* [PATCH 1/5] net/mlx5: fix race condition in counter pool resizing
2022-10-31 16:08 [PATCH 0/5] net/mlx5: some counter fixes Michael Baum
@ 2022-10-31 16:08 ` Michael Baum
2022-10-31 16:08 ` [PATCH 2/5] net/mlx5: fix accessing the wrong counter Michael Baum
` (4 subsequent siblings)
5 siblings, 0 replies; 7+ messages in thread
From: Michael Baum @ 2022-10-31 16:08 UTC (permalink / raw)
To: dev
Cc: Matan Azrad, Raslan Darawsheh, Viacheslav Ovsiienko, suanmingm, stable
Counter management structure has array of counter pools. This array is
invalid in management structure initialization and grows on demand.
The resizing include:
1. Allocate memory for the new size.
2. Copy the existing data to the new memory.
3. Move the pointer to the new memory.
4. Free the old memory.
The third step can be performed before for this function, and compiler
may do that, but another thread might read the pointer before coping and
read invalid data or even crash.
This patch allocates memory for this array once in management structure
initialization and limit the counters number by 16M.
Fixes: 3aa279157fa0 ("net/mlx5: synchronize flow counter pool creation")
Cc: suanmingm@nvidia.com
Cc: stable@dpdk.org
Signed-off-by: Michael Baum <michaelba@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
---
drivers/net/mlx5/mlx5.c | 28 +++++++++++++---
drivers/net/mlx5/mlx5.h | 7 ++--
drivers/net/mlx5/mlx5_flow.c | 24 +++++++-------
drivers/net/mlx5/mlx5_flow_dv.c | 53 +++++-------------------------
drivers/net/mlx5/mlx5_flow_verbs.c | 23 +++----------
5 files changed, 52 insertions(+), 83 deletions(-)
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 78234b116c..b85a56ec24 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -561,18 +561,34 @@ mlx5_flow_counter_mode_config(struct rte_eth_dev *dev __rte_unused)
*
* @param[in] sh
* Pointer to mlx5_dev_ctx_shared object to free
+ *
+ * @return
+ * 0 on success, otherwise negative errno value and rte_errno is set.
*/
-static void
+static int
mlx5_flow_counters_mng_init(struct mlx5_dev_ctx_shared *sh)
{
int i, j;
if (sh->config.dv_flow_en < 2) {
+ void *pools;
+
+ pools = mlx5_malloc(MLX5_MEM_ZERO,
+ sizeof(struct mlx5_flow_counter_pool *) *
+ MLX5_COUNTER_POOLS_MAX_NUM,
+ 0, SOCKET_ID_ANY);
+ if (!pools) {
+ DRV_LOG(ERR,
+ "Counter management allocation was failed.");
+ rte_errno = ENOMEM;
+ return -rte_errno;
+ }
memset(&sh->sws_cmng, 0, sizeof(sh->sws_cmng));
TAILQ_INIT(&sh->sws_cmng.flow_counters);
sh->sws_cmng.min_id = MLX5_CNT_BATCH_OFFSET;
sh->sws_cmng.max_id = -1;
sh->sws_cmng.last_pool_idx = POOL_IDX_INVALID;
+ sh->sws_cmng.pools = pools;
rte_spinlock_init(&sh->sws_cmng.pool_update_sl);
for (i = 0; i < MLX5_COUNTER_TYPE_MAX; i++) {
TAILQ_INIT(&sh->sws_cmng.counters[i]);
@@ -598,6 +614,7 @@ mlx5_flow_counters_mng_init(struct mlx5_dev_ctx_shared *sh)
sh->hws_max_log_bulk_sz = log_dcs;
sh->hws_max_nb_counters = max_nb_cnts;
}
+ return 0;
}
/**
@@ -655,8 +672,7 @@ mlx5_flow_counters_mng_close(struct mlx5_dev_ctx_shared *sh)
claim_zero
(mlx5_flow_os_destroy_flow_action
(cnt->action));
- if (fallback && MLX5_POOL_GET_CNT
- (pool, j)->dcs_when_free)
+ if (fallback && cnt->dcs_when_free)
claim_zero(mlx5_devx_cmd_destroy
(cnt->dcs_when_free));
}
@@ -1572,8 +1588,12 @@ mlx5_alloc_shared_dev_ctx(const struct mlx5_dev_spawn_data *spawn,
if (err)
goto error;
}
+ err = mlx5_flow_counters_mng_init(sh);
+ if (err) {
+ DRV_LOG(ERR, "Fail to initialize counters manage.");
+ goto error;
+ }
mlx5_flow_aging_init(sh);
- mlx5_flow_counters_mng_init(sh);
mlx5_flow_ipool_create(sh);
/* Add context to the global device list. */
LIST_INSERT_HEAD(&mlx5_dev_ctx_list, sh, next);
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index c9fcb71b69..cbe2d88b9e 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -386,11 +386,10 @@ struct mlx5_hw_q {
} __rte_cache_aligned;
-
-
+#define MLX5_COUNTER_POOLS_MAX_NUM (1 << 15)
#define MLX5_COUNTERS_PER_POOL 512
#define MLX5_MAX_PENDING_QUERIES 4
-#define MLX5_CNT_CONTAINER_RESIZE 64
+#define MLX5_CNT_MR_ALLOC_BULK 64
#define MLX5_CNT_SHARED_OFFSET 0x80000000
#define IS_BATCH_CNT(cnt) (((cnt) & (MLX5_CNT_SHARED_OFFSET - 1)) >= \
MLX5_CNT_BATCH_OFFSET)
@@ -549,7 +548,6 @@ TAILQ_HEAD(mlx5_counter_pools, mlx5_flow_counter_pool);
/* Counter global management structure. */
struct mlx5_flow_counter_mng {
volatile uint16_t n_valid; /* Number of valid pools. */
- uint16_t n; /* Number of pools. */
uint16_t last_pool_idx; /* Last used pool index */
int min_id; /* The minimum counter ID in the pools. */
int max_id; /* The maximum counter ID in the pools. */
@@ -621,6 +619,7 @@ struct mlx5_aso_age_action {
};
#define MLX5_ASO_AGE_ACTIONS_PER_POOL 512
+#define MLX5_ASO_AGE_CONTAINER_RESIZE 64
struct mlx5_aso_age_pool {
struct mlx5_devx_obj *flow_hit_aso_obj;
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index 8e7d649d15..e25154199f 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -9063,7 +9063,7 @@ mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
{
struct mlx5_counter_stats_mem_mng *mem_mng;
volatile struct flow_counter_stats *raw_data;
- int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
+ int raws_n = MLX5_CNT_MR_ALLOC_BULK + MLX5_MAX_PENDING_QUERIES;
int size = (sizeof(struct flow_counter_stats) *
MLX5_COUNTERS_PER_POOL +
sizeof(struct mlx5_counter_stats_raw)) * raws_n +
@@ -9101,7 +9101,7 @@ mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
}
for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
LIST_INSERT_HEAD(&sh->sws_cmng.free_stat_raws,
- mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
+ mem_mng->raws + MLX5_CNT_MR_ALLOC_BULK + i,
next);
LIST_INSERT_HEAD(&sh->sws_cmng.mem_mngs, mem_mng, next);
sh->sws_cmng.mem_mng = mem_mng;
@@ -9125,14 +9125,13 @@ mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
{
struct mlx5_flow_counter_mng *cmng = &sh->sws_cmng;
/* Resize statistic memory once used out. */
- if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
+ if (!(pool->index % MLX5_CNT_MR_ALLOC_BULK) &&
mlx5_flow_create_counter_stat_mem_mng(sh)) {
DRV_LOG(ERR, "Cannot resize counter stat mem.");
return -1;
}
rte_spinlock_lock(&pool->sl);
- pool->raw = cmng->mem_mng->raws + pool->index %
- MLX5_CNT_CONTAINER_RESIZE;
+ pool->raw = cmng->mem_mng->raws + pool->index % MLX5_CNT_MR_ALLOC_BULK;
rte_spinlock_unlock(&pool->sl);
pool->raw_hw = NULL;
return 0;
@@ -9174,13 +9173,13 @@ void
mlx5_flow_query_alarm(void *arg)
{
struct mlx5_dev_ctx_shared *sh = arg;
- int ret;
- uint16_t pool_index = sh->sws_cmng.pool_index;
struct mlx5_flow_counter_mng *cmng = &sh->sws_cmng;
+ uint16_t pool_index = cmng->pool_index;
struct mlx5_flow_counter_pool *pool;
uint16_t n_valid;
+ int ret;
- if (sh->sws_cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
+ if (cmng->pending_queries >= MLX5_MAX_PENDING_QUERIES)
goto set_alarm;
rte_spinlock_lock(&cmng->pool_update_sl);
pool = cmng->pools[pool_index];
@@ -9192,8 +9191,7 @@ mlx5_flow_query_alarm(void *arg)
if (pool->raw_hw)
/* There is a pool query in progress. */
goto set_alarm;
- pool->raw_hw =
- LIST_FIRST(&sh->sws_cmng.free_stat_raws);
+ pool->raw_hw = LIST_FIRST(&cmng->free_stat_raws);
if (!pool->raw_hw)
/* No free counter statistics raw memory. */
goto set_alarm;
@@ -9219,12 +9217,12 @@ mlx5_flow_query_alarm(void *arg)
goto set_alarm;
}
LIST_REMOVE(pool->raw_hw, next);
- sh->sws_cmng.pending_queries++;
+ cmng->pending_queries++;
pool_index++;
if (pool_index >= n_valid)
pool_index = 0;
set_alarm:
- sh->sws_cmng.pool_index = pool_index;
+ cmng->pool_index = pool_index;
mlx5_set_query_alarm(sh);
}
@@ -9755,7 +9753,7 @@ mlx5_flow_dev_dump_sh_all(struct rte_eth_dev *dev,
}
/* get counter */
- MLX5_ASSERT(cmng->n_valid <= cmng->n);
+ MLX5_ASSERT(cmng->n_valid <= MLX5_COUNTER_POOLS_MAX_NUM);
max = MLX5_COUNTERS_PER_POOL * cmng->n_valid;
for (j = 1; j <= max; j++) {
action = NULL;
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index 1e52278191..e77cbb862b 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -6091,7 +6091,7 @@ flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
/* Decrease to original index and clear shared bit. */
idx = (idx - 1) & (MLX5_CNT_SHARED_OFFSET - 1);
- MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cmng->n);
+ MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < MLX5_COUNTER_POOLS_MAX_NUM);
pool = cmng->pools[idx / MLX5_COUNTERS_PER_POOL];
MLX5_ASSERT(pool);
if (ppool)
@@ -6167,39 +6167,6 @@ flow_dv_find_pool_by_id(struct mlx5_flow_counter_mng *cmng, int id)
return pool;
}
-/**
- * Resize a counter container.
- *
- * @param[in] dev
- * Pointer to the Ethernet device structure.
- *
- * @return
- * 0 on success, otherwise negative errno value and rte_errno is set.
- */
-static int
-flow_dv_container_resize(struct rte_eth_dev *dev)
-{
- struct mlx5_priv *priv = dev->data->dev_private;
- struct mlx5_flow_counter_mng *cmng = &priv->sh->sws_cmng;
- void *old_pools = cmng->pools;
- uint32_t resize = cmng->n + MLX5_CNT_CONTAINER_RESIZE;
- uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
- void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
-
- if (!pools) {
- rte_errno = ENOMEM;
- return -ENOMEM;
- }
- if (old_pools)
- memcpy(pools, old_pools, cmng->n *
- sizeof(struct mlx5_flow_counter_pool *));
- cmng->n = resize;
- cmng->pools = pools;
- if (old_pools)
- mlx5_free(old_pools);
- return 0;
-}
-
/**
* Query a devx flow counter.
*
@@ -6251,8 +6218,6 @@ _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
* The devX counter handle.
* @param[in] age
* Whether the pool is for counter that was allocated for aging.
- * @param[in/out] cont_cur
- * Pointer to the container pointer, it will be update in pool resize.
*
* @return
* The pool container pointer on success, NULL otherwise and rte_errno is set.
@@ -6264,9 +6229,14 @@ flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
struct mlx5_priv *priv = dev->data->dev_private;
struct mlx5_flow_counter_pool *pool;
struct mlx5_flow_counter_mng *cmng = &priv->sh->sws_cmng;
- bool fallback = priv->sh->sws_cmng.counter_fallback;
+ bool fallback = cmng->counter_fallback;
uint32_t size = sizeof(*pool);
+ if (cmng->n_valid == MLX5_COUNTER_POOLS_MAX_NUM) {
+ DRV_LOG(ERR, "All counter is in used, try again later.");
+ rte_errno = EAGAIN;
+ return NULL;
+ }
size += MLX5_COUNTERS_PER_POOL * MLX5_CNT_SIZE;
size += (!age ? 0 : MLX5_COUNTERS_PER_POOL * MLX5_AGE_SIZE);
pool = mlx5_malloc(MLX5_MEM_ZERO, size, 0, SOCKET_ID_ANY);
@@ -6285,11 +6255,6 @@ flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
pool->time_of_last_age_check = MLX5_CURR_TIME_SEC;
rte_spinlock_lock(&cmng->pool_update_sl);
pool->index = cmng->n_valid;
- if (pool->index == cmng->n && flow_dv_container_resize(dev)) {
- mlx5_free(pool);
- rte_spinlock_unlock(&cmng->pool_update_sl);
- return NULL;
- }
cmng->pools[pool->index] = pool;
cmng->n_valid++;
if (unlikely(fallback)) {
@@ -12511,7 +12476,7 @@ flow_dv_aso_age_release(struct rte_eth_dev *dev, uint32_t age_idx)
}
/**
- * Resize the ASO age pools array by MLX5_CNT_CONTAINER_RESIZE pools.
+ * Resize the ASO age pools array by MLX5_ASO_AGE_CONTAINER_RESIZE pools.
*
* @param[in] dev
* Pointer to the Ethernet device structure.
@@ -12525,7 +12490,7 @@ flow_dv_aso_age_pools_resize(struct rte_eth_dev *dev)
struct mlx5_priv *priv = dev->data->dev_private;
struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
void *old_pools = mng->pools;
- uint32_t resize = mng->n + MLX5_CNT_CONTAINER_RESIZE;
+ uint32_t resize = mng->n + MLX5_ASO_AGE_CONTAINER_RESIZE;
uint32_t mem_size = sizeof(struct mlx5_aso_age_pool *) * resize;
void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
diff --git a/drivers/net/mlx5/mlx5_flow_verbs.c b/drivers/net/mlx5/mlx5_flow_verbs.c
index 81a33ddf09..4bca685674 100644
--- a/drivers/net/mlx5/mlx5_flow_verbs.c
+++ b/drivers/net/mlx5/mlx5_flow_verbs.c
@@ -232,27 +232,14 @@ flow_verbs_counter_new(struct rte_eth_dev *dev, uint32_t id __rte_unused)
break;
}
if (!cnt) {
- struct mlx5_flow_counter_pool **pools;
uint32_t size;
- if (n_valid == cmng->n) {
- /* Resize the container pool array. */
- size = sizeof(struct mlx5_flow_counter_pool *) *
- (n_valid + MLX5_CNT_CONTAINER_RESIZE);
- pools = mlx5_malloc(MLX5_MEM_ZERO, size, 0,
- SOCKET_ID_ANY);
- if (!pools)
- return 0;
- if (n_valid) {
- memcpy(pools, cmng->pools,
- sizeof(struct mlx5_flow_counter_pool *) *
- n_valid);
- mlx5_free(cmng->pools);
- }
- cmng->pools = pools;
- cmng->n += MLX5_CNT_CONTAINER_RESIZE;
+ if (n_valid == MLX5_COUNTER_POOLS_MAX_NUM) {
+ DRV_LOG(ERR, "All counter is in used, try again later.");
+ rte_errno = EAGAIN;
+ return 0;
}
- /* Allocate memory for new pool*/
+ /* Allocate memory for new pool */
size = sizeof(*pool) + sizeof(*cnt) * MLX5_COUNTERS_PER_POOL;
pool = mlx5_malloc(MLX5_MEM_ZERO, size, 0, SOCKET_ID_ANY);
if (!pool)
--
2.25.1
^ permalink raw reply [flat|nested] 7+ messages in thread* [PATCH 5/5] net/mlx5: assert for enough space in counter rings
2022-10-31 16:08 [PATCH 0/5] net/mlx5: some counter fixes Michael Baum
` (3 preceding siblings ...)
2022-10-31 16:08 ` [PATCH 4/5] net/mlx5: add assertions in counter get/put Michael Baum
@ 2022-10-31 16:08 ` Michael Baum
2022-11-03 11:40 ` [PATCH 0/5] net/mlx5: some counter fixes Raslan Darawsheh
5 siblings, 0 replies; 7+ messages in thread
From: Michael Baum @ 2022-10-31 16:08 UTC (permalink / raw)
To: dev; +Cc: Matan Azrad, Raslan Darawsheh, Viacheslav Ovsiienko, Xiaoyu Min
There is a by-design assumption in the code that the global counter
rings can contain all the port counters.
So, enqueuing to these global rings should always succeed.
Add assertions to help for debugging this assumption.
In addition, change mlx5_hws_cnt_pool_put() function to return void due
to those assumptions.
Signed-off-by: Michael Baum <michaelba@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
Acked-by: Xiaoyu Min <jackmin@nvidia.com>
---
drivers/net/mlx5/mlx5_flow_hw.c | 2 +-
drivers/net/mlx5/mlx5_hws_cnt.c | 25 ++++----
drivers/net/mlx5/mlx5_hws_cnt.h | 106 +++++++++++++++++---------------
3 files changed, 72 insertions(+), 61 deletions(-)
diff --git a/drivers/net/mlx5/mlx5_flow_hw.c b/drivers/net/mlx5/mlx5_flow_hw.c
index 2d275ad111..54a0afe45f 100644
--- a/drivers/net/mlx5/mlx5_flow_hw.c
+++ b/drivers/net/mlx5/mlx5_flow_hw.c
@@ -7874,7 +7874,7 @@ flow_hw_action_handle_destroy(struct rte_eth_dev *dev, uint32_t queue,
* time to update the AGE.
*/
mlx5_hws_age_nb_cnt_decrease(priv, age_idx);
- ret = mlx5_hws_cnt_shared_put(priv->hws_cpool, &act_idx);
+ mlx5_hws_cnt_shared_put(priv->hws_cpool, &act_idx);
break;
case MLX5_INDIRECT_ACTION_TYPE_CT:
ret = flow_hw_conntrack_destroy(dev, act_idx, error);
diff --git a/drivers/net/mlx5/mlx5_hws_cnt.c b/drivers/net/mlx5/mlx5_hws_cnt.c
index b8ce69af57..24c01eace0 100644
--- a/drivers/net/mlx5/mlx5_hws_cnt.c
+++ b/drivers/net/mlx5/mlx5_hws_cnt.c
@@ -58,13 +58,14 @@ __hws_cnt_id_load(struct mlx5_hws_cnt_pool *cpool)
static void
__mlx5_hws_cnt_svc(struct mlx5_dev_ctx_shared *sh,
- struct mlx5_hws_cnt_pool *cpool)
+ struct mlx5_hws_cnt_pool *cpool)
{
struct rte_ring *reset_list = cpool->wait_reset_list;
struct rte_ring *reuse_list = cpool->reuse_list;
uint32_t reset_cnt_num;
struct rte_ring_zc_data zcdr = {0};
struct rte_ring_zc_data zcdu = {0};
+ uint32_t ret __rte_unused;
reset_cnt_num = rte_ring_count(reset_list);
do {
@@ -72,17 +73,19 @@ __mlx5_hws_cnt_svc(struct mlx5_dev_ctx_shared *sh,
mlx5_aso_cnt_query(sh, cpool);
zcdr.n1 = 0;
zcdu.n1 = 0;
- rte_ring_enqueue_zc_burst_elem_start(reuse_list,
- sizeof(cnt_id_t), reset_cnt_num, &zcdu,
- NULL);
- rte_ring_dequeue_zc_burst_elem_start(reset_list,
- sizeof(cnt_id_t), reset_cnt_num, &zcdr,
- NULL);
+ ret = rte_ring_enqueue_zc_burst_elem_start(reuse_list,
+ sizeof(cnt_id_t),
+ reset_cnt_num, &zcdu,
+ NULL);
+ MLX5_ASSERT(ret == reset_cnt_num);
+ ret = rte_ring_dequeue_zc_burst_elem_start(reset_list,
+ sizeof(cnt_id_t),
+ reset_cnt_num, &zcdr,
+ NULL);
+ MLX5_ASSERT(ret == reset_cnt_num);
__hws_cnt_r2rcpy(&zcdu, &zcdr, reset_cnt_num);
- rte_ring_dequeue_zc_elem_finish(reset_list,
- reset_cnt_num);
- rte_ring_enqueue_zc_elem_finish(reuse_list,
- reset_cnt_num);
+ rte_ring_dequeue_zc_elem_finish(reset_list, reset_cnt_num);
+ rte_ring_enqueue_zc_elem_finish(reuse_list, reset_cnt_num);
reset_cnt_num = rte_ring_count(reset_list);
} while (reset_cnt_num > 0);
}
diff --git a/drivers/net/mlx5/mlx5_hws_cnt.h b/drivers/net/mlx5/mlx5_hws_cnt.h
index 338ee4d688..030dcead86 100644
--- a/drivers/net/mlx5/mlx5_hws_cnt.h
+++ b/drivers/net/mlx5/mlx5_hws_cnt.h
@@ -116,7 +116,7 @@ enum {
HWS_AGE_CANDIDATE_INSIDE_RING,
/*
* AGE assigned to flows but it still in ring. It was aged-out but the
- * timeout was changed, so it in ring but stiil candidate.
+ * timeout was changed, so it in ring but still candidate.
*/
HWS_AGE_AGED_OUT_REPORTED,
/*
@@ -182,7 +182,7 @@ mlx5_hws_cnt_id_valid(cnt_id_t cnt_id)
*
* @param cpool
* The pointer to counter pool
- * @param index
+ * @param iidx
* The internal counter index.
*
* @return
@@ -231,32 +231,32 @@ __hws_cnt_query_raw(struct mlx5_hws_cnt_pool *cpool, cnt_id_t cnt_id,
}
/**
- * Copy elems from one zero-copy ring to zero-copy ring in place.
+ * Copy elements from one zero-copy ring to zero-copy ring in place.
*
* The input is a rte ring zero-copy data struct, which has two pointer.
* in case of the wrapper happened, the ptr2 will be meaningful.
*
- * So this rountin needs to consider the situation that the address given by
+ * So this routine needs to consider the situation that the address given by
* source and destination could be both wrapped.
* First, calculate the first number of element needs to be copied until wrapped
* address, which could be in source or destination.
* Second, copy left number of element until second wrapped address. If in first
* step the wrapped address is source, then this time it must be in destination.
- * and vice-vers.
- * Third, copy all left numbe of element.
+ * and vice-versa.
+ * Third, copy all left number of element.
*
* In worst case, we need copy three pieces of continuous memory.
*
* @param zcdd
- * A pointer to zero-copy data of dest ring.
+ * A pointer to zero-copy data of destination ring.
* @param zcds
* A pointer to zero-copy data of source ring.
* @param n
- * Number of elems to copy.
+ * Number of elements to copy.
*/
static __rte_always_inline void
__hws_cnt_r2rcpy(struct rte_ring_zc_data *zcdd, struct rte_ring_zc_data *zcds,
- unsigned int n)
+ unsigned int n)
{
unsigned int n1, n2, n3;
void *s1, *s2, *s3;
@@ -291,22 +291,23 @@ static __rte_always_inline int
mlx5_hws_cnt_pool_cache_flush(struct mlx5_hws_cnt_pool *cpool,
uint32_t queue_id)
{
- unsigned int ret;
+ unsigned int ret __rte_unused;
struct rte_ring_zc_data zcdr = {0};
struct rte_ring_zc_data zcdc = {0};
struct rte_ring *reset_list = NULL;
struct rte_ring *qcache = cpool->cache->qcache[queue_id];
+ uint32_t ring_size = rte_ring_count(qcache);
- ret = rte_ring_dequeue_zc_burst_elem_start(qcache,
- sizeof(cnt_id_t), rte_ring_count(qcache), &zcdc,
- NULL);
- MLX5_ASSERT(ret);
+ ret = rte_ring_dequeue_zc_burst_elem_start(qcache, sizeof(cnt_id_t),
+ ring_size, &zcdc, NULL);
+ MLX5_ASSERT(ret == ring_size);
reset_list = cpool->wait_reset_list;
- rte_ring_enqueue_zc_burst_elem_start(reset_list,
- sizeof(cnt_id_t), ret, &zcdr, NULL);
- __hws_cnt_r2rcpy(&zcdr, &zcdc, ret);
- rte_ring_enqueue_zc_elem_finish(reset_list, ret);
- rte_ring_dequeue_zc_elem_finish(qcache, ret);
+ ret = rte_ring_enqueue_zc_burst_elem_start(reset_list, sizeof(cnt_id_t),
+ ring_size, &zcdr, NULL);
+ MLX5_ASSERT(ret == ring_size);
+ __hws_cnt_r2rcpy(&zcdr, &zcdc, ring_size);
+ rte_ring_enqueue_zc_elem_finish(reset_list, ring_size);
+ rte_ring_dequeue_zc_elem_finish(qcache, ring_size);
return 0;
}
@@ -323,7 +324,7 @@ mlx5_hws_cnt_pool_cache_fetch(struct mlx5_hws_cnt_pool *cpool,
struct rte_ring_zc_data zcdu = {0};
struct rte_ring_zc_data zcds = {0};
struct mlx5_hws_cnt_pool_caches *cache = cpool->cache;
- unsigned int ret;
+ unsigned int ret, actual_fetch_size __rte_unused;
reuse_list = cpool->reuse_list;
ret = rte_ring_dequeue_zc_burst_elem_start(reuse_list,
@@ -334,7 +335,9 @@ mlx5_hws_cnt_pool_cache_fetch(struct mlx5_hws_cnt_pool *cpool,
rte_ring_dequeue_zc_elem_finish(reuse_list, 0);
free_list = cpool->free_list;
ret = rte_ring_dequeue_zc_burst_elem_start(free_list,
- sizeof(cnt_id_t), cache->fetch_sz, &zcdf, NULL);
+ sizeof(cnt_id_t),
+ cache->fetch_sz,
+ &zcdf, NULL);
zcds = zcdf;
list = free_list;
if (unlikely(ret == 0)) { /* no free counter. */
@@ -344,8 +347,10 @@ mlx5_hws_cnt_pool_cache_fetch(struct mlx5_hws_cnt_pool *cpool,
return -ENOENT;
}
}
- rte_ring_enqueue_zc_burst_elem_start(qcache, sizeof(cnt_id_t),
- ret, &zcdc, NULL);
+ actual_fetch_size = ret;
+ ret = rte_ring_enqueue_zc_burst_elem_start(qcache, sizeof(cnt_id_t),
+ ret, &zcdc, NULL);
+ MLX5_ASSERT(ret == actual_fetch_size);
__hws_cnt_r2rcpy(&zcdc, &zcds, ret);
rte_ring_dequeue_zc_elem_finish(list, ret);
rte_ring_enqueue_zc_elem_finish(qcache, ret);
@@ -378,15 +383,14 @@ __mlx5_hws_cnt_pool_enqueue_revert(struct rte_ring *r, unsigned int n,
*
* @param cpool
* A pointer to the counter pool structure.
+ * @param queue
+ * A pointer to HWS queue. If null, it means put into common pool.
* @param cnt_id
* A counter id to be added.
- * @return
- * - 0: Success; object taken
- * - -ENOENT: not enough entry in pool
*/
-static __rte_always_inline int
-mlx5_hws_cnt_pool_put(struct mlx5_hws_cnt_pool *cpool,
- uint32_t *queue, cnt_id_t *cnt_id)
+static __rte_always_inline void
+mlx5_hws_cnt_pool_put(struct mlx5_hws_cnt_pool *cpool, uint32_t *queue,
+ cnt_id_t *cnt_id)
{
unsigned int ret = 0;
struct rte_ring_zc_data zcdc = {0};
@@ -404,25 +408,29 @@ mlx5_hws_cnt_pool_put(struct mlx5_hws_cnt_pool *cpool,
qcache = cpool->cache->qcache[*queue];
if (unlikely(qcache == NULL)) {
ret = rte_ring_enqueue_elem(cpool->wait_reset_list, cnt_id,
- sizeof(cnt_id_t));
+ sizeof(cnt_id_t));
MLX5_ASSERT(ret == 0);
- return ret;
+ return;
}
ret = rte_ring_enqueue_burst_elem(qcache, cnt_id, sizeof(cnt_id_t), 1,
NULL);
if (unlikely(ret == 0)) { /* cache is full. */
+ struct rte_ring *reset_list = cpool->wait_reset_list;
+
wb_num = rte_ring_count(qcache) - cpool->cache->threshold;
MLX5_ASSERT(wb_num < rte_ring_count(qcache));
__mlx5_hws_cnt_pool_enqueue_revert(qcache, wb_num, &zcdc);
- rte_ring_enqueue_zc_burst_elem_start(cpool->wait_reset_list,
- sizeof(cnt_id_t), wb_num, &zcdr, NULL);
- __hws_cnt_r2rcpy(&zcdr, &zcdc, wb_num);
- rte_ring_enqueue_zc_elem_finish(cpool->wait_reset_list, wb_num);
+ ret = rte_ring_enqueue_zc_burst_elem_start(reset_list,
+ sizeof(cnt_id_t),
+ wb_num, &zcdr, NULL);
+ MLX5_ASSERT(ret == wb_num);
+ __hws_cnt_r2rcpy(&zcdr, &zcdc, ret);
+ rte_ring_enqueue_zc_elem_finish(reset_list, ret);
/* write-back THIS counter too */
- ret = rte_ring_enqueue_burst_elem(cpool->wait_reset_list,
- cnt_id, sizeof(cnt_id_t), 1, NULL);
+ ret = rte_ring_enqueue_burst_elem(reset_list, cnt_id,
+ sizeof(cnt_id_t), 1, NULL);
}
- return ret == 1 ? 0 : -ENOENT;
+ MLX5_ASSERT(ret == 1);
}
/**
@@ -482,15 +490,17 @@ mlx5_hws_cnt_pool_get(struct mlx5_hws_cnt_pool *cpool, uint32_t *queue,
return 0;
}
ret = rte_ring_dequeue_zc_burst_elem_start(qcache, sizeof(cnt_id_t), 1,
- &zcdc, NULL);
+ &zcdc, NULL);
if (unlikely(ret == 0)) { /* local cache is empty. */
rte_ring_dequeue_zc_elem_finish(qcache, 0);
/* let's fetch from global free list. */
ret = mlx5_hws_cnt_pool_cache_fetch(cpool, *queue);
if (unlikely(ret != 0))
return ret;
- rte_ring_dequeue_zc_burst_elem_start(qcache, sizeof(cnt_id_t),
- 1, &zcdc, NULL);
+ ret = rte_ring_dequeue_zc_burst_elem_start(qcache,
+ sizeof(cnt_id_t), 1,
+ &zcdc, NULL);
+ MLX5_ASSERT(ret == 1);
}
/* get one from local cache. */
*cnt_id = (*(cnt_id_t *)zcdc.ptr1);
@@ -504,8 +514,10 @@ mlx5_hws_cnt_pool_get(struct mlx5_hws_cnt_pool *cpool, uint32_t *queue,
ret = mlx5_hws_cnt_pool_cache_fetch(cpool, *queue);
if (unlikely(ret != 0))
return ret;
- rte_ring_dequeue_zc_burst_elem_start(qcache, sizeof(cnt_id_t),
- 1, &zcdc, NULL);
+ ret = rte_ring_dequeue_zc_burst_elem_start(qcache,
+ sizeof(cnt_id_t), 1,
+ &zcdc, NULL);
+ MLX5_ASSERT(ret == 1);
*cnt_id = *(cnt_id_t *)zcdc.ptr1;
iidx = mlx5_hws_cnt_iidx(cpool, *cnt_id);
}
@@ -553,17 +565,13 @@ mlx5_hws_cnt_shared_get(struct mlx5_hws_cnt_pool *cpool, cnt_id_t *cnt_id,
return 0;
}
-static __rte_always_inline int
+static __rte_always_inline void
mlx5_hws_cnt_shared_put(struct mlx5_hws_cnt_pool *cpool, cnt_id_t *cnt_id)
{
- int ret;
uint32_t iidx = mlx5_hws_cnt_iidx(cpool, *cnt_id);
cpool->pool[iidx].share = 0;
- ret = mlx5_hws_cnt_pool_put(cpool, NULL, cnt_id);
- if (unlikely(ret != 0))
- cpool->pool[iidx].share = 1; /* fail to release, restore. */
- return ret;
+ mlx5_hws_cnt_pool_put(cpool, NULL, cnt_id);
}
static __rte_always_inline bool
--
2.25.1
^ permalink raw reply [flat|nested] 7+ messages in thread