.pos = tail;
+ rte_atomic_store_explicit(&sht->tail.raw, ot.raw,
+ rte_memory_order_release);
+
+ return i;
+}
+
+static __rte_always_inline uint32_t
+__rte_soring_move_prod_head(struct rte_soring *r, uint32_t num,
+ enum rte_ring_queue_behavior behavior, enum rte_ring_sync_type st,
+ uint32_t *head, uint32_t *next, uint32_t *free)
+{
+ uint32_t n;
+
+ switch (st) {
+ case RTE_RING_SYNC_ST:
+ case RTE_RING_SYNC_MT:
+ n = __rte_ring_headtail_move_head(&r->prod.ht, &r->cons.ht,
+ r->capacity, st, num, behavior, head, next, free);
+ break;
+ case RTE_RING_SYNC_MT_RTS:
+ n = __rte_ring_rts_move_head(&r->prod.rts, &r->cons.ht,
+ r->capacity, num, behavior, head, free);
+ *next = *head + n;
+ break;
+ case RTE_RING_SYNC_MT_HTS:
+ n = __rte_ring_hts_move_head(&r->prod.hts, &r->cons.ht,
+ r->capacity, num, behavior, head, free);
+ *next = *head + n;
+ break;
+ default:
+ /* unsupported mode, shouldn't be here */
+ RTE_ASSERT(0);
+ *free = 0;
+ n = 0;
+ }
+
+ return n;
+}
+
+static __rte_always_inline uint32_t
+__rte_soring_move_cons_head(struct rte_soring *r, uint32_t stage, uint32_t num,
+ enum rte_ring_queue_behavior behavior, enum rte_ring_sync_type st,
+ uint32_t *head, uint32_t *next, uint32_t *avail)
+{
+ uint32_t n;
+
+ switch (st) {
+ case RTE_RING_SYNC_ST:
+ case RTE_RING_SYNC_MT:
+ n = __rte_ring_headtail_move_head(&r->cons.ht,
+ &r->stage[stage].ht, 0, st, num, behavior,
+ head, next, avail);
+ break;
+ case RTE_RING_SYNC_MT_RTS:
+ n = __rte_ring_rts_move_head(&r->cons.rts, &r->stage[stage].ht,
+ 0, num, behavior, head, avail);
+ *next = *head + n;
+ break;
+ case RTE_RING_SYNC_MT_HTS:
+ n = __rte_ring_hts_move_head(&r->cons.hts, &r->stage[stage].ht,
+ 0, num, behavior, head, avail);
+ *next = *head + n;
+ break;
+ default:
+ /* unsupported mode, shouldn't be here */
+ RTE_ASSERT(0);
+ *avail = 0;
+ n = 0;
+ }
+
+ return n;
+}
+
+static __rte_always_inline void
+__rte_soring_update_tail(struct __rte_ring_headtail *rht,
+ enum rte_ring_sync_type st, uint32_t head, uint32_t next, uint32_t enq)
+{
+ uint32_t n;
+
+ switch (st) {
+ case RTE_RING_SYNC_ST:
+ case RTE_RING_SYNC_MT:
+ __rte_ring_update_tail(&rht->ht, head, next, st, enq);
+ break;
+ case RTE_RING_SYNC_MT_RTS:
+ __rte_ring_rts_update_tail(&rht->rts);
+ break;
+ case RTE_RING_SYNC_MT_HTS:
+ n = next - head;
+ __rte_ring_hts_update_tail(&rht->hts, head, n, enq);
+ break;
+ default:
+ /* unsupported mode, shouldn't be here */
+ RTE_ASSERT(0);
+ }
+}
+
+static __rte_always_inline uint32_t
+__rte_soring_stage_move_head(struct soring_stage_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity, uint32_t num,
+ enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head, uint32_t *avail)
+{
+ uint32_t n, tail;
+
+ *old_head = rte_atomic_load_explicit(&d->head,
+ rte_memory_order_acquire);
+
+ do {
+ n = num;
+ tail = rte_atomic_load_explicit(&s->tail,
+ rte_memory_order_relaxed);
+ *avail = capacity + tail - *old_head;
+ if (n > *avail)
+ n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *avail;
+ if (n == 0)
+ return 0;
+ *new_head = *old_head + n;
+ } while (rte_atomic_compare_exchange_strong_explicit(&d->head,
+ old_head, *new_head, rte_memory_order_acquire,
+ rte_memory_order_acquire) == 0);
+
+ return n;
+}
+
+static __rte_always_inline uint32_t
+soring_enqueue(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, enum rte_ring_queue_behavior behavior,
+ uint32_t *free_space)
+{
+ enum rte_ring_sync_type st;
+ uint32_t nb_free, prod_head, prod_next;
+
+ RTE_ASSERT(r != NULL && r->nb_stage > 0);
+ RTE_ASSERT(meta == NULL || r->meta != NULL);
+
+ st = r->prod.ht.sync_type;
+
+ n = __rte_soring_move_prod_head(r, n, behavior, st,
+ &prod_head, &prod_next, &nb_free);
+ if (n != 0) {
+ __rte_ring_do_enqueue_elems(&r[1], objs, r->size,
+ prod_head & r->mask, r->esize, n);
+ if (meta != NULL)
+ __rte_ring_do_enqueue_elems(r->meta, meta, r->size,
+ prod_head & r->mask, r->msize, n);
+ __rte_soring_update_tail(&r->prod, st, prod_head, prod_next, 1);
+ }
+
+ if (free_space != NULL)
+ *free_space = nb_free - n;
+ return n;
+}
+
+static __rte_always_inline uint32_t
+soring_dequeue(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, enum rte_ring_queue_behavior behavior,
+ uint32_t *available)
+{
+ enum rte_ring_sync_type st;
+ uint32_t entries, cons_head, cons_next, n, ns, reqn;
+
+ RTE_ASSERT(r != NULL && r->nb_stage > 0);
+ RTE_ASSERT(meta == NULL || r->meta != NULL);
+
+ ns = r->nb_stage - 1;
+ st = r->cons.ht.sync_type;
+
+ /* try to grab exactly @num elems first */
+ n = __rte_soring_move_cons_head(r, ns, num, RTE_RING_QUEUE_FIXED, st,
+ &cons_head, &cons_next, &entries);
+ if (n == 0) {
+ /* try to finalize some elems from previous stage */
+ n = __rte_soring_stage_finalize(&r->stage[ns].sht,
+ r->state, r->mask, 2 * num);
+ entries += n;
+
+ /* repeat attempt to grab elems */
+ reqn = (behavior == RTE_RING_QUEUE_FIXED) ? num : 0;
+ if (entries >= reqn)
+ n = __rte_soring_move_cons_head(r, ns, num, behavior,
+ st, &cons_head, &cons_next, &entries);
+ else
+ n = 0;
+ }
+
+ /* we have some elems to consume */
+ if (n != 0) {
+ __rte_ring_do_dequeue_elems(objs, &r[1], r->size,
+ cons_head & r->mask, r->esize, n);
+ if (meta != NULL)
+ __rte_ring_do_dequeue_elems(meta, r->meta, r->size,
+ cons_head & r->mask, r->msize, n);
+ __rte_soring_update_tail(&r->cons, st, cons_head, cons_next, 0);
+ }
+
+ if (available != NULL)
+ *available = entries - n;
+ return n;
+}
+
+static __rte_always_inline uint32_t
+soring_acquire(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, enum rte_ring_queue_behavior behavior,
+ uint32_t *ftoken, uint32_t *available)
+{
+ uint32_t avail, head, idx, n, next, reqn;
+ union soring_state st;
+ struct soring_stage *pstg;
+ struct soring_stage_headtail *cons;
+
+ RTE_ASSERT(r != NULL && stage < r->nb_stage);
+ RTE_ASSERT(meta == NULL || r->meta != NULL);
+
+ cons = &r->stage[stage].sht;
+
+ if (stage == 0)
+ n = __rte_soring_stage_move_head(cons, &r->prod.ht, 0, num,
+ behavior, &head, &next, &avail);
+ else {
+ pstg = r->stage + stage - 1;
+
+ /* try to grab exactly @num elems */
+ n = __rte_soring_stage_move_head(cons, &pstg->ht, 0, num,
+ RTE_RING_QUEUE_FIXED, &head, &next, &avail);
+ if (n == 0) {
+ /* try to finalize some elems from previous stage */
+ n = __rte_soring_stage_finalize(&pstg->sht, r->state,
+ r->mask, 2 * num);
+ avail += n;
+
+ /* repeat attempt to grab elems */
+ reqn = (behavior == RTE_RING_QUEUE_FIXED) ? num : 0;
+ if (avail >= reqn)
+ n = __rte_soring_stage_move_head(cons,
+ &pstg->ht, 0, num, behavior, &head,
+ &next, &avail);
+ else
+ n = 0;
+ }
+ }
+
+ if (n != 0) {
+
+ idx = head & r->mask;
+
+ /* copy elems that are ready for given stage */
+ __rte_ring_do_dequeue_elems(objs, &r[1], r->size, idx,
+ r->esize, n);
+ if (meta != NULL)
+ __rte_ring_do_dequeue_elems(meta, r->meta,
+ r->size, idx, r->msize, n);
+
+ /* update status ring */
+ st.ftoken = head;
+ st.stnum = (RTE_SORING_ST_START | n);
+
+ rte_atomic_store_explicit(&r->state[idx].raw, st.raw,
+ rte_memory_order_relaxed);
+ *ftoken = head;
+ }
+
+ if (available != NULL)
+ *available = avail - n;
+ return n;
+}
+
+static __rte_always_inline void
+soring_release(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t stage, uint32_t n, uint32_t ftoken)
+{
+ uint32_t idx, tail;
+ struct soring_stage *stg;
+ union soring_state st;
+
+ RTE_ASSERT(r != NULL && stage < r->nb_stage);
+ RTE_ASSERT(meta == NULL || r->meta != NULL);
+
+ stg = r->stage + stage;
+ tail = rte_atomic_load_explicit(&stg->sht.tail.pos,
+ rte_memory_order_relaxed);
+
+ idx = ftoken & r->mask;
+ st.raw = rte_atomic_load_explicit(&r->state[idx].raw,
+ rte_memory_order_acquire);
+
+ /* check state ring contents */
+ RTE_VERIFY(st.stnum == (RTE_SORING_ST_START | n) &&
+ st.ftoken == ftoken);
+
+ /* update contents of the ring, if necessary */
+ if (objs != NULL)
+ __rte_ring_do_enqueue_elems(&r[1], objs, r->size, idx,
+ r->esize, n);
+ if (meta != NULL)
+ __rte_ring_do_enqueue_elems(r->meta, meta, r->size, idx,
+ r->msize, n);
+
+ /* set state to FINISH, make sure it is not reordered */
+ st.stnum = RTE_SORING_ST_FINISH | n;
+ rte_atomic_store_explicit(&r->state[idx].raw, st.raw,
+ rte_memory_order_release);
+
+ if (tail == ftoken)
+ __rte_soring_stage_finalize(&stg->sht, r->state, r->mask,
+ r->capacity);
+}
+
+/*
+ * Public functions (data-path) start here.
+ */
+
+void
+rte_soring_release(struct rte_soring *r, const void *objs,
+ uint32_t stage, uint32_t n, uint32_t ftoken)
+{
+ return soring_release(r, objs, NULL, stage, n, ftoken);
+}
+
+
+void
+rte_soring_releasx(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t stage, uint32_t n, uint32_t ftoken)
+{
+ return soring_release(r, objs, meta, stage, n, ftoken);
+}
+
+uint32_t
+rte_soring_enqueue_bulk(struct rte_soring *r, const void *objs, uint32_t n,
+ uint32_t *free_space)
+{
+ return soring_enqueue(r, objs, NULL, n, RTE_RING_QUEUE_FIXED,
+ free_space);
+}
+
+uint32_t
+rte_soring_enqueux_bulk(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, uint32_t *free_space)
+{
+ return soring_enqueue(r, objs, meta, n, RTE_RING_QUEUE_FIXED,
+ free_space);
+}
+
+uint32_t
+rte_soring_enqueue_burst(struct rte_soring *r, const void *objs, uint32_t n,
+ uint32_t *free_space)
+{
+ return soring_enqueue(r, objs, NULL, n, RTE_RING_QUEUE_VARIABLE,
+ free_space);
+}
+
+uint32_t
+rte_soring_enqueux_burst(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, uint32_t *free_space)
+{
+ return soring_enqueue(r, objs, meta, n, RTE_RING_QUEUE_VARIABLE,
+ free_space);
+}
+
+uint32_t
+rte_soring_dequeue_bulk(struct rte_soring *r, void *objs, uint32_t num,
+ uint32_t *available)
+{
+ return soring_dequeue(r, objs, NULL, num, RTE_RING_QUEUE_FIXED,
+ available);
+}
+
+uint32_t
+rte_soring_dequeux_bulk(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, uint32_t *available)
+{
+ return soring_dequeue(r, objs, meta, num, RTE_RING_QUEUE_FIXED,
+ available);
+}
+
+uint32_t
+rte_soring_dequeue_burst(struct rte_soring *r, void *objs, uint32_t num,
+ uint32_t *available)
+{
+ return soring_dequeue(r, objs, NULL, num, RTE_RING_QUEUE_VARIABLE,
+ available);
+}
+
+uint32_t
+rte_soring_dequeux_burst(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, uint32_t *available)
+{
+ return soring_dequeue(r, objs, meta, num, RTE_RING_QUEUE_VARIABLE,
+ available);
+}
+
+uint32_t
+rte_soring_acquire_bulk(struct rte_soring *r, void *objs,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available)
+{
+ return soring_acquire(r, objs, NULL, stage, num,
+ RTE_RING_QUEUE_FIXED, ftoken, available);
+}
+
+uint32_t
+rte_soring_acquirx_bulk(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available)
+{
+ return soring_acquire(r, objs, meta, stage, num,
+ RTE_RING_QUEUE_FIXED, ftoken, available);
+}
+
+uint32_t
+rte_soring_acquire_burst(struct rte_soring *r, void *objs,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available)
+{
+ return soring_acquire(r, objs, NULL, stage, num,
+ RTE_RING_QUEUE_VARIABLE, ftoken, available);
+}
+
+uint32_t
+rte_soring_acquirx_burst(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available)
+{
+ return soring_acquire(r, objs, meta, stage, num,
+ RTE_RING_QUEUE_VARIABLE, ftoken, available);
+}
+
+unsigned int
+rte_soring_count(const struct rte_soring *r)
+{
+ uint32_t prod_tail = r->prod.ht.tail;
+ uint32_t cons_tail = r->cons.ht.tail;
+ uint32_t count = (prod_tail - cons_tail) & r->mask;
+ return (count > r->capacity) ? r->capacity : count;
+}
+
+unsigned int
+rte_soring_free_count(const struct rte_soring *r)
+{
+ return r->capacity - rte_soring_count(r);
+}
diff --git a/lib/ring/soring.h b/lib/ring/soring.h
new file mode 100644
index 0000000000..c44d8912d9
--- /dev/null
+++ b/lib/ring/soring.h
@@ -0,0 +1,124 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#ifndef _SORING_H_
+#define _SORING_H_
+
+/**
+ * @file
+ * This file contains internal strctures of RTE soring: Staged Ordered Ring.
+ * Sort of extension of conventional RTE ring.
+ * Internal structure:
+ * In addition to 'normal' ring of elems, it also has a ring of states of the
+ * same size. Each state[] corresponds to exactly one elem[].
+ * state[] will be used by acquire/release/dequeue functions to store internal
+ * information and should not be accessed by the user directly.
+ * For actual implementation details, please refer to soring.c
+ */
+
+#include <rte_soring.h>
+
+union soring_state {
+ alignas(sizeof(uint64_t)) RTE_ATOMIC(uint64_t) raw;
+ struct {
+ RTE_ATOMIC(uint32_t) ftoken;
+ RTE_ATOMIC(uint32_t) stnum;
+ };
+};
+
+/* upper 2 bits are used for status */
+#define RTE_SORING_ST_START RTE_SHIFT_VAL32(1, RTE_SORING_ST_BIT)
+#define RTE_SORING_ST_FINISH RTE_SHIFT_VAL32(2, RTE_SORING_ST_BIT)
+
+#define RTE_SORING_ST_MASK \
+ RTE_GENMASK32(sizeof(uint32_t) * CHAR_BIT - 1, RTE_SORING_ST_BIT)
+
+/**
+ * SORING re-uses rte_ring internal structures and implementation
+ * for enqueue/dequeue operations.
+ */
+struct __rte_ring_headtail {
+
+ union __rte_cache_aligned {
+ struct rte_ring_headtail ht;
+ struct rte_ring_hts_headtail hts;
+ struct rte_ring_rts_headtail rts;
+ };
+
+ RTE_CACHE_GUARD;
+};
+
+union soring_stage_tail {
+ /** raw 8B value to read/write *sync* and *pos* as one atomic op */
+ alignas(sizeof(uint64_t)) RTE_ATOMIC(uint64_t) raw;
+ struct {
+ RTE_ATOMIC(uint32_t) sync;
+ RTE_ATOMIC(uint32_t) pos;
+ };
+};
+
+struct soring_stage_headtail {
+ volatile union soring_stage_tail tail;
+ enum rte_ring_sync_type unused; /**< unused */
+ volatile RTE_ATOMIC(uint32_t) head;
+};
+
+/**
+ * SORING stage head_tail structure is 'compatible' with rte_ring ones.
+ * rte_ring internal functions for moving producer/consumer head
+ * can work transparently with stage head_tail and visa-versa
+ * (no modifications required).
+ */
+struct soring_stage {
+
+ union __rte_cache_aligned {
+ struct rte_ring_headtail ht;
+ struct soring_stage_headtail sht;
+ };
+
+ RTE_CACHE_GUARD;
+};
+
+/**
+ * RTE soring internal structure.
+ * As with rte_ring actual elements array supposed to be located direclty
+ * after the rte_soring structure.
+ */
+struct __rte_cache_aligned rte_soring {
+ uint32_t size; /**< Size of ring. */
+ uint32_t mask; /**< Mask (size-1) of ring. */
+ uint32_t capacity; /**< Usable size of ring */
+ uint32_t esize;
+ /**< size of elements in the ring, must be a multiple of 4*/
+ uint32_t msize;
+ /**< size of metadata value for each elem, must be a multiple of 4 */
+
+ /** Ring stages */
+ struct soring_stage *stage;
+ uint32_t nb_stage;
+
+ /** Ring of states (one per element) */
+ union soring_state *state;
+
+ /** Pointer to the buffer where metadata values for each elements
+ * are stored. This is supplementary and optional information that
+ * user can attach to each element of the ring.
+ * While it is possible to incorporate this information inside
+ * user-defined element, in many cases it is plausible to maintain it
+ * as a separate array (mainly for performance reasons).
+ */
+ void *meta;
+
+ RTE_CACHE_GUARD;
+
+ /** Ring producer status. */
+ struct __rte_ring_headtail prod;
+
+ /** Ring consumer status. */
+ struct __rte_ring_headtail cons;
+
+ char name[RTE_RING_NAMESIZE]; /**< Name of the ring. */
+};
+
+#endif /* _SORING_H_ */
diff --git a/lib/ring/version.map b/lib/ring/version.map
index 61f7464f5a..081034c30e 100644
--- a/lib/ring/version.map
+++ b/lib/ring/version.map
@@ -20,4 +20,23 @@ EXPERIMENTAL {
# added in 24.11
rte_ring_headtail_dump;
+ rte_soring_acquire_bulk;
+ rte_soring_acquire_burst;
+ rte_soring_acquirx_bulk;
+ rte_soring_acquirx_burst;
+ rte_soring_count;
+ rte_soring_dequeue_bulk;
+ rte_soring_dequeue_burst;
+ rte_soring_dequeux_bulk;
+ rte_soring_dequeux_burst;
+ rte_soring_enqueue_bulk;
+ rte_soring_enqueue_burst;
+ rte_soring_enqueux_bulk;
+ rte_soring_enqueux_burst;
+ rte_soring_dump;
+ rte_soring_free_count;
+ rte_soring_get_memsize;
+ rte_soring_init;
+ rte_soring_release;
+ rte_soring_releasx;
};
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [PATCH v3 5/5] app/test: add unit tests for soring API
2024-09-16 12:37 ` [PATCH v3 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
` (3 preceding siblings ...)
2024-09-16 12:37 ` [PATCH v3 4/5] ring/soring: introduce Staged Ordered Ring Konstantin Ananyev
@ 2024-09-16 12:37 ` Konstantin Ananyev
2024-09-17 12:09 ` [PATCH v4 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
5 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-16 12:37 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, Konstantin Ananyev, Eimear Morrissey
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
Add both functional and stess test-cases for soring API.
Stress test serves as both functional and performance test of soring
enqueue/dequeue/acquire/release operations under high contention
(for both over committed and non-over committed scenarios).
Signed-off-by: Eimear Morrissey <eimear.morrissey@huawei.com>
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
---
app/test/meson.build | 3 +
app/test/test_soring.c | 442 +++++++++++++++
app/test/test_soring_mt_stress.c | 40 ++
app/test/test_soring_stress.c | 48 ++
app/test/test_soring_stress.h | 35 ++
app/test/test_soring_stress_impl.h | 827 +++++++++++++++++++++++++++++
6 files changed, 1395 insertions(+)
create mode 100644 app/test/test_soring.c
create mode 100644 app/test/test_soring_mt_stress.c
create mode 100644 app/test/test_soring_stress.c
create mode 100644 app/test/test_soring_stress.h
create mode 100644 app/test/test_soring_stress_impl.h
diff --git a/app/test/meson.build b/app/test/meson.build
index e29258e6ec..c290162e43 100644
--- a/app/test/meson.build
+++ b/app/test/meson.build
@@ -175,6 +175,9 @@ source_file_deps = {
'test_security_proto.c' : ['cryptodev', 'security'],
'test_seqlock.c': [],
'test_service_cores.c': [],
+ 'test_soring.c': [],
+ 'test_soring_mt_stress.c': [],
+ 'test_soring_stress.c': [],
'test_spinlock.c': [],
'test_stack.c': ['stack'],
'test_stack_perf.c': ['stack'],
diff --git a/app/test/test_soring.c b/app/test/test_soring.c
new file mode 100644
index 0000000000..b2110305a7
--- /dev/null
+++ b/app/test/test_soring.c
@@ -0,0 +1,442 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#include <string.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <inttypes.h>
+#include <errno.h>
+#include <sys/queue.h>
+
+#include <rte_common.h>
+#include <rte_log.h>
+#include <rte_memory.h>
+#include <rte_launch.h>
+#include <rte_cycles.h>
+#include <rte_eal.h>
+#include <rte_per_lcore.h>
+#include <rte_lcore.h>
+#include <rte_branch_prediction.h>
+#include <rte_malloc.h>
+#include <rte_random.h>
+#include <rte_errno.h>
+#include <rte_hexdump.h>
+
+#include <rte_soring.h>
+
+#include "test.h"
+
+#define MAX_ACQUIRED 20
+
+#define SORING_TEST_ASSERT(val, expected) do { \
+ RTE_TEST_ASSERT(expected == val, \
+ "%s: expected %u got %u\n", #val, expected, val); \
+} while (0)
+
+static void
+set_soring_init_param(struct rte_soring_param *prm,
+ const char *name, uint32_t esize, uint32_t elems,
+ uint32_t stages, uint32_t stsize,
+ enum rte_ring_sync_type rst_prod,
+ enum rte_ring_sync_type rst_cons)
+{
+ prm->name = name;
+ prm->elem_size = esize;
+ prm->elems = elems;
+ prm->stages = stages;
+ prm->meta_size = stsize;
+ prm->prod_synt = rst_prod;
+ prm->cons_synt = rst_cons;
+}
+
+static int
+move_forward_stage(struct rte_soring *sor,
+ uint32_t num_packets, uint32_t stage)
+{
+ uint32_t acquired;
+ uint32_t ftoken;
+ uint32_t *acquired_objs[MAX_ACQUIRED];
+
+ acquired = rte_soring_acquire_bulk(sor, acquired_objs, stage,
+ num_packets, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, num_packets);
+ rte_soring_release(sor, NULL, stage, num_packets,
+ ftoken);
+
+ return 0;
+}
+
+/*
+ * struct rte_soring_param param checking.
+ */
+static int
+test_soring_init(void)
+{
+ struct rte_soring *sor = NULL;
+ struct rte_soring_param prm;
+ int rc;
+ size_t sz;
+ memset(&prm, 0, sizeof(prm));
+
+/*init memory*/
+ set_soring_init_param(&prm, "alloc_memory", sizeof(uintptr_t),
+ 4, 1, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ sor = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ RTE_TEST_ASSERT_NOT_NULL(sor, "could not allocate memory for soring");
+
+ set_soring_init_param(&prm, "test_invalid_stages", sizeof(uintptr_t),
+ 4, 0, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with invalid num stages");
+
+ set_soring_init_param(&prm, "test_invalid_esize", 0,
+ 4, 1, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with 0 esize");
+
+ set_soring_init_param(&prm, "test_invalid_esize", 9,
+ 4, 1, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with esize not multiple of 4");
+
+ set_soring_init_param(&prm, "test_invalid_rsize", sizeof(uintptr_t),
+ 4, 1, 3, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with rcsize not multiple of 4");
+
+ set_soring_init_param(&prm, "test_invalid_elems", sizeof(uintptr_t),
+ RTE_SORING_ELEM_MAX + 1, 1, 4, RTE_RING_SYNC_MT,
+ RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with invalid num elements");
+
+ rte_free(sor);
+ return 0;
+}
+
+static int
+test_soring_get_memsize(void)
+{
+
+ struct rte_soring_param prm;
+ ssize_t sz;
+
+ set_soring_init_param(&prm, "memsize", sizeof(uint32_t *),
+ 10, 1, 0, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ RTE_TEST_ASSERT(sz > 0, "failed to calculate size");
+
+ set_soring_init_param(&prm, "memsize", sizeof(uint8_t),
+ 16, UINT32_MAX, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ RTE_TEST_ASSERT_EQUAL(sz, -EINVAL, "calcuated size incorrect");
+
+ set_soring_init_param(&prm, "memsize", 0,
+ 16, UINT32_MAX, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ RTE_TEST_ASSERT_EQUAL(sz, -EINVAL, "calculated size incorrect");
+
+ return 0;
+
+}
+
+static int
+test_soring_stages(void)
+{
+ struct rte_soring *sor = NULL;
+ struct rte_soring_param prm;
+ uint32_t objs[32];
+ uint32_t rcs[32];
+ uint32_t acquired_objs[32];
+ uint32_t acquired_rcs[32];
+ uint32_t dequeued_rcs[32];
+ uint32_t dequeued_objs[32];
+ size_t ssz;
+ uint32_t stage, enqueued, dequeued, acquired;
+ uint32_t i, ftoken;
+
+ memset(&prm, 0, sizeof(prm));
+ set_soring_init_param(&prm, "stages", sizeof(uint32_t), 32,
+ 10000, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ ssz = rte_soring_get_memsize(&prm);
+ RTE_TEST_ASSERT(ssz > 0, "parameter error calculating ring size");
+ sor = rte_zmalloc(NULL, ssz, RTE_CACHE_LINE_SIZE);
+ RTE_TEST_ASSERT_NOT_NULL(sor, "couldn't allocate memory for soring");
+ rte_soring_init(sor, &prm);
+
+ for (i = 0; i < 32; i++) {
+ rcs[i] = i;
+ objs[i] = i + 32;
+ }
+
+ enqueued = rte_soring_enqueux_burst(sor, objs, rcs, 32, NULL);
+ SORING_TEST_ASSERT(enqueued, 32);
+
+ for (stage = 0; stage < 10000; stage++) {
+ int j;
+ dequeued = rte_soring_dequeue_bulk(sor, dequeued_objs,
+ 32, NULL);
+ /* check none at end stage */
+ SORING_TEST_ASSERT(dequeued, 0);
+
+ acquired = rte_soring_acquirx_bulk(sor, acquired_objs,
+ acquired_rcs, stage, 32, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, 32);
+
+ for (j = 0; j < 32; j++) {
+ SORING_TEST_ASSERT(acquired_rcs[j], j + stage);
+ SORING_TEST_ASSERT(acquired_objs[j], j + stage + 32);
+ /* modify both queue object and rc */
+ acquired_objs[j]++;
+ acquired_rcs[j]++;
+ }
+
+ rte_soring_releasx(sor, acquired_objs,
+ acquired_rcs, stage, 32,
+ ftoken);
+ }
+
+ dequeued = rte_soring_dequeux_bulk(sor, dequeued_objs, dequeued_rcs,
+ 32, NULL);
+ SORING_TEST_ASSERT(dequeued, 32);
+ for (i = 0; i < 32; i++) {
+ /* ensure we ended up with the expected values in order*/
+ SORING_TEST_ASSERT(dequeued_rcs[i], i + 10000);
+ SORING_TEST_ASSERT(dequeued_objs[i], i + 32 + 10000);
+ }
+ rte_free(sor);
+ return 0;
+}
+
+static int
+test_soring_enqueue_dequeue(void)
+{
+ struct rte_soring *sor = NULL;
+ struct rte_soring_param prm;
+ int rc;
+ uint32_t i;
+ size_t sz;
+ uint32_t queue_objs[10];
+ uint32_t *queue_objs_p[10];
+ uint32_t free_space;
+ uint32_t enqueued, dequeued;
+ uint32_t *dequeued_objs[10];
+
+ memset(&prm, 0, sizeof(prm));
+ for (i = 0; i < 10; i++) {
+ queue_objs[i] = i + 1;
+ queue_objs_p[i] = &queue_objs[i];
+ }
+
+/*init memory*/
+ set_soring_init_param(&prm, "enqueue/dequeue", sizeof(uint32_t *),
+ 10, 1, 0, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ sor = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ RTE_TEST_ASSERT_NOT_NULL(sor, "alloc failed for soring");
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_SUCCESS(rc, "Failed to init soring");
+
+ free_space = 0;
+
+ enqueued = rte_soring_enqueue_burst(sor, queue_objs_p, 5, &free_space);
+
+ SORING_TEST_ASSERT(free_space, 5);
+ SORING_TEST_ASSERT(enqueued, 5);
+
+ /* fixed amount enqueue */
+ enqueued = rte_soring_enqueue_bulk(sor, queue_objs_p, 7, &free_space);
+
+ SORING_TEST_ASSERT(free_space, 5);
+ SORING_TEST_ASSERT(enqueued, 0);
+
+ /* variable amount enqueue */
+ enqueued = rte_soring_enqueue_burst(sor, queue_objs_p + 5, 7,
+ &free_space);
+ SORING_TEST_ASSERT(free_space, 0);
+ SORING_TEST_ASSERT(enqueued, 5);
+
+ /* test no dequeue while stage 0 has not completed */
+ dequeued = rte_soring_dequeue_bulk(sor, dequeued_objs, 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 0);
+ dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 0);
+
+ move_forward_stage(sor, 8, 0);
+
+ /* can only dequeue as many as have completed stage*/
+ dequeued = rte_soring_dequeue_bulk(sor, dequeued_objs, 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 0);
+ dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 8);
+ /* packets remain in order */
+ for (i = 0; i < dequeued; i++) {
+ RTE_TEST_ASSERT_EQUAL(dequeued_objs[i],
+ queue_objs_p[i], "dequeued != enqueued");
+ }
+
+ dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 1, NULL);
+ SORING_TEST_ASSERT(dequeued, 0);
+
+ move_forward_stage(sor, 2, 0);
+ dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 2, NULL);
+ SORING_TEST_ASSERT(dequeued, 2);
+ /* packets remain in order */
+ for (i = 0; i < dequeued; i++) {
+ RTE_TEST_ASSERT_EQUAL(dequeued_objs[i],
+ queue_objs_p[i + 8], "dequeued != enqueued");
+ }
+
+ rte_soring_dump(stdout, sor);
+ rte_free(sor);
+ return 0;
+}
+
+static int
+test_soring_acquire_release(void)
+{
+
+ struct rte_soring *sor = NULL;
+ struct rte_soring_param prm;
+ int rc, i;
+ size_t sz;
+
+ memset(&prm, 0, sizeof(prm));
+ uint32_t queue_objs[10];
+ uint32_t rc_objs[10];
+ uint32_t acquired_objs[10];
+ uint32_t dequeued_objs[10];
+ uint32_t dequeued_rcs[10];
+ uint32_t s1_acquired_rcs[10];
+ uint32_t free_space, enqueued, ftoken, acquired, dequeued;
+
+ for (i = 0; i < 10; i++) {
+ queue_objs[i] = i + 5;
+ rc_objs[i] = i + 10;
+ }
+
+/*init memory*/
+ set_soring_init_param(&prm, "test_acquire_release", sizeof(uint32_t),
+ 20, 2, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ sor = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ if (sor == NULL) {
+ printf("%s: alloc(%zu) for FIFO with %u elems failed",
+ __func__, sz, prm.elems);
+ return -ENOMEM;
+ }
+
+ /* init ring */
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_SUCCESS(rc, "failed to init soring");
+
+ /* enqueue with associated rc */
+ enqueued = rte_soring_enqueux_burst(sor, queue_objs, rc_objs, 5,
+ &free_space);
+ SORING_TEST_ASSERT(enqueued, 5);
+ /* enqueue without associated rc */
+ enqueued = rte_soring_enqueue_burst(sor, queue_objs + 5, 5,
+ &free_space);
+ SORING_TEST_ASSERT(enqueued, 5);
+
+ /* acquire the objects with rc's and ensure they are as expected */
+ acquired = rte_soring_acquirx_burst(sor, acquired_objs,
+ s1_acquired_rcs, 0, 5, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, 5);
+ for (i = 0; i < 5; i++) {
+ RTE_TEST_ASSERT_EQUAL(s1_acquired_rcs[i], rc_objs[i],
+ "acquired rc[%d]: %u != enqueued rc: %u",
+ i, s1_acquired_rcs[i], rc_objs[i]);
+ RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i],
+ "acquired obj[%d]: %u != enqueued obj %u",
+ i, acquired_objs[i], queue_objs[i]);
+ }
+ rte_soring_release(sor, NULL, 0, 5, ftoken);
+
+ /* acquire the objects without rc's and ensure they are as expected */
+ acquired = rte_soring_acquirx_burst(sor, acquired_objs,
+ s1_acquired_rcs, 0, 5, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, 5);
+ for (i = 0; i < 5; i++) {
+ /* as the rc area of memory is zero'd at init this is true
+ * but this is a detail of implementation rather than
+ * a guarantee.
+ */
+ RTE_TEST_ASSERT_EQUAL(s1_acquired_rcs[i], 0,
+ "acquired rc not empty");
+ RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i + 5],
+ "acquired obj[%d]: %u != enqueued obj %u",
+ i, acquired_objs[i], queue_objs[i + 5]);
+ }
+ /*release the objects, adding rc's */
+ rte_soring_releasx(sor, NULL, rc_objs + 5, 0, 5,
+ ftoken);
+
+ acquired = rte_soring_acquirx_burst(sor, acquired_objs,
+ s1_acquired_rcs, 1, 10, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, 10);
+ for (i = 0; i < 10; i++) {
+ /* ensure the associated rc's are the ones added at relase */
+ RTE_TEST_ASSERT_EQUAL(s1_acquired_rcs[i], rc_objs[i],
+ "acquired rc[%d]: %u != enqueued rc: %u",
+ i, s1_acquired_rcs[i], rc_objs[i]);
+ RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i],
+ "acquired obj[%d]: %u != enqueued obj %u",
+ i, acquired_objs[i], queue_objs[i]);
+ }
+ /*release the objects, with rc's set to NULL */
+ rte_soring_release(sor, NULL, 1, 10, ftoken);
+
+ dequeued = rte_soring_dequeux_burst(sor, dequeued_objs, dequeued_rcs,
+ 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 10);
+ for (i = 0; i < 10; i++) {
+ /* ensure the associated rc's are the ones added at relase */
+ RTE_TEST_ASSERT_EQUAL(dequeued_rcs[i], rc_objs[i],
+ "dequeued rc[%d]: %u != enqueued rc: %u",
+ i, dequeued_rcs[i], rc_objs[i]);
+ RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i],
+ "acquired obj[%d]: %u != enqueued obj %u",
+ i, acquired_objs[i], queue_objs[i]);
+ }
+ rte_soring_dump(stdout, sor);
+ rte_free(sor);
+ return 0;
+}
+
+static int
+test_soring(void)
+{
+
+ /* Negative test cases */
+ if (test_soring_init() < 0)
+ goto test_fail;
+
+ /* Memory calculations */
+ if (test_soring_get_memsize() < 0)
+ goto test_fail;
+
+ /* Basic enqueue/dequeue operations */
+ if (test_soring_enqueue_dequeue() < 0)
+ goto test_fail;
+
+ /* Acquire/release */
+ if (test_soring_acquire_release() < 0)
+ goto test_fail;
+
+ /* Test large number of stages */
+ if (test_soring_stages() < 0)
+ goto test_fail;
+
+ return 0;
+
+test_fail:
+ return -1;
+}
+
+REGISTER_FAST_TEST(soring_autotest, true, true, test_soring);
+
diff --git a/app/test/test_soring_mt_stress.c b/app/test/test_soring_mt_stress.c
new file mode 100644
index 0000000000..2f90bb4598
--- /dev/null
+++ b/app/test/test_soring_mt_stress.c
@@ -0,0 +1,40 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#include "test_soring_stress_impl.h"
+
+static inline uint32_t
+_st_ring_dequeue_burst(struct rte_soring *r, void **obj, uint32_t n,
+ uint32_t *avail)
+{
+ return rte_soring_dequeue_burst(r, obj, n, avail);
+}
+
+static inline uint32_t
+_st_ring_enqueue_bulk(struct rte_soring *r, void * const *obj, uint32_t n,
+ uint32_t *free)
+{
+ return rte_soring_enqueue_bulk(r, obj, n, free);
+}
+
+static inline uint32_t
+_st_ring_acquire_burst(struct rte_soring *r, uint32_t stage, void **obj,
+ uint32_t num, uint32_t *token, uint32_t *avail)
+{
+ return rte_soring_acquire_burst(r, obj, stage, num, token, avail);
+}
+
+static inline void
+_st_ring_release(struct rte_soring *r, uint32_t stage, uint32_t token,
+ void * const *obj, uint32_t num)
+{
+ RTE_SET_USED(obj);
+ rte_soring_release(r, NULL, stage, num, token);
+}
+
+const struct test test_soring_mt_stress = {
+ .name = "MT",
+ .nb_case = RTE_DIM(tests),
+ .cases = tests,
+};
diff --git a/app/test/test_soring_stress.c b/app/test/test_soring_stress.c
new file mode 100644
index 0000000000..334af6a29c
--- /dev/null
+++ b/app/test/test_soring_stress.c
@@ -0,0 +1,48 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#include "test_soring_stress.h"
+
+static int
+run_test(const struct test *test)
+{
+ int32_t rc;
+ uint32_t i, k;
+
+ for (i = 0, k = 0; i != test->nb_case; i++) {
+
+ printf("TEST-CASE %s %s START\n",
+ test->name, test->cases[i].name);
+
+ rc = test->cases[i].func(test->cases[i].wfunc);
+ k += (rc == 0);
+
+ if (rc != 0)
+ printf("TEST-CASE %s %s FAILED\n",
+ test->name, test->cases[i].name);
+ else
+ printf("TEST-CASE %s %s OK\n",
+ test->name, test->cases[i].name);
+ }
+
+ return k;
+}
+
+static int
+test_ring_stress(void)
+{
+ uint32_t n, k;
+
+ n = 0;
+ k = 0;
+
+ n += test_soring_mt_stress.nb_case;
+ k += run_test(&test_soring_mt_stress);
+
+ printf("Number of tests:\t%u\nSuccess:\t%u\nFailed:\t%u\n",
+ n, k, n - k);
+ return (k != n);
+}
+
+REGISTER_TEST_COMMAND(soring_stress_autotest, test_ring_stress);
diff --git a/app/test/test_soring_stress.h b/app/test/test_soring_stress.h
new file mode 100644
index 0000000000..6190e96117
--- /dev/null
+++ b/app/test/test_soring_stress.h
@@ -0,0 +1,35 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+
+#include <inttypes.h>
+#include <stddef.h>
+#include <stdalign.h>
+#include <string.h>
+#include <stdio.h>
+#include <unistd.h>
+
+#include <rte_soring.h>
+#include <rte_cycles.h>
+#include <rte_launch.h>
+#include <rte_pause.h>
+#include <rte_random.h>
+#include <rte_malloc.h>
+#include <rte_spinlock.h>
+
+#include "test.h"
+
+struct test_case {
+ const char *name;
+ int (*func)(int (*)(void *));
+ int (*wfunc)(void *arg);
+};
+
+struct test {
+ const char *name;
+ uint32_t nb_case;
+ const struct test_case *cases;
+};
+
+extern const struct test test_soring_mt_stress;
diff --git a/app/test/test_soring_stress_impl.h b/app/test/test_soring_stress_impl.h
new file mode 100644
index 0000000000..1ba78a31f5
--- /dev/null
+++ b/app/test/test_soring_stress_impl.h
@@ -0,0 +1,827 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#include <stdalign.h>
+
+#include "test_soring_stress.h"
+
+/**
+ * Stress test for soring enqueue/dequeue/acquire/release operations.
+ * Depending on the role, performs at least one of the following patterns
+ * on each worker:
+ * - dequeue/read-write data from/to the dequeued objects/enqueue.
+ * - acquire/read-write data from/to the acquired objects/release.
+ * Serves as both functional and performance test of soring
+ * data-path API under high contention
+ * (for both over committed and non-over committed scenarios).
+ */
+
+#define RING_NAME "SORING_STRESS"
+#define BULK_NUM 32
+#define RING_SIZE (2 * BULK_NUM * RTE_MAX_LCORE)
+
+#define MAX_STAGES 16
+
+enum {
+ WRK_CMD_STOP,
+ WRK_CMD_RUN,
+};
+
+static uint32_t wrk_cmd __rte_cache_aligned = WRK_CMD_STOP;
+
+/* test run-time in seconds */
+static const uint32_t run_time = 60;
+static const uint32_t verbose;
+
+struct lcore_op_stat {
+ uint64_t nb_lcore;
+ uint64_t nb_call;
+ uint64_t nb_obj;
+ uint64_t nb_cycle;
+ uint64_t max_cycle;
+ uint64_t min_cycle;
+};
+
+struct lcore_stat {
+ uint64_t nb_cycle;
+ struct lcore_op_stat deqenq;
+ uint32_t role_mask;
+ uint32_t nb_stage;
+ struct lcore_op_stat stage[MAX_STAGES];
+};
+
+#define ROLE_DEQENQ RTE_BIT32(0)
+#define ROLE_STAGE(n) RTE_BIT32(n + 1)
+
+struct lcore_arg {
+ struct rte_soring *rng;
+ struct lcore_stat stats;
+} __rte_cache_aligned;
+
+struct ring_elem {
+ uint32_t cnt[RTE_CACHE_LINE_SIZE / sizeof(uint32_t)];
+} __rte_cache_aligned;
+
+/*
+ * redefinable functions
+ */
+
+static uint32_t
+_st_ring_dequeue_burst(struct rte_soring *r, void **obj, uint32_t n,
+ uint32_t *avail);
+
+static uint32_t
+_st_ring_enqueue_bulk(struct rte_soring *r, void * const *obj, uint32_t n,
+ uint32_t *free);
+
+static uint32_t
+_st_ring_acquire_burst(struct rte_soring *r, uint32_t stage, void **obj,
+ uint32_t num, uint32_t *token, uint32_t *avail);
+
+static void
+_st_ring_release(struct rte_soring *r, uint32_t stage, uint32_t token,
+ void * const *obj, uint32_t num);
+
+static void
+lcore_op_stat_update(struct lcore_op_stat *ls, uint64_t call, uint64_t obj,
+ uint64_t tm, int32_t prcs)
+{
+ ls->nb_call += call;
+ ls->nb_obj += obj;
+ ls->nb_cycle += tm;
+ if (prcs) {
+ ls->max_cycle = RTE_MAX(ls->max_cycle, tm);
+ ls->min_cycle = RTE_MIN(ls->min_cycle, tm);
+ }
+}
+
+static void
+lcore_stat_update(struct lcore_stat *ls, uint64_t call, uint64_t obj,
+ uint64_t tm, int32_t prcs)
+{
+ uint32_t i;
+
+ ls->nb_cycle += tm;
+ lcore_op_stat_update(&ls->deqenq, call, obj, tm, prcs);
+ for (i = 0; i != ls->nb_stage; i++)
+ lcore_op_stat_update(ls->stage + i, call, obj, tm, prcs);
+}
+
+static void
+lcore_op_stat_aggr(struct lcore_op_stat *ms, const struct lcore_op_stat *ls)
+{
+ ms->nb_call += ls->nb_call;
+ ms->nb_obj += ls->nb_obj;
+ ms->nb_cycle += ls->nb_cycle;
+ ms->max_cycle = RTE_MAX(ms->max_cycle, ls->max_cycle);
+ ms->min_cycle = RTE_MIN(ms->min_cycle, ls->min_cycle);
+}
+
+static void
+lcore_stat_aggr(struct lcore_stat *ms, const struct lcore_stat *ls)
+{
+ uint32_t i;
+
+ ms->nb_cycle = RTE_MAX(ms->nb_cycle, ls->nb_cycle);
+ lcore_op_stat_aggr(&ms->deqenq, &ls->deqenq);
+ ms->deqenq.nb_lcore += ((ls->role_mask & ROLE_DEQENQ) != 0);
+ for (i = 0; i != ms->nb_stage; i++) {
+ lcore_op_stat_aggr(ms->stage + i, ls->stage + i);
+ ms->stage[i].nb_lcore += ((ls->role_mask & ROLE_STAGE(i)) != 0);
+ }
+}
+
+static void
+lcore_op_stat_dump(FILE *f, const struct lcore_op_stat *ls, const char *cap,
+ long double st)
+{
+ fprintf(f, "\t%s={\n", cap);
+
+ fprintf(f, "\t\tnb_lcore=%" PRIu64 ",\n", ls->nb_lcore);
+ fprintf(f, "\t\tnb_call=%" PRIu64 ",\n", ls->nb_call);
+ fprintf(f, "\t\tnb_obj=%" PRIu64 ",\n", ls->nb_obj);
+ fprintf(f, "\t\tnb_cycle=%" PRIu64 ",\n", ls->nb_cycle);
+ fprintf(f, "\t\tobj/call(avg): %.2Lf\n",
+ (long double)ls->nb_obj / ls->nb_call);
+ fprintf(f, "\t\tcycles/obj(avg): %.2Lf\n",
+ (long double)ls->nb_cycle / ls->nb_obj);
+ fprintf(f, "\t\tcycles/call(avg): %.2Lf\n",
+ (long double)ls->nb_cycle / ls->nb_call);
+
+ /* if min/max cycles per call stats was collected */
+ if (ls->min_cycle != UINT64_MAX) {
+ fprintf(f, "\t\tmax cycles/call=%" PRIu64 "(%.2Lf usec),\n",
+ ls->max_cycle,
+ (long double)ls->max_cycle / st);
+ fprintf(f, "\t\tmin cycles/call=%" PRIu64 "(%.2Lf usec),\n",
+ ls->min_cycle,
+ (long double)ls->min_cycle / st);
+ }
+
+ fprintf(f, "\t},\n");
+}
+
+static void
+lcore_stat_dump(FILE *f, uint32_t lc, const struct lcore_stat *ls)
+{
+ uint32_t i;
+ long double st;
+ char cap[64];
+
+ st = (long double)rte_get_timer_hz() / US_PER_S;
+
+ if (lc == UINT32_MAX)
+ fprintf(f, "%s(AGGREGATE)={\n", __func__);
+ else
+ fprintf(f, "%s(lcore=%u)={\n", __func__, lc);
+
+ fprintf(f, "\tnb_cycle=%" PRIu64 "(%.2Lf usec),\n",
+ ls->nb_cycle, (long double)ls->nb_cycle / st);
+
+ lcore_op_stat_dump(f, &ls->deqenq, "DEQ+ENQ", st);
+ for (i = 0; i != ls->nb_stage; i++) {
+ snprintf(cap, sizeof(cap), "%s#%u", "STAGE", i);
+ lcore_op_stat_dump(f, ls->stage + i, cap, st);
+ }
+
+ fprintf(f, "};\n");
+}
+
+static void
+fill_ring_elm(struct ring_elem *elm, uint32_t fill)
+{
+ uint32_t i;
+
+ for (i = 0; i != RTE_DIM(elm->cnt); i++)
+ elm->cnt[i] = fill;
+}
+
+static int32_t
+check_updt_elem(struct ring_elem *elm[], uint32_t num,
+ const struct ring_elem *check, const struct ring_elem *fill)
+{
+ uint32_t i;
+
+ static rte_spinlock_t dump_lock;
+
+ for (i = 0; i != num; i++) {
+ if (memcmp(check, elm[i], sizeof(*check)) != 0) {
+ rte_spinlock_lock(&dump_lock);
+ printf("%s(lc=%u, num=%u) failed at %u-th iter, "
+ "offending object: %p\n",
+ __func__, rte_lcore_id(), num, i, elm[i]);
+ rte_memdump(stdout, "expected", check, sizeof(*check));
+ rte_memdump(stdout, "result", elm[i], sizeof(*elm[i]));
+ rte_spinlock_unlock(&dump_lock);
+ return -EINVAL;
+ }
+ memcpy(elm[i], fill, sizeof(*elm[i]));
+ }
+
+ return 0;
+}
+
+static int
+check_ring_op(uint32_t exp, uint32_t res, uint32_t lc,
+ enum rte_ring_queue_behavior bhv, const char *fname, const char *opname)
+{
+ if (bhv == RTE_RING_QUEUE_FIXED && exp != res) {
+ printf("%s(lc=%u) failure: %s expected: %u, returned %u\n",
+ fname, lc, opname, exp, res);
+ return -ENOSPC;
+ }
+ if (bhv == RTE_RING_QUEUE_VARIABLE && exp < res) {
+ printf("%s(lc=%u) failure: %s expected: %u, returned %u\n",
+ fname, lc, opname, exp, res);
+ return -ENOSPC;
+ }
+ return 0;
+}
+
+/* num in interval [7/8, 11/8] of BULK_NUM */
+static inline uint32_t
+rand_elem_num(void)
+{
+ uint32_t num;
+
+ num = 7 * BULK_NUM / 8 + rte_rand() % (BULK_NUM / 2);
+ return num;
+}
+
+/*
+ * for each enabled stage do:
+ * acquire burst of objects
+ * read and check their contents
+ * update and check their contents
+ * release burst of objects
+ * done
+ */
+static int32_t
+test_worker_stages(struct lcore_arg *la, uint32_t lc, const char *fname,
+ struct ring_elem *obj[2 * BULK_NUM],
+ const struct ring_elem *def_elm, const struct ring_elem *loc_elm,
+ const struct ring_elem stg_elm[MAX_STAGES], int32_t prcs)
+{
+ int32_t rc;
+ uint32_t i, n, num, tkn;
+ uint64_t tm0, tm1;
+ const struct ring_elem *celm, *pelm;
+
+ num = rand_elem_num();
+
+ rc = 0;
+ tkn = 0;
+ for (i = 0, pelm = def_elm; i != la->stats.nb_stage; pelm = celm, i++) {
+
+ celm = stg_elm + i;
+
+ /* given stage is not enabled on that lcore */
+ if ((la->stats.role_mask & ROLE_STAGE(i)) == 0)
+ continue;
+
+ /* reset all pointer values */
+ memset(obj, 0, sizeof(*obj) * num);
+
+ /* acquire num elems */
+ tm0 = (prcs != 0) ? rte_rdtsc_precise() : 0;
+ n = _st_ring_acquire_burst(la->rng, i, (void **)obj, num,
+ &tkn, NULL);
+ tm0 = (prcs != 0) ? rte_rdtsc_precise() - tm0 : 0;
+
+ /* check return value and objects */
+ rc = check_ring_op(num, n, lc, RTE_RING_QUEUE_VARIABLE, fname,
+ RTE_STR(_st_ring_stage_acquire));
+ if (rc == 0)
+ rc = check_updt_elem(obj, n, pelm, loc_elm);
+ if (rc != 0)
+ break;
+
+ /* release num elems */
+ rte_compiler_barrier();
+ rc = check_updt_elem(obj, n, loc_elm, celm);
+ if (rc != 0)
+ break;
+
+ if (n == 0)
+ tm1 = 0;
+ else {
+ tm1 = (prcs != 0) ? rte_rdtsc_precise() : 0;
+ _st_ring_release(la->rng, i, tkn,
+ (void **)obj, n);
+ tm1 = (prcs != 0) ? rte_rdtsc_precise() - tm1 : 0;
+ }
+ lcore_op_stat_update(la->stats.stage + i, 1, n, tm0 + tm1,
+ prcs);
+ }
+
+ return rc;
+}
+
+static int32_t
+test_worker_deqenq(struct lcore_arg *la, uint32_t lc, const char *fname,
+ struct ring_elem *obj[2 * BULK_NUM],
+ const struct ring_elem *def_elm, const struct ring_elem *loc_elm,
+ const struct ring_elem *pelm, int32_t prcs)
+{
+ int32_t rc;
+ uint32_t k, n, num;
+ uint64_t tm0, tm1;
+
+ num = rand_elem_num();
+
+ /* reset all pointer values */
+ memset(obj, 0, sizeof(*obj) * num);
+
+ /* dequeue num elems */
+ tm0 = (prcs != 0) ? rte_rdtsc_precise() : 0;
+ n = _st_ring_dequeue_burst(la->rng, (void **)obj, num, NULL);
+
+ tm0 = (prcs != 0) ? rte_rdtsc_precise() - tm0 : 0;
+
+ /* check return value and objects */
+ rc = check_ring_op(num, n, lc, RTE_RING_QUEUE_VARIABLE, fname,
+ RTE_STR(_st_ring_dequeue_bulk));
+ if (rc == 0)
+ rc = check_updt_elem(obj, n, pelm, loc_elm);
+ if (rc != 0)
+ return rc;
+
+ /* enqueue n elems */
+ rte_compiler_barrier();
+ rc = check_updt_elem(obj, n, loc_elm, def_elm);
+ if (rc != 0)
+ return rc;
+
+ tm1 = (prcs != 0) ? rte_rdtsc_precise() : 0;
+ k = _st_ring_enqueue_bulk(la->rng, (void **)obj, n, NULL);
+ tm1 = (prcs != 0) ? rte_rdtsc_precise() - tm1 : 0;
+
+ /* check return value */
+ rc = check_ring_op(n, k, lc, RTE_RING_QUEUE_FIXED, fname,
+ RTE_STR(_st_ring_enqueue_bulk));
+ if (rc != 0)
+ return rc;
+
+ lcore_op_stat_update(&la->stats.deqenq, 1, n, tm0 + tm1, prcs);
+ return 0;
+}
+
+static int
+test_worker(void *arg, const char *fname, int32_t prcs)
+{
+ int32_t rc;
+ uint32_t i, lc;
+ uint64_t cl;
+ struct lcore_arg *la;
+ struct ring_elem *obj[2 * BULK_NUM];
+ struct ring_elem *pelm, def_elm, loc_elm, stg_elm[MAX_STAGES];
+
+ la = arg;
+ lc = rte_lcore_id();
+
+ fill_ring_elm(&def_elm, UINT32_MAX);
+ fill_ring_elm(&loc_elm, lc);
+
+ for (i = 0; i != RTE_DIM(stg_elm); i++)
+ fill_ring_elm(stg_elm + i, (i + 1) << 24);
+
+ pelm = stg_elm + la->stats.nb_stage - 1;
+
+ /* Acquire ordering is not required as the main is not
+ * really releasing any data through 'wrk_cmd' to
+ * the worker.
+ */
+ while (__atomic_load_n(&wrk_cmd, __ATOMIC_RELAXED) != WRK_CMD_RUN)
+ rte_pause();
+
+ cl = rte_rdtsc_precise();
+
+ do {
+ if ((la->stats.role_mask & ~ROLE_DEQENQ) != 0) {
+ rc = test_worker_stages(la, lc, fname, obj,
+ &def_elm, &loc_elm, stg_elm, prcs);
+ if (rc != 0)
+ break;
+ }
+
+ if ((la->stats.role_mask & ROLE_DEQENQ) != 0) {
+ rc = test_worker_deqenq(la, lc, fname, obj,
+ &def_elm, &loc_elm, pelm, prcs);
+ if (rc != 0)
+ break;
+ }
+
+ } while (__atomic_load_n(&wrk_cmd, __ATOMIC_RELAXED) == WRK_CMD_RUN);
+
+ cl = rte_rdtsc_precise() - cl;
+ if (prcs == 0)
+ lcore_stat_update(&la->stats, 0, 0, cl, 0);
+ la->stats.nb_cycle = cl;
+ return rc;
+}
+static int
+test_worker_prcs(void *arg)
+{
+ return test_worker(arg, __func__, 1);
+}
+
+static int
+test_worker_avg(void *arg)
+{
+ return test_worker(arg, __func__, 0);
+}
+
+static void
+mt1_fini(struct rte_soring *rng, void *data)
+{
+ rte_free(rng);
+ rte_free(data);
+}
+
+static int
+mt1_init(struct rte_soring **rng, void **data, uint32_t num,
+ enum rte_ring_sync_type prod_synt, enum rte_ring_sync_type cons_synt,
+ uint32_t nb_stages)
+{
+ int32_t rc;
+ size_t sz;
+ uint32_t i;
+ struct rte_soring *r;
+ struct ring_elem *elm;
+ void *p;
+ struct rte_soring_param prm;
+
+ *rng = NULL;
+ *data = NULL;
+
+ sz = num * sizeof(*elm);
+ elm = rte_zmalloc(NULL, sz, alignof(typeof(*elm)));
+ if (elm == NULL) {
+ printf("%s: alloc(%zu) for %u elems data failed",
+ __func__, sz, num);
+ return -ENOMEM;
+ }
+
+ *data = elm;
+
+ /* alloc soring */
+ memset(&prm, 0, sizeof(prm));
+
+ prm.name = __func__;
+ prm.elems = num;
+ prm.elem_size = sizeof(uintptr_t);
+ prm.stages = nb_stages;
+ prm.prod_synt = prod_synt;
+ prm.cons_synt = cons_synt;
+
+ sz = rte_soring_get_memsize(&prm);
+ r = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ if (r == NULL) {
+ printf("%s: alloc(%zu) for FIFO with %u elems failed",
+ __func__, sz, prm.elems);
+ return -ENOMEM;
+ }
+
+ *rng = r;
+
+ rc = rte_soring_init(r, &prm);
+ if (rc != 0) {
+ printf("%s: rte_soring_init(r=%p,elems=%u,stages=%u) failed, "
+ "error: %d(%s)\n",
+ __func__, r, prm.elems, prm.stages, rc, strerror(-rc));
+ return rc;
+ }
+
+ for (i = 0; i != num; i++) {
+ fill_ring_elm(elm + i, UINT32_MAX);
+ p = elm + i;
+ if (_st_ring_enqueue_bulk(r, &p, 1, NULL) != 1)
+ break;
+ }
+
+ if (i != num) {
+ printf("%s: _st_ring_enqueue(%p, %u) returned %u\n",
+ __func__, r, num, i);
+ return -ENOSPC;
+ }
+
+ return 0;
+}
+
+static int
+test_mt(int (*test)(void *), enum rte_ring_sync_type prod_synt,
+ enum rte_ring_sync_type cons_synt, uint32_t nb_stage,
+ const uint32_t role_mask[RTE_MAX_LCORE])
+{
+ int32_t rc;
+ uint32_t i, lc, mc;
+ struct rte_soring *r;
+ void *data;
+ struct lcore_arg arg[RTE_MAX_LCORE];
+
+ static const struct lcore_op_stat init_stat = {
+ .min_cycle = UINT64_MAX,
+ };
+
+ rc = mt1_init(&r, &data, RING_SIZE, prod_synt, cons_synt, nb_stage);
+
+ if (rc != 0) {
+ mt1_fini(r, data);
+ return rc;
+ }
+
+ memset(arg, 0, sizeof(arg));
+
+ /* launch on all workers */
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ arg[lc].rng = r;
+ arg[lc].stats.deqenq = init_stat;
+ arg[lc].stats.nb_stage = nb_stage;
+ arg[lc].stats.role_mask = role_mask[lc];
+ for (i = 0; i != arg[lc].stats.nb_stage; i++)
+ arg[lc].stats.stage[i] = init_stat;
+ rte_eal_remote_launch(test, &arg[lc], lc);
+ }
+
+ /* signal workers to start test */
+ __atomic_store_n(&wrk_cmd, WRK_CMD_RUN, __ATOMIC_RELEASE);
+
+ rte_delay_us(run_time * US_PER_S);
+
+ /* signal workers to stop test */
+ __atomic_store_n(&wrk_cmd, WRK_CMD_STOP, __ATOMIC_RELEASE);
+
+ /* wait for workers and collect stats. */
+ mc = rte_lcore_id();
+ arg[mc].stats.deqenq = init_stat;
+ arg[mc].stats.nb_stage = nb_stage;
+ for (i = 0; i != arg[mc].stats.nb_stage; i++)
+ arg[mc].stats.stage[i] = init_stat;
+
+ rc = 0;
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ rc |= rte_eal_wait_lcore(lc);
+ lcore_stat_aggr(&arg[mc].stats, &arg[lc].stats);
+ if (verbose != 0)
+ lcore_stat_dump(stdout, lc, &arg[lc].stats);
+ }
+
+ lcore_stat_dump(stdout, UINT32_MAX, &arg[mc].stats);
+ rte_soring_dump(stdout, r);
+ mt1_fini(r, data);
+ return rc;
+}
+
+/*
+ * launch all stages and deq+enq on all worker lcores
+ */
+static void
+role_mask_sym(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE])
+{
+ uint32_t lc;
+ const uint32_t msk = RTE_BIT32(nb_stage + 2) - 1;
+
+ memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE);
+ RTE_LCORE_FOREACH_WORKER(lc)
+ role_mask[lc] = msk;
+}
+
+/*
+ * Divide all workers in two (nearly) equal groups:
+ * - workers from 'even' group do deque+enque
+ * - workers from 'odd' group do acquire/release (for all stages)
+ */
+static void
+role_mask_even_odd(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE])
+{
+ uint32_t i, lc;
+ const uint32_t msk[2] = {
+ [0] = ROLE_DEQENQ,
+ [1] = RTE_GENMASK32(nb_stage + 1, 1),
+ };
+
+ memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE);
+
+ i = 0;
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ role_mask[lc] = msk[i & 1];
+ i++;
+ }
+ if (i == 1) {
+ lc = rte_get_next_lcore(-1, 1, 0);
+ role_mask[lc] |= msk[i & 1];
+ }
+}
+
+/*
+ * Divide all workers (nearly) evenly among all possible stages
+ */
+static void
+role_mask_div(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE])
+{
+ uint32_t i, lc;
+ uint32_t msk[nb_stage + 1];
+
+ memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE);
+
+ for (i = 0; i != RTE_DIM(msk); i++) {
+ msk[i] = RTE_BIT32(i);
+ };
+
+ i = 0;
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ role_mask[lc] = msk[i % RTE_DIM(msk)];
+ i++;
+ }
+ if (i < RTE_DIM(msk)) {
+ lc = rte_get_next_lcore(-1, 1, 0);
+ for (; i != RTE_DIM(msk); i++)
+ role_mask[lc] |= msk[i % RTE_DIM(msk)];
+ }
+}
+
+/*
+ * one worker does ST enqueue+dequeue, while all others - stages processing.
+ */
+static void
+role_mask_denq_st(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE])
+{
+ uint32_t i, lc;
+ const uint32_t msk[2] = {
+ [0] = ROLE_DEQENQ,
+ [1] = RTE_GENMASK32(nb_stage + 1, 1),
+ };
+
+ memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE);
+
+ i = 0;
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ if (i == 0)
+ role_mask[lc] = msk[0];
+ else
+ role_mask[lc] = msk[1];
+ i++;
+ }
+ if (i == 1) {
+ lc = rte_get_next_lcore(-1, 1, 0);
+ role_mask[lc] |= msk[1];
+ }
+}
+
+
+static int
+test_sym_mt1(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+ const uint32_t nb_stage = 1;
+
+ role_mask_sym(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT,
+ nb_stage, role_mask);
+}
+
+static int
+test_sym_mt4(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 4;
+
+ role_mask_sym(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT,
+ nb_stage, role_mask);
+}
+
+static int
+test_sym_mt_rts4(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 4;
+
+ role_mask_sym(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT_RTS, RTE_RING_SYNC_MT_RTS,
+ nb_stage, role_mask);
+}
+
+static int
+test_sym_mt_hts4(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 4;
+
+ role_mask_sym(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT_HTS, RTE_RING_SYNC_MT_HTS,
+ nb_stage, role_mask);
+}
+
+static int
+test_stdenq_stage4(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 4;
+
+ role_mask_denq_st(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_ST, RTE_RING_SYNC_ST,
+ nb_stage, role_mask);
+}
+
+
+static int
+test_even_odd_mt5(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 5;
+
+ role_mask_even_odd(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT,
+ nb_stage, role_mask);
+}
+
+static int
+test_div_mt3(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 3;
+
+ role_mask_div(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT,
+ nb_stage, role_mask);
+}
+
+static const struct test_case tests[] = {
+ {
+ .name = "MT_DEQENQ-MT_STG1-PRCS",
+ .func = test_sym_mt1,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG1-AVG",
+ .func = test_sym_mt1,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "ST_DEQENQ-MT_STG4-PRCS",
+ .func = test_stdenq_stage4,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "ST_DEQENQ-MT_STG4-AVG",
+ .func = test_stdenq_stage4,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG4-PRCS",
+ .func = test_sym_mt4,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG4-AVG",
+ .func = test_sym_mt4,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MTRTS_DEQENQ-MT_STG4-PRCS",
+ .func = test_sym_mt_rts4,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MTRTS_DEQENQ-MT_STG4-AVG",
+ .func = test_sym_mt_rts4,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MTHTS_DEQENQ-MT_STG4-PRCS",
+ .func = test_sym_mt_hts4,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MTHTS_DEQENQ-MT_STG4-AVG",
+ .func = test_sym_mt_hts4,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG5-1:1-PRCS",
+ .func = test_even_odd_mt5,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG5-1:1-AVG",
+ .func = test_even_odd_mt5,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG3-1:3-PRCS",
+ .func = test_div_mt3,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MT_DEQENQ_MT_STG3-1:3-AVG",
+ .func = test_div_mt3,
+ .wfunc = test_worker_avg,
+ },
+};
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [PATCH v4 0/5] Stage-Ordered API and other extensions for ring library
2024-09-16 12:37 ` [PATCH v3 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
` (4 preceding siblings ...)
2024-09-16 12:37 ` [PATCH v3 5/5] app/test: add unit tests for soring API Konstantin Ananyev
@ 2024-09-17 12:09 ` Konstantin Ananyev
2024-09-17 12:09 ` [PATCH v4 1/5] ring: common functions for 'move head' ops Konstantin Ananyev
` (4 more replies)
5 siblings, 5 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-17 12:09 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, eimear.morrissey, Konstantin Ananyev
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
v2 -> v3:
- fix compilation/doxygen complains (attempt #2)
- updated release notes
v2 -> v3:
- fix compilation/doxygen complains
- dropped patch:
"examples/l3fwd: make ACL work in pipeline and eventdev modes": [2]
As was mentioned in the patch desctiption it was way too big,
controversial and incomplete. If the community is ok to introduce
pipeline model into the l3fwd, then it is propbably worth to be
a separate patch series.
v1 -> v2:
- rename 'elmst/objst' to 'meta' (Morten)
- introduce new data-path APIs set: one with both meta{} and objs[],
second with just objs[] (Morten)
- split data-path APIs into burst/bulk flavours (same as rte_ring)
- added dump function for te_soring and improved dump() for rte_ring.
- dropped patch:
" ring: minimize reads of the counterpart cache-line"
- no performance gain observed
- actually it does change behavior of conventional rte_ring
enqueue/dequeue APIs -
it could return available/free less then actually exist in the ring.
As in some other libs we reliy on that information - it will
introduce problems.
The main aim of these series is to extend ring library with
new API that allows user to create/use Staged-Ordered-Ring (SORING)
abstraction. In addition to that there are few other patches that serve
different purposes:
- first two patches are just code reordering to de-duplicate
and generalize existing rte_ring code.
- patch #3 extends rte_ring_dump() to correctly print head/tail metadata
for different sync modes.
- next two patches introduce SORING API into the ring library and
provide UT for it.
SORING overview
===============
Staged-Ordered-Ring (SORING) provides a SW abstraction for 'ordered' queues
with multiple processing 'stages'. It is based on conventional DPDK
rte_ring, re-uses many of its concepts, and even substantial part of
its code.
It can be viewed as an 'extension' of rte_ring functionality.
In particular, main SORING properties:
- circular ring buffer with fixed size objects
- producer, consumer plus multiple processing stages in between.
- allows to split objects processing into multiple stages.
- objects remain in the same ring while moving from one stage to the other,
initial order is preserved, no extra copying needed.
- preserves the ingress order of objects within the queue across multiple
stages
- each stage (and producer/consumer) can be served by single and/or
multiple threads.
- number of stages, size and number of objects in the ring are
configurable at ring initialization time.
Data-path API provides four main operations:
- enqueue/dequeue works in the same manner as for conventional rte_ring,
all rte_ring synchronization types are supported.
- acquire/release - for each stage there is an acquire (start) and
release (finish) operation. After some objects are 'acquired' -
given thread can safely assume that it has exclusive ownership of
these objects till it will invoke 'release' for them.
After 'release', objects can be 'acquired' by next stage and/or dequeued
by the consumer (in case of last stage).
Expected use-case: applications that uses pipeline model
(probably with multiple stages) for packet processing, when preserving
incoming packet order is important.
The concept of ‘ring with stages’ is similar to DPDK OPDL eventdev PMD [1],
but the internals are different.
In particular, SORING maintains internal array of 'states' for each element
in the ring that is shared by all threads/processes that access the ring.
That allows 'release' to avoid excessive waits on the tail value and helps
to improve performancei and scalability.
In terms of performance, with our measurements rte_soring and
conventional rte_ring provide nearly identical numbers.
As an example, on our SUT: Intel ICX CPU @ 2.00GHz,
l3fwd (--lookup=acl) in pipeline mode [2] both
rte_ring and rte_soring reach ~20Mpps for single I/O lcore and same
number of worker lcores.
[1] https://www.dpdk.org/wp-content/uploads/sites/35/2018/06/DPDK-China2017-Ma-OPDL.pdf
[2] https://patchwork.dpdk.org/project/dpdk/patch/20240906131348.804-7-konstantin.v.ananyev@yandex.ru/
Eimear Morrissey (1):
ring: make dump function more verbose
Konstantin Ananyev (4):
ring: common functions for 'move head' ops
ring: make copying functions generic
ring/soring: introduce Staged Ordered Ring
app/test: add unit tests for soring API
.mailmap | 1 +
app/test/meson.build | 3 +
app/test/test_ring_stress_impl.h | 1 +
app/test/test_soring.c | 442 +++++++++++++
app/test/test_soring_mt_stress.c | 40 ++
app/test/test_soring_stress.c | 48 ++
app/test/test_soring_stress.h | 35 ++
app/test/test_soring_stress_impl.h | 827 +++++++++++++++++++++++++
doc/guides/rel_notes/release_24_11.rst | 8 +
lib/ring/meson.build | 4 +-
lib/ring/rte_ring.c | 87 ++-
lib/ring/rte_ring.h | 15 +
lib/ring/rte_ring_c11_pvt.h | 134 +---
lib/ring/rte_ring_elem_pvt.h | 181 ++++--
lib/ring/rte_ring_generic_pvt.h | 121 +---
lib/ring/rte_ring_hts_elem_pvt.h | 85 +--
lib/ring/rte_ring_rts_elem_pvt.h | 85 +--
lib/ring/rte_soring.c | 182 ++++++
lib/ring/rte_soring.h | 543 ++++++++++++++++
lib/ring/soring.c | 548 ++++++++++++++++
lib/ring/soring.h | 124 ++++
lib/ring/version.map | 26 +
22 files changed, 3144 insertions(+), 396 deletions(-)
create mode 100644 app/test/test_soring.c
create mode 100644 app/test/test_soring_mt_stress.c
create mode 100644 app/test/test_soring_stress.c
create mode 100644 app/test/test_soring_stress.h
create mode 100644 app/test/test_soring_stress_impl.h
create mode 100644 lib/ring/rte_soring.c
create mode 100644 lib/ring/rte_soring.h
create mode 100644 lib/ring/soring.c
create mode 100644 lib/ring/soring.h
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [PATCH v4 1/5] ring: common functions for 'move head' ops
2024-09-17 12:09 ` [PATCH v4 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
@ 2024-09-17 12:09 ` Konstantin Ananyev
2024-09-17 12:09 ` [PATCH v4 2/5] ring: make copying functions generic Konstantin Ananyev
` (3 subsequent siblings)
4 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-17 12:09 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, eimear.morrissey, Konstantin Ananyev
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
Note upfront: that change doesn't introduce any functional or
performance changes.
It is just a code-reordering for:
- code deduplication
- ability in future to re-use the same code to introduce new functionality
For each sync mode corresponding move_prod_head() and
move_cons_head() are nearly identical to each other,
the only differences are:
- do we need to use a @capacity to calculate number of entries or not.
- what we need to update (prod/cons) and what is used as
read-only counterpart.
So instead of having 2 copies of nearly identical functions,
introduce a new common one that could be used by both functions:
move_prod_head() and move_cons_head().
As another positive thing - we can get rid of referencing whole rte_ring
structure in that new common sub-function.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
---
lib/ring/rte_ring_c11_pvt.h | 134 +++++--------------------------
lib/ring/rte_ring_elem_pvt.h | 66 +++++++++++++++
lib/ring/rte_ring_generic_pvt.h | 121 ++++------------------------
lib/ring/rte_ring_hts_elem_pvt.h | 85 ++++++--------------
lib/ring/rte_ring_rts_elem_pvt.h | 85 ++++++--------------
5 files changed, 149 insertions(+), 342 deletions(-)
diff --git a/lib/ring/rte_ring_c11_pvt.h b/lib/ring/rte_ring_c11_pvt.h
index 629b2d9288..048933ddc6 100644
--- a/lib/ring/rte_ring_c11_pvt.h
+++ b/lib/ring/rte_ring_c11_pvt.h
@@ -28,41 +28,19 @@ __rte_ring_update_tail(struct rte_ring_headtail *ht, uint32_t old_val,
rte_atomic_store_explicit(&ht->tail, new_val, rte_memory_order_release);
}
-/**
- * @internal This function updates the producer head for enqueue
- *
- * @param r
- * A pointer to the ring structure
- * @param is_sp
- * Indicates whether multi-producer path is needed or not
- * @param n
- * The number of elements we will want to enqueue, i.e. how far should the
- * head be moved
- * @param behavior
- * RTE_RING_QUEUE_FIXED: Enqueue a fixed number of items from a ring
- * RTE_RING_QUEUE_VARIABLE: Enqueue as many items as possible from ring
- * @param old_head
- * Returns head value as it was before the move, i.e. where enqueue starts
- * @param new_head
- * Returns the current/new head value i.e. where enqueue finishes
- * @param free_entries
- * Returns the amount of free space in the ring BEFORE head was moved
- * @return
- * Actual number of objects enqueued.
- * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
- */
static __rte_always_inline unsigned int
-__rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
- unsigned int n, enum rte_ring_queue_behavior behavior,
- uint32_t *old_head, uint32_t *new_head,
- uint32_t *free_entries)
+__rte_ring_headtail_move_head(struct rte_ring_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity,
+ unsigned int is_st, unsigned int n,
+ enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head, uint32_t *entries)
{
- const uint32_t capacity = r->capacity;
- uint32_t cons_tail;
- unsigned int max = n;
+ uint32_t stail;
int success;
+ unsigned int max = n;
- *old_head = rte_atomic_load_explicit(&r->prod.head, rte_memory_order_relaxed);
+ *old_head = rte_atomic_load_explicit(&d->head,
+ rte_memory_order_relaxed);
do {
/* Reset n to the initial burst count */
n = max;
@@ -73,112 +51,36 @@ __rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
/* load-acquire synchronize with store-release of ht->tail
* in update_tail.
*/
- cons_tail = rte_atomic_load_explicit(&r->cons.tail,
+ stail = rte_atomic_load_explicit(&s->tail,
rte_memory_order_acquire);
/* The subtraction is done between two unsigned 32bits value
* (the result is always modulo 32 bits even if we have
- * *old_head > cons_tail). So 'free_entries' is always between 0
+ * *old_head > s->tail). So 'free_entries' is always between 0
* and capacity (which is < size).
*/
- *free_entries = (capacity + cons_tail - *old_head);
+ *entries = (capacity + stail - *old_head);
/* check that we have enough room in ring */
- if (unlikely(n > *free_entries))
+ if (unlikely(n > *entries))
n = (behavior == RTE_RING_QUEUE_FIXED) ?
- 0 : *free_entries;
+ 0 : *entries;
if (n == 0)
return 0;
*new_head = *old_head + n;
- if (is_sp) {
- r->prod.head = *new_head;
+ if (is_st) {
+ d->head = *new_head;
success = 1;
} else
/* on failure, *old_head is updated */
- success = rte_atomic_compare_exchange_strong_explicit(&r->prod.head,
- old_head, *new_head,
+ success = rte_atomic_compare_exchange_strong_explicit(
+ &d->head, old_head, *new_head,
rte_memory_order_relaxed,
rte_memory_order_relaxed);
} while (unlikely(success == 0));
return n;
}
-/**
- * @internal This function updates the consumer head for dequeue
- *
- * @param r
- * A pointer to the ring structure
- * @param is_sc
- * Indicates whether multi-consumer path is needed or not
- * @param n
- * The number of elements we will want to dequeue, i.e. how far should the
- * head be moved
- * @param behavior
- * RTE_RING_QUEUE_FIXED: Dequeue a fixed number of items from a ring
- * RTE_RING_QUEUE_VARIABLE: Dequeue as many items as possible from ring
- * @param old_head
- * Returns head value as it was before the move, i.e. where dequeue starts
- * @param new_head
- * Returns the current/new head value i.e. where dequeue finishes
- * @param entries
- * Returns the number of entries in the ring BEFORE head was moved
- * @return
- * - Actual number of objects dequeued.
- * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
- */
-static __rte_always_inline unsigned int
-__rte_ring_move_cons_head(struct rte_ring *r, int is_sc,
- unsigned int n, enum rte_ring_queue_behavior behavior,
- uint32_t *old_head, uint32_t *new_head,
- uint32_t *entries)
-{
- unsigned int max = n;
- uint32_t prod_tail;
- int success;
-
- /* move cons.head atomically */
- *old_head = rte_atomic_load_explicit(&r->cons.head, rte_memory_order_relaxed);
- do {
- /* Restore n as it may change every loop */
- n = max;
-
- /* Ensure the head is read before tail */
- rte_atomic_thread_fence(rte_memory_order_acquire);
-
- /* this load-acquire synchronize with store-release of ht->tail
- * in update_tail.
- */
- prod_tail = rte_atomic_load_explicit(&r->prod.tail,
- rte_memory_order_acquire);
-
- /* The subtraction is done between two unsigned 32bits value
- * (the result is always modulo 32 bits even if we have
- * cons_head > prod_tail). So 'entries' is always between 0
- * and size(ring)-1.
- */
- *entries = (prod_tail - *old_head);
-
- /* Set the actual entries for dequeue */
- if (n > *entries)
- n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries;
-
- if (unlikely(n == 0))
- return 0;
-
- *new_head = *old_head + n;
- if (is_sc) {
- r->cons.head = *new_head;
- success = 1;
- } else
- /* on failure, *old_head will be updated */
- success = rte_atomic_compare_exchange_strong_explicit(&r->cons.head,
- old_head, *new_head,
- rte_memory_order_relaxed,
- rte_memory_order_relaxed);
- } while (unlikely(success == 0));
- return n;
-}
-
#endif /* _RTE_RING_C11_PVT_H_ */
diff --git a/lib/ring/rte_ring_elem_pvt.h b/lib/ring/rte_ring_elem_pvt.h
index 4b80f58980..3a83668a08 100644
--- a/lib/ring/rte_ring_elem_pvt.h
+++ b/lib/ring/rte_ring_elem_pvt.h
@@ -293,6 +293,72 @@ __rte_ring_dequeue_elems(struct rte_ring *r, uint32_t cons_head,
#include "rte_ring_generic_pvt.h"
#endif
+/**
+ * @internal This function updates the producer head for enqueue
+ *
+ * @param r
+ * A pointer to the ring structure
+ * @param is_sp
+ * Indicates whether multi-producer path is needed or not
+ * @param n
+ * The number of elements we will want to enqueue, i.e. how far should the
+ * head be moved
+ * @param behavior
+ * RTE_RING_QUEUE_FIXED: Enqueue a fixed number of items from a ring
+ * RTE_RING_QUEUE_VARIABLE: Enqueue as many items as possible from ring
+ * @param old_head
+ * Returns head value as it was before the move, i.e. where enqueue starts
+ * @param new_head
+ * Returns the current/new head value i.e. where enqueue finishes
+ * @param free_entries
+ * Returns the amount of free space in the ring BEFORE head was moved
+ * @return
+ * Actual number of objects enqueued.
+ * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
+ */
+static __rte_always_inline unsigned int
+__rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
+ unsigned int n, enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head,
+ uint32_t *free_entries)
+{
+ return __rte_ring_headtail_move_head(&r->prod, &r->cons, r->capacity,
+ is_sp, n, behavior, old_head, new_head, free_entries);
+}
+
+/**
+ * @internal This function updates the consumer head for dequeue
+ *
+ * @param r
+ * A pointer to the ring structure
+ * @param is_sc
+ * Indicates whether multi-consumer path is needed or not
+ * @param n
+ * The number of elements we will want to dequeue, i.e. how far should the
+ * head be moved
+ * @param behavior
+ * RTE_RING_QUEUE_FIXED: Dequeue a fixed number of items from a ring
+ * RTE_RING_QUEUE_VARIABLE: Dequeue as many items as possible from ring
+ * @param old_head
+ * Returns head value as it was before the move, i.e. where dequeue starts
+ * @param new_head
+ * Returns the current/new head value i.e. where dequeue finishes
+ * @param entries
+ * Returns the number of entries in the ring BEFORE head was moved
+ * @return
+ * - Actual number of objects dequeued.
+ * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
+ */
+static __rte_always_inline unsigned int
+__rte_ring_move_cons_head(struct rte_ring *r, unsigned int is_sc,
+ unsigned int n, enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head,
+ uint32_t *entries)
+{
+ return __rte_ring_headtail_move_head(&r->cons, &r->prod, 0,
+ is_sc, n, behavior, old_head, new_head, entries);
+}
+
/**
* @internal Enqueue several objects on the ring
*
diff --git a/lib/ring/rte_ring_generic_pvt.h b/lib/ring/rte_ring_generic_pvt.h
index 457f41dab3..12f3595926 100644
--- a/lib/ring/rte_ring_generic_pvt.h
+++ b/lib/ring/rte_ring_generic_pvt.h
@@ -29,36 +29,13 @@ __rte_ring_update_tail(struct rte_ring_headtail *ht, uint32_t old_val,
ht->tail = new_val;
}
-/**
- * @internal This function updates the producer head for enqueue
- *
- * @param r
- * A pointer to the ring structure
- * @param is_sp
- * Indicates whether multi-producer path is needed or not
- * @param n
- * The number of elements we will want to enqueue, i.e. how far should the
- * head be moved
- * @param behavior
- * RTE_RING_QUEUE_FIXED: Enqueue a fixed number of items from a ring
- * RTE_RING_QUEUE_VARIABLE: Enqueue as many items as possible from ring
- * @param old_head
- * Returns head value as it was before the move, i.e. where enqueue starts
- * @param new_head
- * Returns the current/new head value i.e. where enqueue finishes
- * @param free_entries
- * Returns the amount of free space in the ring BEFORE head was moved
- * @return
- * Actual number of objects enqueued.
- * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
- */
static __rte_always_inline unsigned int
-__rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
- unsigned int n, enum rte_ring_queue_behavior behavior,
- uint32_t *old_head, uint32_t *new_head,
- uint32_t *free_entries)
+__rte_ring_headtail_move_head(struct rte_ring_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity,
+ unsigned int is_st, unsigned int n,
+ enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head, uint32_t *entries)
{
- const uint32_t capacity = r->capacity;
unsigned int max = n;
int success;
@@ -66,7 +43,7 @@ __rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
/* Reset n to the initial burst count */
n = max;
- *old_head = r->prod.head;
+ *old_head = d->head;
/* add rmb barrier to avoid load/load reorder in weak
* memory model. It is noop on x86
@@ -76,97 +53,27 @@ __rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp,
/*
* The subtraction is done between two unsigned 32bits value
* (the result is always modulo 32 bits even if we have
- * *old_head > cons_tail). So 'free_entries' is always between 0
+ * *old_head > s->tail). So 'free_entries' is always between 0
* and capacity (which is < size).
*/
- *free_entries = (capacity + r->cons.tail - *old_head);
+ *entries = (capacity + s->tail - *old_head);
/* check that we have enough room in ring */
- if (unlikely(n > *free_entries))
+ if (unlikely(n > *entries))
n = (behavior == RTE_RING_QUEUE_FIXED) ?
- 0 : *free_entries;
+ 0 : *entries;
if (n == 0)
return 0;
*new_head = *old_head + n;
- if (is_sp) {
- r->prod.head = *new_head;
+ if (is_st) {
+ d->head = *new_head;
success = 1;
} else
- success = rte_atomic32_cmpset((uint32_t *)(uintptr_t)&r->prod.head,
- *old_head, *new_head);
- } while (unlikely(success == 0));
- return n;
-}
-
-/**
- * @internal This function updates the consumer head for dequeue
- *
- * @param r
- * A pointer to the ring structure
- * @param is_sc
- * Indicates whether multi-consumer path is needed or not
- * @param n
- * The number of elements we will want to dequeue, i.e. how far should the
- * head be moved
- * @param behavior
- * RTE_RING_QUEUE_FIXED: Dequeue a fixed number of items from a ring
- * RTE_RING_QUEUE_VARIABLE: Dequeue as many items as possible from ring
- * @param old_head
- * Returns head value as it was before the move, i.e. where dequeue starts
- * @param new_head
- * Returns the current/new head value i.e. where dequeue finishes
- * @param entries
- * Returns the number of entries in the ring BEFORE head was moved
- * @return
- * - Actual number of objects dequeued.
- * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only.
- */
-static __rte_always_inline unsigned int
-__rte_ring_move_cons_head(struct rte_ring *r, unsigned int is_sc,
- unsigned int n, enum rte_ring_queue_behavior behavior,
- uint32_t *old_head, uint32_t *new_head,
- uint32_t *entries)
-{
- unsigned int max = n;
- int success;
-
- /* move cons.head atomically */
- do {
- /* Restore n as it may change every loop */
- n = max;
-
- *old_head = r->cons.head;
-
- /* add rmb barrier to avoid load/load reorder in weak
- * memory model. It is noop on x86
- */
- rte_smp_rmb();
-
- /* The subtraction is done between two unsigned 32bits value
- * (the result is always modulo 32 bits even if we have
- * cons_head > prod_tail). So 'entries' is always between 0
- * and size(ring)-1.
- */
- *entries = (r->prod.tail - *old_head);
-
- /* Set the actual entries for dequeue */
- if (n > *entries)
- n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries;
-
- if (unlikely(n == 0))
- return 0;
-
- *new_head = *old_head + n;
- if (is_sc) {
- r->cons.head = *new_head;
- rte_smp_rmb();
- success = 1;
- } else {
- success = rte_atomic32_cmpset((uint32_t *)(uintptr_t)&r->cons.head,
+ success = rte_atomic32_cmpset(
+ (uint32_t *)(uintptr_t)&d->head,
*old_head, *new_head);
- }
} while (unlikely(success == 0));
return n;
}
diff --git a/lib/ring/rte_ring_hts_elem_pvt.h b/lib/ring/rte_ring_hts_elem_pvt.h
index 91f5eeccb9..ed5f16879f 100644
--- a/lib/ring/rte_ring_hts_elem_pvt.h
+++ b/lib/ring/rte_ring_hts_elem_pvt.h
@@ -50,20 +50,16 @@ __rte_ring_hts_head_wait(const struct rte_ring_hts_headtail *ht,
}
}
-/**
- * @internal This function updates the producer head for enqueue
- */
-static __rte_always_inline unsigned int
-__rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
+static __rte_always_inline uint32_t
+__rte_ring_hts_move_head(struct rte_ring_hts_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity, unsigned int num,
enum rte_ring_queue_behavior behavior, uint32_t *old_head,
- uint32_t *free_entries)
+ uint32_t *entries)
{
uint32_t n;
union __rte_ring_hts_pos np, op;
- const uint32_t capacity = r->capacity;
-
- op.raw = rte_atomic_load_explicit(&r->hts_prod.ht.raw, rte_memory_order_acquire);
+ op.raw = rte_atomic_load_explicit(&d->ht.raw, rte_memory_order_acquire);
do {
/* Reset n to the initial burst count */
@@ -74,7 +70,7 @@ __rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
* make sure that we read prod head/tail *before*
* reading cons tail.
*/
- __rte_ring_hts_head_wait(&r->hts_prod, &op);
+ __rte_ring_hts_head_wait(d, &op);
/*
* The subtraction is done between two unsigned 32bits value
@@ -82,12 +78,12 @@ __rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
* *old_head > cons_tail). So 'free_entries' is always between 0
* and capacity (which is < size).
*/
- *free_entries = capacity + r->cons.tail - op.pos.head;
+ *entries = capacity + s->tail - op.pos.head;
/* check that we have enough room in ring */
- if (unlikely(n > *free_entries))
+ if (unlikely(n > *entries))
n = (behavior == RTE_RING_QUEUE_FIXED) ?
- 0 : *free_entries;
+ 0 : *entries;
if (n == 0)
break;
@@ -100,13 +96,25 @@ __rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
* - OOO reads of cons tail value
* - OOO copy of elems from the ring
*/
- } while (rte_atomic_compare_exchange_strong_explicit(&r->hts_prod.ht.raw,
+ } while (rte_atomic_compare_exchange_strong_explicit(&d->ht.raw,
(uint64_t *)(uintptr_t)&op.raw, np.raw,
- rte_memory_order_acquire, rte_memory_order_acquire) == 0);
+ rte_memory_order_acquire,
+ rte_memory_order_acquire) == 0);
*old_head = op.pos.head;
return n;
}
+/**
+ * @internal This function updates the producer head for enqueue
+ */
+static __rte_always_inline unsigned int
+__rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num,
+ enum rte_ring_queue_behavior behavior, uint32_t *old_head,
+ uint32_t *free_entries)
+{
+ return __rte_ring_hts_move_head(&r->hts_prod, &r->cons,
+ r->capacity, num, behavior, old_head, free_entries);
+}
/**
* @internal This function updates the consumer head for dequeue
@@ -116,51 +124,8 @@ __rte_ring_hts_move_cons_head(struct rte_ring *r, unsigned int num,
enum rte_ring_queue_behavior behavior, uint32_t *old_head,
uint32_t *entries)
{
- uint32_t n;
- union __rte_ring_hts_pos np, op;
-
- op.raw = rte_atomic_load_explicit(&r->hts_cons.ht.raw, rte_memory_order_acquire);
-
- /* move cons.head atomically */
- do {
- /* Restore n as it may change every loop */
- n = num;
-
- /*
- * wait for tail to be equal to head,
- * make sure that we read cons head/tail *before*
- * reading prod tail.
- */
- __rte_ring_hts_head_wait(&r->hts_cons, &op);
-
- /* The subtraction is done between two unsigned 32bits value
- * (the result is always modulo 32 bits even if we have
- * cons_head > prod_tail). So 'entries' is always between 0
- * and size(ring)-1.
- */
- *entries = r->prod.tail - op.pos.head;
-
- /* Set the actual entries for dequeue */
- if (n > *entries)
- n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries;
-
- if (unlikely(n == 0))
- break;
-
- np.pos.tail = op.pos.tail;
- np.pos.head = op.pos.head + n;
-
- /*
- * this CAS(ACQUIRE, ACQUIRE) serves as a hoist barrier to prevent:
- * - OOO reads of prod tail value
- * - OOO copy of elems from the ring
- */
- } while (rte_atomic_compare_exchange_strong_explicit(&r->hts_cons.ht.raw,
- (uint64_t *)(uintptr_t)&op.raw, np.raw,
- rte_memory_order_acquire, rte_memory_order_acquire) == 0);
-
- *old_head = op.pos.head;
- return n;
+ return __rte_ring_hts_move_head(&r->hts_cons, &r->prod,
+ 0, num, behavior, old_head, entries);
}
/**
diff --git a/lib/ring/rte_ring_rts_elem_pvt.h b/lib/ring/rte_ring_rts_elem_pvt.h
index 122650346b..027409a3fa 100644
--- a/lib/ring/rte_ring_rts_elem_pvt.h
+++ b/lib/ring/rte_ring_rts_elem_pvt.h
@@ -64,20 +64,17 @@ __rte_ring_rts_head_wait(const struct rte_ring_rts_headtail *ht,
}
}
-/**
- * @internal This function updates the producer head for enqueue.
- */
static __rte_always_inline uint32_t
-__rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
+__rte_ring_rts_move_head(struct rte_ring_rts_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity, uint32_t num,
enum rte_ring_queue_behavior behavior, uint32_t *old_head,
- uint32_t *free_entries)
+ uint32_t *entries)
{
uint32_t n;
union __rte_ring_rts_poscnt nh, oh;
- const uint32_t capacity = r->capacity;
-
- oh.raw = rte_atomic_load_explicit(&r->rts_prod.head.raw, rte_memory_order_acquire);
+ oh.raw = rte_atomic_load_explicit(&d->head.raw,
+ rte_memory_order_acquire);
do {
/* Reset n to the initial burst count */
@@ -88,7 +85,7 @@ __rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
* make sure that we read prod head *before*
* reading cons tail.
*/
- __rte_ring_rts_head_wait(&r->rts_prod, &oh);
+ __rte_ring_rts_head_wait(d, &oh);
/*
* The subtraction is done between two unsigned 32bits value
@@ -96,12 +93,12 @@ __rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
* *old_head > cons_tail). So 'free_entries' is always between 0
* and capacity (which is < size).
*/
- *free_entries = capacity + r->cons.tail - oh.val.pos;
+ *entries = capacity + s->tail - oh.val.pos;
/* check that we have enough room in ring */
- if (unlikely(n > *free_entries))
+ if (unlikely(n > *entries))
n = (behavior == RTE_RING_QUEUE_FIXED) ?
- 0 : *free_entries;
+ 0 : *entries;
if (n == 0)
break;
@@ -114,14 +111,27 @@ __rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
* - OOO reads of cons tail value
* - OOO copy of elems to the ring
*/
- } while (rte_atomic_compare_exchange_strong_explicit(&r->rts_prod.head.raw,
+ } while (rte_atomic_compare_exchange_strong_explicit(&d->head.raw,
(uint64_t *)(uintptr_t)&oh.raw, nh.raw,
- rte_memory_order_acquire, rte_memory_order_acquire) == 0);
+ rte_memory_order_acquire,
+ rte_memory_order_acquire) == 0);
*old_head = oh.val.pos;
return n;
}
+/**
+ * @internal This function updates the producer head for enqueue.
+ */
+static __rte_always_inline uint32_t
+__rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num,
+ enum rte_ring_queue_behavior behavior, uint32_t *old_head,
+ uint32_t *free_entries)
+{
+ return __rte_ring_rts_move_head(&r->rts_prod, &r->cons,
+ r->capacity, num, behavior, old_head, free_entries);
+}
+
/**
* @internal This function updates the consumer head for dequeue
*/
@@ -130,51 +140,8 @@ __rte_ring_rts_move_cons_head(struct rte_ring *r, uint32_t num,
enum rte_ring_queue_behavior behavior, uint32_t *old_head,
uint32_t *entries)
{
- uint32_t n;
- union __rte_ring_rts_poscnt nh, oh;
-
- oh.raw = rte_atomic_load_explicit(&r->rts_cons.head.raw, rte_memory_order_acquire);
-
- /* move cons.head atomically */
- do {
- /* Restore n as it may change every loop */
- n = num;
-
- /*
- * wait for cons head/tail distance,
- * make sure that we read cons head *before*
- * reading prod tail.
- */
- __rte_ring_rts_head_wait(&r->rts_cons, &oh);
-
- /* The subtraction is done between two unsigned 32bits value
- * (the result is always modulo 32 bits even if we have
- * cons_head > prod_tail). So 'entries' is always between 0
- * and size(ring)-1.
- */
- *entries = r->prod.tail - oh.val.pos;
-
- /* Set the actual entries for dequeue */
- if (n > *entries)
- n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries;
-
- if (unlikely(n == 0))
- break;
-
- nh.val.pos = oh.val.pos + n;
- nh.val.cnt = oh.val.cnt + 1;
-
- /*
- * this CAS(ACQUIRE, ACQUIRE) serves as a hoist barrier to prevent:
- * - OOO reads of prod tail value
- * - OOO copy of elems from the ring
- */
- } while (rte_atomic_compare_exchange_strong_explicit(&r->rts_cons.head.raw,
- (uint64_t *)(uintptr_t)&oh.raw, nh.raw,
- rte_memory_order_acquire, rte_memory_order_acquire) == 0);
-
- *old_head = oh.val.pos;
- return n;
+ return __rte_ring_rts_move_head(&r->rts_cons, &r->prod,
+ 0, num, behavior, old_head, entries);
}
/**
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [PATCH v4 2/5] ring: make copying functions generic
2024-09-17 12:09 ` [PATCH v4 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
2024-09-17 12:09 ` [PATCH v4 1/5] ring: common functions for 'move head' ops Konstantin Ananyev
@ 2024-09-17 12:09 ` Konstantin Ananyev
2024-09-17 12:09 ` [PATCH v4 3/5] ring: make dump function more verbose Konstantin Ananyev
` (2 subsequent siblings)
4 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-17 12:09 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, eimear.morrissey, Konstantin Ananyev
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
Note upfront: that change doesn't introduce any functional
or performance changes.
It is just a code-reordering for:
- improve code modularity and re-usability
- ability in future to re-use the same code to introduce new functionality
There is no real need for enqueue_elems()/dequeue_elems()
to get pointer to actual rte_ring structure, instead it is enough to pass
a pointer to actual elements buffer inside the ring.
In return, we'll get a copying functions that could be used for other
queueing abstractions that do have circular ring buffer inside.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
---
lib/ring/rte_ring_elem_pvt.h | 115 ++++++++++++++++++++---------------
1 file changed, 67 insertions(+), 48 deletions(-)
diff --git a/lib/ring/rte_ring_elem_pvt.h b/lib/ring/rte_ring_elem_pvt.h
index 3a83668a08..6eafae121f 100644
--- a/lib/ring/rte_ring_elem_pvt.h
+++ b/lib/ring/rte_ring_elem_pvt.h
@@ -17,12 +17,14 @@
#endif
static __rte_always_inline void
-__rte_ring_enqueue_elems_32(struct rte_ring *r, const uint32_t size,
- uint32_t idx, const void *obj_table, uint32_t n)
+__rte_ring_enqueue_elems_32(void *ring_table, const void *obj_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- uint32_t *ring = (uint32_t *)&r[1];
+
+ uint32_t *ring = (uint32_t *)ring_table;
const uint32_t *obj = (const uint32_t *)obj_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x7); i += 8, idx += 8) {
ring[idx] = obj[i];
@@ -60,14 +62,14 @@ __rte_ring_enqueue_elems_32(struct rte_ring *r, const uint32_t size,
}
static __rte_always_inline void
-__rte_ring_enqueue_elems_64(struct rte_ring *r, uint32_t prod_head,
- const void *obj_table, uint32_t n)
+__rte_ring_enqueue_elems_64(void *ring_table, const void *obj_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- const uint32_t size = r->size;
- uint32_t idx = prod_head & r->mask;
- uint64_t *ring = (uint64_t *)&r[1];
+
+ uint64_t *ring = (uint64_t *)ring_table;
const unaligned_uint64_t *obj = (const unaligned_uint64_t *)obj_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x3); i += 4, idx += 4) {
ring[idx] = obj[i];
@@ -93,14 +95,14 @@ __rte_ring_enqueue_elems_64(struct rte_ring *r, uint32_t prod_head,
}
static __rte_always_inline void
-__rte_ring_enqueue_elems_128(struct rte_ring *r, uint32_t prod_head,
- const void *obj_table, uint32_t n)
+__rte_ring_enqueue_elems_128(void *ring_table, const void *obj_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- const uint32_t size = r->size;
- uint32_t idx = prod_head & r->mask;
- rte_int128_t *ring = (rte_int128_t *)&r[1];
+
+ rte_int128_t *ring = (rte_int128_t *)ring_table;
const rte_int128_t *obj = (const rte_int128_t *)obj_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x1); i += 2, idx += 2)
memcpy((void *)(ring + idx),
@@ -126,37 +128,47 @@ __rte_ring_enqueue_elems_128(struct rte_ring *r, uint32_t prod_head,
* single and multi producer enqueue functions.
*/
static __rte_always_inline void
-__rte_ring_enqueue_elems(struct rte_ring *r, uint32_t prod_head,
- const void *obj_table, uint32_t esize, uint32_t num)
+__rte_ring_do_enqueue_elems(void *ring_table, const void *obj_table,
+ uint32_t size, uint32_t idx, uint32_t esize, uint32_t num)
{
/* 8B and 16B copies implemented individually to retain
* the current performance.
*/
if (esize == 8)
- __rte_ring_enqueue_elems_64(r, prod_head, obj_table, num);
+ __rte_ring_enqueue_elems_64(ring_table, obj_table, size,
+ idx, num);
else if (esize == 16)
- __rte_ring_enqueue_elems_128(r, prod_head, obj_table, num);
+ __rte_ring_enqueue_elems_128(ring_table, obj_table, size,
+ idx, num);
else {
- uint32_t idx, scale, nr_idx, nr_num, nr_size;
+ uint32_t scale, nr_idx, nr_num, nr_size;
/* Normalize to uint32_t */
scale = esize / sizeof(uint32_t);
nr_num = num * scale;
- idx = prod_head & r->mask;
nr_idx = idx * scale;
- nr_size = r->size * scale;
- __rte_ring_enqueue_elems_32(r, nr_size, nr_idx,
- obj_table, nr_num);
+ nr_size = size * scale;
+ __rte_ring_enqueue_elems_32(ring_table, obj_table, nr_size,
+ nr_idx, nr_num);
}
}
static __rte_always_inline void
-__rte_ring_dequeue_elems_32(struct rte_ring *r, const uint32_t size,
- uint32_t idx, void *obj_table, uint32_t n)
+__rte_ring_enqueue_elems(struct rte_ring *r, uint32_t prod_head,
+ const void *obj_table, uint32_t esize, uint32_t num)
+{
+ __rte_ring_do_enqueue_elems(&r[1], obj_table, r->size,
+ prod_head & r->mask, esize, num);
+}
+
+static __rte_always_inline void
+__rte_ring_dequeue_elems_32(void *obj_table, const void *ring_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- uint32_t *ring = (uint32_t *)&r[1];
uint32_t *obj = (uint32_t *)obj_table;
+ const uint32_t *ring = (const uint32_t *)ring_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x7); i += 8, idx += 8) {
obj[i] = ring[idx];
@@ -194,14 +206,13 @@ __rte_ring_dequeue_elems_32(struct rte_ring *r, const uint32_t size,
}
static __rte_always_inline void
-__rte_ring_dequeue_elems_64(struct rte_ring *r, uint32_t cons_head,
- void *obj_table, uint32_t n)
+__rte_ring_dequeue_elems_64(void *obj_table, const void *ring_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- const uint32_t size = r->size;
- uint32_t idx = cons_head & r->mask;
- uint64_t *ring = (uint64_t *)&r[1];
unaligned_uint64_t *obj = (unaligned_uint64_t *)obj_table;
+ const uint64_t *ring = (const uint64_t *)ring_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x3); i += 4, idx += 4) {
obj[i] = ring[idx];
@@ -227,27 +238,26 @@ __rte_ring_dequeue_elems_64(struct rte_ring *r, uint32_t cons_head,
}
static __rte_always_inline void
-__rte_ring_dequeue_elems_128(struct rte_ring *r, uint32_t cons_head,
- void *obj_table, uint32_t n)
+__rte_ring_dequeue_elems_128(void *obj_table, const void *ring_table,
+ uint32_t size, uint32_t idx, uint32_t n)
{
unsigned int i;
- const uint32_t size = r->size;
- uint32_t idx = cons_head & r->mask;
- rte_int128_t *ring = (rte_int128_t *)&r[1];
rte_int128_t *obj = (rte_int128_t *)obj_table;
+ const rte_int128_t *ring = (const rte_int128_t *)ring_table;
+
if (likely(idx + n <= size)) {
for (i = 0; i < (n & ~0x1); i += 2, idx += 2)
- memcpy((void *)(obj + i), (void *)(ring + idx), 32);
+ memcpy((obj + i), (const void *)(ring + idx), 32);
switch (n & 0x1) {
case 1:
- memcpy((void *)(obj + i), (void *)(ring + idx), 16);
+ memcpy((obj + i), (const void *)(ring + idx), 16);
}
} else {
for (i = 0; idx < size; i++, idx++)
- memcpy((void *)(obj + i), (void *)(ring + idx), 16);
+ memcpy((obj + i), (const void *)(ring + idx), 16);
/* Start at the beginning */
for (idx = 0; i < n; i++, idx++)
- memcpy((void *)(obj + i), (void *)(ring + idx), 16);
+ memcpy((obj + i), (const void *)(ring + idx), 16);
}
}
@@ -256,30 +266,39 @@ __rte_ring_dequeue_elems_128(struct rte_ring *r, uint32_t cons_head,
* single and multi producer enqueue functions.
*/
static __rte_always_inline void
-__rte_ring_dequeue_elems(struct rte_ring *r, uint32_t cons_head,
- void *obj_table, uint32_t esize, uint32_t num)
+__rte_ring_do_dequeue_elems(void *obj_table, const void *ring_table,
+ uint32_t size, uint32_t idx, uint32_t esize, uint32_t num)
{
/* 8B and 16B copies implemented individually to retain
* the current performance.
*/
if (esize == 8)
- __rte_ring_dequeue_elems_64(r, cons_head, obj_table, num);
+ __rte_ring_dequeue_elems_64(obj_table, ring_table, size,
+ idx, num);
else if (esize == 16)
- __rte_ring_dequeue_elems_128(r, cons_head, obj_table, num);
+ __rte_ring_dequeue_elems_128(obj_table, ring_table, size,
+ idx, num);
else {
- uint32_t idx, scale, nr_idx, nr_num, nr_size;
+ uint32_t scale, nr_idx, nr_num, nr_size;
/* Normalize to uint32_t */
scale = esize / sizeof(uint32_t);
nr_num = num * scale;
- idx = cons_head & r->mask;
nr_idx = idx * scale;
- nr_size = r->size * scale;
- __rte_ring_dequeue_elems_32(r, nr_size, nr_idx,
- obj_table, nr_num);
+ nr_size = size * scale;
+ __rte_ring_dequeue_elems_32(obj_table, ring_table, nr_size,
+ nr_idx, nr_num);
}
}
+static __rte_always_inline void
+__rte_ring_dequeue_elems(struct rte_ring *r, uint32_t cons_head,
+ void *obj_table, uint32_t esize, uint32_t num)
+{
+ __rte_ring_do_dequeue_elems(obj_table, &r[1], r->size,
+ cons_head & r->mask, esize, num);
+}
+
/* Between load and load. there might be cpu reorder in weak model
* (powerpc/arm).
* There are 2 choices for the users
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [PATCH v4 3/5] ring: make dump function more verbose
2024-09-17 12:09 ` [PATCH v4 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
2024-09-17 12:09 ` [PATCH v4 1/5] ring: common functions for 'move head' ops Konstantin Ananyev
2024-09-17 12:09 ` [PATCH v4 2/5] ring: make copying functions generic Konstantin Ananyev
@ 2024-09-17 12:09 ` Konstantin Ananyev
2024-09-17 12:09 ` [PATCH v4 4/5] ring/soring: introduce Staged Ordered Ring Konstantin Ananyev
2024-09-17 12:09 ` [PATCH v4 5/5] app/test: add unit tests for soring API Konstantin Ananyev
4 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-17 12:09 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, eimear.morrissey
From: Eimear Morrissey <eimear.morrissey@huawei.com>
The current rte_ring_dump function uses the generic rte_ring_headtail
structure to access head/tail positions. This is incorrect for the RTS
case where the head is stored in a different offset in the union of
structs. Switching to a separate function for each sync type allows
to dump correct head/tail values and extra metadata.
Signed-off-by: Eimear Morrissey <eimear.morrissey@huawei.com>
---
.mailmap | 1 +
app/test/test_ring_stress_impl.h | 1 +
lib/ring/rte_ring.c | 87 ++++++++++++++++++++++++++++++--
lib/ring/rte_ring.h | 15 ++++++
lib/ring/version.map | 7 +++
5 files changed, 107 insertions(+), 4 deletions(-)
diff --git a/.mailmap b/.mailmap
index 4a508bafad..3da86393c0 100644
--- a/.mailmap
+++ b/.mailmap
@@ -379,6 +379,7 @@ Eduard Serra <eserra@vmware.com>
Edward Makarov <makarov@kraftway.ru>
Edwin Brossette <edwin.brossette@6wind.com>
Eelco Chaudron <echaudro@redhat.com>
+Eimear Morrissey <eimear.morrissey@huawei.com>
Elad Nachman <eladv6@gmail.com>
Elad Persiko <eladpe@mellanox.com>
Elena Agostini <eagostini@nvidia.com>
diff --git a/app/test/test_ring_stress_impl.h b/app/test/test_ring_stress_impl.h
index 8b0bfb11fe..8449cb4b15 100644
--- a/app/test/test_ring_stress_impl.h
+++ b/app/test/test_ring_stress_impl.h
@@ -380,6 +380,7 @@ test_mt1(int (*test)(void *))
}
lcore_stat_dump(stdout, UINT32_MAX, &arg[mc].stats);
+ rte_ring_dump(stdout, r);
mt1_fini(r, data);
return rc;
}
diff --git a/lib/ring/rte_ring.c b/lib/ring/rte_ring.c
index aebb6d6728..261f2a06db 100644
--- a/lib/ring/rte_ring.c
+++ b/lib/ring/rte_ring.c
@@ -364,20 +364,99 @@ rte_ring_free(struct rte_ring *r)
rte_free(te);
}
+static const char *
+ring_get_sync_type(const enum rte_ring_sync_type st)
+{
+ switch (st) {
+ case RTE_RING_SYNC_ST:
+ return "single thread";
+ case RTE_RING_SYNC_MT:
+ return "multi thread";
+ case RTE_RING_SYNC_MT_RTS:
+ return "multi thread - RTS";
+ case RTE_RING_SYNC_MT_HTS:
+ return "multi thread - HTS";
+ default:
+ return "unknown";
+ }
+}
+
+static void
+ring_dump_ht_headtail(FILE *f, const char *prefix,
+ const struct rte_ring_headtail *ht)
+{
+ fprintf(f, "%ssync_type=%s\n", prefix,
+ ring_get_sync_type(ht->sync_type));
+ fprintf(f, "%shead=%"PRIu32"\n", prefix, ht->head);
+ fprintf(f, "%stail=%"PRIu32"\n", prefix, ht->tail);
+}
+
+static void
+ring_dump_rts_headtail(FILE *f, const char *prefix,
+ const struct rte_ring_rts_headtail *rts)
+{
+ fprintf(f, "%ssync_type=%s\n", prefix,
+ ring_get_sync_type(rts->sync_type));
+ fprintf(f, "%shead.pos=%"PRIu32"\n", prefix, rts->head.val.pos);
+ fprintf(f, "%shead.cnt=%"PRIu32"\n", prefix, rts->head.val.cnt);
+ fprintf(f, "%stail.pos=%"PRIu32"\n", prefix, rts->tail.val.pos);
+ fprintf(f, "%stail.cnt=%"PRIu32"\n", prefix, rts->tail.val.cnt);
+ fprintf(f, "%shtd_max=%"PRIu32"\n", prefix, rts->htd_max);
+}
+
+static void
+ring_dump_hts_headtail(FILE *f, const char *prefix,
+ const struct rte_ring_hts_headtail *hts)
+{
+ fprintf(f, "%ssync_type=%s\n", prefix,
+ ring_get_sync_type(hts->sync_type));
+ fprintf(f, "%shead=%"PRIu32"\n", prefix, hts->ht.pos.head);
+ fprintf(f, "%stail=%"PRIu32"\n", prefix, hts->ht.pos.tail);
+}
+
+void
+rte_ring_headtail_dump(FILE *f, const char *prefix,
+ const struct rte_ring_headtail *r)
+{
+ if (f == NULL || r == NULL)
+ return;
+
+ prefix = (prefix != NULL) ? prefix : "";
+
+ switch (r->sync_type) {
+ case RTE_RING_SYNC_ST:
+ case RTE_RING_SYNC_MT:
+ ring_dump_ht_headtail(f, prefix, r);
+ break;
+ case RTE_RING_SYNC_MT_RTS:
+ ring_dump_rts_headtail(f, prefix,
+ (const struct rte_ring_rts_headtail *)r);
+ break;
+ case RTE_RING_SYNC_MT_HTS:
+ ring_dump_hts_headtail(f, prefix,
+ (const struct rte_ring_hts_headtail *)r);
+ break;
+ default:
+ RING_LOG(ERR, "Invalid ring sync type detected");
+ }
+}
+
/* dump the status of the ring on the console */
void
rte_ring_dump(FILE *f, const struct rte_ring *r)
{
+ if (f == NULL || r == NULL)
+ return;
+
fprintf(f, "ring <%s>@%p\n", r->name, r);
fprintf(f, " flags=%x\n", r->flags);
fprintf(f, " size=%"PRIu32"\n", r->size);
fprintf(f, " capacity=%"PRIu32"\n", r->capacity);
- fprintf(f, " ct=%"PRIu32"\n", r->cons.tail);
- fprintf(f, " ch=%"PRIu32"\n", r->cons.head);
- fprintf(f, " pt=%"PRIu32"\n", r->prod.tail);
- fprintf(f, " ph=%"PRIu32"\n", r->prod.head);
fprintf(f, " used=%u\n", rte_ring_count(r));
fprintf(f, " avail=%u\n", rte_ring_free_count(r));
+
+ rte_ring_headtail_dump(f, " cons.", &(r->cons));
+ rte_ring_headtail_dump(f, " prod.", &(r->prod));
}
/* dump the status of all rings on the console */
diff --git a/lib/ring/rte_ring.h b/lib/ring/rte_ring.h
index c709f30497..d6f9e85c64 100644
--- a/lib/ring/rte_ring.h
+++ b/lib/ring/rte_ring.h
@@ -204,6 +204,21 @@ void rte_ring_free(struct rte_ring *r);
*/
void rte_ring_dump(FILE *f, const struct rte_ring *r);
+/**
+ * Dump the status of a headtail to a file.
+ *
+ * @param f
+ * A pointer to a file for output
+ * @param prefix
+ * A string to prefix each output line with
+ * @param r
+ * A pointer to a ring headtail structure.
+ */
+__rte_experimental
+void
+rte_ring_headtail_dump(FILE *f, const char *prefix,
+ const struct rte_ring_headtail *r);
+
/**
* Enqueue several objects on the ring (multi-producers safe).
*
diff --git a/lib/ring/version.map b/lib/ring/version.map
index 8da094a69a..61f7464f5a 100644
--- a/lib/ring/version.map
+++ b/lib/ring/version.map
@@ -14,3 +14,10 @@ DPDK_25 {
local: *;
};
+
+EXPERIMENTAL {
+ global:
+
+ # added in 24.11
+ rte_ring_headtail_dump;
+};
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [PATCH v4 4/5] ring/soring: introduce Staged Ordered Ring
2024-09-17 12:09 ` [PATCH v4 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
` (2 preceding siblings ...)
2024-09-17 12:09 ` [PATCH v4 3/5] ring: make dump function more verbose Konstantin Ananyev
@ 2024-09-17 12:09 ` Konstantin Ananyev
2024-09-17 12:09 ` [PATCH v4 5/5] app/test: add unit tests for soring API Konstantin Ananyev
4 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-17 12:09 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, eimear.morrissey, Konstantin Ananyev
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
Staged-Ordered-Ring (SORING) provides a SW abstraction for 'ordered' queues
with multiple processing 'stages'.
It is based on conventional DPDK rte_ring, re-uses many of its concepts,
and even substantial part of its code.
It can be viewed as an 'extension' of rte_ring functionality.
In particular, main SORING properties:
- circular ring buffer with fixed size objects
- producer, consumer plus multiple processing stages in the middle.
- allows to split objects processing into multiple stages.
- objects remain in the same ring while moving from one stage to the other,
initial order is preserved, no extra copying needed.
- preserves the ingress order of objects within the queue across multiple
stages, i.e.:
at the same stage multiple threads can process objects from the ring in
any order, but for the next stage objects will always appear in the
original order.
- each stage (and producer/consumer) can be served by single and/or
multiple threads.
- number of stages, size and number of objects in the ring are
configurable at ring initialization time.
Data-path API provides four main operations:
- enqueue/dequeue works in the same manner as for conventional rte_ring,
all rte_ring synchronization types are supported.
- acquire/release - for each stage there is an acquire (start) and
release (finish) operation.
after some objects are 'acquired' - given thread can safely assume that
it has exclusive possession of these objects till 'release' for them is
invoked.
Note that right now user has to release exactly the same number of
objects that was acquired before.
After 'release', objects can be 'acquired' by next stage and/or dequeued
by the consumer (in case of last stage).
Expected use-case: applications that uses pipeline model
(probably with multiple stages) for packet processing, when preserving
incoming packet order is important. I.E.: IPsec processing, etc.
Signed-off-by: Eimear Morrissey <eimear.morrissey@huawei.com>
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
---
doc/guides/rel_notes/release_24_11.rst | 8 +
lib/ring/meson.build | 4 +-
lib/ring/rte_soring.c | 182 ++++++++
lib/ring/rte_soring.h | 543 ++++++++++++++++++++++++
lib/ring/soring.c | 548 +++++++++++++++++++++++++
lib/ring/soring.h | 124 ++++++
lib/ring/version.map | 19 +
7 files changed, 1426 insertions(+), 2 deletions(-)
create mode 100644 lib/ring/rte_soring.c
create mode 100644 lib/ring/rte_soring.h
create mode 100644 lib/ring/soring.c
create mode 100644 lib/ring/soring.h
diff --git a/doc/guides/rel_notes/release_24_11.rst b/doc/guides/rel_notes/release_24_11.rst
index 0ff70d9057..3ceeef7eb6 100644
--- a/doc/guides/rel_notes/release_24_11.rst
+++ b/doc/guides/rel_notes/release_24_11.rst
@@ -55,6 +55,14 @@ New Features
Also, make sure to start the actual text at the margin.
=======================================================
+* **Add Staged-Ordered-Ring (SORING) API to the rte_ring library.**
+
+ New API to the ring library to provide a SW abstraction for
+ 'ordered' queues with multiple processing 'stages'.
+ It is based on conventional DPDK rte_ring, re-uses many of its concepts,
+ and even substantial part of its code.
+ It can be viewed as an 'extension' of rte_ring functionality.
+
Removed Items
-------------
diff --git a/lib/ring/meson.build b/lib/ring/meson.build
index 7fca958ed7..21f2c12989 100644
--- a/lib/ring/meson.build
+++ b/lib/ring/meson.build
@@ -1,8 +1,8 @@
# SPDX-License-Identifier: BSD-3-Clause
# Copyright(c) 2017 Intel Corporation
-sources = files('rte_ring.c')
-headers = files('rte_ring.h')
+sources = files('rte_ring.c', 'rte_soring.c', 'soring.c')
+headers = files('rte_ring.h', 'rte_soring.h')
# most sub-headers are not for direct inclusion
indirect_headers += files (
'rte_ring_core.h',
diff --git a/lib/ring/rte_soring.c b/lib/ring/rte_soring.c
new file mode 100644
index 0000000000..b6bc71b8c9
--- /dev/null
+++ b/lib/ring/rte_soring.c
@@ -0,0 +1,182 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#include <inttypes.h>
+
+#include "soring.h"
+#include <rte_string_fns.h>
+
+RTE_LOG_REGISTER_DEFAULT(soring_logtype, INFO);
+#define RTE_LOGTYPE_SORING soring_logtype
+#define SORING_LOG(level, ...) \
+ RTE_LOG_LINE(level, SORING, "" __VA_ARGS__)
+
+static uint32_t
+soring_calc_elem_num(uint32_t count)
+{
+ return rte_align32pow2(count + 1);
+}
+
+static int
+soring_check_param(uint32_t esize, uint32_t stsize, uint32_t count,
+ uint32_t stages)
+{
+ if (stages == 0) {
+ SORING_LOG(ERR, "invalid number of stages: %u", stages);
+ return -EINVAL;
+ }
+
+ /* Check if element size is a multiple of 4B */
+ if (esize == 0 || esize % 4 != 0) {
+ SORING_LOG(ERR, "invalid element size: %u", esize);
+ return -EINVAL;
+ }
+
+ /* Check if ret-code size is a multiple of 4B */
+ if (stsize % 4 != 0) {
+ SORING_LOG(ERR, "invalid retcode size: %u", stsize);
+ return -EINVAL;
+ }
+
+ /* count must be a power of 2 */
+ if (rte_is_power_of_2(count) == 0 ||
+ (count > RTE_SORING_ELEM_MAX + 1)) {
+ SORING_LOG(ERR, "invalid number of elements: %u", count);
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+/*
+ * Calculate size offsets for SORING internal data layout.
+ */
+static size_t
+soring_get_szofs(uint32_t esize, uint32_t stsize, uint32_t count,
+ uint32_t stages, size_t *elst_ofs, size_t *state_ofs,
+ size_t *stage_ofs)
+{
+ size_t sz;
+ const struct rte_soring * const r = NULL;
+
+ sz = sizeof(r[0]) + (size_t)count * esize;
+ sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE);
+
+ if (elst_ofs != NULL)
+ *elst_ofs = sz;
+
+ sz = sz + (size_t)count * stsize;
+ sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE);
+
+ if (state_ofs != NULL)
+ *state_ofs = sz;
+
+ sz += sizeof(r->state[0]) * count;
+ sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE);
+
+ if (stage_ofs != NULL)
+ *stage_ofs = sz;
+
+ sz += sizeof(r->stage[0]) * stages;
+ sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE);
+
+ return sz;
+}
+
+static void
+soring_dump_stage_headtail(FILE *f, const char *prefix,
+ struct soring_stage *st)
+{
+ fprintf(f, "%stail.pos=%"PRIu32"\n", prefix, st->sht.tail.pos);
+ fprintf(f, "%stail.sync=%"PRIu32"\n", prefix, st->sht.tail.sync);
+ fprintf(f, "%shead=%"PRIu32"\n", prefix, st->sht.head);
+}
+
+void
+rte_soring_dump(FILE *f, const struct rte_soring *r)
+{
+ uint32_t i;
+ char buf[32];
+
+ if (f == NULL || r == NULL)
+ return;
+
+ fprintf(f, "soring <%s>@%p\n", r->name, r);
+ fprintf(f, " size=%"PRIu32"\n", r->size);
+ fprintf(f, " capacity=%"PRIu32"\n", r->capacity);
+ fprintf(f, " esize=%"PRIu32"\n", r->esize);
+ fprintf(f, " msize=%"PRIu32"\n", r->msize);
+ fprintf(f, " used=%u\n", rte_soring_count(r));
+ fprintf(f, " avail=%u\n", rte_soring_free_count(r));
+
+ rte_ring_headtail_dump(f, " cons.", &(r->cons.ht));
+ rte_ring_headtail_dump(f, " prod.", &(r->prod.ht));
+
+ fprintf(f, " nb_stage=%"PRIu32"\n", r->nb_stage);
+ for (i = 0; i < r->nb_stage; i++) {
+ snprintf(buf, sizeof(buf), " stage[%u].", i);
+ soring_dump_stage_headtail(f, buf, r->stage + i);
+ }
+}
+
+ssize_t
+rte_soring_get_memsize(const struct rte_soring_param *prm)
+{
+ int32_t rc;
+ uint32_t count;
+
+ count = soring_calc_elem_num(prm->elems);
+ rc = soring_check_param(prm->elem_size, prm->meta_size, count,
+ prm->stages);
+ if (rc != 0)
+ return rc;
+
+ return soring_get_szofs(prm->elem_size, prm->meta_size, count,
+ prm->stages, NULL, NULL, NULL);
+}
+
+int
+rte_soring_init(struct rte_soring *r, const struct rte_soring_param *prm)
+{
+ int32_t rc;
+ uint32_t n;
+ size_t meta_ofs, stage_ofs, state_ofs;
+
+ if (r == NULL || prm == NULL)
+ return -EINVAL;
+
+ n = soring_calc_elem_num(prm->elems);
+ rc = soring_check_param(prm->elem_size, prm->meta_size, n, prm->stages);
+ if (rc != 0)
+ return rc;
+
+ soring_get_szofs(prm->elem_size, prm->meta_size, n, prm->stages,
+ &meta_ofs, &state_ofs, &stage_ofs);
+
+ memset(r, 0, sizeof(*r));
+ rc = strlcpy(r->name, prm->name, sizeof(r->name));
+ if (rc < 0 || rc >= (int)sizeof(r->name))
+ return -ENAMETOOLONG;
+
+ r->size = n;
+ r->mask = r->size - 1;
+ r->capacity = prm->elems;
+ r->esize = prm->elem_size;
+ r->msize = prm->meta_size;
+
+ r->prod.ht.sync_type = prm->prod_synt;
+ r->cons.ht.sync_type = prm->cons_synt;
+
+ r->state = (union soring_state *)((uintptr_t)r + state_ofs);
+ memset(r->state, 0, sizeof(r->state[0]) * r->size);
+
+ r->stage = (struct soring_stage *)((uintptr_t)r + stage_ofs);
+ r->nb_stage = prm->stages;
+ memset(r->stage, 0, r->nb_stage * sizeof(r->stage[0]));
+
+ if (r->msize != 0)
+ r->meta = (void *)((uintptr_t)r + meta_ofs);
+
+ return 0;
+}
diff --git a/lib/ring/rte_soring.h b/lib/ring/rte_soring.h
new file mode 100644
index 0000000000..9c35b4a18c
--- /dev/null
+++ b/lib/ring/rte_soring.h
@@ -0,0 +1,543 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#ifndef _RTE_SORING_H_
+#define _RTE_SORING_H_
+
+/**
+ * @file
+ * This file contains definition of RTE soring (Staged Ordered Ring) public API.
+ * Brief description:
+ * enqueue/dequeue works the same as for conventional rte_ring:
+ * any rte_ring sync types can be used, etc.
+ * Plus there could be multiple 'stages'.
+ * For each stage there is an acquire (start) and release (finish) operation.
+ * after some elems are 'acquired' - user can safely assume that he has
+ * exclusive possession of these elems till 'release' for them is done.
+ * Note that right now user has to release exactly the same number of elems
+ * he acquired before.
+ * After 'release', elems can be 'acquired' by next stage and/or dequeued
+ * (in case of last stage).
+ */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <rte_ring.h>
+
+/* upper 2 bits are used for status */
+#define RTE_SORING_ST_BIT 30
+
+/* max possible number of elements in the soring */
+#define RTE_SORING_ELEM_MAX (RTE_BIT32(RTE_SORING_ST_BIT) - 1)
+
+struct rte_soring_param {
+ /** expected name of the ring */
+ const char *name;
+ /** number of elemnts in the ring */
+ uint32_t elems;
+ /** size of elements in the ring, must be a multiple of 4 */
+ uint32_t elem_size;
+ /**
+ * size of metadata for each elem, must be a multiple of 4.
+ * This parameter defines a size of supplementary and optional
+ * array of metadata associated with each object in the soring.
+ * While element size is configurable (see 'elem_size' parameter above),
+ * so user can specify it big enough to hold both object and its
+ * metadata together, for performance reasons it might be plausible
+ * to access them as separate arrays.
+ * Common usage scenario when such separation helps:
+ * enqueue() - writes to objects array
+ * acquire() - reads from objects array
+ * release() - writes to metadata array (as an example: return code)
+ * dequeue() - reads both objects and metadata array
+ */
+ uint32_t meta_size;
+ /** number of stages in the ring */
+ uint32_t stages;
+ /** sync type for producer */
+ enum rte_ring_sync_type prod_synt;
+ /** sync type for consumer */
+ enum rte_ring_sync_type cons_synt;
+};
+
+struct rte_soring;
+
+/**
+ * Calculate the memory size needed for a soring
+ *
+ * This function returns the number of bytes needed for a ring, given
+ * the expected parameters for it. This value is the sum of the size of
+ * the internal metadata and the size of the memory needed by the
+ * actual ring elements and theri rec-codes. The value is aligned to a cache
+ * line size.
+ *
+ * @param prm
+ * Pointer to the structure that contains soring creation paramers.
+ * @return
+ * - The memory size needed for the soring on success.
+ * - -EINVAL if provided paramer values are invalid.
+ */
+__rte_experimental
+ssize_t
+rte_soring_get_memsize(const struct rte_soring_param *prm);
+
+/**
+ * Initialize a soring structure.
+ *
+ * Initialize a soring structure in memory pointed by "r".
+ * The size of the memory area must be large enough to store the soring
+ * internal structures plus the objects and ret-code tables.
+ * It is strongly advised to use rte_ring_get_memsize() to get the
+ * appropriate size.
+ *
+ * @param r
+ * Pointer to the soring structure.
+ * @param prm
+ * Pointer to the structure that contains soring creation paramers.
+ * @return
+ * - 0 on success, or a negative error code.
+ */
+__rte_experimental
+int
+rte_soring_init(struct rte_soring *r, const struct rte_soring_param *prm);
+
+/**
+ * Return the total number of filled entries in a ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @return
+ * The number of entries in the ring.
+ */
+__rte_experimental
+unsigned int
+rte_soring_count(const struct rte_soring *r);
+
+/**
+ * Return the total number of unfilled entries in a ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @return
+ * The number of free entries in the ring.
+ */
+__rte_experimental
+unsigned int
+rte_soring_free_count(const struct rte_soring *r);
+
+/**
+ * Dump the status of the soring
+ *
+ * @param f
+ * A pointer to a file for output
+ * @param r
+ * Pointer to the soring structure.
+ */
+__rte_experimental
+void
+rte_soring_dump(FILE *f, const struct rte_soring *r);
+
+/**
+ * Enqueue several objects on the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to enqueue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param n
+ * The number of objects to add in the ring from the 'objs'.
+ * @param free_space
+ * if non-NULL, returns the amount of space in the ring after the
+ * enqueue operation has finished.
+ * @return
+ * - Actual number of objects enqueued, either 0 or @n.
+ */
+__rte_experimental
+uint32_t
+rte_soring_enqueue_bulk(struct rte_soring *r, const void *objs,
+ uint32_t n, uint32_t *free_space);
+
+/**
+ * Enqueue several objects plus metadata on the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to enqueue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each object to enqueue.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then 'meta' parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param n
+ * The number of objects to add in the ring from the 'objs'.
+ * @param free_space
+ * if non-NULL, returns the amount of space in the ring after the
+ * enqueue operation has finished.
+ * @return
+ * - Actual number of objects enqueued, either 0 or @n.
+ */
+__rte_experimental
+uint32_t
+rte_soring_enqueux_bulk(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, uint32_t *free_space);
+
+/**
+ * Enqueue several objects on the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to enqueue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param n
+ * The number of objects to add in the ring from the 'objs'.
+ * @param free_space
+ * if non-NULL, returns the amount of space in the ring after the
+ * enqueue operation has finished.
+ * @return
+ * - Actual number of objects enqueued.
+ */
+__rte_experimental
+uint32_t
+rte_soring_enqueue_burst(struct rte_soring *r, const void *objs,
+ uint32_t n, uint32_t *free_space);
+
+/**
+ * Enqueue several objects plus metadata on the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to enqueue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each object to enqueue.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then 'meta' parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param n
+ * The number of objects to add in the ring from the 'objs'.
+ * @param free_space
+ * if non-NULL, returns the amount of space in the ring after the
+ * enqueue operation has finished.
+ * @return
+ * - Actual number of objects enqueued.
+ */
+__rte_experimental
+uint32_t
+rte_soring_enqueux_burst(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, uint32_t *free_space);
+
+/**
+ * Dequeue several objects from the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to dequeue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param num
+ * The number of objects to dequeue from the ring into the objs.
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries after the
+ * dequeue has finished.
+ * @return
+ * - Actual number of objects dequeued, either 0 or 'num'.
+ */
+__rte_experimental
+uint32_t
+rte_soring_dequeue_bulk(struct rte_soring *r, void *objs,
+ uint32_t num, uint32_t *available);
+
+/**
+ * Dequeue several objects plus metadata from the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to dequeue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to array of metadata values for each object to dequeue.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then 'meta' parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param num
+ * The number of objects to dequeue from the ring into the objs.
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries after the
+ * dequeue has finished.
+ * @return
+ * - Actual number of objects dequeued, either 0 or 'num'.
+ */
+__rte_experimental
+uint32_t
+rte_soring_dequeux_bulk(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, uint32_t *available);
+
+/**
+ * Dequeue several objects from the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to dequeue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param num
+ * The number of objects to dequeue from the ring into the objs.
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries after the
+ * dequeue has finished.
+ * @return
+ * - Actual number of objects dequeued.
+ */
+__rte_experimental
+uint32_t
+rte_soring_dequeue_burst(struct rte_soring *r, void *objs,
+ uint32_t num, uint32_t *available);
+
+/**
+ * Dequeue several objects plus metadata from the ring.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to dequeue.
+ * Size of objects to enqueue must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to array of metadata values for each object to dequeue.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then 'meta' parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param num
+ * The number of objects to dequeue from the ring into the objs.
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries after the
+ * dequeue has finished.
+ * @return
+ * - Actual number of objects dequeued.
+ */
+__rte_experimental
+uint32_t
+rte_soring_dequeux_burst(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, uint32_t *available);
+
+/**
+ * Acquire several objects from the ring for given stage.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to acquire.
+ * Size of objects must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param stage
+ * Stage to acquire objects for.
+ * @param num
+ * The number of objects to acquire.
+ * @param ftoken
+ * Pointer to the opaque 'token' value used by release() op.
+ * User has to store this value somewhere, and later provide to the
+ * release().
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries for given stage
+ * after the acquire has finished.
+ * @return
+ * - Actual number of objects acquired, either 0 or 'num'.
+ */
+__rte_experimental
+uint32_t
+rte_soring_acquire_bulk(struct rte_soring *r, void *objs,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available);
+
+/**
+ * Acquire several objects plus metadata from the ring for given stage.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to acquire.
+ * Size of objects must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each for each acquired object.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then 'meta' parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param stage
+ * Stage to acquire objects for.
+ * @param num
+ * The number of objects to acquire.
+ * @param ftoken
+ * Pointer to the opaque 'token' value used by release() op.
+ * User has to store this value somewhere, and later provide to the
+ * release().
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries for given stage
+ * after the acquire has finished.
+ * @return
+ * - Actual number of objects acquired, either 0 or 'num'.
+ */
+__rte_experimental
+uint32_t
+rte_soring_acquirx_bulk(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available);
+
+/**
+ * Acquire several objects from the ring for given stage.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to acquire.
+ * Size of objects must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param stage
+ * Stage to acquire objects for.
+ * @param num
+ * The number of objects to acquire.
+ * @param ftoken
+ * Pointer to the opaque 'token' value used by release() op.
+ * User has to store this value somewhere, and later provide to the
+ * release().
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries for given stage
+ * after the acquire has finished.
+ * @return
+ * - Actual number of objects acquired.
+ */
+__rte_experimental
+uint32_t
+rte_soring_acquire_burst(struct rte_soring *r, void *objs,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available);
+
+/**
+ * Acquire several objects plus metadata from the ring for given stage.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to acquire.
+ * Size of objects must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each for each acquired object.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then 'meta' parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param stage
+ * Stage to acquire objects for.
+ * @param num
+ * The number of objects to acquire.
+ * @param ftoken
+ * Pointer to the opaque 'token' value used by release() op.
+ * User has to store this value somewhere, and later provide to the
+ * release().
+ * @param available
+ * If non-NULL, returns the number of remaining ring entries for given stage
+ * after the acquire has finished.
+ * @return
+ * - Actual number of objects acquired.
+ */
+__rte_experimental
+uint32_t
+rte_soring_acquirx_burst(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available);
+
+/**
+ * Release several objects for given stage back to the ring.
+ * Note that it means these objects become avaialble for next stage or
+ * dequeue.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to relase.
+ * Note that unless user needs to overwrite ring objects this parameter
+ * can be NULL.
+ * Size of objects must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param stage
+ * Current stage.
+ * @param n
+ * The number of objects to release.
+ * Has to be the same value as returned by acquire() op.
+ * @param ftoken
+ * Opaque 'token' value obtained from acquire() op.
+ */
+__rte_experimental
+void
+rte_soring_release(struct rte_soring *r, const void *objs,
+ uint32_t stage, uint32_t n, uint32_t ftoken);
+
+/**
+ * Release several objects plus metadata for given stage back to the ring.
+ * Note that it means these objects become avaialble for next stage or
+ * dequeue.
+ *
+ * @param r
+ * A pointer to the soring structure.
+ * @param objs
+ * A pointer to an array of objects to relase.
+ * Note that unless user needs to overwrite ring objects this parameter
+ * can be NULL.
+ * Size of objects must be the same value as 'elem_size' parameter
+ * used while creating the ring. Otherwise the results are undefined.
+ * @param meta
+ * A pointer to an array of metadata values for each object to release.
+ * Note that if user not using object metadata values, then this parameter
+ * can be NULL.
+ * Size of elements in this array must be the same value as 'meta_size'
+ * parameter used while creating the ring. If user created the soring with
+ * 'meta_size' value equals zero, then meta parameter should be NULL.
+ * Otherwise the results are undefined.
+ * @param stage
+ * Current stage.
+ * @param n
+ * The number of objects to release.
+ * Has to be the same value as returned by acquire() op.
+ * @param ftoken
+ * Opaque 'token' value obtained from acquire() op.
+ */
+__rte_experimental
+void
+rte_soring_releasx(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t stage, uint32_t n, uint32_t ftoken);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _RTE_SORING_H_ */
diff --git a/lib/ring/soring.c b/lib/ring/soring.c
new file mode 100644
index 0000000000..a4439f00fa
--- /dev/null
+++ b/lib/ring/soring.c
@@ -0,0 +1,548 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+/**
+ * @file
+ * This file contains implementation of SORING 'datapath' functions.
+ * Brief description:
+ * ==================
+ * enqueue/dequeue works the same as for conventional rte_ring:
+ * any rte_ring sync types can be used, etc.
+ * Plus there could be multiple 'stages'.
+ * For each stage there is an acquire (start) and release (finish) operation.
+ * After some elems are 'acquired' - user can safely assume that he has
+ * exclusive possession of these elems till 'release' for them is done.
+ * Note that right now user has to release exactly the same number of elems
+ * he acquired before.
+ * After 'release', elems can be 'acquired' by next stage and/or dequeued
+ * (in case of last stage).
+ * Internal structure:
+ * ===================
+ * In addition to 'normal' ring of elems, it also has a ring of states of the
+ * same size. Each state[] corresponds to exactly one elem[].
+ * state[] will be used by acquire/release/dequeue functions to store internal
+ * information and should not be accessed by the user directly.
+ * How it works:
+ * =============
+ * 'acquire()' just moves stage's head (same as rte_ring move_head does),
+ * plus it saves in state[stage.cur_head] information about how many elems
+ * were acquired, current head position and special flag value to indicate
+ * that elems are acquired (RTE_SORING_ST_START).
+ * Note that 'acquire()' returns to the user a special 'ftoken' that user has
+ * to provide for 'release()' (in fact it is just a position for current head).
+ * 'release()' extracts old head value from provided ftoken and checks that
+ * corresponding 'state[]' contains expected values(mostly for sanity
+ * purposes).
+ * * Then it marks this state[] with 'RTE_SORING_ST_FINISH' flag to indicate
+ * that given subset of objects was released.
+ * After that, it checks does old head value equals to current tail value?
+ * If yes, then it performs 'finalize()' operation, otherwise 'release()'
+ * just returns (without spinning on stage tail value).
+ * As updated state[] is shared by all threads, some other thread can do
+ * 'finalize()' for given stage.
+ * That allows 'release()' to avoid excessive waits on the tail value.
+ * Main purpose of 'finalize()' operation is to walk through 'state[]'
+ * from current stage tail up to its head, check state[] and move stage tail
+ * through elements that already are in RTE_SORING_ST_FINISH state.
+ * Along with that, corresponding state[] values are reset to zero.
+ * Note that 'finalize()' for given stage can be done from multiple places:
+ * 'release()' for that stage or from 'acquire()' for next stage
+ * even from consumer's 'dequeue()' - in case given stage is the last one.
+ * So 'finalize()' has to be MT-safe and inside it we have to
+ * guarantee that only one thread will update state[] and stage's tail values.
+ */
+
+#include "soring.h"
+
+/*
+ * Inline functions (fastpath) start here.
+ */
+static __rte_always_inline uint32_t
+__rte_soring_stage_finalize(struct soring_stage_headtail *sht,
+ union soring_state *rstate, uint32_t rmask, uint32_t maxn)
+{
+ int32_t rc;
+ uint32_t head, i, idx, k, n, tail;
+ union soring_stage_tail nt, ot;
+ union soring_state st;
+
+ /* try to grab exclusive right to update tail value */
+ ot.raw = rte_atomic_load_explicit(&sht->tail.raw,
+ rte_memory_order_acquire);
+
+ /* other thread already finalizing it for us */
+ if (ot.sync != 0)
+ return 0;
+
+ nt.pos = ot.pos;
+ nt.sync = 1;
+ rc = rte_atomic_compare_exchange_strong_explicit(&sht->tail.raw,
+ (uint64_t *)(uintptr_t)&ot.raw, nt.raw,
+ rte_memory_order_acquire, rte_memory_order_relaxed);
+
+ /* other thread won the race */
+ if (rc == 0)
+ return 0;
+
+ /*
+ * start with current tail and walk through states that are
+ * already finished.
+ */
+
+ head = rte_atomic_load_explicit(&sht->head, rte_memory_order_relaxed);
+ n = RTE_MIN(head - ot.pos, maxn);
+ for (i = 0, tail = ot.pos; i < n; i += k, tail += k) {
+
+ idx = tail & rmask;
+ st.raw = rte_atomic_load_explicit(&rstate[idx].raw,
+ rte_memory_order_relaxed);
+ if ((st.stnum & RTE_SORING_ST_MASK) != RTE_SORING_ST_FINISH ||
+ st.ftoken != tail)
+ break;
+
+ k = st.stnum & ~RTE_SORING_ST_MASK;
+ rte_atomic_store_explicit(&rstate[idx].raw, 0,
+ rte_memory_order_relaxed);
+ }
+
+
+ /* release exclusive right to update along with new tail value */
+ ot.pos = tail;
+ rte_atomic_store_explicit(&sht->tail.raw, ot.raw,
+ rte_memory_order_release);
+
+ return i;
+}
+
+static __rte_always_inline uint32_t
+__rte_soring_move_prod_head(struct rte_soring *r, uint32_t num,
+ enum rte_ring_queue_behavior behavior, enum rte_ring_sync_type st,
+ uint32_t *head, uint32_t *next, uint32_t *free)
+{
+ uint32_t n;
+
+ switch (st) {
+ case RTE_RING_SYNC_ST:
+ case RTE_RING_SYNC_MT:
+ n = __rte_ring_headtail_move_head(&r->prod.ht, &r->cons.ht,
+ r->capacity, st, num, behavior, head, next, free);
+ break;
+ case RTE_RING_SYNC_MT_RTS:
+ n = __rte_ring_rts_move_head(&r->prod.rts, &r->cons.ht,
+ r->capacity, num, behavior, head, free);
+ *next = *head + n;
+ break;
+ case RTE_RING_SYNC_MT_HTS:
+ n = __rte_ring_hts_move_head(&r->prod.hts, &r->cons.ht,
+ r->capacity, num, behavior, head, free);
+ *next = *head + n;
+ break;
+ default:
+ /* unsupported mode, shouldn't be here */
+ RTE_ASSERT(0);
+ *free = 0;
+ n = 0;
+ }
+
+ return n;
+}
+
+static __rte_always_inline uint32_t
+__rte_soring_move_cons_head(struct rte_soring *r, uint32_t stage, uint32_t num,
+ enum rte_ring_queue_behavior behavior, enum rte_ring_sync_type st,
+ uint32_t *head, uint32_t *next, uint32_t *avail)
+{
+ uint32_t n;
+
+ switch (st) {
+ case RTE_RING_SYNC_ST:
+ case RTE_RING_SYNC_MT:
+ n = __rte_ring_headtail_move_head(&r->cons.ht,
+ &r->stage[stage].ht, 0, st, num, behavior,
+ head, next, avail);
+ break;
+ case RTE_RING_SYNC_MT_RTS:
+ n = __rte_ring_rts_move_head(&r->cons.rts, &r->stage[stage].ht,
+ 0, num, behavior, head, avail);
+ *next = *head + n;
+ break;
+ case RTE_RING_SYNC_MT_HTS:
+ n = __rte_ring_hts_move_head(&r->cons.hts, &r->stage[stage].ht,
+ 0, num, behavior, head, avail);
+ *next = *head + n;
+ break;
+ default:
+ /* unsupported mode, shouldn't be here */
+ RTE_ASSERT(0);
+ *avail = 0;
+ n = 0;
+ }
+
+ return n;
+}
+
+static __rte_always_inline void
+__rte_soring_update_tail(struct __rte_ring_headtail *rht,
+ enum rte_ring_sync_type st, uint32_t head, uint32_t next, uint32_t enq)
+{
+ uint32_t n;
+
+ switch (st) {
+ case RTE_RING_SYNC_ST:
+ case RTE_RING_SYNC_MT:
+ __rte_ring_update_tail(&rht->ht, head, next, st, enq);
+ break;
+ case RTE_RING_SYNC_MT_RTS:
+ __rte_ring_rts_update_tail(&rht->rts);
+ break;
+ case RTE_RING_SYNC_MT_HTS:
+ n = next - head;
+ __rte_ring_hts_update_tail(&rht->hts, head, n, enq);
+ break;
+ default:
+ /* unsupported mode, shouldn't be here */
+ RTE_ASSERT(0);
+ }
+}
+
+static __rte_always_inline uint32_t
+__rte_soring_stage_move_head(struct soring_stage_headtail *d,
+ const struct rte_ring_headtail *s, uint32_t capacity, uint32_t num,
+ enum rte_ring_queue_behavior behavior,
+ uint32_t *old_head, uint32_t *new_head, uint32_t *avail)
+{
+ uint32_t n, tail;
+
+ *old_head = rte_atomic_load_explicit(&d->head,
+ rte_memory_order_acquire);
+
+ do {
+ n = num;
+ tail = rte_atomic_load_explicit(&s->tail,
+ rte_memory_order_relaxed);
+ *avail = capacity + tail - *old_head;
+ if (n > *avail)
+ n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *avail;
+ if (n == 0)
+ return 0;
+ *new_head = *old_head + n;
+ } while (rte_atomic_compare_exchange_strong_explicit(&d->head,
+ old_head, *new_head, rte_memory_order_acquire,
+ rte_memory_order_acquire) == 0);
+
+ return n;
+}
+
+static __rte_always_inline uint32_t
+soring_enqueue(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, enum rte_ring_queue_behavior behavior,
+ uint32_t *free_space)
+{
+ enum rte_ring_sync_type st;
+ uint32_t nb_free, prod_head, prod_next;
+
+ RTE_ASSERT(r != NULL && r->nb_stage > 0);
+ RTE_ASSERT(meta == NULL || r->meta != NULL);
+
+ st = r->prod.ht.sync_type;
+
+ n = __rte_soring_move_prod_head(r, n, behavior, st,
+ &prod_head, &prod_next, &nb_free);
+ if (n != 0) {
+ __rte_ring_do_enqueue_elems(&r[1], objs, r->size,
+ prod_head & r->mask, r->esize, n);
+ if (meta != NULL)
+ __rte_ring_do_enqueue_elems(r->meta, meta, r->size,
+ prod_head & r->mask, r->msize, n);
+ __rte_soring_update_tail(&r->prod, st, prod_head, prod_next, 1);
+ }
+
+ if (free_space != NULL)
+ *free_space = nb_free - n;
+ return n;
+}
+
+static __rte_always_inline uint32_t
+soring_dequeue(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, enum rte_ring_queue_behavior behavior,
+ uint32_t *available)
+{
+ enum rte_ring_sync_type st;
+ uint32_t entries, cons_head, cons_next, n, ns, reqn;
+
+ RTE_ASSERT(r != NULL && r->nb_stage > 0);
+ RTE_ASSERT(meta == NULL || r->meta != NULL);
+
+ ns = r->nb_stage - 1;
+ st = r->cons.ht.sync_type;
+
+ /* try to grab exactly @num elems first */
+ n = __rte_soring_move_cons_head(r, ns, num, RTE_RING_QUEUE_FIXED, st,
+ &cons_head, &cons_next, &entries);
+ if (n == 0) {
+ /* try to finalize some elems from previous stage */
+ n = __rte_soring_stage_finalize(&r->stage[ns].sht,
+ r->state, r->mask, 2 * num);
+ entries += n;
+
+ /* repeat attempt to grab elems */
+ reqn = (behavior == RTE_RING_QUEUE_FIXED) ? num : 0;
+ if (entries >= reqn)
+ n = __rte_soring_move_cons_head(r, ns, num, behavior,
+ st, &cons_head, &cons_next, &entries);
+ else
+ n = 0;
+ }
+
+ /* we have some elems to consume */
+ if (n != 0) {
+ __rte_ring_do_dequeue_elems(objs, &r[1], r->size,
+ cons_head & r->mask, r->esize, n);
+ if (meta != NULL)
+ __rte_ring_do_dequeue_elems(meta, r->meta, r->size,
+ cons_head & r->mask, r->msize, n);
+ __rte_soring_update_tail(&r->cons, st, cons_head, cons_next, 0);
+ }
+
+ if (available != NULL)
+ *available = entries - n;
+ return n;
+}
+
+static __rte_always_inline uint32_t
+soring_acquire(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, enum rte_ring_queue_behavior behavior,
+ uint32_t *ftoken, uint32_t *available)
+{
+ uint32_t avail, head, idx, n, next, reqn;
+ union soring_state st;
+ struct soring_stage *pstg;
+ struct soring_stage_headtail *cons;
+
+ RTE_ASSERT(r != NULL && stage < r->nb_stage);
+ RTE_ASSERT(meta == NULL || r->meta != NULL);
+
+ cons = &r->stage[stage].sht;
+
+ if (stage == 0)
+ n = __rte_soring_stage_move_head(cons, &r->prod.ht, 0, num,
+ behavior, &head, &next, &avail);
+ else {
+ pstg = r->stage + stage - 1;
+
+ /* try to grab exactly @num elems */
+ n = __rte_soring_stage_move_head(cons, &pstg->ht, 0, num,
+ RTE_RING_QUEUE_FIXED, &head, &next, &avail);
+ if (n == 0) {
+ /* try to finalize some elems from previous stage */
+ n = __rte_soring_stage_finalize(&pstg->sht, r->state,
+ r->mask, 2 * num);
+ avail += n;
+
+ /* repeat attempt to grab elems */
+ reqn = (behavior == RTE_RING_QUEUE_FIXED) ? num : 0;
+ if (avail >= reqn)
+ n = __rte_soring_stage_move_head(cons,
+ &pstg->ht, 0, num, behavior, &head,
+ &next, &avail);
+ else
+ n = 0;
+ }
+ }
+
+ if (n != 0) {
+
+ idx = head & r->mask;
+
+ /* copy elems that are ready for given stage */
+ __rte_ring_do_dequeue_elems(objs, &r[1], r->size, idx,
+ r->esize, n);
+ if (meta != NULL)
+ __rte_ring_do_dequeue_elems(meta, r->meta,
+ r->size, idx, r->msize, n);
+
+ /* update status ring */
+ st.ftoken = head;
+ st.stnum = (RTE_SORING_ST_START | n);
+
+ rte_atomic_store_explicit(&r->state[idx].raw, st.raw,
+ rte_memory_order_relaxed);
+ *ftoken = head;
+ }
+
+ if (available != NULL)
+ *available = avail - n;
+ return n;
+}
+
+static __rte_always_inline void
+soring_release(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t stage, uint32_t n, uint32_t ftoken)
+{
+ uint32_t idx, tail;
+ struct soring_stage *stg;
+ union soring_state st;
+
+ RTE_ASSERT(r != NULL && stage < r->nb_stage);
+ RTE_ASSERT(meta == NULL || r->meta != NULL);
+
+ stg = r->stage + stage;
+ tail = rte_atomic_load_explicit(&stg->sht.tail.pos,
+ rte_memory_order_relaxed);
+
+ idx = ftoken & r->mask;
+ st.raw = rte_atomic_load_explicit(&r->state[idx].raw,
+ rte_memory_order_acquire);
+
+ /* check state ring contents */
+ RTE_VERIFY(st.stnum == (RTE_SORING_ST_START | n) &&
+ st.ftoken == ftoken);
+
+ /* update contents of the ring, if necessary */
+ if (objs != NULL)
+ __rte_ring_do_enqueue_elems(&r[1], objs, r->size, idx,
+ r->esize, n);
+ if (meta != NULL)
+ __rte_ring_do_enqueue_elems(r->meta, meta, r->size, idx,
+ r->msize, n);
+
+ /* set state to FINISH, make sure it is not reordered */
+ st.stnum = RTE_SORING_ST_FINISH | n;
+ rte_atomic_store_explicit(&r->state[idx].raw, st.raw,
+ rte_memory_order_release);
+
+ if (tail == ftoken)
+ __rte_soring_stage_finalize(&stg->sht, r->state, r->mask,
+ r->capacity);
+}
+
+/*
+ * Public functions (data-path) start here.
+ */
+
+void
+rte_soring_release(struct rte_soring *r, const void *objs,
+ uint32_t stage, uint32_t n, uint32_t ftoken)
+{
+ return soring_release(r, objs, NULL, stage, n, ftoken);
+}
+
+
+void
+rte_soring_releasx(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t stage, uint32_t n, uint32_t ftoken)
+{
+ return soring_release(r, objs, meta, stage, n, ftoken);
+}
+
+uint32_t
+rte_soring_enqueue_bulk(struct rte_soring *r, const void *objs, uint32_t n,
+ uint32_t *free_space)
+{
+ return soring_enqueue(r, objs, NULL, n, RTE_RING_QUEUE_FIXED,
+ free_space);
+}
+
+uint32_t
+rte_soring_enqueux_bulk(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, uint32_t *free_space)
+{
+ return soring_enqueue(r, objs, meta, n, RTE_RING_QUEUE_FIXED,
+ free_space);
+}
+
+uint32_t
+rte_soring_enqueue_burst(struct rte_soring *r, const void *objs, uint32_t n,
+ uint32_t *free_space)
+{
+ return soring_enqueue(r, objs, NULL, n, RTE_RING_QUEUE_VARIABLE,
+ free_space);
+}
+
+uint32_t
+rte_soring_enqueux_burst(struct rte_soring *r, const void *objs,
+ const void *meta, uint32_t n, uint32_t *free_space)
+{
+ return soring_enqueue(r, objs, meta, n, RTE_RING_QUEUE_VARIABLE,
+ free_space);
+}
+
+uint32_t
+rte_soring_dequeue_bulk(struct rte_soring *r, void *objs, uint32_t num,
+ uint32_t *available)
+{
+ return soring_dequeue(r, objs, NULL, num, RTE_RING_QUEUE_FIXED,
+ available);
+}
+
+uint32_t
+rte_soring_dequeux_bulk(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, uint32_t *available)
+{
+ return soring_dequeue(r, objs, meta, num, RTE_RING_QUEUE_FIXED,
+ available);
+}
+
+uint32_t
+rte_soring_dequeue_burst(struct rte_soring *r, void *objs, uint32_t num,
+ uint32_t *available)
+{
+ return soring_dequeue(r, objs, NULL, num, RTE_RING_QUEUE_VARIABLE,
+ available);
+}
+
+uint32_t
+rte_soring_dequeux_burst(struct rte_soring *r, void *objs, void *meta,
+ uint32_t num, uint32_t *available)
+{
+ return soring_dequeue(r, objs, meta, num, RTE_RING_QUEUE_VARIABLE,
+ available);
+}
+
+uint32_t
+rte_soring_acquire_bulk(struct rte_soring *r, void *objs,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available)
+{
+ return soring_acquire(r, objs, NULL, stage, num,
+ RTE_RING_QUEUE_FIXED, ftoken, available);
+}
+
+uint32_t
+rte_soring_acquirx_bulk(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available)
+{
+ return soring_acquire(r, objs, meta, stage, num,
+ RTE_RING_QUEUE_FIXED, ftoken, available);
+}
+
+uint32_t
+rte_soring_acquire_burst(struct rte_soring *r, void *objs,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available)
+{
+ return soring_acquire(r, objs, NULL, stage, num,
+ RTE_RING_QUEUE_VARIABLE, ftoken, available);
+}
+
+uint32_t
+rte_soring_acquirx_burst(struct rte_soring *r, void *objs, void *meta,
+ uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available)
+{
+ return soring_acquire(r, objs, meta, stage, num,
+ RTE_RING_QUEUE_VARIABLE, ftoken, available);
+}
+
+unsigned int
+rte_soring_count(const struct rte_soring *r)
+{
+ uint32_t prod_tail = r->prod.ht.tail;
+ uint32_t cons_tail = r->cons.ht.tail;
+ uint32_t count = (prod_tail - cons_tail) & r->mask;
+ return (count > r->capacity) ? r->capacity : count;
+}
+
+unsigned int
+rte_soring_free_count(const struct rte_soring *r)
+{
+ return r->capacity - rte_soring_count(r);
+}
diff --git a/lib/ring/soring.h b/lib/ring/soring.h
new file mode 100644
index 0000000000..c44d8912d9
--- /dev/null
+++ b/lib/ring/soring.h
@@ -0,0 +1,124 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#ifndef _SORING_H_
+#define _SORING_H_
+
+/**
+ * @file
+ * This file contains internal strctures of RTE soring: Staged Ordered Ring.
+ * Sort of extension of conventional RTE ring.
+ * Internal structure:
+ * In addition to 'normal' ring of elems, it also has a ring of states of the
+ * same size. Each state[] corresponds to exactly one elem[].
+ * state[] will be used by acquire/release/dequeue functions to store internal
+ * information and should not be accessed by the user directly.
+ * For actual implementation details, please refer to soring.c
+ */
+
+#include <rte_soring.h>
+
+union soring_state {
+ alignas(sizeof(uint64_t)) RTE_ATOMIC(uint64_t) raw;
+ struct {
+ RTE_ATOMIC(uint32_t) ftoken;
+ RTE_ATOMIC(uint32_t) stnum;
+ };
+};
+
+/* upper 2 bits are used for status */
+#define RTE_SORING_ST_START RTE_SHIFT_VAL32(1, RTE_SORING_ST_BIT)
+#define RTE_SORING_ST_FINISH RTE_SHIFT_VAL32(2, RTE_SORING_ST_BIT)
+
+#define RTE_SORING_ST_MASK \
+ RTE_GENMASK32(sizeof(uint32_t) * CHAR_BIT - 1, RTE_SORING_ST_BIT)
+
+/**
+ * SORING re-uses rte_ring internal structures and implementation
+ * for enqueue/dequeue operations.
+ */
+struct __rte_ring_headtail {
+
+ union __rte_cache_aligned {
+ struct rte_ring_headtail ht;
+ struct rte_ring_hts_headtail hts;
+ struct rte_ring_rts_headtail rts;
+ };
+
+ RTE_CACHE_GUARD;
+};
+
+union soring_stage_tail {
+ /** raw 8B value to read/write *sync* and *pos* as one atomic op */
+ alignas(sizeof(uint64_t)) RTE_ATOMIC(uint64_t) raw;
+ struct {
+ RTE_ATOMIC(uint32_t) sync;
+ RTE_ATOMIC(uint32_t) pos;
+ };
+};
+
+struct soring_stage_headtail {
+ volatile union soring_stage_tail tail;
+ enum rte_ring_sync_type unused; /**< unused */
+ volatile RTE_ATOMIC(uint32_t) head;
+};
+
+/**
+ * SORING stage head_tail structure is 'compatible' with rte_ring ones.
+ * rte_ring internal functions for moving producer/consumer head
+ * can work transparently with stage head_tail and visa-versa
+ * (no modifications required).
+ */
+struct soring_stage {
+
+ union __rte_cache_aligned {
+ struct rte_ring_headtail ht;
+ struct soring_stage_headtail sht;
+ };
+
+ RTE_CACHE_GUARD;
+};
+
+/**
+ * RTE soring internal structure.
+ * As with rte_ring actual elements array supposed to be located direclty
+ * after the rte_soring structure.
+ */
+struct __rte_cache_aligned rte_soring {
+ uint32_t size; /**< Size of ring. */
+ uint32_t mask; /**< Mask (size-1) of ring. */
+ uint32_t capacity; /**< Usable size of ring */
+ uint32_t esize;
+ /**< size of elements in the ring, must be a multiple of 4*/
+ uint32_t msize;
+ /**< size of metadata value for each elem, must be a multiple of 4 */
+
+ /** Ring stages */
+ struct soring_stage *stage;
+ uint32_t nb_stage;
+
+ /** Ring of states (one per element) */
+ union soring_state *state;
+
+ /** Pointer to the buffer where metadata values for each elements
+ * are stored. This is supplementary and optional information that
+ * user can attach to each element of the ring.
+ * While it is possible to incorporate this information inside
+ * user-defined element, in many cases it is plausible to maintain it
+ * as a separate array (mainly for performance reasons).
+ */
+ void *meta;
+
+ RTE_CACHE_GUARD;
+
+ /** Ring producer status. */
+ struct __rte_ring_headtail prod;
+
+ /** Ring consumer status. */
+ struct __rte_ring_headtail cons;
+
+ char name[RTE_RING_NAMESIZE]; /**< Name of the ring. */
+};
+
+#endif /* _SORING_H_ */
diff --git a/lib/ring/version.map b/lib/ring/version.map
index 61f7464f5a..081034c30e 100644
--- a/lib/ring/version.map
+++ b/lib/ring/version.map
@@ -20,4 +20,23 @@ EXPERIMENTAL {
# added in 24.11
rte_ring_headtail_dump;
+ rte_soring_acquire_bulk;
+ rte_soring_acquire_burst;
+ rte_soring_acquirx_bulk;
+ rte_soring_acquirx_burst;
+ rte_soring_count;
+ rte_soring_dequeue_bulk;
+ rte_soring_dequeue_burst;
+ rte_soring_dequeux_bulk;
+ rte_soring_dequeux_burst;
+ rte_soring_enqueue_bulk;
+ rte_soring_enqueue_burst;
+ rte_soring_enqueux_bulk;
+ rte_soring_enqueux_burst;
+ rte_soring_dump;
+ rte_soring_free_count;
+ rte_soring_get_memsize;
+ rte_soring_init;
+ rte_soring_release;
+ rte_soring_releasx;
};
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread* [PATCH v4 5/5] app/test: add unit tests for soring API
2024-09-17 12:09 ` [PATCH v4 0/5] Stage-Ordered API and other extensions for ring library Konstantin Ananyev
` (3 preceding siblings ...)
2024-09-17 12:09 ` [PATCH v4 4/5] ring/soring: introduce Staged Ordered Ring Konstantin Ananyev
@ 2024-09-17 12:09 ` Konstantin Ananyev
4 siblings, 0 replies; 31+ messages in thread
From: Konstantin Ananyev @ 2024-09-17 12:09 UTC (permalink / raw)
To: dev
Cc: honnappa.nagarahalli, jerinj, hemant.agrawal, bruce.richardson,
drc, ruifeng.wang, mb, eimear.morrissey, Konstantin Ananyev
From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
Add both functional and stess test-cases for soring API.
Stress test serves as both functional and performance test of soring
enqueue/dequeue/acquire/release operations under high contention
(for both over committed and non-over committed scenarios).
Signed-off-by: Eimear Morrissey <eimear.morrissey@huawei.com>
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
---
app/test/meson.build | 3 +
app/test/test_soring.c | 442 +++++++++++++++
app/test/test_soring_mt_stress.c | 40 ++
app/test/test_soring_stress.c | 48 ++
app/test/test_soring_stress.h | 35 ++
app/test/test_soring_stress_impl.h | 827 +++++++++++++++++++++++++++++
6 files changed, 1395 insertions(+)
create mode 100644 app/test/test_soring.c
create mode 100644 app/test/test_soring_mt_stress.c
create mode 100644 app/test/test_soring_stress.c
create mode 100644 app/test/test_soring_stress.h
create mode 100644 app/test/test_soring_stress_impl.h
diff --git a/app/test/meson.build b/app/test/meson.build
index e29258e6ec..c290162e43 100644
--- a/app/test/meson.build
+++ b/app/test/meson.build
@@ -175,6 +175,9 @@ source_file_deps = {
'test_security_proto.c' : ['cryptodev', 'security'],
'test_seqlock.c': [],
'test_service_cores.c': [],
+ 'test_soring.c': [],
+ 'test_soring_mt_stress.c': [],
+ 'test_soring_stress.c': [],
'test_spinlock.c': [],
'test_stack.c': ['stack'],
'test_stack_perf.c': ['stack'],
diff --git a/app/test/test_soring.c b/app/test/test_soring.c
new file mode 100644
index 0000000000..b2110305a7
--- /dev/null
+++ b/app/test/test_soring.c
@@ -0,0 +1,442 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#include <string.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <inttypes.h>
+#include <errno.h>
+#include <sys/queue.h>
+
+#include <rte_common.h>
+#include <rte_log.h>
+#include <rte_memory.h>
+#include <rte_launch.h>
+#include <rte_cycles.h>
+#include <rte_eal.h>
+#include <rte_per_lcore.h>
+#include <rte_lcore.h>
+#include <rte_branch_prediction.h>
+#include <rte_malloc.h>
+#include <rte_random.h>
+#include <rte_errno.h>
+#include <rte_hexdump.h>
+
+#include <rte_soring.h>
+
+#include "test.h"
+
+#define MAX_ACQUIRED 20
+
+#define SORING_TEST_ASSERT(val, expected) do { \
+ RTE_TEST_ASSERT(expected == val, \
+ "%s: expected %u got %u\n", #val, expected, val); \
+} while (0)
+
+static void
+set_soring_init_param(struct rte_soring_param *prm,
+ const char *name, uint32_t esize, uint32_t elems,
+ uint32_t stages, uint32_t stsize,
+ enum rte_ring_sync_type rst_prod,
+ enum rte_ring_sync_type rst_cons)
+{
+ prm->name = name;
+ prm->elem_size = esize;
+ prm->elems = elems;
+ prm->stages = stages;
+ prm->meta_size = stsize;
+ prm->prod_synt = rst_prod;
+ prm->cons_synt = rst_cons;
+}
+
+static int
+move_forward_stage(struct rte_soring *sor,
+ uint32_t num_packets, uint32_t stage)
+{
+ uint32_t acquired;
+ uint32_t ftoken;
+ uint32_t *acquired_objs[MAX_ACQUIRED];
+
+ acquired = rte_soring_acquire_bulk(sor, acquired_objs, stage,
+ num_packets, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, num_packets);
+ rte_soring_release(sor, NULL, stage, num_packets,
+ ftoken);
+
+ return 0;
+}
+
+/*
+ * struct rte_soring_param param checking.
+ */
+static int
+test_soring_init(void)
+{
+ struct rte_soring *sor = NULL;
+ struct rte_soring_param prm;
+ int rc;
+ size_t sz;
+ memset(&prm, 0, sizeof(prm));
+
+/*init memory*/
+ set_soring_init_param(&prm, "alloc_memory", sizeof(uintptr_t),
+ 4, 1, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ sor = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ RTE_TEST_ASSERT_NOT_NULL(sor, "could not allocate memory for soring");
+
+ set_soring_init_param(&prm, "test_invalid_stages", sizeof(uintptr_t),
+ 4, 0, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with invalid num stages");
+
+ set_soring_init_param(&prm, "test_invalid_esize", 0,
+ 4, 1, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with 0 esize");
+
+ set_soring_init_param(&prm, "test_invalid_esize", 9,
+ 4, 1, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with esize not multiple of 4");
+
+ set_soring_init_param(&prm, "test_invalid_rsize", sizeof(uintptr_t),
+ 4, 1, 3, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with rcsize not multiple of 4");
+
+ set_soring_init_param(&prm, "test_invalid_elems", sizeof(uintptr_t),
+ RTE_SORING_ELEM_MAX + 1, 1, 4, RTE_RING_SYNC_MT,
+ RTE_RING_SYNC_MT);
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_FAIL(rc, "initted soring with invalid num elements");
+
+ rte_free(sor);
+ return 0;
+}
+
+static int
+test_soring_get_memsize(void)
+{
+
+ struct rte_soring_param prm;
+ ssize_t sz;
+
+ set_soring_init_param(&prm, "memsize", sizeof(uint32_t *),
+ 10, 1, 0, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ RTE_TEST_ASSERT(sz > 0, "failed to calculate size");
+
+ set_soring_init_param(&prm, "memsize", sizeof(uint8_t),
+ 16, UINT32_MAX, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ RTE_TEST_ASSERT_EQUAL(sz, -EINVAL, "calcuated size incorrect");
+
+ set_soring_init_param(&prm, "memsize", 0,
+ 16, UINT32_MAX, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ RTE_TEST_ASSERT_EQUAL(sz, -EINVAL, "calculated size incorrect");
+
+ return 0;
+
+}
+
+static int
+test_soring_stages(void)
+{
+ struct rte_soring *sor = NULL;
+ struct rte_soring_param prm;
+ uint32_t objs[32];
+ uint32_t rcs[32];
+ uint32_t acquired_objs[32];
+ uint32_t acquired_rcs[32];
+ uint32_t dequeued_rcs[32];
+ uint32_t dequeued_objs[32];
+ size_t ssz;
+ uint32_t stage, enqueued, dequeued, acquired;
+ uint32_t i, ftoken;
+
+ memset(&prm, 0, sizeof(prm));
+ set_soring_init_param(&prm, "stages", sizeof(uint32_t), 32,
+ 10000, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ ssz = rte_soring_get_memsize(&prm);
+ RTE_TEST_ASSERT(ssz > 0, "parameter error calculating ring size");
+ sor = rte_zmalloc(NULL, ssz, RTE_CACHE_LINE_SIZE);
+ RTE_TEST_ASSERT_NOT_NULL(sor, "couldn't allocate memory for soring");
+ rte_soring_init(sor, &prm);
+
+ for (i = 0; i < 32; i++) {
+ rcs[i] = i;
+ objs[i] = i + 32;
+ }
+
+ enqueued = rte_soring_enqueux_burst(sor, objs, rcs, 32, NULL);
+ SORING_TEST_ASSERT(enqueued, 32);
+
+ for (stage = 0; stage < 10000; stage++) {
+ int j;
+ dequeued = rte_soring_dequeue_bulk(sor, dequeued_objs,
+ 32, NULL);
+ /* check none at end stage */
+ SORING_TEST_ASSERT(dequeued, 0);
+
+ acquired = rte_soring_acquirx_bulk(sor, acquired_objs,
+ acquired_rcs, stage, 32, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, 32);
+
+ for (j = 0; j < 32; j++) {
+ SORING_TEST_ASSERT(acquired_rcs[j], j + stage);
+ SORING_TEST_ASSERT(acquired_objs[j], j + stage + 32);
+ /* modify both queue object and rc */
+ acquired_objs[j]++;
+ acquired_rcs[j]++;
+ }
+
+ rte_soring_releasx(sor, acquired_objs,
+ acquired_rcs, stage, 32,
+ ftoken);
+ }
+
+ dequeued = rte_soring_dequeux_bulk(sor, dequeued_objs, dequeued_rcs,
+ 32, NULL);
+ SORING_TEST_ASSERT(dequeued, 32);
+ for (i = 0; i < 32; i++) {
+ /* ensure we ended up with the expected values in order*/
+ SORING_TEST_ASSERT(dequeued_rcs[i], i + 10000);
+ SORING_TEST_ASSERT(dequeued_objs[i], i + 32 + 10000);
+ }
+ rte_free(sor);
+ return 0;
+}
+
+static int
+test_soring_enqueue_dequeue(void)
+{
+ struct rte_soring *sor = NULL;
+ struct rte_soring_param prm;
+ int rc;
+ uint32_t i;
+ size_t sz;
+ uint32_t queue_objs[10];
+ uint32_t *queue_objs_p[10];
+ uint32_t free_space;
+ uint32_t enqueued, dequeued;
+ uint32_t *dequeued_objs[10];
+
+ memset(&prm, 0, sizeof(prm));
+ for (i = 0; i < 10; i++) {
+ queue_objs[i] = i + 1;
+ queue_objs_p[i] = &queue_objs[i];
+ }
+
+/*init memory*/
+ set_soring_init_param(&prm, "enqueue/dequeue", sizeof(uint32_t *),
+ 10, 1, 0, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ sor = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ RTE_TEST_ASSERT_NOT_NULL(sor, "alloc failed for soring");
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_SUCCESS(rc, "Failed to init soring");
+
+ free_space = 0;
+
+ enqueued = rte_soring_enqueue_burst(sor, queue_objs_p, 5, &free_space);
+
+ SORING_TEST_ASSERT(free_space, 5);
+ SORING_TEST_ASSERT(enqueued, 5);
+
+ /* fixed amount enqueue */
+ enqueued = rte_soring_enqueue_bulk(sor, queue_objs_p, 7, &free_space);
+
+ SORING_TEST_ASSERT(free_space, 5);
+ SORING_TEST_ASSERT(enqueued, 0);
+
+ /* variable amount enqueue */
+ enqueued = rte_soring_enqueue_burst(sor, queue_objs_p + 5, 7,
+ &free_space);
+ SORING_TEST_ASSERT(free_space, 0);
+ SORING_TEST_ASSERT(enqueued, 5);
+
+ /* test no dequeue while stage 0 has not completed */
+ dequeued = rte_soring_dequeue_bulk(sor, dequeued_objs, 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 0);
+ dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 0);
+
+ move_forward_stage(sor, 8, 0);
+
+ /* can only dequeue as many as have completed stage*/
+ dequeued = rte_soring_dequeue_bulk(sor, dequeued_objs, 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 0);
+ dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 8);
+ /* packets remain in order */
+ for (i = 0; i < dequeued; i++) {
+ RTE_TEST_ASSERT_EQUAL(dequeued_objs[i],
+ queue_objs_p[i], "dequeued != enqueued");
+ }
+
+ dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 1, NULL);
+ SORING_TEST_ASSERT(dequeued, 0);
+
+ move_forward_stage(sor, 2, 0);
+ dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 2, NULL);
+ SORING_TEST_ASSERT(dequeued, 2);
+ /* packets remain in order */
+ for (i = 0; i < dequeued; i++) {
+ RTE_TEST_ASSERT_EQUAL(dequeued_objs[i],
+ queue_objs_p[i + 8], "dequeued != enqueued");
+ }
+
+ rte_soring_dump(stdout, sor);
+ rte_free(sor);
+ return 0;
+}
+
+static int
+test_soring_acquire_release(void)
+{
+
+ struct rte_soring *sor = NULL;
+ struct rte_soring_param prm;
+ int rc, i;
+ size_t sz;
+
+ memset(&prm, 0, sizeof(prm));
+ uint32_t queue_objs[10];
+ uint32_t rc_objs[10];
+ uint32_t acquired_objs[10];
+ uint32_t dequeued_objs[10];
+ uint32_t dequeued_rcs[10];
+ uint32_t s1_acquired_rcs[10];
+ uint32_t free_space, enqueued, ftoken, acquired, dequeued;
+
+ for (i = 0; i < 10; i++) {
+ queue_objs[i] = i + 5;
+ rc_objs[i] = i + 10;
+ }
+
+/*init memory*/
+ set_soring_init_param(&prm, "test_acquire_release", sizeof(uint32_t),
+ 20, 2, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT);
+ sz = rte_soring_get_memsize(&prm);
+ sor = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ if (sor == NULL) {
+ printf("%s: alloc(%zu) for FIFO with %u elems failed",
+ __func__, sz, prm.elems);
+ return -ENOMEM;
+ }
+
+ /* init ring */
+ rc = rte_soring_init(sor, &prm);
+ RTE_TEST_ASSERT_SUCCESS(rc, "failed to init soring");
+
+ /* enqueue with associated rc */
+ enqueued = rte_soring_enqueux_burst(sor, queue_objs, rc_objs, 5,
+ &free_space);
+ SORING_TEST_ASSERT(enqueued, 5);
+ /* enqueue without associated rc */
+ enqueued = rte_soring_enqueue_burst(sor, queue_objs + 5, 5,
+ &free_space);
+ SORING_TEST_ASSERT(enqueued, 5);
+
+ /* acquire the objects with rc's and ensure they are as expected */
+ acquired = rte_soring_acquirx_burst(sor, acquired_objs,
+ s1_acquired_rcs, 0, 5, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, 5);
+ for (i = 0; i < 5; i++) {
+ RTE_TEST_ASSERT_EQUAL(s1_acquired_rcs[i], rc_objs[i],
+ "acquired rc[%d]: %u != enqueued rc: %u",
+ i, s1_acquired_rcs[i], rc_objs[i]);
+ RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i],
+ "acquired obj[%d]: %u != enqueued obj %u",
+ i, acquired_objs[i], queue_objs[i]);
+ }
+ rte_soring_release(sor, NULL, 0, 5, ftoken);
+
+ /* acquire the objects without rc's and ensure they are as expected */
+ acquired = rte_soring_acquirx_burst(sor, acquired_objs,
+ s1_acquired_rcs, 0, 5, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, 5);
+ for (i = 0; i < 5; i++) {
+ /* as the rc area of memory is zero'd at init this is true
+ * but this is a detail of implementation rather than
+ * a guarantee.
+ */
+ RTE_TEST_ASSERT_EQUAL(s1_acquired_rcs[i], 0,
+ "acquired rc not empty");
+ RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i + 5],
+ "acquired obj[%d]: %u != enqueued obj %u",
+ i, acquired_objs[i], queue_objs[i + 5]);
+ }
+ /*release the objects, adding rc's */
+ rte_soring_releasx(sor, NULL, rc_objs + 5, 0, 5,
+ ftoken);
+
+ acquired = rte_soring_acquirx_burst(sor, acquired_objs,
+ s1_acquired_rcs, 1, 10, &ftoken, NULL);
+ SORING_TEST_ASSERT(acquired, 10);
+ for (i = 0; i < 10; i++) {
+ /* ensure the associated rc's are the ones added at relase */
+ RTE_TEST_ASSERT_EQUAL(s1_acquired_rcs[i], rc_objs[i],
+ "acquired rc[%d]: %u != enqueued rc: %u",
+ i, s1_acquired_rcs[i], rc_objs[i]);
+ RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i],
+ "acquired obj[%d]: %u != enqueued obj %u",
+ i, acquired_objs[i], queue_objs[i]);
+ }
+ /*release the objects, with rc's set to NULL */
+ rte_soring_release(sor, NULL, 1, 10, ftoken);
+
+ dequeued = rte_soring_dequeux_burst(sor, dequeued_objs, dequeued_rcs,
+ 10, NULL);
+ SORING_TEST_ASSERT(dequeued, 10);
+ for (i = 0; i < 10; i++) {
+ /* ensure the associated rc's are the ones added at relase */
+ RTE_TEST_ASSERT_EQUAL(dequeued_rcs[i], rc_objs[i],
+ "dequeued rc[%d]: %u != enqueued rc: %u",
+ i, dequeued_rcs[i], rc_objs[i]);
+ RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i],
+ "acquired obj[%d]: %u != enqueued obj %u",
+ i, acquired_objs[i], queue_objs[i]);
+ }
+ rte_soring_dump(stdout, sor);
+ rte_free(sor);
+ return 0;
+}
+
+static int
+test_soring(void)
+{
+
+ /* Negative test cases */
+ if (test_soring_init() < 0)
+ goto test_fail;
+
+ /* Memory calculations */
+ if (test_soring_get_memsize() < 0)
+ goto test_fail;
+
+ /* Basic enqueue/dequeue operations */
+ if (test_soring_enqueue_dequeue() < 0)
+ goto test_fail;
+
+ /* Acquire/release */
+ if (test_soring_acquire_release() < 0)
+ goto test_fail;
+
+ /* Test large number of stages */
+ if (test_soring_stages() < 0)
+ goto test_fail;
+
+ return 0;
+
+test_fail:
+ return -1;
+}
+
+REGISTER_FAST_TEST(soring_autotest, true, true, test_soring);
+
diff --git a/app/test/test_soring_mt_stress.c b/app/test/test_soring_mt_stress.c
new file mode 100644
index 0000000000..2f90bb4598
--- /dev/null
+++ b/app/test/test_soring_mt_stress.c
@@ -0,0 +1,40 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#include "test_soring_stress_impl.h"
+
+static inline uint32_t
+_st_ring_dequeue_burst(struct rte_soring *r, void **obj, uint32_t n,
+ uint32_t *avail)
+{
+ return rte_soring_dequeue_burst(r, obj, n, avail);
+}
+
+static inline uint32_t
+_st_ring_enqueue_bulk(struct rte_soring *r, void * const *obj, uint32_t n,
+ uint32_t *free)
+{
+ return rte_soring_enqueue_bulk(r, obj, n, free);
+}
+
+static inline uint32_t
+_st_ring_acquire_burst(struct rte_soring *r, uint32_t stage, void **obj,
+ uint32_t num, uint32_t *token, uint32_t *avail)
+{
+ return rte_soring_acquire_burst(r, obj, stage, num, token, avail);
+}
+
+static inline void
+_st_ring_release(struct rte_soring *r, uint32_t stage, uint32_t token,
+ void * const *obj, uint32_t num)
+{
+ RTE_SET_USED(obj);
+ rte_soring_release(r, NULL, stage, num, token);
+}
+
+const struct test test_soring_mt_stress = {
+ .name = "MT",
+ .nb_case = RTE_DIM(tests),
+ .cases = tests,
+};
diff --git a/app/test/test_soring_stress.c b/app/test/test_soring_stress.c
new file mode 100644
index 0000000000..334af6a29c
--- /dev/null
+++ b/app/test/test_soring_stress.c
@@ -0,0 +1,48 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+#include "test_soring_stress.h"
+
+static int
+run_test(const struct test *test)
+{
+ int32_t rc;
+ uint32_t i, k;
+
+ for (i = 0, k = 0; i != test->nb_case; i++) {
+
+ printf("TEST-CASE %s %s START\n",
+ test->name, test->cases[i].name);
+
+ rc = test->cases[i].func(test->cases[i].wfunc);
+ k += (rc == 0);
+
+ if (rc != 0)
+ printf("TEST-CASE %s %s FAILED\n",
+ test->name, test->cases[i].name);
+ else
+ printf("TEST-CASE %s %s OK\n",
+ test->name, test->cases[i].name);
+ }
+
+ return k;
+}
+
+static int
+test_ring_stress(void)
+{
+ uint32_t n, k;
+
+ n = 0;
+ k = 0;
+
+ n += test_soring_mt_stress.nb_case;
+ k += run_test(&test_soring_mt_stress);
+
+ printf("Number of tests:\t%u\nSuccess:\t%u\nFailed:\t%u\n",
+ n, k, n - k);
+ return (k != n);
+}
+
+REGISTER_TEST_COMMAND(soring_stress_autotest, test_ring_stress);
diff --git a/app/test/test_soring_stress.h b/app/test/test_soring_stress.h
new file mode 100644
index 0000000000..6190e96117
--- /dev/null
+++ b/app/test/test_soring_stress.h
@@ -0,0 +1,35 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 Huawei Technologies Co., Ltd
+ */
+
+
+#include <inttypes.h>
+#include <stddef.h>
+#include <stdalign.h>
+#include <string.h>
+#include <stdio.h>
+#include <unistd.h>
+
+#include <rte_soring.h>
+#include <rte_cycles.h>
+#include <rte_launch.h>
+#include <rte_pause.h>
+#include <rte_random.h>
+#include <rte_malloc.h>
+#include <rte_spinlock.h>
+
+#include "test.h"
+
+struct test_case {
+ const char *name;
+ int (*func)(int (*)(void *));
+ int (*wfunc)(void *arg);
+};
+
+struct test {
+ const char *name;
+ uint32_t nb_case;
+ const struct test_case *cases;
+};
+
+extern const struct test test_soring_mt_stress;
diff --git a/app/test/test_soring_stress_impl.h b/app/test/test_soring_stress_impl.h
new file mode 100644
index 0000000000..1ba78a31f5
--- /dev/null
+++ b/app/test/test_soring_stress_impl.h
@@ -0,0 +1,827 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#include <stdalign.h>
+
+#include "test_soring_stress.h"
+
+/**
+ * Stress test for soring enqueue/dequeue/acquire/release operations.
+ * Depending on the role, performs at least one of the following patterns
+ * on each worker:
+ * - dequeue/read-write data from/to the dequeued objects/enqueue.
+ * - acquire/read-write data from/to the acquired objects/release.
+ * Serves as both functional and performance test of soring
+ * data-path API under high contention
+ * (for both over committed and non-over committed scenarios).
+ */
+
+#define RING_NAME "SORING_STRESS"
+#define BULK_NUM 32
+#define RING_SIZE (2 * BULK_NUM * RTE_MAX_LCORE)
+
+#define MAX_STAGES 16
+
+enum {
+ WRK_CMD_STOP,
+ WRK_CMD_RUN,
+};
+
+static uint32_t wrk_cmd __rte_cache_aligned = WRK_CMD_STOP;
+
+/* test run-time in seconds */
+static const uint32_t run_time = 60;
+static const uint32_t verbose;
+
+struct lcore_op_stat {
+ uint64_t nb_lcore;
+ uint64_t nb_call;
+ uint64_t nb_obj;
+ uint64_t nb_cycle;
+ uint64_t max_cycle;
+ uint64_t min_cycle;
+};
+
+struct lcore_stat {
+ uint64_t nb_cycle;
+ struct lcore_op_stat deqenq;
+ uint32_t role_mask;
+ uint32_t nb_stage;
+ struct lcore_op_stat stage[MAX_STAGES];
+};
+
+#define ROLE_DEQENQ RTE_BIT32(0)
+#define ROLE_STAGE(n) RTE_BIT32(n + 1)
+
+struct lcore_arg {
+ struct rte_soring *rng;
+ struct lcore_stat stats;
+} __rte_cache_aligned;
+
+struct ring_elem {
+ uint32_t cnt[RTE_CACHE_LINE_SIZE / sizeof(uint32_t)];
+} __rte_cache_aligned;
+
+/*
+ * redefinable functions
+ */
+
+static uint32_t
+_st_ring_dequeue_burst(struct rte_soring *r, void **obj, uint32_t n,
+ uint32_t *avail);
+
+static uint32_t
+_st_ring_enqueue_bulk(struct rte_soring *r, void * const *obj, uint32_t n,
+ uint32_t *free);
+
+static uint32_t
+_st_ring_acquire_burst(struct rte_soring *r, uint32_t stage, void **obj,
+ uint32_t num, uint32_t *token, uint32_t *avail);
+
+static void
+_st_ring_release(struct rte_soring *r, uint32_t stage, uint32_t token,
+ void * const *obj, uint32_t num);
+
+static void
+lcore_op_stat_update(struct lcore_op_stat *ls, uint64_t call, uint64_t obj,
+ uint64_t tm, int32_t prcs)
+{
+ ls->nb_call += call;
+ ls->nb_obj += obj;
+ ls->nb_cycle += tm;
+ if (prcs) {
+ ls->max_cycle = RTE_MAX(ls->max_cycle, tm);
+ ls->min_cycle = RTE_MIN(ls->min_cycle, tm);
+ }
+}
+
+static void
+lcore_stat_update(struct lcore_stat *ls, uint64_t call, uint64_t obj,
+ uint64_t tm, int32_t prcs)
+{
+ uint32_t i;
+
+ ls->nb_cycle += tm;
+ lcore_op_stat_update(&ls->deqenq, call, obj, tm, prcs);
+ for (i = 0; i != ls->nb_stage; i++)
+ lcore_op_stat_update(ls->stage + i, call, obj, tm, prcs);
+}
+
+static void
+lcore_op_stat_aggr(struct lcore_op_stat *ms, const struct lcore_op_stat *ls)
+{
+ ms->nb_call += ls->nb_call;
+ ms->nb_obj += ls->nb_obj;
+ ms->nb_cycle += ls->nb_cycle;
+ ms->max_cycle = RTE_MAX(ms->max_cycle, ls->max_cycle);
+ ms->min_cycle = RTE_MIN(ms->min_cycle, ls->min_cycle);
+}
+
+static void
+lcore_stat_aggr(struct lcore_stat *ms, const struct lcore_stat *ls)
+{
+ uint32_t i;
+
+ ms->nb_cycle = RTE_MAX(ms->nb_cycle, ls->nb_cycle);
+ lcore_op_stat_aggr(&ms->deqenq, &ls->deqenq);
+ ms->deqenq.nb_lcore += ((ls->role_mask & ROLE_DEQENQ) != 0);
+ for (i = 0; i != ms->nb_stage; i++) {
+ lcore_op_stat_aggr(ms->stage + i, ls->stage + i);
+ ms->stage[i].nb_lcore += ((ls->role_mask & ROLE_STAGE(i)) != 0);
+ }
+}
+
+static void
+lcore_op_stat_dump(FILE *f, const struct lcore_op_stat *ls, const char *cap,
+ long double st)
+{
+ fprintf(f, "\t%s={\n", cap);
+
+ fprintf(f, "\t\tnb_lcore=%" PRIu64 ",\n", ls->nb_lcore);
+ fprintf(f, "\t\tnb_call=%" PRIu64 ",\n", ls->nb_call);
+ fprintf(f, "\t\tnb_obj=%" PRIu64 ",\n", ls->nb_obj);
+ fprintf(f, "\t\tnb_cycle=%" PRIu64 ",\n", ls->nb_cycle);
+ fprintf(f, "\t\tobj/call(avg): %.2Lf\n",
+ (long double)ls->nb_obj / ls->nb_call);
+ fprintf(f, "\t\tcycles/obj(avg): %.2Lf\n",
+ (long double)ls->nb_cycle / ls->nb_obj);
+ fprintf(f, "\t\tcycles/call(avg): %.2Lf\n",
+ (long double)ls->nb_cycle / ls->nb_call);
+
+ /* if min/max cycles per call stats was collected */
+ if (ls->min_cycle != UINT64_MAX) {
+ fprintf(f, "\t\tmax cycles/call=%" PRIu64 "(%.2Lf usec),\n",
+ ls->max_cycle,
+ (long double)ls->max_cycle / st);
+ fprintf(f, "\t\tmin cycles/call=%" PRIu64 "(%.2Lf usec),\n",
+ ls->min_cycle,
+ (long double)ls->min_cycle / st);
+ }
+
+ fprintf(f, "\t},\n");
+}
+
+static void
+lcore_stat_dump(FILE *f, uint32_t lc, const struct lcore_stat *ls)
+{
+ uint32_t i;
+ long double st;
+ char cap[64];
+
+ st = (long double)rte_get_timer_hz() / US_PER_S;
+
+ if (lc == UINT32_MAX)
+ fprintf(f, "%s(AGGREGATE)={\n", __func__);
+ else
+ fprintf(f, "%s(lcore=%u)={\n", __func__, lc);
+
+ fprintf(f, "\tnb_cycle=%" PRIu64 "(%.2Lf usec),\n",
+ ls->nb_cycle, (long double)ls->nb_cycle / st);
+
+ lcore_op_stat_dump(f, &ls->deqenq, "DEQ+ENQ", st);
+ for (i = 0; i != ls->nb_stage; i++) {
+ snprintf(cap, sizeof(cap), "%s#%u", "STAGE", i);
+ lcore_op_stat_dump(f, ls->stage + i, cap, st);
+ }
+
+ fprintf(f, "};\n");
+}
+
+static void
+fill_ring_elm(struct ring_elem *elm, uint32_t fill)
+{
+ uint32_t i;
+
+ for (i = 0; i != RTE_DIM(elm->cnt); i++)
+ elm->cnt[i] = fill;
+}
+
+static int32_t
+check_updt_elem(struct ring_elem *elm[], uint32_t num,
+ const struct ring_elem *check, const struct ring_elem *fill)
+{
+ uint32_t i;
+
+ static rte_spinlock_t dump_lock;
+
+ for (i = 0; i != num; i++) {
+ if (memcmp(check, elm[i], sizeof(*check)) != 0) {
+ rte_spinlock_lock(&dump_lock);
+ printf("%s(lc=%u, num=%u) failed at %u-th iter, "
+ "offending object: %p\n",
+ __func__, rte_lcore_id(), num, i, elm[i]);
+ rte_memdump(stdout, "expected", check, sizeof(*check));
+ rte_memdump(stdout, "result", elm[i], sizeof(*elm[i]));
+ rte_spinlock_unlock(&dump_lock);
+ return -EINVAL;
+ }
+ memcpy(elm[i], fill, sizeof(*elm[i]));
+ }
+
+ return 0;
+}
+
+static int
+check_ring_op(uint32_t exp, uint32_t res, uint32_t lc,
+ enum rte_ring_queue_behavior bhv, const char *fname, const char *opname)
+{
+ if (bhv == RTE_RING_QUEUE_FIXED && exp != res) {
+ printf("%s(lc=%u) failure: %s expected: %u, returned %u\n",
+ fname, lc, opname, exp, res);
+ return -ENOSPC;
+ }
+ if (bhv == RTE_RING_QUEUE_VARIABLE && exp < res) {
+ printf("%s(lc=%u) failure: %s expected: %u, returned %u\n",
+ fname, lc, opname, exp, res);
+ return -ENOSPC;
+ }
+ return 0;
+}
+
+/* num in interval [7/8, 11/8] of BULK_NUM */
+static inline uint32_t
+rand_elem_num(void)
+{
+ uint32_t num;
+
+ num = 7 * BULK_NUM / 8 + rte_rand() % (BULK_NUM / 2);
+ return num;
+}
+
+/*
+ * for each enabled stage do:
+ * acquire burst of objects
+ * read and check their contents
+ * update and check their contents
+ * release burst of objects
+ * done
+ */
+static int32_t
+test_worker_stages(struct lcore_arg *la, uint32_t lc, const char *fname,
+ struct ring_elem *obj[2 * BULK_NUM],
+ const struct ring_elem *def_elm, const struct ring_elem *loc_elm,
+ const struct ring_elem stg_elm[MAX_STAGES], int32_t prcs)
+{
+ int32_t rc;
+ uint32_t i, n, num, tkn;
+ uint64_t tm0, tm1;
+ const struct ring_elem *celm, *pelm;
+
+ num = rand_elem_num();
+
+ rc = 0;
+ tkn = 0;
+ for (i = 0, pelm = def_elm; i != la->stats.nb_stage; pelm = celm, i++) {
+
+ celm = stg_elm + i;
+
+ /* given stage is not enabled on that lcore */
+ if ((la->stats.role_mask & ROLE_STAGE(i)) == 0)
+ continue;
+
+ /* reset all pointer values */
+ memset(obj, 0, sizeof(*obj) * num);
+
+ /* acquire num elems */
+ tm0 = (prcs != 0) ? rte_rdtsc_precise() : 0;
+ n = _st_ring_acquire_burst(la->rng, i, (void **)obj, num,
+ &tkn, NULL);
+ tm0 = (prcs != 0) ? rte_rdtsc_precise() - tm0 : 0;
+
+ /* check return value and objects */
+ rc = check_ring_op(num, n, lc, RTE_RING_QUEUE_VARIABLE, fname,
+ RTE_STR(_st_ring_stage_acquire));
+ if (rc == 0)
+ rc = check_updt_elem(obj, n, pelm, loc_elm);
+ if (rc != 0)
+ break;
+
+ /* release num elems */
+ rte_compiler_barrier();
+ rc = check_updt_elem(obj, n, loc_elm, celm);
+ if (rc != 0)
+ break;
+
+ if (n == 0)
+ tm1 = 0;
+ else {
+ tm1 = (prcs != 0) ? rte_rdtsc_precise() : 0;
+ _st_ring_release(la->rng, i, tkn,
+ (void **)obj, n);
+ tm1 = (prcs != 0) ? rte_rdtsc_precise() - tm1 : 0;
+ }
+ lcore_op_stat_update(la->stats.stage + i, 1, n, tm0 + tm1,
+ prcs);
+ }
+
+ return rc;
+}
+
+static int32_t
+test_worker_deqenq(struct lcore_arg *la, uint32_t lc, const char *fname,
+ struct ring_elem *obj[2 * BULK_NUM],
+ const struct ring_elem *def_elm, const struct ring_elem *loc_elm,
+ const struct ring_elem *pelm, int32_t prcs)
+{
+ int32_t rc;
+ uint32_t k, n, num;
+ uint64_t tm0, tm1;
+
+ num = rand_elem_num();
+
+ /* reset all pointer values */
+ memset(obj, 0, sizeof(*obj) * num);
+
+ /* dequeue num elems */
+ tm0 = (prcs != 0) ? rte_rdtsc_precise() : 0;
+ n = _st_ring_dequeue_burst(la->rng, (void **)obj, num, NULL);
+
+ tm0 = (prcs != 0) ? rte_rdtsc_precise() - tm0 : 0;
+
+ /* check return value and objects */
+ rc = check_ring_op(num, n, lc, RTE_RING_QUEUE_VARIABLE, fname,
+ RTE_STR(_st_ring_dequeue_bulk));
+ if (rc == 0)
+ rc = check_updt_elem(obj, n, pelm, loc_elm);
+ if (rc != 0)
+ return rc;
+
+ /* enqueue n elems */
+ rte_compiler_barrier();
+ rc = check_updt_elem(obj, n, loc_elm, def_elm);
+ if (rc != 0)
+ return rc;
+
+ tm1 = (prcs != 0) ? rte_rdtsc_precise() : 0;
+ k = _st_ring_enqueue_bulk(la->rng, (void **)obj, n, NULL);
+ tm1 = (prcs != 0) ? rte_rdtsc_precise() - tm1 : 0;
+
+ /* check return value */
+ rc = check_ring_op(n, k, lc, RTE_RING_QUEUE_FIXED, fname,
+ RTE_STR(_st_ring_enqueue_bulk));
+ if (rc != 0)
+ return rc;
+
+ lcore_op_stat_update(&la->stats.deqenq, 1, n, tm0 + tm1, prcs);
+ return 0;
+}
+
+static int
+test_worker(void *arg, const char *fname, int32_t prcs)
+{
+ int32_t rc;
+ uint32_t i, lc;
+ uint64_t cl;
+ struct lcore_arg *la;
+ struct ring_elem *obj[2 * BULK_NUM];
+ struct ring_elem *pelm, def_elm, loc_elm, stg_elm[MAX_STAGES];
+
+ la = arg;
+ lc = rte_lcore_id();
+
+ fill_ring_elm(&def_elm, UINT32_MAX);
+ fill_ring_elm(&loc_elm, lc);
+
+ for (i = 0; i != RTE_DIM(stg_elm); i++)
+ fill_ring_elm(stg_elm + i, (i + 1) << 24);
+
+ pelm = stg_elm + la->stats.nb_stage - 1;
+
+ /* Acquire ordering is not required as the main is not
+ * really releasing any data through 'wrk_cmd' to
+ * the worker.
+ */
+ while (__atomic_load_n(&wrk_cmd, __ATOMIC_RELAXED) != WRK_CMD_RUN)
+ rte_pause();
+
+ cl = rte_rdtsc_precise();
+
+ do {
+ if ((la->stats.role_mask & ~ROLE_DEQENQ) != 0) {
+ rc = test_worker_stages(la, lc, fname, obj,
+ &def_elm, &loc_elm, stg_elm, prcs);
+ if (rc != 0)
+ break;
+ }
+
+ if ((la->stats.role_mask & ROLE_DEQENQ) != 0) {
+ rc = test_worker_deqenq(la, lc, fname, obj,
+ &def_elm, &loc_elm, pelm, prcs);
+ if (rc != 0)
+ break;
+ }
+
+ } while (__atomic_load_n(&wrk_cmd, __ATOMIC_RELAXED) == WRK_CMD_RUN);
+
+ cl = rte_rdtsc_precise() - cl;
+ if (prcs == 0)
+ lcore_stat_update(&la->stats, 0, 0, cl, 0);
+ la->stats.nb_cycle = cl;
+ return rc;
+}
+static int
+test_worker_prcs(void *arg)
+{
+ return test_worker(arg, __func__, 1);
+}
+
+static int
+test_worker_avg(void *arg)
+{
+ return test_worker(arg, __func__, 0);
+}
+
+static void
+mt1_fini(struct rte_soring *rng, void *data)
+{
+ rte_free(rng);
+ rte_free(data);
+}
+
+static int
+mt1_init(struct rte_soring **rng, void **data, uint32_t num,
+ enum rte_ring_sync_type prod_synt, enum rte_ring_sync_type cons_synt,
+ uint32_t nb_stages)
+{
+ int32_t rc;
+ size_t sz;
+ uint32_t i;
+ struct rte_soring *r;
+ struct ring_elem *elm;
+ void *p;
+ struct rte_soring_param prm;
+
+ *rng = NULL;
+ *data = NULL;
+
+ sz = num * sizeof(*elm);
+ elm = rte_zmalloc(NULL, sz, alignof(typeof(*elm)));
+ if (elm == NULL) {
+ printf("%s: alloc(%zu) for %u elems data failed",
+ __func__, sz, num);
+ return -ENOMEM;
+ }
+
+ *data = elm;
+
+ /* alloc soring */
+ memset(&prm, 0, sizeof(prm));
+
+ prm.name = __func__;
+ prm.elems = num;
+ prm.elem_size = sizeof(uintptr_t);
+ prm.stages = nb_stages;
+ prm.prod_synt = prod_synt;
+ prm.cons_synt = cons_synt;
+
+ sz = rte_soring_get_memsize(&prm);
+ r = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE);
+ if (r == NULL) {
+ printf("%s: alloc(%zu) for FIFO with %u elems failed",
+ __func__, sz, prm.elems);
+ return -ENOMEM;
+ }
+
+ *rng = r;
+
+ rc = rte_soring_init(r, &prm);
+ if (rc != 0) {
+ printf("%s: rte_soring_init(r=%p,elems=%u,stages=%u) failed, "
+ "error: %d(%s)\n",
+ __func__, r, prm.elems, prm.stages, rc, strerror(-rc));
+ return rc;
+ }
+
+ for (i = 0; i != num; i++) {
+ fill_ring_elm(elm + i, UINT32_MAX);
+ p = elm + i;
+ if (_st_ring_enqueue_bulk(r, &p, 1, NULL) != 1)
+ break;
+ }
+
+ if (i != num) {
+ printf("%s: _st_ring_enqueue(%p, %u) returned %u\n",
+ __func__, r, num, i);
+ return -ENOSPC;
+ }
+
+ return 0;
+}
+
+static int
+test_mt(int (*test)(void *), enum rte_ring_sync_type prod_synt,
+ enum rte_ring_sync_type cons_synt, uint32_t nb_stage,
+ const uint32_t role_mask[RTE_MAX_LCORE])
+{
+ int32_t rc;
+ uint32_t i, lc, mc;
+ struct rte_soring *r;
+ void *data;
+ struct lcore_arg arg[RTE_MAX_LCORE];
+
+ static const struct lcore_op_stat init_stat = {
+ .min_cycle = UINT64_MAX,
+ };
+
+ rc = mt1_init(&r, &data, RING_SIZE, prod_synt, cons_synt, nb_stage);
+
+ if (rc != 0) {
+ mt1_fini(r, data);
+ return rc;
+ }
+
+ memset(arg, 0, sizeof(arg));
+
+ /* launch on all workers */
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ arg[lc].rng = r;
+ arg[lc].stats.deqenq = init_stat;
+ arg[lc].stats.nb_stage = nb_stage;
+ arg[lc].stats.role_mask = role_mask[lc];
+ for (i = 0; i != arg[lc].stats.nb_stage; i++)
+ arg[lc].stats.stage[i] = init_stat;
+ rte_eal_remote_launch(test, &arg[lc], lc);
+ }
+
+ /* signal workers to start test */
+ __atomic_store_n(&wrk_cmd, WRK_CMD_RUN, __ATOMIC_RELEASE);
+
+ rte_delay_us(run_time * US_PER_S);
+
+ /* signal workers to stop test */
+ __atomic_store_n(&wrk_cmd, WRK_CMD_STOP, __ATOMIC_RELEASE);
+
+ /* wait for workers and collect stats. */
+ mc = rte_lcore_id();
+ arg[mc].stats.deqenq = init_stat;
+ arg[mc].stats.nb_stage = nb_stage;
+ for (i = 0; i != arg[mc].stats.nb_stage; i++)
+ arg[mc].stats.stage[i] = init_stat;
+
+ rc = 0;
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ rc |= rte_eal_wait_lcore(lc);
+ lcore_stat_aggr(&arg[mc].stats, &arg[lc].stats);
+ if (verbose != 0)
+ lcore_stat_dump(stdout, lc, &arg[lc].stats);
+ }
+
+ lcore_stat_dump(stdout, UINT32_MAX, &arg[mc].stats);
+ rte_soring_dump(stdout, r);
+ mt1_fini(r, data);
+ return rc;
+}
+
+/*
+ * launch all stages and deq+enq on all worker lcores
+ */
+static void
+role_mask_sym(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE])
+{
+ uint32_t lc;
+ const uint32_t msk = RTE_BIT32(nb_stage + 2) - 1;
+
+ memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE);
+ RTE_LCORE_FOREACH_WORKER(lc)
+ role_mask[lc] = msk;
+}
+
+/*
+ * Divide all workers in two (nearly) equal groups:
+ * - workers from 'even' group do deque+enque
+ * - workers from 'odd' group do acquire/release (for all stages)
+ */
+static void
+role_mask_even_odd(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE])
+{
+ uint32_t i, lc;
+ const uint32_t msk[2] = {
+ [0] = ROLE_DEQENQ,
+ [1] = RTE_GENMASK32(nb_stage + 1, 1),
+ };
+
+ memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE);
+
+ i = 0;
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ role_mask[lc] = msk[i & 1];
+ i++;
+ }
+ if (i == 1) {
+ lc = rte_get_next_lcore(-1, 1, 0);
+ role_mask[lc] |= msk[i & 1];
+ }
+}
+
+/*
+ * Divide all workers (nearly) evenly among all possible stages
+ */
+static void
+role_mask_div(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE])
+{
+ uint32_t i, lc;
+ uint32_t msk[nb_stage + 1];
+
+ memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE);
+
+ for (i = 0; i != RTE_DIM(msk); i++) {
+ msk[i] = RTE_BIT32(i);
+ };
+
+ i = 0;
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ role_mask[lc] = msk[i % RTE_DIM(msk)];
+ i++;
+ }
+ if (i < RTE_DIM(msk)) {
+ lc = rte_get_next_lcore(-1, 1, 0);
+ for (; i != RTE_DIM(msk); i++)
+ role_mask[lc] |= msk[i % RTE_DIM(msk)];
+ }
+}
+
+/*
+ * one worker does ST enqueue+dequeue, while all others - stages processing.
+ */
+static void
+role_mask_denq_st(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE])
+{
+ uint32_t i, lc;
+ const uint32_t msk[2] = {
+ [0] = ROLE_DEQENQ,
+ [1] = RTE_GENMASK32(nb_stage + 1, 1),
+ };
+
+ memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE);
+
+ i = 0;
+ RTE_LCORE_FOREACH_WORKER(lc) {
+ if (i == 0)
+ role_mask[lc] = msk[0];
+ else
+ role_mask[lc] = msk[1];
+ i++;
+ }
+ if (i == 1) {
+ lc = rte_get_next_lcore(-1, 1, 0);
+ role_mask[lc] |= msk[1];
+ }
+}
+
+
+static int
+test_sym_mt1(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+ const uint32_t nb_stage = 1;
+
+ role_mask_sym(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT,
+ nb_stage, role_mask);
+}
+
+static int
+test_sym_mt4(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 4;
+
+ role_mask_sym(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT,
+ nb_stage, role_mask);
+}
+
+static int
+test_sym_mt_rts4(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 4;
+
+ role_mask_sym(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT_RTS, RTE_RING_SYNC_MT_RTS,
+ nb_stage, role_mask);
+}
+
+static int
+test_sym_mt_hts4(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 4;
+
+ role_mask_sym(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT_HTS, RTE_RING_SYNC_MT_HTS,
+ nb_stage, role_mask);
+}
+
+static int
+test_stdenq_stage4(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 4;
+
+ role_mask_denq_st(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_ST, RTE_RING_SYNC_ST,
+ nb_stage, role_mask);
+}
+
+
+static int
+test_even_odd_mt5(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 5;
+
+ role_mask_even_odd(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT,
+ nb_stage, role_mask);
+}
+
+static int
+test_div_mt3(int (*test)(void *))
+{
+ uint32_t role_mask[RTE_MAX_LCORE];
+
+ const uint32_t nb_stage = 3;
+
+ role_mask_div(nb_stage, role_mask);
+ return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT,
+ nb_stage, role_mask);
+}
+
+static const struct test_case tests[] = {
+ {
+ .name = "MT_DEQENQ-MT_STG1-PRCS",
+ .func = test_sym_mt1,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG1-AVG",
+ .func = test_sym_mt1,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "ST_DEQENQ-MT_STG4-PRCS",
+ .func = test_stdenq_stage4,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "ST_DEQENQ-MT_STG4-AVG",
+ .func = test_stdenq_stage4,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG4-PRCS",
+ .func = test_sym_mt4,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG4-AVG",
+ .func = test_sym_mt4,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MTRTS_DEQENQ-MT_STG4-PRCS",
+ .func = test_sym_mt_rts4,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MTRTS_DEQENQ-MT_STG4-AVG",
+ .func = test_sym_mt_rts4,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MTHTS_DEQENQ-MT_STG4-PRCS",
+ .func = test_sym_mt_hts4,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MTHTS_DEQENQ-MT_STG4-AVG",
+ .func = test_sym_mt_hts4,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG5-1:1-PRCS",
+ .func = test_even_odd_mt5,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG5-1:1-AVG",
+ .func = test_even_odd_mt5,
+ .wfunc = test_worker_avg,
+ },
+ {
+ .name = "MT_DEQENQ-MT_STG3-1:3-PRCS",
+ .func = test_div_mt3,
+ .wfunc = test_worker_prcs,
+ },
+ {
+ .name = "MT_DEQENQ_MT_STG3-1:3-AVG",
+ .func = test_div_mt3,
+ .wfunc = test_worker_avg,
+ },
+};
--
2.35.3
^ permalink raw reply [flat|nested] 31+ messages in thread