* [PATCH] cnxk: use stdatomic API
@ 2025-03-27 11:11 pbhagavatula
2025-05-12 13:58 ` [PATCH v2] " pbhagavatula
0 siblings, 1 reply; 3+ messages in thread
From: pbhagavatula @ 2025-03-27 11:11 UTC (permalink / raw)
To: jerinj, Ankur Dwivedi, Anoob Joseph, Tejasree Kondoj,
Nithin Dabilpuram, Kiran Kumar K, Sunil Kumar Kori, Satha Rao,
Harman Kalra
Cc: dev, Pavan Nikhilesh
From: Pavan Nikhilesh <pbhagavatula@marvell.com>
Replace GCC inbuilt atomic functions with rte_atomic_xxx API.
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
drivers/crypto/cnxk/cn10k_cryptodev_ops.c | 36 ++++++++++---------
drivers/crypto/cnxk/cn9k_cryptodev_ops.c | 20 +++++------
drivers/net/cnxk/cn10k_tx.h | 42 +++++++++++++----------
drivers/net/cnxk/cn20k_tx.h | 39 ++++++++++++---------
drivers/net/cnxk/cn9k_tx.h | 5 +--
5 files changed, 79 insertions(+), 63 deletions(-)
diff --git a/drivers/crypto/cnxk/cn10k_cryptodev_ops.c b/drivers/crypto/cnxk/cn10k_cryptodev_ops.c
index dbebc5aef1..fe28c7529d 100644
--- a/drivers/crypto/cnxk/cn10k_cryptodev_ops.c
+++ b/drivers/crypto/cnxk/cn10k_cryptodev_ops.c
@@ -292,7 +292,7 @@ cn10k_cpt_fill_inst(struct cnxk_cpt_qp *qp, struct rte_crypto_op *ops[], struct
}
inst[0].res_addr = (uint64_t)&infl_req->res;
- __atomic_store_n(&infl_req->res.u64[0], res.u64[0], __ATOMIC_RELAXED);
+ rte_atomic_store_explicit(&infl_req->res.u64[0], res.u64[0], rte_memory_order_relaxed);
infl_req->cop = op;
inst[0].w7.u64 = w7;
@@ -361,7 +361,7 @@ cn10k_cpt_enqueue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops,
inst = (struct cpt_inst_s *)lmt_base;
again:
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (unlikely(fc.s.qsize > fc_thresh)) {
i = 0;
goto pend_q_commit;
@@ -392,7 +392,7 @@ cn10k_cpt_enqueue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops,
}
pend_q_commit:
- rte_atomic_thread_fence(__ATOMIC_RELEASE);
+ rte_atomic_thread_fence(rte_memory_order_release);
pend_q->head = head;
pend_q->time_out = rte_get_timer_cycles() +
@@ -541,7 +541,7 @@ cn10k_cpt_vec_inst_fill(struct vec_request *vec_req, struct cpt_inst_s *inst,
infl_req->qp = qp;
inst->res_addr = (uint64_t)&infl_req->res;
- __atomic_store_n(&infl_req->res.u64[0], res.u64[0], __ATOMIC_RELAXED);
+ rte_atomic_store_explicit(&infl_req->res.u64[0], res.u64[0], rte_memory_order_relaxed);
inst->w0.u64 = 0;
inst->w2.u64 = vec_req->w2;
@@ -593,7 +593,7 @@ cn10k_cpt_vec_submit(struct vec_request vec_tbl[], uint16_t vec_tbl_len, struct
cn10k_cpt_vec_inst_fill(&vec_tbl[i], &inst[i], qp, vec_tbl[0].w7);
do {
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (likely(fc.s.qsize < fc_thresh))
break;
if (unlikely(rte_get_timer_cycles() > timeout))
@@ -644,7 +644,7 @@ ca_lmtst_vec_submit(struct ops_burst *burst, struct vec_request vec_tbl[], uint1
#endif
/* Perform fc check before putting packets into vectors */
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (unlikely(fc.s.qsize > fc_thresh)) {
rte_errno = EAGAIN;
return 0;
@@ -789,7 +789,7 @@ ca_lmtst_burst_submit(struct ops_burst *burst, const bool is_sg_ver2)
inst->w3.u64 = CNXK_CPT_INST_W3(1, infl_req);
}
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (unlikely(fc.s.qsize > fc_thresh)) {
rte_errno = EAGAIN;
for (j = 0; j < i; j++) {
@@ -1237,7 +1237,7 @@ cn10k_cpt_crypto_adapter_dequeue(uintptr_t get_work1)
cop = infl_req->cop;
qp = infl_req->qp;
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0], __ATOMIC_RELAXED);
+ res.u64[0] = rte_atomic_load_explicit(&infl_req->res.u64[0], rte_memory_order_relaxed);
cn10k_cpt_dequeue_post_process(qp, infl_req->cop, infl_req, &res.cn10k);
@@ -1267,7 +1267,7 @@ cn10k_cpt_crypto_adapter_vector_dequeue(uintptr_t get_work1)
req_mp = qp->ca.req_mp;
#ifdef CNXK_CRYPTODEV_DEBUG
- res.u64[0] = __atomic_load_n(&vec_infl_req->res.u64[0], __ATOMIC_RELAXED);
+ res.u64[0] = rte_atomic_load_explicit(&vec_infl_req->res.u64[0], rte_memory_order_relaxed);
PLT_ASSERT(res.cn10k.compcode == CPT_COMP_GOOD);
PLT_ASSERT(res.cn10k.uc_compcode == 0);
#endif
@@ -1276,7 +1276,8 @@ cn10k_cpt_crypto_adapter_vector_dequeue(uintptr_t get_work1)
infl_req = vec->ptrs[i];
cop = infl_req->cop;
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0], __ATOMIC_RELAXED);
+ res.u64[0] =
+ rte_atomic_load_explicit(&infl_req->res.u64[0], rte_memory_order_relaxed);
cn10k_cpt_dequeue_post_process(qp, cop, infl_req, &res.cn10k);
vec->ptrs[i] = cop;
@@ -1316,8 +1317,8 @@ cn10k_cpt_dequeue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops)
for (i = 0; i < nb_ops; i++) {
infl_req = &pend_q->req_queue[pq_tail];
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0],
- __ATOMIC_RELAXED);
+ res.u64[0] =
+ rte_atomic_load_explicit(&infl_req->res.u64[0], rte_memory_order_relaxed);
if (unlikely(res.cn10k.compcode == CPT_COMP_NOT_DONE)) {
if (unlikely(rte_get_timer_cycles() >
@@ -1569,7 +1570,7 @@ cn10k_cpt_raw_fill_inst(struct cnxk_iov *iov, struct cnxk_cpt_qp *qp,
return 0;
inst[0].res_addr = (uint64_t)&infl_req->res;
- __atomic_store_n(&infl_req->res.u64[0], res.u64[0], __ATOMIC_RELAXED);
+ rte_atomic_store_explicit(&infl_req->res.u64[0], res.u64[0], rte_memory_order_relaxed);
infl_req->opaque = opaque;
inst[0].w7.u64 = sess->cpt_inst_w7;
@@ -1615,7 +1616,7 @@ cn10k_cpt_raw_enqueue_burst(void *qpair, uint8_t *drv_ctx, struct rte_crypto_sym
dp_ctx = (struct cnxk_sym_dp_ctx *)drv_ctx;
again:
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (unlikely(fc.s.qsize > fc_thresh)) {
i = 0;
goto pend_q_commit;
@@ -1716,7 +1717,7 @@ cn10k_cpt_raw_enqueue(void *qpair, uint8_t *drv_ctx, struct rte_crypto_vec *data
ROC_LMT_BASE_ID_GET(lmt_base, lmt_id);
inst = (struct cpt_inst_s *)lmt_base;
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (unlikely(fc.s.qsize > fc_thresh))
return -1;
@@ -1819,7 +1820,8 @@ cn10k_cpt_sym_raw_dequeue_burst(void *qptr, uint8_t *drv_ctx,
is_op_success = 0;
infl_req = &pend_q->req_queue[pq_tail];
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0], __ATOMIC_RELAXED);
+ res.u64[0] =
+ rte_atomic_load_explicit(&infl_req->res.u64[0], rte_memory_order_relaxed);
if (unlikely(res.cn10k.compcode == CPT_COMP_NOT_DONE)) {
if (unlikely(rte_get_timer_cycles() > pend_q->time_out)) {
@@ -1880,7 +1882,7 @@ cn10k_cpt_sym_raw_dequeue(void *qptr, uint8_t *drv_ctx, int *dequeue_status,
infl_req = &pend_q->req_queue[pq_tail];
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0], __ATOMIC_RELAXED);
+ res.u64[0] = rte_atomic_load_explicit(&infl_req->res.u64[0], rte_memory_order_relaxed);
if (unlikely(res.cn10k.compcode == CPT_COMP_NOT_DONE)) {
if (unlikely(rte_get_timer_cycles() > pend_q->time_out)) {
diff --git a/drivers/crypto/cnxk/cn9k_cryptodev_ops.c b/drivers/crypto/cnxk/cn9k_cryptodev_ops.c
index 2b1bd70c6d..e55f5690a3 100644
--- a/drivers/crypto/cnxk/cn9k_cryptodev_ops.c
+++ b/drivers/crypto/cnxk/cn9k_cryptodev_ops.c
@@ -194,12 +194,12 @@ cn9k_cpt_enqueue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops)
infl_req_1->op_flags = 0;
infl_req_2->op_flags = 0;
- __atomic_store_n(&infl_req_1->res.u64[0], res.u64[0],
- __ATOMIC_RELAXED);
+ rte_atomic_store_explicit(&infl_req_1->res.u64[0], res.u64[0],
+ rte_memory_order_relaxed);
inst[0].res_addr = (uint64_t)&infl_req_1->res;
- __atomic_store_n(&infl_req_2->res.u64[0], res.u64[0],
- __ATOMIC_RELAXED);
+ rte_atomic_store_explicit(&infl_req_2->res.u64[0], res.u64[0],
+ rte_memory_order_relaxed);
inst[1].res_addr = (uint64_t)&infl_req_2->res;
ret = cn9k_cpt_inst_prep(qp, op_1, infl_req_1, &inst[0]);
@@ -223,7 +223,7 @@ cn9k_cpt_enqueue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops)
count += 2;
}
- rte_atomic_thread_fence(__ATOMIC_RELEASE);
+ rte_atomic_thread_fence(rte_memory_order_release);
pend_q->head = head;
pend_q->time_out = rte_get_timer_cycles() +
@@ -385,7 +385,7 @@ cn9k_cpt_crypto_adapter_enqueue(uintptr_t base, struct rte_crypto_op *op)
const uint32_t fc_thresh = qp->lmtline.fc_thresh;
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (unlikely(fc.s.qsize > fc_thresh)) {
rte_mempool_put(qp->ca.req_mp, infl_req);
rte_errno = EAGAIN;
@@ -604,7 +604,7 @@ cn9k_cpt_crypto_adapter_dequeue(uintptr_t get_work1)
cop = infl_req->cop;
qp = infl_req->qp;
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0], __ATOMIC_RELAXED);
+ res.u64[0] = rte_atomic_load_explicit(&infl_req->res.u64[0], rte_memory_order_relaxed);
cn9k_cpt_dequeue_post_process(qp, infl_req->cop, infl_req, &res.cn9k);
@@ -635,13 +635,13 @@ cn9k_cpt_dequeue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops)
nb_ops = RTE_MIN(nb_ops, infl_cnt);
/* Ensure infl_cnt isn't read before data lands */
- rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
+ rte_atomic_thread_fence(rte_memory_order_acquire);
for (i = 0; i < nb_ops; i++) {
infl_req = &pend_q->req_queue[pq_tail];
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0],
- __ATOMIC_RELAXED);
+ res.u64[0] =
+ rte_atomic_load_explicit(&infl_req->res.u64[0], rte_memory_order_relaxed);
if (unlikely(res.cn9k.compcode == CPT_COMP_NOT_DONE)) {
if (unlikely(rte_get_timer_cycles() >
diff --git a/drivers/net/cnxk/cn10k_tx.h b/drivers/net/cnxk/cn10k_tx.h
index 809fafb2f7..f6dc69983c 100644
--- a/drivers/net/cnxk/cn10k_tx.h
+++ b/drivers/net/cnxk/cn10k_tx.h
@@ -55,17 +55,19 @@
int64_t val, newval; \
retry: \
/* Reduce the cached count */ \
- val = (int64_t)__atomic_fetch_sub(fc_cache, pkts, __ATOMIC_RELAXED); \
+ val = (int64_t)rte_atomic_fetch_sub_explicit(fc_cache, pkts, \
+ rte_memory_order_relaxed); \
val -= pkts; \
/* Cached value is low, Update the fc_cache_pkts */ \
if (unlikely(val < 0)) { \
/* Multiply with sqe_per_sqb to express in pkts */ \
- newval = txq->nb_sqb_bufs_adj - __atomic_load_n(txq->fc_mem, \
- __ATOMIC_RELAXED); \
+ newval = txq->nb_sqb_bufs_adj - \
+ rte_atomic_load_explicit(txq->fc_mem, rte_memory_order_relaxed); \
newval = (newval << (txq)->sqes_per_sqb_log2) - newval; \
newval -= pkts; \
- if (!__atomic_compare_exchange_n(fc_cache, &val, newval, false, \
- __ATOMIC_RELAXED, __ATOMIC_RELAXED)) { \
+ if (!rte_atomic_compare_exchange_strong_explicit( \
+ fc_cache, &val, newval, rte_memory_order_relaxed, \
+ rte_memory_order_relaxed)) { \
if (retry_count) { \
retry_count--; \
goto retry; \
@@ -164,10 +166,11 @@ cn10k_nix_vwqe_wait_fc(struct cn10k_eth_txq *txq, uint16_t req)
: "memory");
#else
RTE_SET_USED(pkts);
- while (__atomic_load_n(&txq->fc_cache_pkts, __ATOMIC_RELAXED) < 0)
+ while (rte_atomic_load_explicit(&txq->fc_cache_pkts, rte_memory_order_relaxed) < 0)
;
#endif
- cached = __atomic_fetch_sub(&txq->fc_cache_pkts, req, __ATOMIC_ACQUIRE) - req;
+ cached = rte_atomic_fetch_sub_explicit(&txq->fc_cache_pkts, req, rte_memory_order_acquire) -
+ req;
/* Check if there is enough space, else update and retry. */
if (cached >= 0)
return;
@@ -200,14 +203,15 @@ cn10k_nix_vwqe_wait_fc(struct cn10k_eth_txq *txq, uint16_t req)
: "memory");
#else
do {
- refill = (txq->nb_sqb_bufs_adj - __atomic_load_n(txq->fc_mem, __ATOMIC_RELAXED));
+ refill = (txq->nb_sqb_bufs_adj -
+ rte_atomic_load_explicit(txq->fc_mem, rte_memory_order_relaxed));
refill = (refill << txq->sqes_per_sqb_log2) - refill;
refill -= req;
} while (refill < 0);
#endif
- if (!__atomic_compare_exchange(&txq->fc_cache_pkts, &cached, &refill,
- 0, __ATOMIC_RELEASE,
- __ATOMIC_RELAXED))
+ if (!rte_atomic_compare_exchange_strong_explicit(&txq->fc_cache_pkts, &cached, refill,
+ rte_memory_order_release,
+ rte_memory_order_relaxed))
goto retry;
}
@@ -365,7 +369,7 @@ cn10k_nix_sec_fc_wait_one(struct cn10k_eth_txq *txq)
: "memory");
#else
RTE_SET_USED(fc);
- while (nb_desc <= __atomic_load_n(txq->cpt_fc, __ATOMIC_RELAXED))
+ while (nb_desc <= rte_atomic_load_explicit(txq->cpt_fc, rte_memory_order_relaxed))
;
#endif
}
@@ -397,11 +401,11 @@ cn10k_nix_sec_fc_wait(struct cn10k_eth_txq *txq, uint16_t nb_pkts)
: "memory");
#else
/* Wait for primary core to refill FC. */
- while (__atomic_load_n(fc_sw, __ATOMIC_RELAXED) < 0)
+ while (rte_atomic_load_explicit(fc_sw, rte_memory_order_relaxed) < 0)
;
#endif
- val = __atomic_fetch_sub(fc_sw, nb_pkts, __ATOMIC_ACQUIRE) - nb_pkts;
+ val = rte_atomic_fetch_sub_explicit(fc_sw, nb_pkts, rte_memory_order_acquire) - nb_pkts;
if (likely(val >= 0))
return;
@@ -427,15 +431,16 @@ cn10k_nix_sec_fc_wait(struct cn10k_eth_txq *txq, uint16_t nb_pkts)
: "memory");
#else
while (true) {
- newval = nb_desc - __atomic_load_n(fc, __ATOMIC_RELAXED);
+ newval = nb_desc - rte_atomic_load_explicit(fc, rte_memory_order_relaxed);
newval -= nb_pkts;
if (newval >= 0)
break;
}
#endif
- if (!__atomic_compare_exchange_n(fc_sw, &val, newval, false, __ATOMIC_RELEASE,
- __ATOMIC_RELAXED))
+ if (!rte_atomic_compare_exchange_strong_explicit(fc_sw, &val, newval,
+ rte_memory_order_release,
+ rte_memory_order_relaxed))
goto again;
}
@@ -763,7 +768,8 @@ cn10k_nix_prefree_seg(struct rte_mbuf *m, struct rte_mbuf **extm, struct cn10k_e
m->next = prev;
txq->tx_compl.ptr[sqe_id] = m;
} else {
- sqe_id = __atomic_fetch_add(&txq->tx_compl.sqe_id, 1, __ATOMIC_RELAXED);
+ sqe_id = rte_atomic_fetch_add_explicit(&txq->tx_compl.sqe_id, 1,
+ rte_memory_order_relaxed);
send_hdr->w0.pnc = 1;
send_hdr->w1.sqe_id = sqe_id &
txq->tx_compl.nb_desc_mask;
diff --git a/drivers/net/cnxk/cn20k_tx.h b/drivers/net/cnxk/cn20k_tx.h
index c419778970..8c3cb05e7d 100644
--- a/drivers/net/cnxk/cn20k_tx.h
+++ b/drivers/net/cnxk/cn20k_tx.h
@@ -53,17 +53,19 @@
int64_t val, newval; \
retry: \
/* Reduce the cached count */ \
- val = (int64_t)__atomic_fetch_sub(fc_cache, pkts, __ATOMIC_RELAXED); \
+ val = (int64_t)rte_atomic_fetch_sub_explicit(fc_cache, pkts, \
+ rte_memory_order_relaxed); \
val -= pkts; \
/* Cached value is low, Update the fc_cache_pkts */ \
if (unlikely(val < 0)) { \
/* Multiply with sqe_per_sqb to express in pkts */ \
newval = txq->nb_sqb_bufs_adj - \
- __atomic_load_n(txq->fc_mem, __ATOMIC_RELAXED); \
+ rte_atomic_load_explicit(txq->fc_mem, rte_memory_order_relaxed); \
newval = (newval << (txq)->sqes_per_sqb_log2) - newval; \
newval -= pkts; \
- if (!__atomic_compare_exchange_n(fc_cache, &val, newval, false, \
- __ATOMIC_RELAXED, __ATOMIC_RELAXED)) { \
+ if (!rte_atomic_compare_exchange_strong_explicit( \
+ fc_cache, &val, newval, rte_memory_order_relaxed, \
+ rte_memory_order_relaxed)) { \
if (retry_count) { \
retry_count--; \
goto retry; \
@@ -162,10 +164,11 @@ cn20k_nix_vwqe_wait_fc(struct cn20k_eth_txq *txq, uint16_t req)
: "memory");
#else
RTE_SET_USED(pkts);
- while (__atomic_load_n(&txq->fc_cache_pkts, __ATOMIC_RELAXED) < 0)
+ while (rte_atomic_load_explicit(&txq->fc_cache_pkts, rte_memory_order_relaxed) < 0)
;
#endif
- cached = __atomic_fetch_sub(&txq->fc_cache_pkts, req, __ATOMIC_ACQUIRE) - req;
+ cached = rte_atomic_fetch_sub_explicit(&txq->fc_cache_pkts, req, rte_memory_order_acquire) -
+ req;
/* Check if there is enough space, else update and retry. */
if (cached >= 0)
return;
@@ -198,13 +201,15 @@ cn20k_nix_vwqe_wait_fc(struct cn20k_eth_txq *txq, uint16_t req)
: "memory");
#else
do {
- refill = (txq->nb_sqb_bufs_adj - __atomic_load_n(txq->fc_mem, __ATOMIC_RELAXED));
+ refill = (txq->nb_sqb_bufs_adj -
+ rte_atomic_load_explicit(txq->fc_mem, rte_memory_order_relaxed));
refill = (refill << txq->sqes_per_sqb_log2) - refill;
refill -= req;
} while (refill < 0);
#endif
- if (!__atomic_compare_exchange(&txq->fc_cache_pkts, &cached, &refill, 0, __ATOMIC_RELEASE,
- __ATOMIC_RELAXED))
+ if (!rte_atomic_compare_exchange_strong_explicit(&txq->fc_cache_pkts, &cached, refill,
+ rte_memory_order_release,
+ rte_memory_order_relaxed))
goto retry;
}
@@ -354,7 +359,7 @@ cn20k_nix_sec_fc_wait_one(struct cn20k_eth_txq *txq)
: "memory");
#else
RTE_SET_USED(fc);
- while (nb_desc <= __atomic_load_n(txq->cpt_fc, __ATOMIC_RELAXED))
+ while (nb_desc <= rte_atomic_load_explicit(txq->cpt_fc, rte_memory_order_relaxed))
;
#endif
}
@@ -386,11 +391,11 @@ cn20k_nix_sec_fc_wait(struct cn20k_eth_txq *txq, uint16_t nb_pkts)
: "memory");
#else
/* Wait for primary core to refill FC. */
- while (__atomic_load_n(fc_sw, __ATOMIC_RELAXED) < 0)
+ while (rte_atomic_load_explicit(fc_sw, rte_memory_order_relaxed) < 0)
;
#endif
- val = __atomic_fetch_sub(fc_sw, nb_pkts, __ATOMIC_ACQUIRE) - nb_pkts;
+ val = rte_atomic_fetch_sub_explicit(fc_sw, nb_pkts, rte_memory_order_acquire) - nb_pkts;
if (likely(val >= 0))
return;
@@ -416,15 +421,16 @@ cn20k_nix_sec_fc_wait(struct cn20k_eth_txq *txq, uint16_t nb_pkts)
: "memory");
#else
while (true) {
- newval = nb_desc - __atomic_load_n(fc, __ATOMIC_RELAXED);
+ newval = nb_desc - rte_atomic_load_explicit(fc, rte_memory_order_relaxed);
newval -= nb_pkts;
if (newval >= 0)
break;
}
#endif
- if (!__atomic_compare_exchange_n(fc_sw, &val, newval, false, __ATOMIC_RELEASE,
- __ATOMIC_RELAXED))
+ if (!rte_atomic_compare_exchange_strong_explicit(fc_sw, &val, newval,
+ rte_memory_order_release,
+ rte_memory_order_relaxed))
goto again;
}
@@ -747,7 +753,8 @@ cn20k_nix_prefree_seg(struct rte_mbuf *m, struct rte_mbuf **extm, struct cn20k_e
m->next = prev;
txq->tx_compl.ptr[sqe_id] = m;
} else {
- sqe_id = __atomic_fetch_add(&txq->tx_compl.sqe_id, 1, __ATOMIC_RELAXED);
+ sqe_id = rte_atomic_fetch_add_explicit(&txq->tx_compl.sqe_id, 1,
+ rte_memory_order_relaxed);
send_hdr->w0.pnc = 1;
send_hdr->w1.sqe_id = sqe_id & txq->tx_compl.nb_desc_mask;
txq->tx_compl.ptr[send_hdr->w1.sqe_id] = m;
diff --git a/drivers/net/cnxk/cn9k_tx.h b/drivers/net/cnxk/cn9k_tx.h
index 902a17860c..af7a5bdfe9 100644
--- a/drivers/net/cnxk/cn9k_tx.h
+++ b/drivers/net/cnxk/cn9k_tx.h
@@ -112,7 +112,8 @@ cn9k_nix_prefree_seg(struct rte_mbuf *m, struct rte_mbuf **extm, struct cn9k_eth
m->next = prev;
txq->tx_compl.ptr[sqe_id] = m;
} else {
- sqe_id = __atomic_fetch_add(&txq->tx_compl.sqe_id, 1, __ATOMIC_RELAXED);
+ sqe_id = rte_atomic_fetch_add_explicit(&txq->tx_compl.sqe_id, 1,
+ rte_memory_order_relaxed);
send_hdr->w0.pnc = 1;
send_hdr->w1.sqe_id = sqe_id &
txq->tx_compl.nb_desc_mask;
@@ -599,7 +600,7 @@ cn9k_nix_sec_fc_wait_one(const struct cn9k_eth_txq *txq)
uint64_t nb_desc = txq->cpt_desc;
uint64_t *fc = txq->cpt_fc;
- while (nb_desc <= __atomic_load_n(fc, __ATOMIC_RELAXED))
+ while (nb_desc <= rte_atomic_load_explicit(fc, rte_memory_order_relaxed))
;
}
--
2.43.0
^ permalink raw reply [flat|nested] 3+ messages in thread
* [PATCH v2] cnxk: use stdatomic API
2025-03-27 11:11 [PATCH] cnxk: use stdatomic API pbhagavatula
@ 2025-05-12 13:58 ` pbhagavatula
2025-05-27 4:58 ` [EXTERNAL] " Jerin Jacob
0 siblings, 1 reply; 3+ messages in thread
From: pbhagavatula @ 2025-05-12 13:58 UTC (permalink / raw)
To: jerinj, Ankur Dwivedi, Anoob Joseph, Tejasree Kondoj,
Nithin Dabilpuram, Kiran Kumar K, Sunil Kumar Kori, Satha Rao,
Harman Kalra
Cc: dev, Pavan Nikhilesh
From: Pavan Nikhilesh <pbhagavatula@marvell.com>
Replace GCC inbuilt atomic functions with rte_atomic_xxx API.
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
drivers/crypto/cnxk/cn10k_cryptodev_ops.c | 68 +++++++++++++----------
drivers/crypto/cnxk/cn9k_cryptodev_ops.c | 25 +++++----
drivers/net/cnxk/cn10k_ethdev.c | 6 +-
drivers/net/cnxk/cn10k_rxtx.h | 8 +--
drivers/net/cnxk/cn10k_tx.h | 48 +++++++++-------
drivers/net/cnxk/cn20k_ethdev.c | 6 +-
drivers/net/cnxk/cn20k_rxtx.h | 8 +--
drivers/net/cnxk/cn20k_tx.h | 45 ++++++++-------
drivers/net/cnxk/cn9k_ethdev.c | 2 +-
drivers/net/cnxk/cn9k_ethdev.h | 2 +-
drivers/net/cnxk/cn9k_tx.h | 7 ++-
drivers/net/cnxk/cnxk_ethdev_dp.h | 2 +-
12 files changed, 125 insertions(+), 102 deletions(-)
diff --git a/drivers/crypto/cnxk/cn10k_cryptodev_ops.c b/drivers/crypto/cnxk/cn10k_cryptodev_ops.c
index 851e6f0a88..fa8c4ebb01 100644
--- a/drivers/crypto/cnxk/cn10k_cryptodev_ops.c
+++ b/drivers/crypto/cnxk/cn10k_cryptodev_ops.c
@@ -293,7 +293,8 @@ cn10k_cpt_fill_inst(struct cnxk_cpt_qp *qp, struct rte_crypto_op *ops[], struct
}
inst[0].res_addr = (uint64_t)&infl_req->res;
- __atomic_store_n(&infl_req->res.u64[0], res.u64[0], __ATOMIC_RELAXED);
+ rte_atomic_store_explicit((uint64_t __rte_atomic *)&infl_req->res.u64[0], res.u64[0],
+ rte_memory_order_relaxed);
infl_req->cop = op;
inst[0].w7.u64 = w7;
@@ -333,11 +334,11 @@ cn10k_cpt_enqueue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops,
struct cpt_inflight_req *infl_req;
uint64_t head, lmt_base, io_addr;
uint16_t nb_allowed, count = 0;
+ uint64_t __rte_atomic *fc_addr;
struct cnxk_cpt_qp *qp = qptr;
struct pending_queue *pend_q;
struct cpt_inst_s *inst;
union cpt_fc_write_s fc;
- uint64_t *fc_addr;
uint16_t lmt_id;
int ret, i;
@@ -354,7 +355,7 @@ cn10k_cpt_enqueue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops,
lmt_base = qp->lmtline.lmt_base;
io_addr = qp->lmtline.io_addr;
- fc_addr = qp->lmtline.fc_addr;
+ fc_addr = (uint64_t __rte_atomic *)qp->lmtline.fc_addr;
const uint32_t fc_thresh = qp->lmtline.fc_thresh;
@@ -362,7 +363,7 @@ cn10k_cpt_enqueue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops,
inst = (struct cpt_inst_s *)lmt_base;
again:
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (unlikely(fc.s.qsize > fc_thresh)) {
i = 0;
goto pend_q_commit;
@@ -393,7 +394,7 @@ cn10k_cpt_enqueue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops,
}
pend_q_commit:
- rte_atomic_thread_fence(__ATOMIC_RELEASE);
+ rte_atomic_thread_fence(rte_memory_order_release);
pend_q->head = head;
pend_q->time_out = rte_get_timer_cycles() +
@@ -542,7 +543,8 @@ cn10k_cpt_vec_inst_fill(struct vec_request *vec_req, struct cpt_inst_s *inst,
infl_req->qp = qp;
inst->res_addr = (uint64_t)&infl_req->res;
- __atomic_store_n(&infl_req->res.u64[0], res.u64[0], __ATOMIC_RELAXED);
+ rte_atomic_store_explicit((uint64_t __rte_atomic *)&infl_req->res.u64[0], res.u64[0],
+ rte_memory_order_relaxed);
inst->w0.u64 = 0;
inst->w2.u64 = vec_req->w2;
@@ -564,10 +566,10 @@ static inline void
cn10k_cpt_vec_submit(struct vec_request vec_tbl[], uint16_t vec_tbl_len, struct cnxk_cpt_qp *qp)
{
uint64_t lmt_base, lmt_id, io_addr;
+ uint64_t __rte_atomic *fc_addr;
union cpt_fc_write_s fc;
struct cpt_inst_s *inst;
uint16_t burst_size;
- uint64_t *fc_addr;
int i;
if (vec_tbl_len == 0)
@@ -584,7 +586,7 @@ cn10k_cpt_vec_submit(struct vec_request vec_tbl[], uint16_t vec_tbl_len, struct
lmt_base = qp->lmtline.lmt_base;
io_addr = qp->lmtline.io_addr;
- fc_addr = qp->lmtline.fc_addr;
+ fc_addr = (uint64_t __rte_atomic *)qp->lmtline.fc_addr;
ROC_LMT_BASE_ID_GET(lmt_base, lmt_id);
inst = (struct cpt_inst_s *)lmt_base;
@@ -594,7 +596,7 @@ cn10k_cpt_vec_submit(struct vec_request vec_tbl[], uint16_t vec_tbl_len, struct
cn10k_cpt_vec_inst_fill(&vec_tbl[i], &inst[i], qp, vec_tbl[0].w7);
do {
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (likely(fc.s.qsize < fc_thresh))
break;
if (unlikely(rte_get_timer_cycles() > timeout))
@@ -619,18 +621,18 @@ ca_lmtst_vec_submit(struct ops_burst *burst, struct vec_request vec_tbl[], uint1
uint16_t lmt_id, len = *vec_tbl_len;
struct cpt_inst_s *inst, *inst_base;
struct cpt_inflight_req *infl_req;
+ uint64_t __rte_atomic *fc_addr;
struct rte_event_vector *vec;
uint64_t lmt_base, io_addr;
union cpt_fc_write_s fc;
struct cnxk_cpt_qp *qp;
- uint64_t *fc_addr;
int ret, i, vi;
qp = burst->qp;
lmt_base = qp->lmtline.lmt_base;
io_addr = qp->lmtline.io_addr;
- fc_addr = qp->lmtline.fc_addr;
+ fc_addr = (uint64_t __rte_atomic *)qp->lmtline.fc_addr;
const uint32_t fc_thresh = qp->lmtline.fc_thresh;
@@ -645,7 +647,7 @@ ca_lmtst_vec_submit(struct ops_burst *burst, struct vec_request vec_tbl[], uint1
#endif
/* Perform fc check before putting packets into vectors */
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (unlikely(fc.s.qsize > fc_thresh)) {
rte_errno = EAGAIN;
return 0;
@@ -738,10 +740,10 @@ ca_lmtst_burst_submit(struct ops_burst *burst, const bool is_sg_ver2)
struct cpt_inflight_req *infl_reqs[CN10K_CPT_PKTS_PER_LOOP];
struct cpt_inst_s *inst, *inst_base;
struct cpt_inflight_req *infl_req;
+ uint64_t __rte_atomic *fc_addr;
uint64_t lmt_base, io_addr;
union cpt_fc_write_s fc;
struct cnxk_cpt_qp *qp;
- uint64_t *fc_addr;
uint16_t lmt_id;
int ret, i, j;
@@ -749,7 +751,7 @@ ca_lmtst_burst_submit(struct ops_burst *burst, const bool is_sg_ver2)
lmt_base = qp->lmtline.lmt_base;
io_addr = qp->lmtline.io_addr;
- fc_addr = qp->lmtline.fc_addr;
+ fc_addr = (uint64_t __rte_atomic *)qp->lmtline.fc_addr;
const uint32_t fc_thresh = qp->lmtline.fc_thresh;
@@ -790,7 +792,7 @@ ca_lmtst_burst_submit(struct ops_burst *burst, const bool is_sg_ver2)
inst->w3.u64 = CNXK_CPT_INST_W3(1, infl_req);
}
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (unlikely(fc.s.qsize > fc_thresh)) {
rte_errno = EAGAIN;
for (j = 0; j < i; j++) {
@@ -1241,7 +1243,8 @@ cn10k_cpt_crypto_adapter_dequeue(uintptr_t get_work1)
cop = infl_req->cop;
qp = infl_req->qp;
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0], __ATOMIC_RELAXED);
+ res.u64[0] = rte_atomic_load_explicit((uint64_t __rte_atomic *)&infl_req->res.u64[0],
+ rte_memory_order_relaxed);
cn10k_cpt_dequeue_post_process(qp, infl_req->cop, infl_req, &res.cn10k);
@@ -1272,7 +1275,8 @@ cn10k_cpt_crypto_adapter_vector_dequeue(uintptr_t get_work1)
req_mp = qp->ca.req_mp;
#ifdef CNXK_CRYPTODEV_DEBUG
- res.u64[0] = __atomic_load_n(&vec_infl_req->res.u64[0], __ATOMIC_RELAXED);
+ res.u64[0] = rte_atomic_load_explicit((uint64_t __rte_atomic *)&vec_infl_req->res.u64[0],
+ rte_memory_order_relaxed);
PLT_ASSERT(res.cn10k.compcode == CPT_COMP_GOOD);
PLT_ASSERT(res.cn10k.uc_compcode == 0);
#endif
@@ -1281,7 +1285,8 @@ cn10k_cpt_crypto_adapter_vector_dequeue(uintptr_t get_work1)
infl_req = vec->ptrs[i];
cop = infl_req->cop;
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0], __ATOMIC_RELAXED);
+ res.u64[0] = rte_atomic_load_explicit(
+ (uint64_t __rte_atomic *)&infl_req->res.u64[0], rte_memory_order_relaxed);
cn10k_cpt_dequeue_post_process(qp, cop, infl_req, &res.cn10k);
vec->ptrs[i] = cop;
@@ -1321,8 +1326,8 @@ cn10k_cpt_dequeue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops)
for (i = 0; i < nb_ops; i++) {
infl_req = &pend_q->req_queue[pq_tail];
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0],
- __ATOMIC_RELAXED);
+ res.u64[0] = rte_atomic_load_explicit(
+ (uint64_t __rte_atomic *)&infl_req->res.u64[0], rte_memory_order_relaxed);
if (unlikely(res.cn10k.compcode == CPT_COMP_NOT_DONE)) {
if (unlikely(rte_get_timer_cycles() >
@@ -1576,7 +1581,8 @@ cn10k_cpt_raw_fill_inst(struct cnxk_iov *iov, struct cnxk_cpt_qp *qp,
return 0;
inst[0].res_addr = (uint64_t)&infl_req->res;
- __atomic_store_n(&infl_req->res.u64[0], res.u64[0], __ATOMIC_RELAXED);
+ rte_atomic_store_explicit((uint64_t __rte_atomic *)&infl_req->res.u64[0], res.u64[0],
+ rte_memory_order_relaxed);
infl_req->opaque = opaque;
inst[0].w7.u64 = sess->cpt_inst_w7;
@@ -1594,11 +1600,11 @@ cn10k_cpt_raw_enqueue_burst(void *qpair, uint8_t *drv_ctx, struct rte_crypto_sym
uint64_t lmt_base, io_addr, head;
struct cnxk_cpt_qp *qp = qpair;
struct cnxk_sym_dp_ctx *dp_ctx;
+ uint64_t __rte_atomic *fc_addr;
struct pending_queue *pend_q;
uint32_t count = 0, index;
union cpt_fc_write_s fc;
struct cpt_inst_s *inst;
- uint64_t *fc_addr;
int ret, i;
pend_q = &qp->pend_q;
@@ -1613,7 +1619,7 @@ cn10k_cpt_raw_enqueue_burst(void *qpair, uint8_t *drv_ctx, struct rte_crypto_sym
lmt_base = qp->lmtline.lmt_base;
io_addr = qp->lmtline.io_addr;
- fc_addr = qp->lmtline.fc_addr;
+ fc_addr = (uint64_t __rte_atomic *)qp->lmtline.fc_addr;
const uint32_t fc_thresh = qp->lmtline.fc_thresh;
@@ -1622,7 +1628,7 @@ cn10k_cpt_raw_enqueue_burst(void *qpair, uint8_t *drv_ctx, struct rte_crypto_sym
dp_ctx = (struct cnxk_sym_dp_ctx *)drv_ctx;
again:
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (unlikely(fc.s.qsize > fc_thresh)) {
i = 0;
goto pend_q_commit;
@@ -1697,11 +1703,11 @@ cn10k_cpt_raw_enqueue(void *qpair, uint8_t *drv_ctx, struct rte_crypto_vec *data
uint64_t lmt_base, io_addr, head;
struct cnxk_cpt_qp *qp = qpair;
struct cnxk_sym_dp_ctx *dp_ctx;
+ uint64_t __rte_atomic *fc_addr;
uint16_t lmt_id, nb_allowed;
struct cpt_inst_s *inst;
union cpt_fc_write_s fc;
struct cnxk_iov iov;
- uint64_t *fc_addr;
int ret, i = 1;
struct pending_queue *pend_q = &qp->pend_q;
@@ -1718,12 +1724,12 @@ cn10k_cpt_raw_enqueue(void *qpair, uint8_t *drv_ctx, struct rte_crypto_vec *data
lmt_base = qp->lmtline.lmt_base;
io_addr = qp->lmtline.io_addr;
- fc_addr = qp->lmtline.fc_addr;
+ fc_addr = (uint64_t __rte_atomic *)qp->lmtline.fc_addr;
ROC_LMT_BASE_ID_GET(lmt_base, lmt_id);
inst = (struct cpt_inst_s *)lmt_base;
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (unlikely(fc.s.qsize > fc_thresh))
return -1;
@@ -1826,7 +1832,8 @@ cn10k_cpt_sym_raw_dequeue_burst(void *qptr, uint8_t *drv_ctx,
is_op_success = 0;
infl_req = &pend_q->req_queue[pq_tail];
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0], __ATOMIC_RELAXED);
+ res.u64[0] = rte_atomic_load_explicit(
+ (uint64_t __rte_atomic *)&infl_req->res.u64[0], rte_memory_order_relaxed);
if (unlikely(res.cn10k.compcode == CPT_COMP_NOT_DONE)) {
if (unlikely(rte_get_timer_cycles() > pend_q->time_out)) {
@@ -1871,9 +1878,9 @@ cn10k_cpt_sym_raw_dequeue(void *qptr, uint8_t *drv_ctx, int *dequeue_status,
struct cpt_inflight_req *infl_req;
struct cnxk_cpt_qp *qp = qptr;
struct pending_queue *pend_q;
- uint64_t pq_tail;
union cpt_res_s res;
void *opaque = NULL;
+ uint64_t pq_tail;
pend_q = &qp->pend_q;
@@ -1887,7 +1894,8 @@ cn10k_cpt_sym_raw_dequeue(void *qptr, uint8_t *drv_ctx, int *dequeue_status,
infl_req = &pend_q->req_queue[pq_tail];
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0], __ATOMIC_RELAXED);
+ res.u64[0] = rte_atomic_load_explicit((uint64_t __rte_atomic *)&infl_req->res.u64[0],
+ rte_memory_order_relaxed);
if (unlikely(res.cn10k.compcode == CPT_COMP_NOT_DONE)) {
if (unlikely(rte_get_timer_cycles() > pend_q->time_out)) {
diff --git a/drivers/crypto/cnxk/cn9k_cryptodev_ops.c b/drivers/crypto/cnxk/cn9k_cryptodev_ops.c
index ee35ed1eba..69e15a862f 100644
--- a/drivers/crypto/cnxk/cn9k_cryptodev_ops.c
+++ b/drivers/crypto/cnxk/cn9k_cryptodev_ops.c
@@ -195,12 +195,12 @@ cn9k_cpt_enqueue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops)
infl_req_1->op_flags = 0;
infl_req_2->op_flags = 0;
- __atomic_store_n(&infl_req_1->res.u64[0], res.u64[0],
- __ATOMIC_RELAXED);
+ rte_atomic_store_explicit((uint64_t __rte_atomic *)&infl_req_1->res.u64[0],
+ res.u64[0], rte_memory_order_relaxed);
inst[0].res_addr = (uint64_t)&infl_req_1->res;
- __atomic_store_n(&infl_req_2->res.u64[0], res.u64[0],
- __ATOMIC_RELAXED);
+ rte_atomic_store_explicit((uint64_t __rte_atomic *)&infl_req_2->res.u64[0],
+ res.u64[0], rte_memory_order_relaxed);
inst[1].res_addr = (uint64_t)&infl_req_2->res;
ret = cn9k_cpt_inst_prep(qp, op_1, infl_req_1, &inst[0]);
@@ -224,7 +224,7 @@ cn9k_cpt_enqueue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops)
count += 2;
}
- rte_atomic_thread_fence(__ATOMIC_RELEASE);
+ rte_atomic_thread_fence(rte_memory_order_release);
pend_q->head = head;
pend_q->time_out = rte_get_timer_cycles() +
@@ -346,10 +346,10 @@ uint16_t
cn9k_cpt_crypto_adapter_enqueue(uintptr_t base, struct rte_crypto_op *op)
{
struct cpt_inflight_req *infl_req;
+ uint64_t __rte_atomic *fc_addr;
union cpt_fc_write_s fc;
struct cnxk_cpt_qp *qp;
struct cpt_inst_s inst;
- uint64_t *fc_addr;
int ret;
ret = cn9k_ca_meta_info_extract(op, &qp, &inst);
@@ -383,11 +383,11 @@ cn9k_cpt_crypto_adapter_enqueue(uintptr_t base, struct rte_crypto_op *op)
inst.res_addr = (uint64_t)&infl_req->res;
inst.w3.u64 = CNXK_CPT_INST_W3(1, infl_req);
- fc_addr = qp->lmtline.fc_addr;
+ fc_addr = (uint64_t __rte_atomic *)qp->lmtline.fc_addr;
const uint32_t fc_thresh = qp->lmtline.fc_thresh;
- fc.u64[0] = __atomic_load_n(fc_addr, __ATOMIC_RELAXED);
+ fc.u64[0] = rte_atomic_load_explicit(fc_addr, rte_memory_order_relaxed);
if (unlikely(fc.s.qsize > fc_thresh)) {
rte_mempool_put(qp->ca.req_mp, infl_req);
rte_errno = EAGAIN;
@@ -607,7 +607,8 @@ cn9k_cpt_crypto_adapter_dequeue(uintptr_t get_work1)
cop = infl_req->cop;
qp = infl_req->qp;
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0], __ATOMIC_RELAXED);
+ res.u64[0] = rte_atomic_load_explicit((uint64_t __rte_atomic *)&infl_req->res.u64[0],
+ rte_memory_order_relaxed);
cn9k_cpt_dequeue_post_process(qp, infl_req->cop, infl_req, &res.cn9k);
@@ -638,13 +639,13 @@ cn9k_cpt_dequeue_burst(void *qptr, struct rte_crypto_op **ops, uint16_t nb_ops)
nb_ops = RTE_MIN(nb_ops, infl_cnt);
/* Ensure infl_cnt isn't read before data lands */
- rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
+ rte_atomic_thread_fence(rte_memory_order_acquire);
for (i = 0; i < nb_ops; i++) {
infl_req = &pend_q->req_queue[pq_tail];
- res.u64[0] = __atomic_load_n(&infl_req->res.u64[0],
- __ATOMIC_RELAXED);
+ res.u64[0] = rte_atomic_load_explicit(
+ (uint64_t __rte_atomic *)&infl_req->res.u64[0], rte_memory_order_relaxed);
if (unlikely(res.cn9k.compcode == CPT_COMP_NOT_DONE)) {
if (unlikely(rte_get_timer_cycles() >
diff --git a/drivers/net/cnxk/cn10k_ethdev.c b/drivers/net/cnxk/cn10k_ethdev.c
index e491854cb2..9c1621dbfa 100644
--- a/drivers/net/cnxk/cn10k_ethdev.c
+++ b/drivers/net/cnxk/cn10k_ethdev.c
@@ -256,9 +256,9 @@ cn10k_nix_tx_queue_setup(struct rte_eth_dev *eth_dev, uint16_t qid,
inl_lf = dev->outb.lf_base + crypto_qid;
txq->cpt_io_addr = inl_lf->io_addr;
- txq->cpt_fc = inl_lf->fc_addr;
- txq->cpt_fc_sw = (int32_t *)((uintptr_t)dev->outb.fc_sw_mem +
- crypto_qid * RTE_CACHE_LINE_SIZE);
+ txq->cpt_fc = (uint64_t __rte_atomic *)inl_lf->fc_addr;
+ txq->cpt_fc_sw = (int32_t __rte_atomic *)((uintptr_t)dev->outb.fc_sw_mem +
+ crypto_qid * RTE_CACHE_LINE_SIZE);
txq->cpt_desc = inl_lf->nb_desc * 0.7;
txq->sa_base = (uint64_t)dev->outb.sa_base;
diff --git a/drivers/net/cnxk/cn10k_rxtx.h b/drivers/net/cnxk/cn10k_rxtx.h
index 53a6fbd60d..c18e9c8c0e 100644
--- a/drivers/net/cnxk/cn10k_rxtx.h
+++ b/drivers/net/cnxk/cn10k_rxtx.h
@@ -47,8 +47,8 @@
struct cn10k_eth_txq {
uint64_t send_hdr_w0;
- int64_t fc_cache_pkts;
- uint64_t *fc_mem;
+ int64_t __rte_atomic fc_cache_pkts;
+ uint64_t __rte_atomic *fc_mem;
uintptr_t lmt_base;
rte_iova_t io_addr;
uint16_t sqes_per_sqb_log2;
@@ -56,9 +56,9 @@ struct cn10k_eth_txq {
uint8_t flag;
rte_iova_t cpt_io_addr;
uint64_t sa_base;
- uint64_t *cpt_fc;
+ uint64_t __rte_atomic *cpt_fc;
uint16_t cpt_desc;
- int32_t *cpt_fc_sw;
+ int32_t __rte_atomic *cpt_fc_sw;
uint64_t lso_tun_fmt;
uint64_t ts_mem;
uint64_t mark_flag : 8;
diff --git a/drivers/net/cnxk/cn10k_tx.h b/drivers/net/cnxk/cn10k_tx.h
index 809fafb2f7..be9e020ac5 100644
--- a/drivers/net/cnxk/cn10k_tx.h
+++ b/drivers/net/cnxk/cn10k_tx.h
@@ -50,22 +50,24 @@
#define NIX_XMIT_FC_OR_RETURN_MTS(txq, pkts) \
do { \
- int64_t *fc_cache = &(txq)->fc_cache_pkts; \
+ int64_t __rte_atomic *fc_cache = &(txq)->fc_cache_pkts; \
uint8_t retry_count = 8; \
int64_t val, newval; \
retry: \
/* Reduce the cached count */ \
- val = (int64_t)__atomic_fetch_sub(fc_cache, pkts, __ATOMIC_RELAXED); \
+ val = (int64_t)rte_atomic_fetch_sub_explicit(fc_cache, pkts, \
+ rte_memory_order_relaxed); \
val -= pkts; \
/* Cached value is low, Update the fc_cache_pkts */ \
if (unlikely(val < 0)) { \
/* Multiply with sqe_per_sqb to express in pkts */ \
- newval = txq->nb_sqb_bufs_adj - __atomic_load_n(txq->fc_mem, \
- __ATOMIC_RELAXED); \
+ newval = txq->nb_sqb_bufs_adj - \
+ rte_atomic_load_explicit(txq->fc_mem, rte_memory_order_relaxed); \
newval = (newval << (txq)->sqes_per_sqb_log2) - newval; \
newval -= pkts; \
- if (!__atomic_compare_exchange_n(fc_cache, &val, newval, false, \
- __ATOMIC_RELAXED, __ATOMIC_RELAXED)) { \
+ if (!rte_atomic_compare_exchange_strong_explicit( \
+ fc_cache, &val, newval, rte_memory_order_relaxed, \
+ rte_memory_order_relaxed)) { \
if (retry_count) { \
retry_count--; \
goto retry; \
@@ -164,10 +166,11 @@ cn10k_nix_vwqe_wait_fc(struct cn10k_eth_txq *txq, uint16_t req)
: "memory");
#else
RTE_SET_USED(pkts);
- while (__atomic_load_n(&txq->fc_cache_pkts, __ATOMIC_RELAXED) < 0)
+ while (rte_atomic_load_explicit(&txq->fc_cache_pkts, rte_memory_order_relaxed) < 0)
;
#endif
- cached = __atomic_fetch_sub(&txq->fc_cache_pkts, req, __ATOMIC_ACQUIRE) - req;
+ cached = rte_atomic_fetch_sub_explicit(&txq->fc_cache_pkts, req, rte_memory_order_acquire) -
+ req;
/* Check if there is enough space, else update and retry. */
if (cached >= 0)
return;
@@ -200,14 +203,15 @@ cn10k_nix_vwqe_wait_fc(struct cn10k_eth_txq *txq, uint16_t req)
: "memory");
#else
do {
- refill = (txq->nb_sqb_bufs_adj - __atomic_load_n(txq->fc_mem, __ATOMIC_RELAXED));
+ refill = (txq->nb_sqb_bufs_adj -
+ rte_atomic_load_explicit(txq->fc_mem, rte_memory_order_relaxed));
refill = (refill << txq->sqes_per_sqb_log2) - refill;
refill -= req;
} while (refill < 0);
#endif
- if (!__atomic_compare_exchange(&txq->fc_cache_pkts, &cached, &refill,
- 0, __ATOMIC_RELEASE,
- __ATOMIC_RELAXED))
+ if (!rte_atomic_compare_exchange_strong_explicit(&txq->fc_cache_pkts, &cached, refill,
+ rte_memory_order_release,
+ rte_memory_order_relaxed))
goto retry;
}
@@ -365,7 +369,7 @@ cn10k_nix_sec_fc_wait_one(struct cn10k_eth_txq *txq)
: "memory");
#else
RTE_SET_USED(fc);
- while (nb_desc <= __atomic_load_n(txq->cpt_fc, __ATOMIC_RELAXED))
+ while (nb_desc <= rte_atomic_load_explicit(txq->cpt_fc, rte_memory_order_relaxed))
;
#endif
}
@@ -374,8 +378,8 @@ static __rte_always_inline void
cn10k_nix_sec_fc_wait(struct cn10k_eth_txq *txq, uint16_t nb_pkts)
{
int32_t nb_desc, val, newval;
- int32_t *fc_sw;
- uint64_t *fc;
+ int32_t __rte_atomic *fc_sw;
+ uint64_t __rte_atomic *fc;
/* Check if there is any CPT instruction to submit */
if (!nb_pkts)
@@ -397,11 +401,11 @@ cn10k_nix_sec_fc_wait(struct cn10k_eth_txq *txq, uint16_t nb_pkts)
: "memory");
#else
/* Wait for primary core to refill FC. */
- while (__atomic_load_n(fc_sw, __ATOMIC_RELAXED) < 0)
+ while (rte_atomic_load_explicit(fc_sw, rte_memory_order_relaxed) < 0)
;
#endif
- val = __atomic_fetch_sub(fc_sw, nb_pkts, __ATOMIC_ACQUIRE) - nb_pkts;
+ val = rte_atomic_fetch_sub_explicit(fc_sw, nb_pkts, rte_memory_order_acquire) - nb_pkts;
if (likely(val >= 0))
return;
@@ -427,15 +431,16 @@ cn10k_nix_sec_fc_wait(struct cn10k_eth_txq *txq, uint16_t nb_pkts)
: "memory");
#else
while (true) {
- newval = nb_desc - __atomic_load_n(fc, __ATOMIC_RELAXED);
+ newval = nb_desc - rte_atomic_load_explicit(fc, rte_memory_order_relaxed);
newval -= nb_pkts;
if (newval >= 0)
break;
}
#endif
- if (!__atomic_compare_exchange_n(fc_sw, &val, newval, false, __ATOMIC_RELEASE,
- __ATOMIC_RELAXED))
+ if (!rte_atomic_compare_exchange_strong_explicit(fc_sw, &val, newval,
+ rte_memory_order_release,
+ rte_memory_order_relaxed))
goto again;
}
@@ -763,7 +768,8 @@ cn10k_nix_prefree_seg(struct rte_mbuf *m, struct rte_mbuf **extm, struct cn10k_e
m->next = prev;
txq->tx_compl.ptr[sqe_id] = m;
} else {
- sqe_id = __atomic_fetch_add(&txq->tx_compl.sqe_id, 1, __ATOMIC_RELAXED);
+ sqe_id = rte_atomic_fetch_add_explicit(&txq->tx_compl.sqe_id, 1,
+ rte_memory_order_relaxed);
send_hdr->w0.pnc = 1;
send_hdr->w1.sqe_id = sqe_id &
txq->tx_compl.nb_desc_mask;
diff --git a/drivers/net/cnxk/cn20k_ethdev.c b/drivers/net/cnxk/cn20k_ethdev.c
index 740fdb7f76..97e3a8d557 100644
--- a/drivers/net/cnxk/cn20k_ethdev.c
+++ b/drivers/net/cnxk/cn20k_ethdev.c
@@ -248,9 +248,9 @@ cn20k_nix_tx_queue_setup(struct rte_eth_dev *eth_dev, uint16_t qid, uint16_t nb_
inl_lf = dev->outb.lf_base + crypto_qid;
txq->cpt_io_addr = inl_lf->io_addr;
- txq->cpt_fc = inl_lf->fc_addr;
- txq->cpt_fc_sw = (int32_t *)((uintptr_t)dev->outb.fc_sw_mem +
- crypto_qid * RTE_CACHE_LINE_SIZE);
+ txq->cpt_fc = (uint64_t __rte_atomic *)inl_lf->fc_addr;
+ txq->cpt_fc_sw = (int32_t __rte_atomic *)((uintptr_t)dev->outb.fc_sw_mem +
+ crypto_qid * RTE_CACHE_LINE_SIZE);
txq->cpt_desc = inl_lf->nb_desc * 0.7;
txq->sa_base = (uint64_t)dev->outb.sa_base;
diff --git a/drivers/net/cnxk/cn20k_rxtx.h b/drivers/net/cnxk/cn20k_rxtx.h
index f23c16ec07..10da42680f 100644
--- a/drivers/net/cnxk/cn20k_rxtx.h
+++ b/drivers/net/cnxk/cn20k_rxtx.h
@@ -47,8 +47,8 @@
struct cn20k_eth_txq {
uint64_t send_hdr_w0;
- int64_t fc_cache_pkts;
- uint64_t *fc_mem;
+ int64_t __rte_atomic fc_cache_pkts;
+ uint64_t __rte_atomic *fc_mem;
uintptr_t lmt_base;
rte_iova_t io_addr;
uint16_t sqes_per_sqb_log2;
@@ -56,9 +56,9 @@ struct cn20k_eth_txq {
uint8_t flag;
rte_iova_t cpt_io_addr;
uint64_t sa_base;
- uint64_t *cpt_fc;
+ uint64_t __rte_atomic *cpt_fc;
uint16_t cpt_desc;
- int32_t *cpt_fc_sw;
+ int32_t __rte_atomic *cpt_fc_sw;
uint64_t lso_tun_fmt;
uint64_t ts_mem;
uint64_t mark_flag : 8;
diff --git a/drivers/net/cnxk/cn20k_tx.h b/drivers/net/cnxk/cn20k_tx.h
index c419778970..3fb2e1f4e1 100644
--- a/drivers/net/cnxk/cn20k_tx.h
+++ b/drivers/net/cnxk/cn20k_tx.h
@@ -48,22 +48,24 @@
#define NIX_XMIT_FC_OR_RETURN_MTS(txq, pkts) \
do { \
- int64_t *fc_cache = &(txq)->fc_cache_pkts; \
+ int64_t __rte_atomic *fc_cache = &(txq)->fc_cache_pkts; \
uint8_t retry_count = 8; \
int64_t val, newval; \
retry: \
/* Reduce the cached count */ \
- val = (int64_t)__atomic_fetch_sub(fc_cache, pkts, __ATOMIC_RELAXED); \
+ val = (int64_t)rte_atomic_fetch_sub_explicit(fc_cache, pkts, \
+ rte_memory_order_relaxed); \
val -= pkts; \
/* Cached value is low, Update the fc_cache_pkts */ \
if (unlikely(val < 0)) { \
/* Multiply with sqe_per_sqb to express in pkts */ \
newval = txq->nb_sqb_bufs_adj - \
- __atomic_load_n(txq->fc_mem, __ATOMIC_RELAXED); \
+ rte_atomic_load_explicit(txq->fc_mem, rte_memory_order_relaxed); \
newval = (newval << (txq)->sqes_per_sqb_log2) - newval; \
newval -= pkts; \
- if (!__atomic_compare_exchange_n(fc_cache, &val, newval, false, \
- __ATOMIC_RELAXED, __ATOMIC_RELAXED)) { \
+ if (!rte_atomic_compare_exchange_strong_explicit( \
+ fc_cache, &val, newval, rte_memory_order_relaxed, \
+ rte_memory_order_relaxed)) { \
if (retry_count) { \
retry_count--; \
goto retry; \
@@ -162,10 +164,11 @@ cn20k_nix_vwqe_wait_fc(struct cn20k_eth_txq *txq, uint16_t req)
: "memory");
#else
RTE_SET_USED(pkts);
- while (__atomic_load_n(&txq->fc_cache_pkts, __ATOMIC_RELAXED) < 0)
+ while (rte_atomic_load_explicit(&txq->fc_cache_pkts, rte_memory_order_relaxed) < 0)
;
#endif
- cached = __atomic_fetch_sub(&txq->fc_cache_pkts, req, __ATOMIC_ACQUIRE) - req;
+ cached = rte_atomic_fetch_sub_explicit(&txq->fc_cache_pkts, req, rte_memory_order_acquire) -
+ req;
/* Check if there is enough space, else update and retry. */
if (cached >= 0)
return;
@@ -198,13 +201,15 @@ cn20k_nix_vwqe_wait_fc(struct cn20k_eth_txq *txq, uint16_t req)
: "memory");
#else
do {
- refill = (txq->nb_sqb_bufs_adj - __atomic_load_n(txq->fc_mem, __ATOMIC_RELAXED));
+ refill = (txq->nb_sqb_bufs_adj -
+ rte_atomic_load_explicit(txq->fc_mem, rte_memory_order_relaxed));
refill = (refill << txq->sqes_per_sqb_log2) - refill;
refill -= req;
} while (refill < 0);
#endif
- if (!__atomic_compare_exchange(&txq->fc_cache_pkts, &cached, &refill, 0, __ATOMIC_RELEASE,
- __ATOMIC_RELAXED))
+ if (!rte_atomic_compare_exchange_strong_explicit(&txq->fc_cache_pkts, &cached, refill,
+ rte_memory_order_release,
+ rte_memory_order_relaxed))
goto retry;
}
@@ -354,7 +359,7 @@ cn20k_nix_sec_fc_wait_one(struct cn20k_eth_txq *txq)
: "memory");
#else
RTE_SET_USED(fc);
- while (nb_desc <= __atomic_load_n(txq->cpt_fc, __ATOMIC_RELAXED))
+ while (nb_desc <= rte_atomic_load_explicit(txq->cpt_fc, rte_memory_order_relaxed))
;
#endif
}
@@ -363,8 +368,8 @@ static __rte_always_inline void
cn20k_nix_sec_fc_wait(struct cn20k_eth_txq *txq, uint16_t nb_pkts)
{
int32_t nb_desc, val, newval;
- int32_t *fc_sw;
- uint64_t *fc;
+ int32_t __rte_atomic *fc_sw;
+ uint64_t __rte_atomic *fc;
/* Check if there is any CPT instruction to submit */
if (!nb_pkts)
@@ -386,11 +391,11 @@ cn20k_nix_sec_fc_wait(struct cn20k_eth_txq *txq, uint16_t nb_pkts)
: "memory");
#else
/* Wait for primary core to refill FC. */
- while (__atomic_load_n(fc_sw, __ATOMIC_RELAXED) < 0)
+ while (rte_atomic_load_explicit(fc_sw, rte_memory_order_relaxed) < 0)
;
#endif
- val = __atomic_fetch_sub(fc_sw, nb_pkts, __ATOMIC_ACQUIRE) - nb_pkts;
+ val = rte_atomic_fetch_sub_explicit(fc_sw, nb_pkts, rte_memory_order_acquire) - nb_pkts;
if (likely(val >= 0))
return;
@@ -416,15 +421,16 @@ cn20k_nix_sec_fc_wait(struct cn20k_eth_txq *txq, uint16_t nb_pkts)
: "memory");
#else
while (true) {
- newval = nb_desc - __atomic_load_n(fc, __ATOMIC_RELAXED);
+ newval = nb_desc - rte_atomic_load_explicit(fc, rte_memory_order_relaxed);
newval -= nb_pkts;
if (newval >= 0)
break;
}
#endif
- if (!__atomic_compare_exchange_n(fc_sw, &val, newval, false, __ATOMIC_RELEASE,
- __ATOMIC_RELAXED))
+ if (!rte_atomic_compare_exchange_strong_explicit(fc_sw, &val, newval,
+ rte_memory_order_release,
+ rte_memory_order_relaxed))
goto again;
}
@@ -747,7 +753,8 @@ cn20k_nix_prefree_seg(struct rte_mbuf *m, struct rte_mbuf **extm, struct cn20k_e
m->next = prev;
txq->tx_compl.ptr[sqe_id] = m;
} else {
- sqe_id = __atomic_fetch_add(&txq->tx_compl.sqe_id, 1, __ATOMIC_RELAXED);
+ sqe_id = rte_atomic_fetch_add_explicit(&txq->tx_compl.sqe_id, 1,
+ rte_memory_order_relaxed);
send_hdr->w0.pnc = 1;
send_hdr->w1.sqe_id = sqe_id & txq->tx_compl.nb_desc_mask;
txq->tx_compl.ptr[send_hdr->w1.sqe_id] = m;
diff --git a/drivers/net/cnxk/cn9k_ethdev.c b/drivers/net/cnxk/cn9k_ethdev.c
index c419593a23..5f67e8a6ba 100644
--- a/drivers/net/cnxk/cn9k_ethdev.c
+++ b/drivers/net/cnxk/cn9k_ethdev.c
@@ -251,7 +251,7 @@ cn9k_nix_tx_queue_setup(struct rte_eth_dev *eth_dev, uint16_t qid,
inl_lf = dev->outb.lf_base + crypto_qid;
txq->cpt_io_addr = inl_lf->io_addr;
- txq->cpt_fc = inl_lf->fc_addr;
+ txq->cpt_fc = (uint64_t __rte_atomic *)inl_lf->fc_addr;
txq->cpt_desc = inl_lf->nb_desc * 0.7;
txq->sa_base = (uint64_t)dev->outb.sa_base;
txq->sa_base |= eth_dev->data->port_id;
diff --git a/drivers/net/cnxk/cn9k_ethdev.h b/drivers/net/cnxk/cn9k_ethdev.h
index c267c11c46..59ed69acda 100644
--- a/drivers/net/cnxk/cn9k_ethdev.h
+++ b/drivers/net/cnxk/cn9k_ethdev.h
@@ -20,7 +20,7 @@ struct cn9k_eth_txq {
int16_t nb_sqb_bufs_adj;
rte_iova_t cpt_io_addr;
uint64_t sa_base;
- uint64_t *cpt_fc;
+ uint64_t __rte_atomic *cpt_fc;
uint16_t cpt_desc;
uint64_t mark_flag : 8;
uint64_t mark_fmt : 48;
diff --git a/drivers/net/cnxk/cn9k_tx.h b/drivers/net/cnxk/cn9k_tx.h
index 902a17860c..14594b6311 100644
--- a/drivers/net/cnxk/cn9k_tx.h
+++ b/drivers/net/cnxk/cn9k_tx.h
@@ -112,7 +112,8 @@ cn9k_nix_prefree_seg(struct rte_mbuf *m, struct rte_mbuf **extm, struct cn9k_eth
m->next = prev;
txq->tx_compl.ptr[sqe_id] = m;
} else {
- sqe_id = __atomic_fetch_add(&txq->tx_compl.sqe_id, 1, __ATOMIC_RELAXED);
+ sqe_id = rte_atomic_fetch_add_explicit(&txq->tx_compl.sqe_id, 1,
+ rte_memory_order_relaxed);
send_hdr->w0.pnc = 1;
send_hdr->w1.sqe_id = sqe_id &
txq->tx_compl.nb_desc_mask;
@@ -597,9 +598,9 @@ static __rte_always_inline void
cn9k_nix_sec_fc_wait_one(const struct cn9k_eth_txq *txq)
{
uint64_t nb_desc = txq->cpt_desc;
- uint64_t *fc = txq->cpt_fc;
+ uint64_t __rte_atomic *fc = txq->cpt_fc;
- while (nb_desc <= __atomic_load_n(fc, __ATOMIC_RELAXED))
+ while (nb_desc <= rte_atomic_load_explicit(fc, rte_memory_order_relaxed))
;
}
diff --git a/drivers/net/cnxk/cnxk_ethdev_dp.h b/drivers/net/cnxk/cnxk_ethdev_dp.h
index b5836b491e..cd31a36936 100644
--- a/drivers/net/cnxk/cnxk_ethdev_dp.h
+++ b/drivers/net/cnxk/cnxk_ethdev_dp.h
@@ -74,7 +74,7 @@ struct cnxk_eth_txq_comp {
uint32_t qmask;
uint32_t nb_desc_mask;
uint32_t available;
- uint32_t sqe_id;
+ uint32_t __rte_atomic sqe_id;
bool ena;
struct rte_mbuf **ptr;
rte_spinlock_t ext_buf_lock;
--
2.43.0
^ permalink raw reply [flat|nested] 3+ messages in thread
* [EXTERNAL] RE: [PATCH v2] cnxk: use stdatomic API
2025-05-12 13:58 ` [PATCH v2] " pbhagavatula
@ 2025-05-27 4:58 ` Jerin Jacob
0 siblings, 0 replies; 3+ messages in thread
From: Jerin Jacob @ 2025-05-27 4:58 UTC (permalink / raw)
To: Pavan Nikhilesh Bhagavatula, Ankur Dwivedi, Anoob Joseph,
Tejasree Kondoj, Nithin Kumar Dabilpuram,
Kiran Kumar Kokkilagadda, Sunil Kumar Kori,
Satha Koteswara Rao Kottidi, Harman Kalra
Cc: dev, Pavan Nikhilesh Bhagavatula
[-- Attachment #1: Type: text/html, Size: 7493 bytes --]
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2025-05-27 4:58 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2025-03-27 11:11 [PATCH] cnxk: use stdatomic API pbhagavatula
2025-05-12 13:58 ` [PATCH v2] " pbhagavatula
2025-05-27 4:58 ` [EXTERNAL] " Jerin Jacob
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).