DPDK patches and discussions
 help / color / mirror / Atom feed
From: <pbhagavatula@marvell.com>
To: <jerinj@marvell.com>, Pavan Nikhilesh <pbhagavatula@marvell.com>,
	"Shijith Thotton" <sthotton@marvell.com>
Cc: <dev@dpdk.org>
Subject: [PATCH] event/cnxk: update timer arm routine
Date: Thu, 10 Apr 2025 23:35:44 +0530	[thread overview]
Message-ID: <20250410180544.10765-1-pbhagavatula@marvell.com> (raw)

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add additional HW state checks in timer arm routine to ensure
HW-SW synchronization.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 drivers/event/cnxk/cnxk_tim_worker.h | 140 ++++++++++++++-------------
 1 file changed, 73 insertions(+), 67 deletions(-)

diff --git a/drivers/event/cnxk/cnxk_tim_worker.h b/drivers/event/cnxk/cnxk_tim_worker.h
index 6a9ee2fd5284..09f84091abba 100644
--- a/drivers/event/cnxk/cnxk_tim_worker.h
+++ b/drivers/event/cnxk/cnxk_tim_worker.h
@@ -102,7 +102,7 @@ cnxk_tim_bkt_get_nent(uint64_t w1)
 static inline void
 cnxk_tim_bkt_inc_nent(struct cnxk_tim_bkt *bktp)
 {
-	rte_atomic_fetch_add_explicit(&bktp->nb_entry, 1, rte_memory_order_relaxed);
+	rte_atomic_fetch_add_explicit(&bktp->nb_entry, 1, rte_memory_order_release);
 }
 
 static inline void
@@ -132,6 +132,33 @@ cnxk_tim_bkt_fast_mod(uint64_t n, uint64_t d, struct rte_reciprocal_u64 R)
 	return (n - (d * rte_reciprocal_divide_u64(n, &R)));
 }
 
+static inline uint64_t
+cnxk_tim_bkt_wait_hbt(struct cnxk_tim_bkt *bkt)
+{
+	uint64_t hbt_state;
+
+#ifdef RTE_ARCH_ARM64
+	asm volatile(PLT_CPU_FEATURE_PREAMBLE
+		     "		ldxr %[hbt], [%[w1]]	\n"
+		     "		tbz %[hbt], 33, .Ldne%=	\n"
+		     "		sevl			\n"
+		     ".Lrty%=:	wfe			\n"
+		     "		ldxr %[hbt], [%[w1]]	\n"
+		     "		tbnz %[hbt], 33, .Lrty%=\n"
+		     ".Ldne%=:				\n"
+		     : [hbt] "=&r"(hbt_state)
+		     : [w1] "r"((&bkt->w1))
+		     : "memory");
+#else
+	do {
+		hbt_state = rte_atomic_load_explicit(&bkt->w1,
+						     rte_memory_order_relaxed);
+	} while (hbt_state & BIT_ULL(33));
+#endif
+
+	return hbt_state;
+}
+
 static inline void
 cnxk_tim_format_event(const struct rte_event_timer *const tim, struct cnxk_tim_ent *const entry)
 {
@@ -265,31 +292,24 @@ cnxk_tim_add_entry_sp(struct cnxk_tim_ring *const tim_ring,
 	/* Bucket related checks. */
 	if (unlikely(cnxk_tim_bkt_get_hbt(lock_sema))) {
 		if (cnxk_tim_bkt_get_nent(lock_sema) != 0) {
-			uint64_t hbt_state;
-#ifdef RTE_ARCH_ARM64
-			asm volatile(PLT_CPU_FEATURE_PREAMBLE
-				     "		ldxr %[hbt], [%[w1]]	\n"
-				     "		tbz %[hbt], 33, .Ldne%=	\n"
-				     "		sevl			\n"
-				     ".Lrty%=:	wfe			\n"
-				     "		ldxr %[hbt], [%[w1]]	\n"
-				     "		tbnz %[hbt], 33, .Lrty%=\n"
-				     ".Ldne%=:				\n"
-				     : [hbt] "=&r"(hbt_state)
-				     : [w1] "r"((&bkt->w1))
-				     : "memory");
-#else
-			do {
-				hbt_state = rte_atomic_load_explicit(&bkt->w1,
-								     rte_memory_order_relaxed);
-			} while (hbt_state & BIT_ULL(33));
-#endif
+			uint64_t hbt_state = cnxk_tim_bkt_wait_hbt(bkt);
+
+			if (cnxk_tim_bkt_get_nent(lock_sema) != 0 &&
+			    cnxk_tim_bkt_get_nent(hbt_state) == 0) {
+				cnxk_tim_bkt_dec_lock(bkt);
+				goto __retry;
+			}
+
+			if (cnxk_tim_bkt_get_nent(hbt_state) != 0)
+				hbt_state |= BIT_ULL(34);
 
 			if (!(hbt_state & BIT_ULL(34)) ||
 			    !(hbt_state & GENMASK(31, 0))) {
 				cnxk_tim_bkt_dec_lock(bkt);
 				goto __retry;
 			}
+		} else {
+			cnxk_tim_bkt_wait_hbt(bkt);
 		}
 	}
 	/* Insert the work. */
@@ -321,9 +341,9 @@ cnxk_tim_add_entry_sp(struct cnxk_tim_ring *const tim_ring,
 
 	tim->impl_opaque[0] = (uintptr_t)chunk;
 	tim->impl_opaque[1] = (uintptr_t)bkt;
-	rte_atomic_store_explicit(&tim->state, RTE_EVENT_TIMER_ARMED, rte_memory_order_release);
+	tim->state = RTE_EVENT_TIMER_ARMED;
 	cnxk_tim_bkt_inc_nent(bkt);
-	cnxk_tim_bkt_dec_lock_relaxed(bkt);
+	cnxk_tim_bkt_dec_lock(bkt);
 
 	return 0;
 }
@@ -348,31 +368,24 @@ cnxk_tim_add_entry_mp(struct cnxk_tim_ring *const tim_ring,
 	/* Bucket related checks. */
 	if (unlikely(cnxk_tim_bkt_get_hbt(lock_sema))) {
 		if (cnxk_tim_bkt_get_nent(lock_sema) != 0) {
-			uint64_t hbt_state;
-#ifdef RTE_ARCH_ARM64
-			asm volatile(PLT_CPU_FEATURE_PREAMBLE
-				     "		ldxr %[hbt], [%[w1]]	\n"
-				     "		tbz %[hbt], 33, .Ldne%=	\n"
-				     "		sevl			\n"
-				     ".Lrty%=:	wfe			\n"
-				     "		ldxr %[hbt], [%[w1]]	\n"
-				     "		tbnz %[hbt], 33, .Lrty%=\n"
-				     ".Ldne%=:				\n"
-				     : [hbt] "=&r"(hbt_state)
-				     : [w1] "r"((&bkt->w1))
-				     : "memory");
-#else
-			do {
-				hbt_state = rte_atomic_load_explicit(&bkt->w1,
-								     rte_memory_order_relaxed);
-			} while (hbt_state & BIT_ULL(33));
-#endif
+			uint64_t hbt_state = cnxk_tim_bkt_wait_hbt(bkt);
+
+			if (cnxk_tim_bkt_get_nent(lock_sema) != 0 &&
+			    cnxk_tim_bkt_get_nent(hbt_state) == 0) {
+				cnxk_tim_bkt_dec_lock(bkt);
+				goto __retry;
+			}
+
+			if (cnxk_tim_bkt_get_nent(hbt_state) != 0)
+				hbt_state |= BIT_ULL(34);
 
 			if (!(hbt_state & BIT_ULL(34)) ||
 			    !(hbt_state & GENMASK(31, 0))) {
 				cnxk_tim_bkt_dec_lock(bkt);
 				goto __retry;
 			}
+		} else {
+			cnxk_tim_bkt_wait_hbt(bkt);
 		}
 	}
 
@@ -412,13 +425,13 @@ cnxk_tim_add_entry_mp(struct cnxk_tim_ring *const tim_ring,
 			cnxk_tim_bkt_dec_lock(bkt);
 			return -ENOMEM;
 		}
-		*chunk = *pent;
 		if (cnxk_tim_bkt_fetch_lock(lock_sema)) {
 			do {
 				lock_sema = rte_atomic_load_explicit(&bkt->w1,
 								     rte_memory_order_relaxed);
 			} while (cnxk_tim_bkt_fetch_lock(lock_sema) - 1);
 		}
+		cnxk_tim_bkt_inc_nent(bkt);
 		rte_atomic_thread_fence(rte_memory_order_acquire);
 		mirr_bkt->current_chunk = (uintptr_t)chunk;
 		rte_atomic_store_explicit(&bkt->chunk_remainder, tim_ring->nb_chunk_slots - 1,
@@ -426,14 +439,14 @@ cnxk_tim_add_entry_mp(struct cnxk_tim_ring *const tim_ring,
 	} else {
 		chunk = (struct cnxk_tim_ent *)mirr_bkt->current_chunk;
 		chunk += tim_ring->nb_chunk_slots - rem;
-		*chunk = *pent;
+		cnxk_tim_bkt_inc_nent(bkt);
 	}
 
+	*chunk = *pent;
 	tim->impl_opaque[0] = (uintptr_t)chunk;
 	tim->impl_opaque[1] = (uintptr_t)bkt;
-	rte_atomic_store_explicit(&tim->state, RTE_EVENT_TIMER_ARMED, rte_memory_order_release);
-	cnxk_tim_bkt_inc_nent(bkt);
-	cnxk_tim_bkt_dec_lock_relaxed(bkt);
+	tim->state = RTE_EVENT_TIMER_ARMED;
+	cnxk_tim_bkt_dec_lock(bkt);
 
 	return 0;
 }
@@ -480,31 +493,24 @@ cnxk_tim_add_entry_brst(struct cnxk_tim_ring *const tim_ring,
 	/* Bucket related checks. */
 	if (unlikely(cnxk_tim_bkt_get_hbt(lock_sema))) {
 		if (cnxk_tim_bkt_get_nent(lock_sema) != 0) {
-			uint64_t hbt_state;
-#ifdef RTE_ARCH_ARM64
-			asm volatile(PLT_CPU_FEATURE_PREAMBLE
-				     "		ldxr %[hbt], [%[w1]]	\n"
-				     "		tbz %[hbt], 33, .Ldne%=	\n"
-				     "		sevl			\n"
-				     ".Lrty%=:	wfe			\n"
-				     "		ldxr %[hbt], [%[w1]]	\n"
-				     "		tbnz %[hbt], 33, .Lrty%=\n"
-				     ".Ldne%=:				\n"
-				     : [hbt] "=&r"(hbt_state)
-				     : [w1] "r"((&bkt->w1))
-				     : "memory");
-#else
-			do {
-				hbt_state = rte_atomic_load_explicit(&bkt->w1,
-								     rte_memory_order_relaxed);
-			} while (hbt_state & BIT_ULL(33));
-#endif
+			uint64_t hbt_state = cnxk_tim_bkt_wait_hbt(bkt);
+
+			if (cnxk_tim_bkt_get_nent(lock_sema) != 0 &&
+			    cnxk_tim_bkt_get_nent(hbt_state) == 0) {
+				cnxk_tim_bkt_dec_lock(bkt);
+				goto __retry;
+			}
+
+			if (cnxk_tim_bkt_get_nent(hbt_state) != 0)
+				hbt_state |= BIT_ULL(34);
 
 			if (!(hbt_state & BIT_ULL(34)) ||
 			    !(hbt_state & GENMASK(31, 0))) {
 				cnxk_tim_bkt_dec_lock(bkt);
 				goto __retry;
 			}
+		} else {
+			cnxk_tim_bkt_wait_hbt(bkt);
 		}
 	}
 
@@ -553,7 +559,7 @@ cnxk_tim_add_entry_brst(struct cnxk_tim_ring *const tim_ring,
 			chunk = cnxk_tim_insert_chunk(bkt, mirr_bkt, tim_ring);
 
 		if (unlikely(chunk == NULL)) {
-			cnxk_tim_bkt_dec_lock_relaxed(bkt);
+			cnxk_tim_bkt_dec_lock(bkt);
 			rte_errno = ENOMEM;
 			tim[index]->state = RTE_EVENT_TIMER_ERROR;
 			return index;
@@ -575,7 +581,7 @@ cnxk_tim_add_entry_brst(struct cnxk_tim_ring *const tim_ring,
 		cnxk_tim_bkt_add_nent(bkt, nb_timers);
 	}
 
-	cnxk_tim_bkt_dec_lock_relaxed(bkt);
+	cnxk_tim_bkt_dec_lock(bkt);
 
 	return nb_timers;
 }
-- 
2.49.0


                 reply	other threads:[~2025-04-10 18:05 UTC|newest]

Thread overview: [no followups] expand[flat|nested]  mbox.gz  Atom feed

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20250410180544.10765-1-pbhagavatula@marvell.com \
    --to=pbhagavatula@marvell.com \
    --cc=dev@dpdk.org \
    --cc=jerinj@marvell.com \
    --cc=sthotton@marvell.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
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).