From mboxrd@z Thu Jan 1 00:00:00 1970
Return-Path: <dev-bounces@dpdk.org>
Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124])
by inbox.dpdk.org (Postfix) with ESMTP id 0402C44024;
Tue, 14 May 2024 18:39:00 +0200 (CEST)
Received: from mails.dpdk.org (localhost [127.0.0.1])
by mails.dpdk.org (Postfix) with ESMTP id 51C3E410E8;
Tue, 14 May 2024 18:36:36 +0200 (CEST)
Received: from linux.microsoft.com (linux.microsoft.com [13.77.154.182])
by mails.dpdk.org (Postfix) with ESMTP id 422BB406FF
for <dev@dpdk.org>; Tue, 14 May 2024 18:36:06 +0200 (CEST)
Received: by linux.microsoft.com (Postfix, from userid 1086)
id A8DB420B2C9E; Tue, 14 May 2024 09:36:00 -0700 (PDT)
DKIM-Filter: OpenDKIM Filter v2.11.0 linux.microsoft.com A8DB420B2C9E
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.microsoft.com;
s=default; t=1715704561;
bh=pcK/itH1S3yVJp7OnuITB1MwqY//rn9dHPNOayUftj8=;
h=From:To:Cc:Subject:Date:In-Reply-To:References:From;
b=fTPsXPsoKAzIch2rY9Yn82VkoWxIcS0CMJsuTtXGw4GAiCuLexvjEzzmX6wmByyd5
iLK/B1+aj/ZMJe5VdhBm0IAAtsP3zPYfOhT3/+I8qdP+la1WXiGsuy2vGBXL5kbUTf
IShirw5Gie7SoFk1lUOzZMAbYCU2jgAWTsKNfblM=
From: Tyler Retzlaff <roretzla@linux.microsoft.com>
To: dev@dpdk.org
Cc: =?UTF-8?q?Mattias=20R=C3=B6nnblom?= <mattias.ronnblom@ericsson.com>,
=?UTF-8?q?Morten=20Br=C3=B8rup?= <mb@smartsharesystems.com>,
Abdullah Sevincer <abdullah.sevincer@intel.com>,
Ajit Khaparde <ajit.khaparde@broadcom.com>,
Alok Prasad <palok@marvell.com>,
Anatoly Burakov <anatoly.burakov@intel.com>,
Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>,
Anoob Joseph <anoobj@marvell.com>,
Bruce Richardson <bruce.richardson@intel.com>,
Byron Marohn <byron.marohn@intel.com>, Chenbo Xia <chenbox@nvidia.com>,
Chengwen Feng <fengchengwen@huawei.com>,
Ciara Loftus <ciara.loftus@intel.com>, Ciara Power <ciara.power@intel.com>,
Dariusz Sosnowski <dsosnowski@nvidia.com>,
David Hunt <david.hunt@intel.com>,
Devendra Singh Rawat <dsinghrawat@marvell.com>,
Erik Gabriel Carrillo <erik.g.carrillo@intel.com>,
Guoyang Zhou <zhouguoyang@huawei.com>, Harman Kalra <hkalra@marvell.com>,
Harry van Haaren <harry.van.haaren@intel.com>,
Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>,
Jakub Grajciar <jgrajcia@cisco.com>, Jerin Jacob <jerinj@marvell.com>,
Jeroen de Borst <jeroendb@google.com>, Jian Wang <jianwang@trustnetic.com>,
Jiawen Wu <jiawenwu@trustnetic.com>, Jie Hai <haijie1@huawei.com>,
Jingjing Wu <jingjing.wu@intel.com>,
Joshua Washington <joshwash@google.com>, Joyce Kong <joyce.kong@arm.com>,
Junfeng Guo <junfeng.guo@intel.com>, Kevin Laatz <kevin.laatz@intel.com>,
Konstantin Ananyev <konstantin.v.ananyev@yandex.ru>,
Liang Ma <liangma@liangbit.com>, Long Li <longli@microsoft.com>,
Maciej Czekaj <mczekaj@marvell.com>, Matan Azrad <matan@nvidia.com>,
Maxime Coquelin <maxime.coquelin@redhat.com>,
Nicolas Chautru <nicolas.chautru@intel.com>, Ori Kam <orika@nvidia.com>,
Pavan Nikhilesh <pbhagavatula@marvell.com>,
Peter Mccarthy <peter.mccarthy@intel.com>,
Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>,
Reshma Pattan <reshma.pattan@intel.com>, Rosen Xu <rosen.xu@intel.com>,
Ruifeng Wang <ruifeng.wang@arm.com>, Rushil Gupta <rushilg@google.com>,
Sameh Gobriel <sameh.gobriel@intel.com>,
Sivaprasad Tummala <sivaprasad.tummala@amd.com>,
Somnath Kotur <somnath.kotur@broadcom.com>,
Stephen Hemminger <stephen@networkplumber.org>,
Suanming Mou <suanmingm@nvidia.com>, Sunil Kumar Kori <skori@marvell.com>,
Sunil Uttarwar <sunilprakashrao.uttarwar@amd.com>,
Tetsuya Mukawa <mtetsuyah@gmail.com>,
Vamsi Attunuru <vattunuru@marvell.com>,
Viacheslav Ovsiienko <viacheslavo@nvidia.com>,
Vladimir Medvedkin <vladimir.medvedkin@intel.com>,
Xiaoyun Wang <cloud.wangxiaoyun@huawei.com>,
Yipeng Wang <yipeng1.wang@intel.com>,
Yisen Zhuang <yisen.zhuang@huawei.com>,
Ziyang Xuan <xuanziyang2@huawei.com>,
Tyler Retzlaff <roretzla@linux.microsoft.com>
Subject: [PATCH v6 23/45] event/opdl: use rte stdatomic API
Date: Tue, 14 May 2024 09:35:37 -0700
Message-Id: <1715704559-15839-24-git-send-email-roretzla@linux.microsoft.com>
X-Mailer: git-send-email 1.8.3.1
In-Reply-To: <1715704559-15839-1-git-send-email-roretzla@linux.microsoft.com>
References: <1710967892-7046-1-git-send-email-roretzla@linux.microsoft.com>
<1715704559-15839-1-git-send-email-roretzla@linux.microsoft.com>
X-BeenThere: dev@dpdk.org
X-Mailman-Version: 2.1.29
Precedence: list
List-Id: DPDK patches and discussions <dev.dpdk.org>
List-Unsubscribe: <https://mails.dpdk.org/options/dev>,
<mailto:dev-request@dpdk.org?subject=unsubscribe>
List-Archive: <http://mails.dpdk.org/archives/dev/>
List-Post: <mailto:dev@dpdk.org>
List-Help: <mailto:dev-request@dpdk.org?subject=help>
List-Subscribe: <https://mails.dpdk.org/listinfo/dev>,
<mailto:dev-request@dpdk.org?subject=subscribe>
Errors-To: dev-bounces@dpdk.org
Replace the use of gcc builtin __atomic_xxx intrinsics with
corresponding rte_atomic_xxx optional rte stdatomic API.
Signed-off-by: Tyler Retzlaff <roretzla@linux.microsoft.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
---
drivers/event/opdl/opdl_ring.c | 80 +++++++++++++++++++++---------------------
1 file changed, 40 insertions(+), 40 deletions(-)
diff --git a/drivers/event/opdl/opdl_ring.c b/drivers/event/opdl/opdl_ring.c
index e87ffd5..3476f6b 100644
--- a/drivers/event/opdl/opdl_ring.c
+++ b/drivers/event/opdl/opdl_ring.c
@@ -47,12 +47,12 @@ struct __rte_cache_aligned shared_state {
/* Last known minimum sequence number of dependencies, used for multi
* thread operation
*/
- uint32_t available_seq;
+ RTE_ATOMIC(uint32_t) available_seq;
char _pad1[RTE_CACHE_LINE_SIZE * 3];
- uint32_t head; /* Head sequence number (for multi thread operation) */
+ RTE_ATOMIC(uint32_t) head; /* Head sequence number (for multi thread operation) */
char _pad2[RTE_CACHE_LINE_SIZE * 3];
struct opdl_stage *stage; /* back pointer */
- uint32_t tail; /* Tail sequence number */
+ RTE_ATOMIC(uint32_t) tail; /* Tail sequence number */
char _pad3[RTE_CACHE_LINE_SIZE * 2];
};
@@ -149,10 +149,10 @@ struct opdl_ring {
available(const struct opdl_stage *s)
{
if (s->threadsafe == true) {
- uint32_t n = __atomic_load_n(&s->shared.available_seq,
- __ATOMIC_ACQUIRE) -
- __atomic_load_n(&s->shared.head,
- __ATOMIC_ACQUIRE);
+ uint32_t n = rte_atomic_load_explicit(&s->shared.available_seq,
+ rte_memory_order_acquire) -
+ rte_atomic_load_explicit(&s->shared.head,
+ rte_memory_order_acquire);
/* Return 0 if available_seq needs to be updated */
return (n <= s->num_slots) ? n : 0;
@@ -168,7 +168,7 @@ struct opdl_ring {
{
uint32_t i;
uint32_t this_tail = s->shared.tail;
- uint32_t min_seq = __atomic_load_n(&s->deps[0]->tail, __ATOMIC_ACQUIRE);
+ uint32_t min_seq = rte_atomic_load_explicit(&s->deps[0]->tail, rte_memory_order_acquire);
/* Input stage sequence numbers are greater than the sequence numbers of
* its dependencies so an offset of t->num_slots is needed when
* calculating available slots and also the condition which is used to
@@ -179,16 +179,16 @@ struct opdl_ring {
if (is_input_stage(s)) {
wrap = s->num_slots;
for (i = 1; i < s->num_deps; i++) {
- uint32_t seq = __atomic_load_n(&s->deps[i]->tail,
- __ATOMIC_ACQUIRE);
+ uint32_t seq = rte_atomic_load_explicit(&s->deps[i]->tail,
+ rte_memory_order_acquire);
if ((this_tail - seq) > (this_tail - min_seq))
min_seq = seq;
}
} else {
wrap = 0;
for (i = 1; i < s->num_deps; i++) {
- uint32_t seq = __atomic_load_n(&s->deps[i]->tail,
- __ATOMIC_ACQUIRE);
+ uint32_t seq = rte_atomic_load_explicit(&s->deps[i]->tail,
+ rte_memory_order_acquire);
if ((seq - this_tail) < (min_seq - this_tail))
min_seq = seq;
}
@@ -197,8 +197,8 @@ struct opdl_ring {
if (s->threadsafe == false)
s->available_seq = min_seq + wrap;
else
- __atomic_store_n(&s->shared.available_seq, min_seq + wrap,
- __ATOMIC_RELEASE);
+ rte_atomic_store_explicit(&s->shared.available_seq, min_seq + wrap,
+ rte_memory_order_release);
}
/* Wait until the number of available slots reaches number requested */
@@ -298,7 +298,7 @@ struct opdl_ring {
copy_entries_in(t, head, entries, num_entries);
s->head += num_entries;
- __atomic_store_n(&s->shared.tail, s->head, __ATOMIC_RELEASE);
+ rte_atomic_store_explicit(&s->shared.tail, s->head, rte_memory_order_release);
return num_entries;
}
@@ -381,18 +381,18 @@ struct opdl_ring {
/* There should be no race condition here. If shared.tail
* matches, no other core can update it until this one does.
*/
- if (__atomic_load_n(&s->shared.tail, __ATOMIC_ACQUIRE) ==
+ if (rte_atomic_load_explicit(&s->shared.tail, rte_memory_order_acquire) ==
tail) {
if (num_entries >= (head - tail)) {
claim_mgr_remove(disclaims);
- __atomic_store_n(&s->shared.tail, head,
- __ATOMIC_RELEASE);
+ rte_atomic_store_explicit(&s->shared.tail, head,
+ rte_memory_order_release);
num_entries -= (head - tail);
} else {
claim_mgr_move_tail(disclaims, num_entries);
- __atomic_store_n(&s->shared.tail,
+ rte_atomic_store_explicit(&s->shared.tail,
num_entries + tail,
- __ATOMIC_RELEASE);
+ rte_memory_order_release);
num_entries = 0;
}
} else if (block == false)
@@ -420,7 +420,7 @@ struct opdl_ring {
opdl_stage_disclaim_multithread_n(s, disclaims->num_to_disclaim,
false);
- *old_head = __atomic_load_n(&s->shared.head, __ATOMIC_ACQUIRE);
+ *old_head = rte_atomic_load_explicit(&s->shared.head, rte_memory_order_acquire);
while (true) {
bool success;
/* If called by opdl_ring_input(), claim does not need to be
@@ -440,11 +440,10 @@ struct opdl_ring {
if (*num_entries == 0)
return;
- success = __atomic_compare_exchange_n(&s->shared.head, old_head,
+ success = rte_atomic_compare_exchange_weak_explicit(&s->shared.head, old_head,
*old_head + *num_entries,
- true, /* may fail spuriously */
- __ATOMIC_RELEASE, /* memory order on success */
- __ATOMIC_ACQUIRE); /* memory order on fail */
+ rte_memory_order_release, /* memory order on success */
+ rte_memory_order_acquire); /* memory order on fail */
if (likely(success))
break;
rte_pause();
@@ -472,10 +471,11 @@ struct opdl_ring {
/* If another thread started inputting before this one, but hasn't
* finished, we need to wait for it to complete to update the tail.
*/
- rte_wait_until_equal_32(&s->shared.tail, old_head, __ATOMIC_ACQUIRE);
+ rte_wait_until_equal_32((uint32_t *)(uintptr_t)&s->shared.tail, old_head,
+ rte_memory_order_acquire);
- __atomic_store_n(&s->shared.tail, old_head + num_entries,
- __ATOMIC_RELEASE);
+ rte_atomic_store_explicit(&s->shared.tail, old_head + num_entries,
+ rte_memory_order_release);
return num_entries;
}
@@ -525,8 +525,8 @@ struct opdl_ring {
for (j = 0; j < num_entries; j++) {
ev = (struct rte_event *)get_slot(t, s->head+j);
- event = __atomic_load_n(&(ev->event),
- __ATOMIC_ACQUIRE);
+ event = rte_atomic_load_explicit((uint64_t __rte_atomic *)&ev->event,
+ rte_memory_order_acquire);
opa_id = OPDL_OPA_MASK & (event >> OPDL_OPA_OFFSET);
flow_id = OPDL_FLOWID_MASK & event;
@@ -627,8 +627,8 @@ struct opdl_ring {
num_entries, s->head - old_tail);
num_entries = s->head - old_tail;
}
- __atomic_store_n(&s->shared.tail, num_entries + old_tail,
- __ATOMIC_RELEASE);
+ rte_atomic_store_explicit(&s->shared.tail, num_entries + old_tail,
+ rte_memory_order_release);
}
uint32_t
@@ -657,7 +657,7 @@ struct opdl_ring {
copy_entries_in(t, head, entries, num_entries);
s->head += num_entries;
- __atomic_store_n(&s->shared.tail, s->head, __ATOMIC_RELEASE);
+ rte_atomic_store_explicit(&s->shared.tail, s->head, rte_memory_order_release);
return num_entries;
@@ -676,7 +676,7 @@ struct opdl_ring {
copy_entries_out(t, head, entries, num_entries);
s->head += num_entries;
- __atomic_store_n(&s->shared.tail, s->head, __ATOMIC_RELEASE);
+ rte_atomic_store_explicit(&s->shared.tail, s->head, rte_memory_order_release);
return num_entries;
}
@@ -755,7 +755,7 @@ struct opdl_ring {
return 0;
}
if (s->threadsafe == false) {
- __atomic_store_n(&s->shared.tail, s->head, __ATOMIC_RELEASE);
+ rte_atomic_store_explicit(&s->shared.tail, s->head, rte_memory_order_release);
s->seq += s->num_claimed;
s->shadow_head = s->head;
s->num_claimed = 0;
@@ -1008,8 +1008,8 @@ struct opdl_ring *
ev_orig = (struct rte_event *)
get_slot(t, s->shadow_head+i);
- event = __atomic_load_n(&(ev_orig->event),
- __ATOMIC_ACQUIRE);
+ event = rte_atomic_load_explicit((uint64_t __rte_atomic *)&ev_orig->event,
+ rte_memory_order_acquire);
opa_id = OPDL_OPA_MASK & (event >> OPDL_OPA_OFFSET);
flow_id = OPDL_FLOWID_MASK & event;
@@ -1026,9 +1026,9 @@ struct opdl_ring *
if ((event & OPDL_EVENT_MASK) !=
ev_temp) {
- __atomic_store_n(&(ev_orig->event),
- ev_update,
- __ATOMIC_RELEASE);
+ rte_atomic_store_explicit(
+ (uint64_t __rte_atomic *)&ev_orig->event,
+ ev_update, rte_memory_order_release);
ev_updated = true;
}
if (ev_orig->u64 != ev->u64) {
--
1.8.3.1