From patchwork Wed Oct 30 21:22:58 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Konstantin Ananyev X-Patchwork-Id: 147761 X-Patchwork-Delegate: thomas@monjalon.net Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 02F7E45B82; Wed, 30 Oct 2024 21:32:49 +0100 (CET) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id E8A874332C; Wed, 30 Oct 2024 21:32:48 +0100 (CET) Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) by mails.dpdk.org (Postfix) with ESMTP id B5F524332C; Wed, 30 Oct 2024 21:32:47 +0100 (CET) Received: from mail.maildlp.com (unknown [172.18.186.231]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4XdzL76Vv6z6K5mS; Thu, 31 Oct 2024 04:31:27 +0800 (CST) Received: from frapeml500007.china.huawei.com (unknown [7.182.85.172]) by mail.maildlp.com (Postfix) with ESMTPS id ECCAE140B38; Thu, 31 Oct 2024 04:32:46 +0800 (CST) Received: from localhost.localdomain (10.220.239.45) by frapeml500007.china.huawei.com (7.182.85.172) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.1.2507.39; Wed, 30 Oct 2024 21:32:46 +0100 From: Konstantin Ananyev To: CC: , , , , , , , , , Subject: [PATCH v7 1/7] test/ring: fix failure with custom number of lcores Date: Wed, 30 Oct 2024 17:22:58 -0400 Message-ID: <20241030212304.104180-2-konstantin.ananyev@huawei.com> X-Mailer: git-send-email 2.35.3 In-Reply-To: <20241030212304.104180-1-konstantin.ananyev@huawei.com> References: <20241021174745.1843-1-konstantin.ananyev@huawei.com> <20241030212304.104180-1-konstantin.ananyev@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.220.239.45] X-ClientProxiedBy: frapeml100006.china.huawei.com (7.182.85.201) To frapeml500007.china.huawei.com (7.182.85.172) X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org ring_stress_autotest fails to initialize the ring when RTE_MAX_LCORE value is not a number of 2. There is a flaw in calculation required number of elements in the ring. Fix it by aligning number of elements to next power of 2. Fixes: bf28df24e915 ("test/ring: add contention stress test") Cc: stable@dpdk.org Signed-off-by: Konstantin Ananyev Acked-by: Morten Brørup --- app/test/test_ring_stress_impl.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/app/test/test_ring_stress_impl.h b/app/test/test_ring_stress_impl.h index 8b0bfb11fe..ee5274aeef 100644 --- a/app/test/test_ring_stress_impl.h +++ b/app/test/test_ring_stress_impl.h @@ -297,7 +297,7 @@ mt1_init(struct rte_ring **rng, void **data, uint32_t num) *data = elm; /* alloc ring */ - nr = 2 * num; + nr = rte_align32pow2(2 * num); sz = rte_ring_get_memsize(nr); r = rte_zmalloc(NULL, sz, alignof(typeof(*r))); if (r == NULL) { From patchwork Wed Oct 30 21:22:59 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Konstantin Ananyev X-Patchwork-Id: 147762 X-Patchwork-Delegate: thomas@monjalon.net Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 20C8A45B82; Wed, 30 Oct 2024 21:32:55 +0100 (CET) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id 10B9843366; Wed, 30 Oct 2024 21:32:55 +0100 (CET) Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) by mails.dpdk.org (Postfix) with ESMTP id 994ED4335B for ; Wed, 30 Oct 2024 21:32:51 +0100 (CET) Received: from mail.maildlp.com (unknown [172.18.186.31]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4XdzG748mdz6L74N; Thu, 31 Oct 2024 04:27:59 +0800 (CST) Received: from frapeml500007.china.huawei.com (unknown [7.182.85.172]) by mail.maildlp.com (Postfix) with ESMTPS id 2AC501404F4; Thu, 31 Oct 2024 04:32:51 +0800 (CST) Received: from localhost.localdomain (10.220.239.45) by frapeml500007.china.huawei.com (7.182.85.172) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.1.2507.39; Wed, 30 Oct 2024 21:32:50 +0100 From: Konstantin Ananyev To: CC: , , , , , , , , Subject: [PATCH v7 2/7] ring: common functions for 'move head' ops Date: Wed, 30 Oct 2024 17:22:59 -0400 Message-ID: <20241030212304.104180-3-konstantin.ananyev@huawei.com> X-Mailer: git-send-email 2.35.3 In-Reply-To: <20241030212304.104180-1-konstantin.ananyev@huawei.com> References: <20241021174745.1843-1-konstantin.ananyev@huawei.com> <20241030212304.104180-1-konstantin.ananyev@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.220.239.45] X-ClientProxiedBy: frapeml100006.china.huawei.com (7.182.85.201) To frapeml500007.china.huawei.com (7.182.85.172) X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Note upfront: that change doesn't introduce any functional or performance changes. It is just a code-reordering for: - code deduplication - ability in future to re-use the same code to introduce new functionality For each sync mode corresponding move_prod_head() and move_cons_head() are nearly identical to each other, the only differences are: - do we need to use a @capacity to calculate number of entries or not. - what we need to update (prod/cons) and what is used as read-only counterpart. So instead of having 2 copies of nearly identical functions, introduce a new common one that could be used by both functions: move_prod_head() and move_cons_head(). As another positive thing - we can get rid of referencing whole rte_ring structure in that new common sub-function. Signed-off-by: Konstantin Ananyev Acked-by: Morten Brørup --- lib/ring/rte_ring_c11_pvt.h | 156 ++++++++++--------------------- lib/ring/rte_ring_elem_pvt.h | 66 +++++++++++++ lib/ring/rte_ring_generic_pvt.h | 143 +++++++++------------------- lib/ring/rte_ring_hts_elem_pvt.h | 107 ++++++++++----------- lib/ring/rte_ring_rts_elem_pvt.h | 107 ++++++++++----------- 5 files changed, 255 insertions(+), 324 deletions(-) diff --git a/lib/ring/rte_ring_c11_pvt.h b/lib/ring/rte_ring_c11_pvt.h index 629b2d9288..b9388af0da 100644 --- a/lib/ring/rte_ring_c11_pvt.h +++ b/lib/ring/rte_ring_c11_pvt.h @@ -11,6 +11,17 @@ #ifndef _RTE_RING_C11_PVT_H_ #define _RTE_RING_C11_PVT_H_ +/** + * @file rte_ring_c11_pvt.h + * It is not recommended to include this file directly, + * include instead. + * Contains internal helper functions for MP/SP and MC/SC ring modes. + * For more information please refer to . + */ + +/** + * @internal This function updates tail values. + */ static __rte_always_inline void __rte_ring_update_tail(struct rte_ring_headtail *ht, uint32_t old_val, uint32_t new_val, uint32_t single, uint32_t enqueue) @@ -29,40 +40,45 @@ __rte_ring_update_tail(struct rte_ring_headtail *ht, uint32_t old_val, } /** - * @internal This function updates the producer head for enqueue + * @internal This is a helper function that moves the producer/consumer head * - * @param r - * A pointer to the ring structure - * @param is_sp - * Indicates whether multi-producer path is needed or not + * @param d + * A pointer to the headtail structure with head value to be moved + * @param s + * A pointer to the counter-part headtail structure. Note that this + * function only reads tail value from it + * @param capacity + * Either ring capacity value (for producer), or zero (for consumer) + * @param is_st + * Indicates whether multi-thread safe path is needed or not * @param n - * The number of elements we will want to enqueue, i.e. how far should the - * head be moved + * The number of elements we want to move head value on * @param behavior - * RTE_RING_QUEUE_FIXED: Enqueue a fixed number of items from a ring - * RTE_RING_QUEUE_VARIABLE: Enqueue as many items as possible from ring + * RTE_RING_QUEUE_FIXED: Move on a fixed number of items + * RTE_RING_QUEUE_VARIABLE: Move on as many items as possible * @param old_head - * Returns head value as it was before the move, i.e. where enqueue starts + * Returns head value as it was before the move * @param new_head - * Returns the current/new head value i.e. where enqueue finishes - * @param free_entries - * Returns the amount of free space in the ring BEFORE head was moved + * Returns the new head value + * @param entries + * Returns the number of ring entries available BEFORE head was moved * @return - * Actual number of objects enqueued. - * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only. + * Actual number of objects the head was moved on + * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only */ static __rte_always_inline unsigned int -__rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp, - unsigned int n, enum rte_ring_queue_behavior behavior, - uint32_t *old_head, uint32_t *new_head, - uint32_t *free_entries) +__rte_ring_headtail_move_head(struct rte_ring_headtail *d, + const struct rte_ring_headtail *s, uint32_t capacity, + unsigned int is_st, unsigned int n, + enum rte_ring_queue_behavior behavior, + uint32_t *old_head, uint32_t *new_head, uint32_t *entries) { - const uint32_t capacity = r->capacity; - uint32_t cons_tail; - unsigned int max = n; + uint32_t stail; int success; + unsigned int max = n; - *old_head = rte_atomic_load_explicit(&r->prod.head, rte_memory_order_relaxed); + *old_head = rte_atomic_load_explicit(&d->head, + rte_memory_order_relaxed); do { /* Reset n to the initial burst count */ n = max; @@ -73,112 +89,36 @@ __rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp, /* load-acquire synchronize with store-release of ht->tail * in update_tail. */ - cons_tail = rte_atomic_load_explicit(&r->cons.tail, + stail = rte_atomic_load_explicit(&s->tail, rte_memory_order_acquire); /* The subtraction is done between two unsigned 32bits value * (the result is always modulo 32 bits even if we have - * *old_head > cons_tail). So 'free_entries' is always between 0 + * *old_head > s->tail). So 'entries' is always between 0 * and capacity (which is < size). */ - *free_entries = (capacity + cons_tail - *old_head); + *entries = (capacity + stail - *old_head); /* check that we have enough room in ring */ - if (unlikely(n > *free_entries)) + if (unlikely(n > *entries)) n = (behavior == RTE_RING_QUEUE_FIXED) ? - 0 : *free_entries; + 0 : *entries; if (n == 0) return 0; *new_head = *old_head + n; - if (is_sp) { - r->prod.head = *new_head; + if (is_st) { + d->head = *new_head; success = 1; } else /* on failure, *old_head is updated */ - success = rte_atomic_compare_exchange_strong_explicit(&r->prod.head, - old_head, *new_head, + success = rte_atomic_compare_exchange_strong_explicit( + &d->head, old_head, *new_head, rte_memory_order_relaxed, rte_memory_order_relaxed); } while (unlikely(success == 0)); return n; } -/** - * @internal This function updates the consumer head for dequeue - * - * @param r - * A pointer to the ring structure - * @param is_sc - * Indicates whether multi-consumer path is needed or not - * @param n - * The number of elements we will want to dequeue, i.e. how far should the - * head be moved - * @param behavior - * RTE_RING_QUEUE_FIXED: Dequeue a fixed number of items from a ring - * RTE_RING_QUEUE_VARIABLE: Dequeue as many items as possible from ring - * @param old_head - * Returns head value as it was before the move, i.e. where dequeue starts - * @param new_head - * Returns the current/new head value i.e. where dequeue finishes - * @param entries - * Returns the number of entries in the ring BEFORE head was moved - * @return - * - Actual number of objects dequeued. - * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only. - */ -static __rte_always_inline unsigned int -__rte_ring_move_cons_head(struct rte_ring *r, int is_sc, - unsigned int n, enum rte_ring_queue_behavior behavior, - uint32_t *old_head, uint32_t *new_head, - uint32_t *entries) -{ - unsigned int max = n; - uint32_t prod_tail; - int success; - - /* move cons.head atomically */ - *old_head = rte_atomic_load_explicit(&r->cons.head, rte_memory_order_relaxed); - do { - /* Restore n as it may change every loop */ - n = max; - - /* Ensure the head is read before tail */ - rte_atomic_thread_fence(rte_memory_order_acquire); - - /* this load-acquire synchronize with store-release of ht->tail - * in update_tail. - */ - prod_tail = rte_atomic_load_explicit(&r->prod.tail, - rte_memory_order_acquire); - - /* The subtraction is done between two unsigned 32bits value - * (the result is always modulo 32 bits even if we have - * cons_head > prod_tail). So 'entries' is always between 0 - * and size(ring)-1. - */ - *entries = (prod_tail - *old_head); - - /* Set the actual entries for dequeue */ - if (n > *entries) - n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries; - - if (unlikely(n == 0)) - return 0; - - *new_head = *old_head + n; - if (is_sc) { - r->cons.head = *new_head; - success = 1; - } else - /* on failure, *old_head will be updated */ - success = rte_atomic_compare_exchange_strong_explicit(&r->cons.head, - old_head, *new_head, - rte_memory_order_relaxed, - rte_memory_order_relaxed); - } while (unlikely(success == 0)); - return n; -} - #endif /* _RTE_RING_C11_PVT_H_ */ diff --git a/lib/ring/rte_ring_elem_pvt.h b/lib/ring/rte_ring_elem_pvt.h index 4b80f58980..3a83668a08 100644 --- a/lib/ring/rte_ring_elem_pvt.h +++ b/lib/ring/rte_ring_elem_pvt.h @@ -293,6 +293,72 @@ __rte_ring_dequeue_elems(struct rte_ring *r, uint32_t cons_head, #include "rte_ring_generic_pvt.h" #endif +/** + * @internal This function updates the producer head for enqueue + * + * @param r + * A pointer to the ring structure + * @param is_sp + * Indicates whether multi-producer path is needed or not + * @param n + * The number of elements we will want to enqueue, i.e. how far should the + * head be moved + * @param behavior + * RTE_RING_QUEUE_FIXED: Enqueue a fixed number of items from a ring + * RTE_RING_QUEUE_VARIABLE: Enqueue as many items as possible from ring + * @param old_head + * Returns head value as it was before the move, i.e. where enqueue starts + * @param new_head + * Returns the current/new head value i.e. where enqueue finishes + * @param free_entries + * Returns the amount of free space in the ring BEFORE head was moved + * @return + * Actual number of objects enqueued. + * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only. + */ +static __rte_always_inline unsigned int +__rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp, + unsigned int n, enum rte_ring_queue_behavior behavior, + uint32_t *old_head, uint32_t *new_head, + uint32_t *free_entries) +{ + return __rte_ring_headtail_move_head(&r->prod, &r->cons, r->capacity, + is_sp, n, behavior, old_head, new_head, free_entries); +} + +/** + * @internal This function updates the consumer head for dequeue + * + * @param r + * A pointer to the ring structure + * @param is_sc + * Indicates whether multi-consumer path is needed or not + * @param n + * The number of elements we will want to dequeue, i.e. how far should the + * head be moved + * @param behavior + * RTE_RING_QUEUE_FIXED: Dequeue a fixed number of items from a ring + * RTE_RING_QUEUE_VARIABLE: Dequeue as many items as possible from ring + * @param old_head + * Returns head value as it was before the move, i.e. where dequeue starts + * @param new_head + * Returns the current/new head value i.e. where dequeue finishes + * @param entries + * Returns the number of entries in the ring BEFORE head was moved + * @return + * - Actual number of objects dequeued. + * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only. + */ +static __rte_always_inline unsigned int +__rte_ring_move_cons_head(struct rte_ring *r, unsigned int is_sc, + unsigned int n, enum rte_ring_queue_behavior behavior, + uint32_t *old_head, uint32_t *new_head, + uint32_t *entries) +{ + return __rte_ring_headtail_move_head(&r->cons, &r->prod, 0, + is_sc, n, behavior, old_head, new_head, entries); +} + /** * @internal Enqueue several objects on the ring * diff --git a/lib/ring/rte_ring_generic_pvt.h b/lib/ring/rte_ring_generic_pvt.h index 457f41dab3..affd2d5ba7 100644 --- a/lib/ring/rte_ring_generic_pvt.h +++ b/lib/ring/rte_ring_generic_pvt.h @@ -10,6 +10,17 @@ #ifndef _RTE_RING_GENERIC_PVT_H_ #define _RTE_RING_GENERIC_PVT_H_ +/** + * @file rte_ring_generic_pvt.h + * It is not recommended to include this file directly, + * include instead. + * Contains internal helper functions for MP/SP and MC/SC ring modes. + * For more information please refer to . + */ + +/** + * @internal This function updates tail values. + */ static __rte_always_inline void __rte_ring_update_tail(struct rte_ring_headtail *ht, uint32_t old_val, uint32_t new_val, uint32_t single, uint32_t enqueue) @@ -30,35 +41,39 @@ __rte_ring_update_tail(struct rte_ring_headtail *ht, uint32_t old_val, } /** - * @internal This function updates the producer head for enqueue + * @internal This is a helper function that moves the producer/consumer head * - * @param r - * A pointer to the ring structure - * @param is_sp - * Indicates whether multi-producer path is needed or not + * @param d + * A pointer to the headtail structure with head value to be moved + * @param s + * A pointer to the counter-part headtail structure. Note that this + * function only reads tail value from it + * @param capacity + * Either ring capacity value (for producer), or zero (for consumer) + * @param is_st + * Indicates whether multi-thread safe path is needed or not * @param n - * The number of elements we will want to enqueue, i.e. how far should the - * head be moved + * The number of elements we want to move head value on * @param behavior - * RTE_RING_QUEUE_FIXED: Enqueue a fixed number of items from a ring - * RTE_RING_QUEUE_VARIABLE: Enqueue as many items as possible from ring + * RTE_RING_QUEUE_FIXED: Move on a fixed number of items + * RTE_RING_QUEUE_VARIABLE: Move on as many items as possible * @param old_head - * Returns head value as it was before the move, i.e. where enqueue starts + * Returns head value as it was before the move * @param new_head - * Returns the current/new head value i.e. where enqueue finishes - * @param free_entries - * Returns the amount of free space in the ring BEFORE head was moved + * Returns the new head value + * @param entries + * Returns the number of ring entries available BEFORE head was moved * @return - * Actual number of objects enqueued. - * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only. + * Actual number of objects the head was moved on + * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only */ static __rte_always_inline unsigned int -__rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp, - unsigned int n, enum rte_ring_queue_behavior behavior, - uint32_t *old_head, uint32_t *new_head, - uint32_t *free_entries) +__rte_ring_headtail_move_head(struct rte_ring_headtail *d, + const struct rte_ring_headtail *s, uint32_t capacity, + unsigned int is_st, unsigned int n, + enum rte_ring_queue_behavior behavior, + uint32_t *old_head, uint32_t *new_head, uint32_t *entries) { - const uint32_t capacity = r->capacity; unsigned int max = n; int success; @@ -66,7 +81,7 @@ __rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp, /* Reset n to the initial burst count */ n = max; - *old_head = r->prod.head; + *old_head = d->head; /* add rmb barrier to avoid load/load reorder in weak * memory model. It is noop on x86 @@ -76,97 +91,27 @@ __rte_ring_move_prod_head(struct rte_ring *r, unsigned int is_sp, /* * The subtraction is done between two unsigned 32bits value * (the result is always modulo 32 bits even if we have - * *old_head > cons_tail). So 'free_entries' is always between 0 + * *old_head > s->tail). So 'entries' is always between 0 * and capacity (which is < size). */ - *free_entries = (capacity + r->cons.tail - *old_head); + *entries = (capacity + s->tail - *old_head); /* check that we have enough room in ring */ - if (unlikely(n > *free_entries)) + if (unlikely(n > *entries)) n = (behavior == RTE_RING_QUEUE_FIXED) ? - 0 : *free_entries; + 0 : *entries; if (n == 0) return 0; *new_head = *old_head + n; - if (is_sp) { - r->prod.head = *new_head; + if (is_st) { + d->head = *new_head; success = 1; } else - success = rte_atomic32_cmpset((uint32_t *)(uintptr_t)&r->prod.head, - *old_head, *new_head); - } while (unlikely(success == 0)); - return n; -} - -/** - * @internal This function updates the consumer head for dequeue - * - * @param r - * A pointer to the ring structure - * @param is_sc - * Indicates whether multi-consumer path is needed or not - * @param n - * The number of elements we will want to dequeue, i.e. how far should the - * head be moved - * @param behavior - * RTE_RING_QUEUE_FIXED: Dequeue a fixed number of items from a ring - * RTE_RING_QUEUE_VARIABLE: Dequeue as many items as possible from ring - * @param old_head - * Returns head value as it was before the move, i.e. where dequeue starts - * @param new_head - * Returns the current/new head value i.e. where dequeue finishes - * @param entries - * Returns the number of entries in the ring BEFORE head was moved - * @return - * - Actual number of objects dequeued. - * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only. - */ -static __rte_always_inline unsigned int -__rte_ring_move_cons_head(struct rte_ring *r, unsigned int is_sc, - unsigned int n, enum rte_ring_queue_behavior behavior, - uint32_t *old_head, uint32_t *new_head, - uint32_t *entries) -{ - unsigned int max = n; - int success; - - /* move cons.head atomically */ - do { - /* Restore n as it may change every loop */ - n = max; - - *old_head = r->cons.head; - - /* add rmb barrier to avoid load/load reorder in weak - * memory model. It is noop on x86 - */ - rte_smp_rmb(); - - /* The subtraction is done between two unsigned 32bits value - * (the result is always modulo 32 bits even if we have - * cons_head > prod_tail). So 'entries' is always between 0 - * and size(ring)-1. - */ - *entries = (r->prod.tail - *old_head); - - /* Set the actual entries for dequeue */ - if (n > *entries) - n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries; - - if (unlikely(n == 0)) - return 0; - - *new_head = *old_head + n; - if (is_sc) { - r->cons.head = *new_head; - rte_smp_rmb(); - success = 1; - } else { - success = rte_atomic32_cmpset((uint32_t *)(uintptr_t)&r->cons.head, + success = rte_atomic32_cmpset( + (uint32_t *)(uintptr_t)&d->head, *old_head, *new_head); - } } while (unlikely(success == 0)); return n; } diff --git a/lib/ring/rte_ring_hts_elem_pvt.h b/lib/ring/rte_ring_hts_elem_pvt.h index 91f5eeccb9..e2b82dd1e6 100644 --- a/lib/ring/rte_ring_hts_elem_pvt.h +++ b/lib/ring/rte_ring_hts_elem_pvt.h @@ -51,19 +51,39 @@ __rte_ring_hts_head_wait(const struct rte_ring_hts_headtail *ht, } /** - * @internal This function updates the producer head for enqueue + * @internal This is a helper function that moves the producer/consumer head + * + * @param d + * A pointer to the headtail structure with head value to be moved + * @param s + * A pointer to the counter-part headtail structure. Note that this + * function only reads tail value from it + * @param capacity + * Either ring capacity value (for producer), or zero (for consumer) + * Indicates whether multi-thread safe path is needed or not + * @param num + * The number of elements we want to move head value on + * @param behavior + * RTE_RING_QUEUE_FIXED: Move on a fixed number of items + * RTE_RING_QUEUE_VARIABLE: Move on as many items as possible + * @param old_head + * Returns head value as it was before the move + * @param entries + * Returns the number of ring entries available BEFORE head was moved + * @return + * Actual number of objects the head was moved on + * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only */ -static __rte_always_inline unsigned int -__rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num, +static __rte_always_inline uint32_t +__rte_ring_hts_move_head(struct rte_ring_hts_headtail *d, + const struct rte_ring_headtail *s, uint32_t capacity, unsigned int num, enum rte_ring_queue_behavior behavior, uint32_t *old_head, - uint32_t *free_entries) + uint32_t *entries) { uint32_t n; union __rte_ring_hts_pos np, op; - const uint32_t capacity = r->capacity; - - op.raw = rte_atomic_load_explicit(&r->hts_prod.ht.raw, rte_memory_order_acquire); + op.raw = rte_atomic_load_explicit(&d->ht.raw, rte_memory_order_acquire); do { /* Reset n to the initial burst count */ @@ -74,20 +94,20 @@ __rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num, * make sure that we read prod head/tail *before* * reading cons tail. */ - __rte_ring_hts_head_wait(&r->hts_prod, &op); + __rte_ring_hts_head_wait(d, &op); /* * The subtraction is done between two unsigned 32bits value * (the result is always modulo 32 bits even if we have - * *old_head > cons_tail). So 'free_entries' is always between 0 + * *old_head > cons_tail). So 'entries' is always between 0 * and capacity (which is < size). */ - *free_entries = capacity + r->cons.tail - op.pos.head; + *entries = capacity + s->tail - op.pos.head; /* check that we have enough room in ring */ - if (unlikely(n > *free_entries)) + if (unlikely(n > *entries)) n = (behavior == RTE_RING_QUEUE_FIXED) ? - 0 : *free_entries; + 0 : *entries; if (n == 0) break; @@ -100,13 +120,25 @@ __rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num, * - OOO reads of cons tail value * - OOO copy of elems from the ring */ - } while (rte_atomic_compare_exchange_strong_explicit(&r->hts_prod.ht.raw, + } while (rte_atomic_compare_exchange_strong_explicit(&d->ht.raw, (uint64_t *)(uintptr_t)&op.raw, np.raw, - rte_memory_order_acquire, rte_memory_order_acquire) == 0); + rte_memory_order_acquire, + rte_memory_order_acquire) == 0); *old_head = op.pos.head; return n; } +/** + * @internal This function updates the producer head for enqueue + */ +static __rte_always_inline unsigned int +__rte_ring_hts_move_prod_head(struct rte_ring *r, unsigned int num, + enum rte_ring_queue_behavior behavior, uint32_t *old_head, + uint32_t *free_entries) +{ + return __rte_ring_hts_move_head(&r->hts_prod, &r->cons, + r->capacity, num, behavior, old_head, free_entries); +} /** * @internal This function updates the consumer head for dequeue @@ -116,51 +148,8 @@ __rte_ring_hts_move_cons_head(struct rte_ring *r, unsigned int num, enum rte_ring_queue_behavior behavior, uint32_t *old_head, uint32_t *entries) { - uint32_t n; - union __rte_ring_hts_pos np, op; - - op.raw = rte_atomic_load_explicit(&r->hts_cons.ht.raw, rte_memory_order_acquire); - - /* move cons.head atomically */ - do { - /* Restore n as it may change every loop */ - n = num; - - /* - * wait for tail to be equal to head, - * make sure that we read cons head/tail *before* - * reading prod tail. - */ - __rte_ring_hts_head_wait(&r->hts_cons, &op); - - /* The subtraction is done between two unsigned 32bits value - * (the result is always modulo 32 bits even if we have - * cons_head > prod_tail). So 'entries' is always between 0 - * and size(ring)-1. - */ - *entries = r->prod.tail - op.pos.head; - - /* Set the actual entries for dequeue */ - if (n > *entries) - n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries; - - if (unlikely(n == 0)) - break; - - np.pos.tail = op.pos.tail; - np.pos.head = op.pos.head + n; - - /* - * this CAS(ACQUIRE, ACQUIRE) serves as a hoist barrier to prevent: - * - OOO reads of prod tail value - * - OOO copy of elems from the ring - */ - } while (rte_atomic_compare_exchange_strong_explicit(&r->hts_cons.ht.raw, - (uint64_t *)(uintptr_t)&op.raw, np.raw, - rte_memory_order_acquire, rte_memory_order_acquire) == 0); - - *old_head = op.pos.head; - return n; + return __rte_ring_hts_move_head(&r->hts_cons, &r->prod, + 0, num, behavior, old_head, entries); } /** diff --git a/lib/ring/rte_ring_rts_elem_pvt.h b/lib/ring/rte_ring_rts_elem_pvt.h index 122650346b..96825931f8 100644 --- a/lib/ring/rte_ring_rts_elem_pvt.h +++ b/lib/ring/rte_ring_rts_elem_pvt.h @@ -65,19 +65,40 @@ __rte_ring_rts_head_wait(const struct rte_ring_rts_headtail *ht, } /** - * @internal This function updates the producer head for enqueue. + * @internal This is a helper function that moves the producer/consumer head + * + * @param d + * A pointer to the headtail structure with head value to be moved + * @param s + * A pointer to the counter-part headtail structure. Note that this + * function only reads tail value from it + * @param capacity + * Either ring capacity value (for producer), or zero (for consumer) + * Indicates whether multi-thread safe path is needed or not + * @param num + * The number of elements we want to move head value on + * @param behavior + * RTE_RING_QUEUE_FIXED: Move on a fixed number of items + * RTE_RING_QUEUE_VARIABLE: Move on as many items as possible + * @param old_head + * Returns head value as it was before the move + * @param entries + * Returns the number of ring entries available BEFORE head was moved + * @return + * Actual number of objects the head was moved on + * If behavior == RTE_RING_QUEUE_FIXED, this will be 0 or n only */ static __rte_always_inline uint32_t -__rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num, +__rte_ring_rts_move_head(struct rte_ring_rts_headtail *d, + const struct rte_ring_headtail *s, uint32_t capacity, uint32_t num, enum rte_ring_queue_behavior behavior, uint32_t *old_head, - uint32_t *free_entries) + uint32_t *entries) { uint32_t n; union __rte_ring_rts_poscnt nh, oh; - const uint32_t capacity = r->capacity; - - oh.raw = rte_atomic_load_explicit(&r->rts_prod.head.raw, rte_memory_order_acquire); + oh.raw = rte_atomic_load_explicit(&d->head.raw, + rte_memory_order_acquire); do { /* Reset n to the initial burst count */ @@ -88,20 +109,20 @@ __rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num, * make sure that we read prod head *before* * reading cons tail. */ - __rte_ring_rts_head_wait(&r->rts_prod, &oh); + __rte_ring_rts_head_wait(d, &oh); /* * The subtraction is done between two unsigned 32bits value * (the result is always modulo 32 bits even if we have - * *old_head > cons_tail). So 'free_entries' is always between 0 + * *old_head > cons_tail). So 'entries' is always between 0 * and capacity (which is < size). */ - *free_entries = capacity + r->cons.tail - oh.val.pos; + *entries = capacity + s->tail - oh.val.pos; /* check that we have enough room in ring */ - if (unlikely(n > *free_entries)) + if (unlikely(n > *entries)) n = (behavior == RTE_RING_QUEUE_FIXED) ? - 0 : *free_entries; + 0 : *entries; if (n == 0) break; @@ -114,14 +135,27 @@ __rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num, * - OOO reads of cons tail value * - OOO copy of elems to the ring */ - } while (rte_atomic_compare_exchange_strong_explicit(&r->rts_prod.head.raw, + } while (rte_atomic_compare_exchange_strong_explicit(&d->head.raw, (uint64_t *)(uintptr_t)&oh.raw, nh.raw, - rte_memory_order_acquire, rte_memory_order_acquire) == 0); + rte_memory_order_acquire, + rte_memory_order_acquire) == 0); *old_head = oh.val.pos; return n; } +/** + * @internal This function updates the producer head for enqueue. + */ +static __rte_always_inline uint32_t +__rte_ring_rts_move_prod_head(struct rte_ring *r, uint32_t num, + enum rte_ring_queue_behavior behavior, uint32_t *old_head, + uint32_t *free_entries) +{ + return __rte_ring_rts_move_head(&r->rts_prod, &r->cons, + r->capacity, num, behavior, old_head, free_entries); +} + /** * @internal This function updates the consumer head for dequeue */ @@ -130,51 +164,8 @@ __rte_ring_rts_move_cons_head(struct rte_ring *r, uint32_t num, enum rte_ring_queue_behavior behavior, uint32_t *old_head, uint32_t *entries) { - uint32_t n; - union __rte_ring_rts_poscnt nh, oh; - - oh.raw = rte_atomic_load_explicit(&r->rts_cons.head.raw, rte_memory_order_acquire); - - /* move cons.head atomically */ - do { - /* Restore n as it may change every loop */ - n = num; - - /* - * wait for cons head/tail distance, - * make sure that we read cons head *before* - * reading prod tail. - */ - __rte_ring_rts_head_wait(&r->rts_cons, &oh); - - /* The subtraction is done between two unsigned 32bits value - * (the result is always modulo 32 bits even if we have - * cons_head > prod_tail). So 'entries' is always between 0 - * and size(ring)-1. - */ - *entries = r->prod.tail - oh.val.pos; - - /* Set the actual entries for dequeue */ - if (n > *entries) - n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *entries; - - if (unlikely(n == 0)) - break; - - nh.val.pos = oh.val.pos + n; - nh.val.cnt = oh.val.cnt + 1; - - /* - * this CAS(ACQUIRE, ACQUIRE) serves as a hoist barrier to prevent: - * - OOO reads of prod tail value - * - OOO copy of elems from the ring - */ - } while (rte_atomic_compare_exchange_strong_explicit(&r->rts_cons.head.raw, - (uint64_t *)(uintptr_t)&oh.raw, nh.raw, - rte_memory_order_acquire, rte_memory_order_acquire) == 0); - - *old_head = oh.val.pos; - return n; + return __rte_ring_rts_move_head(&r->rts_cons, &r->prod, + 0, num, behavior, old_head, entries); } /** From patchwork Wed Oct 30 21:23:00 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Konstantin Ananyev X-Patchwork-Id: 147763 X-Patchwork-Delegate: thomas@monjalon.net Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 996B345B82; Wed, 30 Oct 2024 21:33:03 +0100 (CET) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id 994C8433BA; Wed, 30 Oct 2024 21:33:02 +0100 (CET) Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) by mails.dpdk.org (Postfix) with ESMTP id A4D764335B for ; Wed, 30 Oct 2024 21:32:54 +0100 (CET) Received: from mail.maildlp.com (unknown [172.18.186.216]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4XdzLH0PvWz6K5mK; Thu, 31 Oct 2024 04:31:35 +0800 (CST) Received: from frapeml500007.china.huawei.com (unknown [7.182.85.172]) by mail.maildlp.com (Postfix) with ESMTPS id 12F3A140B67; Thu, 31 Oct 2024 04:32:54 +0800 (CST) Received: from localhost.localdomain (10.220.239.45) by frapeml500007.china.huawei.com (7.182.85.172) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.1.2507.39; Wed, 30 Oct 2024 21:32:53 +0100 From: Konstantin Ananyev To: CC: , , , , , , , , Subject: [PATCH v7 3/7] ring: make copying functions generic Date: Wed, 30 Oct 2024 17:23:00 -0400 Message-ID: <20241030212304.104180-4-konstantin.ananyev@huawei.com> X-Mailer: git-send-email 2.35.3 In-Reply-To: <20241030212304.104180-1-konstantin.ananyev@huawei.com> References: <20241021174745.1843-1-konstantin.ananyev@huawei.com> <20241030212304.104180-1-konstantin.ananyev@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.220.239.45] X-ClientProxiedBy: frapeml100006.china.huawei.com (7.182.85.201) To frapeml500007.china.huawei.com (7.182.85.172) X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Note upfront: that change doesn't introduce any functional or performance changes. It is just a code-reordering for: - improve code modularity and re-usability - ability in future to re-use the same code to introduce new functionality There is no real need for enqueue_elems()/dequeue_elems() to get pointer to actual rte_ring structure, instead it is enough to pass a pointer to actual elements buffer inside the ring. In return, we'll get a copying functions that could be used for other queueing abstractions that do have circular ring buffer inside. Signed-off-by: Konstantin Ananyev Acked-by: Morten Brørup --- lib/ring/rte_ring_elem_pvt.h | 115 ++++++++++++++++++++--------------- 1 file changed, 67 insertions(+), 48 deletions(-) diff --git a/lib/ring/rte_ring_elem_pvt.h b/lib/ring/rte_ring_elem_pvt.h index 3a83668a08..6eafae121f 100644 --- a/lib/ring/rte_ring_elem_pvt.h +++ b/lib/ring/rte_ring_elem_pvt.h @@ -17,12 +17,14 @@ #endif static __rte_always_inline void -__rte_ring_enqueue_elems_32(struct rte_ring *r, const uint32_t size, - uint32_t idx, const void *obj_table, uint32_t n) +__rte_ring_enqueue_elems_32(void *ring_table, const void *obj_table, + uint32_t size, uint32_t idx, uint32_t n) { unsigned int i; - uint32_t *ring = (uint32_t *)&r[1]; + + uint32_t *ring = (uint32_t *)ring_table; const uint32_t *obj = (const uint32_t *)obj_table; + if (likely(idx + n <= size)) { for (i = 0; i < (n & ~0x7); i += 8, idx += 8) { ring[idx] = obj[i]; @@ -60,14 +62,14 @@ __rte_ring_enqueue_elems_32(struct rte_ring *r, const uint32_t size, } static __rte_always_inline void -__rte_ring_enqueue_elems_64(struct rte_ring *r, uint32_t prod_head, - const void *obj_table, uint32_t n) +__rte_ring_enqueue_elems_64(void *ring_table, const void *obj_table, + uint32_t size, uint32_t idx, uint32_t n) { unsigned int i; - const uint32_t size = r->size; - uint32_t idx = prod_head & r->mask; - uint64_t *ring = (uint64_t *)&r[1]; + + uint64_t *ring = (uint64_t *)ring_table; const unaligned_uint64_t *obj = (const unaligned_uint64_t *)obj_table; + if (likely(idx + n <= size)) { for (i = 0; i < (n & ~0x3); i += 4, idx += 4) { ring[idx] = obj[i]; @@ -93,14 +95,14 @@ __rte_ring_enqueue_elems_64(struct rte_ring *r, uint32_t prod_head, } static __rte_always_inline void -__rte_ring_enqueue_elems_128(struct rte_ring *r, uint32_t prod_head, - const void *obj_table, uint32_t n) +__rte_ring_enqueue_elems_128(void *ring_table, const void *obj_table, + uint32_t size, uint32_t idx, uint32_t n) { unsigned int i; - const uint32_t size = r->size; - uint32_t idx = prod_head & r->mask; - rte_int128_t *ring = (rte_int128_t *)&r[1]; + + rte_int128_t *ring = (rte_int128_t *)ring_table; const rte_int128_t *obj = (const rte_int128_t *)obj_table; + if (likely(idx + n <= size)) { for (i = 0; i < (n & ~0x1); i += 2, idx += 2) memcpy((void *)(ring + idx), @@ -126,37 +128,47 @@ __rte_ring_enqueue_elems_128(struct rte_ring *r, uint32_t prod_head, * single and multi producer enqueue functions. */ static __rte_always_inline void -__rte_ring_enqueue_elems(struct rte_ring *r, uint32_t prod_head, - const void *obj_table, uint32_t esize, uint32_t num) +__rte_ring_do_enqueue_elems(void *ring_table, const void *obj_table, + uint32_t size, uint32_t idx, uint32_t esize, uint32_t num) { /* 8B and 16B copies implemented individually to retain * the current performance. */ if (esize == 8) - __rte_ring_enqueue_elems_64(r, prod_head, obj_table, num); + __rte_ring_enqueue_elems_64(ring_table, obj_table, size, + idx, num); else if (esize == 16) - __rte_ring_enqueue_elems_128(r, prod_head, obj_table, num); + __rte_ring_enqueue_elems_128(ring_table, obj_table, size, + idx, num); else { - uint32_t idx, scale, nr_idx, nr_num, nr_size; + uint32_t scale, nr_idx, nr_num, nr_size; /* Normalize to uint32_t */ scale = esize / sizeof(uint32_t); nr_num = num * scale; - idx = prod_head & r->mask; nr_idx = idx * scale; - nr_size = r->size * scale; - __rte_ring_enqueue_elems_32(r, nr_size, nr_idx, - obj_table, nr_num); + nr_size = size * scale; + __rte_ring_enqueue_elems_32(ring_table, obj_table, nr_size, + nr_idx, nr_num); } } static __rte_always_inline void -__rte_ring_dequeue_elems_32(struct rte_ring *r, const uint32_t size, - uint32_t idx, void *obj_table, uint32_t n) +__rte_ring_enqueue_elems(struct rte_ring *r, uint32_t prod_head, + const void *obj_table, uint32_t esize, uint32_t num) +{ + __rte_ring_do_enqueue_elems(&r[1], obj_table, r->size, + prod_head & r->mask, esize, num); +} + +static __rte_always_inline void +__rte_ring_dequeue_elems_32(void *obj_table, const void *ring_table, + uint32_t size, uint32_t idx, uint32_t n) { unsigned int i; - uint32_t *ring = (uint32_t *)&r[1]; uint32_t *obj = (uint32_t *)obj_table; + const uint32_t *ring = (const uint32_t *)ring_table; + if (likely(idx + n <= size)) { for (i = 0; i < (n & ~0x7); i += 8, idx += 8) { obj[i] = ring[idx]; @@ -194,14 +206,13 @@ __rte_ring_dequeue_elems_32(struct rte_ring *r, const uint32_t size, } static __rte_always_inline void -__rte_ring_dequeue_elems_64(struct rte_ring *r, uint32_t cons_head, - void *obj_table, uint32_t n) +__rte_ring_dequeue_elems_64(void *obj_table, const void *ring_table, + uint32_t size, uint32_t idx, uint32_t n) { unsigned int i; - const uint32_t size = r->size; - uint32_t idx = cons_head & r->mask; - uint64_t *ring = (uint64_t *)&r[1]; unaligned_uint64_t *obj = (unaligned_uint64_t *)obj_table; + const uint64_t *ring = (const uint64_t *)ring_table; + if (likely(idx + n <= size)) { for (i = 0; i < (n & ~0x3); i += 4, idx += 4) { obj[i] = ring[idx]; @@ -227,27 +238,26 @@ __rte_ring_dequeue_elems_64(struct rte_ring *r, uint32_t cons_head, } static __rte_always_inline void -__rte_ring_dequeue_elems_128(struct rte_ring *r, uint32_t cons_head, - void *obj_table, uint32_t n) +__rte_ring_dequeue_elems_128(void *obj_table, const void *ring_table, + uint32_t size, uint32_t idx, uint32_t n) { unsigned int i; - const uint32_t size = r->size; - uint32_t idx = cons_head & r->mask; - rte_int128_t *ring = (rte_int128_t *)&r[1]; rte_int128_t *obj = (rte_int128_t *)obj_table; + const rte_int128_t *ring = (const rte_int128_t *)ring_table; + if (likely(idx + n <= size)) { for (i = 0; i < (n & ~0x1); i += 2, idx += 2) - memcpy((void *)(obj + i), (void *)(ring + idx), 32); + memcpy((obj + i), (const void *)(ring + idx), 32); switch (n & 0x1) { case 1: - memcpy((void *)(obj + i), (void *)(ring + idx), 16); + memcpy((obj + i), (const void *)(ring + idx), 16); } } else { for (i = 0; idx < size; i++, idx++) - memcpy((void *)(obj + i), (void *)(ring + idx), 16); + memcpy((obj + i), (const void *)(ring + idx), 16); /* Start at the beginning */ for (idx = 0; i < n; i++, idx++) - memcpy((void *)(obj + i), (void *)(ring + idx), 16); + memcpy((obj + i), (const void *)(ring + idx), 16); } } @@ -256,30 +266,39 @@ __rte_ring_dequeue_elems_128(struct rte_ring *r, uint32_t cons_head, * single and multi producer enqueue functions. */ static __rte_always_inline void -__rte_ring_dequeue_elems(struct rte_ring *r, uint32_t cons_head, - void *obj_table, uint32_t esize, uint32_t num) +__rte_ring_do_dequeue_elems(void *obj_table, const void *ring_table, + uint32_t size, uint32_t idx, uint32_t esize, uint32_t num) { /* 8B and 16B copies implemented individually to retain * the current performance. */ if (esize == 8) - __rte_ring_dequeue_elems_64(r, cons_head, obj_table, num); + __rte_ring_dequeue_elems_64(obj_table, ring_table, size, + idx, num); else if (esize == 16) - __rte_ring_dequeue_elems_128(r, cons_head, obj_table, num); + __rte_ring_dequeue_elems_128(obj_table, ring_table, size, + idx, num); else { - uint32_t idx, scale, nr_idx, nr_num, nr_size; + uint32_t scale, nr_idx, nr_num, nr_size; /* Normalize to uint32_t */ scale = esize / sizeof(uint32_t); nr_num = num * scale; - idx = cons_head & r->mask; nr_idx = idx * scale; - nr_size = r->size * scale; - __rte_ring_dequeue_elems_32(r, nr_size, nr_idx, - obj_table, nr_num); + nr_size = size * scale; + __rte_ring_dequeue_elems_32(obj_table, ring_table, nr_size, + nr_idx, nr_num); } } +static __rte_always_inline void +__rte_ring_dequeue_elems(struct rte_ring *r, uint32_t cons_head, + void *obj_table, uint32_t esize, uint32_t num) +{ + __rte_ring_do_dequeue_elems(obj_table, &r[1], r->size, + cons_head & r->mask, esize, num); +} + /* Between load and load. there might be cpu reorder in weak model * (powerpc/arm). * There are 2 choices for the users From patchwork Wed Oct 30 21:23:01 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Konstantin Ananyev X-Patchwork-Id: 147764 X-Patchwork-Delegate: thomas@monjalon.net Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 1A5AF45B82; Wed, 30 Oct 2024 21:33:09 +0100 (CET) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id BC3E843387; Wed, 30 Oct 2024 21:33:03 +0100 (CET) Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) by mails.dpdk.org (Postfix) with ESMTP id 0A08143308 for ; Wed, 30 Oct 2024 21:32:57 +0100 (CET) Received: from mail.maildlp.com (unknown [172.18.186.31]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4XdzGF0YSlz6L74N; Thu, 31 Oct 2024 04:28:05 +0800 (CST) Received: from frapeml500007.china.huawei.com (unknown [7.182.85.172]) by mail.maildlp.com (Postfix) with ESMTPS id A412B1404F4; Thu, 31 Oct 2024 04:32:56 +0800 (CST) Received: from localhost.localdomain (10.220.239.45) by frapeml500007.china.huawei.com (7.182.85.172) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.1.2507.39; Wed, 30 Oct 2024 21:32:56 +0100 From: Konstantin Ananyev To: CC: , , , , , , , , Subject: [PATCH v7 4/7] ring: make dump function more verbose Date: Wed, 30 Oct 2024 17:23:01 -0400 Message-ID: <20241030212304.104180-5-konstantin.ananyev@huawei.com> X-Mailer: git-send-email 2.35.3 In-Reply-To: <20241030212304.104180-1-konstantin.ananyev@huawei.com> References: <20241021174745.1843-1-konstantin.ananyev@huawei.com> <20241030212304.104180-1-konstantin.ananyev@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.220.239.45] X-ClientProxiedBy: frapeml100006.china.huawei.com (7.182.85.201) To frapeml500007.china.huawei.com (7.182.85.172) X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org From: Eimear Morrissey The current rte_ring_dump function uses the generic rte_ring_headtail structure to access head/tail positions. This is incorrect for the RTS case where the head is stored in a different offset in the union of structs. Switching to a separate function for each sync type allows to dump correct head/tail values and extra metadata. Signed-off-by: Eimear Morrissey Acked-by: Morten Brørup --- .mailmap | 1 + app/test/test_ring_stress_impl.h | 1 + lib/ring/rte_ring.c | 87 ++++++++++++++++++++++++++++++-- lib/ring/rte_ring.h | 15 ++++++ lib/ring/version.map | 7 +++ 5 files changed, 107 insertions(+), 4 deletions(-) diff --git a/.mailmap b/.mailmap index 5290420258..7d6fb51ca8 100644 --- a/.mailmap +++ b/.mailmap @@ -388,6 +388,7 @@ Eduard Serra Edward Makarov Edwin Brossette Eelco Chaudron +Eimear Morrissey Elad Nachman Elad Persiko Elena Agostini diff --git a/app/test/test_ring_stress_impl.h b/app/test/test_ring_stress_impl.h index ee5274aeef..f99a7ff675 100644 --- a/app/test/test_ring_stress_impl.h +++ b/app/test/test_ring_stress_impl.h @@ -380,6 +380,7 @@ test_mt1(int (*test)(void *)) } lcore_stat_dump(stdout, UINT32_MAX, &arg[mc].stats); + rte_ring_dump(stdout, r); mt1_fini(r, data); return rc; } diff --git a/lib/ring/rte_ring.c b/lib/ring/rte_ring.c index aebb6d6728..261f2a06db 100644 --- a/lib/ring/rte_ring.c +++ b/lib/ring/rte_ring.c @@ -364,20 +364,99 @@ rte_ring_free(struct rte_ring *r) rte_free(te); } +static const char * +ring_get_sync_type(const enum rte_ring_sync_type st) +{ + switch (st) { + case RTE_RING_SYNC_ST: + return "single thread"; + case RTE_RING_SYNC_MT: + return "multi thread"; + case RTE_RING_SYNC_MT_RTS: + return "multi thread - RTS"; + case RTE_RING_SYNC_MT_HTS: + return "multi thread - HTS"; + default: + return "unknown"; + } +} + +static void +ring_dump_ht_headtail(FILE *f, const char *prefix, + const struct rte_ring_headtail *ht) +{ + fprintf(f, "%ssync_type=%s\n", prefix, + ring_get_sync_type(ht->sync_type)); + fprintf(f, "%shead=%"PRIu32"\n", prefix, ht->head); + fprintf(f, "%stail=%"PRIu32"\n", prefix, ht->tail); +} + +static void +ring_dump_rts_headtail(FILE *f, const char *prefix, + const struct rte_ring_rts_headtail *rts) +{ + fprintf(f, "%ssync_type=%s\n", prefix, + ring_get_sync_type(rts->sync_type)); + fprintf(f, "%shead.pos=%"PRIu32"\n", prefix, rts->head.val.pos); + fprintf(f, "%shead.cnt=%"PRIu32"\n", prefix, rts->head.val.cnt); + fprintf(f, "%stail.pos=%"PRIu32"\n", prefix, rts->tail.val.pos); + fprintf(f, "%stail.cnt=%"PRIu32"\n", prefix, rts->tail.val.cnt); + fprintf(f, "%shtd_max=%"PRIu32"\n", prefix, rts->htd_max); +} + +static void +ring_dump_hts_headtail(FILE *f, const char *prefix, + const struct rte_ring_hts_headtail *hts) +{ + fprintf(f, "%ssync_type=%s\n", prefix, + ring_get_sync_type(hts->sync_type)); + fprintf(f, "%shead=%"PRIu32"\n", prefix, hts->ht.pos.head); + fprintf(f, "%stail=%"PRIu32"\n", prefix, hts->ht.pos.tail); +} + +void +rte_ring_headtail_dump(FILE *f, const char *prefix, + const struct rte_ring_headtail *r) +{ + if (f == NULL || r == NULL) + return; + + prefix = (prefix != NULL) ? prefix : ""; + + switch (r->sync_type) { + case RTE_RING_SYNC_ST: + case RTE_RING_SYNC_MT: + ring_dump_ht_headtail(f, prefix, r); + break; + case RTE_RING_SYNC_MT_RTS: + ring_dump_rts_headtail(f, prefix, + (const struct rte_ring_rts_headtail *)r); + break; + case RTE_RING_SYNC_MT_HTS: + ring_dump_hts_headtail(f, prefix, + (const struct rte_ring_hts_headtail *)r); + break; + default: + RING_LOG(ERR, "Invalid ring sync type detected"); + } +} + /* dump the status of the ring on the console */ void rte_ring_dump(FILE *f, const struct rte_ring *r) { + if (f == NULL || r == NULL) + return; + fprintf(f, "ring <%s>@%p\n", r->name, r); fprintf(f, " flags=%x\n", r->flags); fprintf(f, " size=%"PRIu32"\n", r->size); fprintf(f, " capacity=%"PRIu32"\n", r->capacity); - fprintf(f, " ct=%"PRIu32"\n", r->cons.tail); - fprintf(f, " ch=%"PRIu32"\n", r->cons.head); - fprintf(f, " pt=%"PRIu32"\n", r->prod.tail); - fprintf(f, " ph=%"PRIu32"\n", r->prod.head); fprintf(f, " used=%u\n", rte_ring_count(r)); fprintf(f, " avail=%u\n", rte_ring_free_count(r)); + + rte_ring_headtail_dump(f, " cons.", &(r->cons)); + rte_ring_headtail_dump(f, " prod.", &(r->prod)); } /* dump the status of all rings on the console */ diff --git a/lib/ring/rte_ring.h b/lib/ring/rte_ring.h index 11ca69c73d..33ac5e4423 100644 --- a/lib/ring/rte_ring.h +++ b/lib/ring/rte_ring.h @@ -204,6 +204,21 @@ void rte_ring_free(struct rte_ring *r); */ void rte_ring_dump(FILE *f, const struct rte_ring *r); +/** + * Dump the status of a headtail to a file. + * + * @param f + * A pointer to a file for output + * @param prefix + * A string to prefix each output line with + * @param r + * A pointer to a ring headtail structure. + */ +__rte_experimental +void +rte_ring_headtail_dump(FILE *f, const char *prefix, + const struct rte_ring_headtail *r); + /** * Enqueue several objects on the ring (multi-producers safe). * diff --git a/lib/ring/version.map b/lib/ring/version.map index 8da094a69a..61f7464f5a 100644 --- a/lib/ring/version.map +++ b/lib/ring/version.map @@ -14,3 +14,10 @@ DPDK_25 { local: *; }; + +EXPERIMENTAL { + global: + + # added in 24.11 + rte_ring_headtail_dump; +}; From patchwork Wed Oct 30 21:23:02 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Konstantin Ananyev X-Patchwork-Id: 147765 X-Patchwork-Delegate: thomas@monjalon.net Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id C542745B82; Wed, 30 Oct 2024 21:33:14 +0100 (CET) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id E2C54433C9; Wed, 30 Oct 2024 21:33:04 +0100 (CET) Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) by mails.dpdk.org (Postfix) with ESMTP id C0E344334A for ; Wed, 30 Oct 2024 21:32:59 +0100 (CET) Received: from mail.maildlp.com (unknown [172.18.186.31]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4XdzGH5cf5z6L74N; Thu, 31 Oct 2024 04:28:07 +0800 (CST) Received: from frapeml500007.china.huawei.com (unknown [7.182.85.172]) by mail.maildlp.com (Postfix) with ESMTPS id 594881404F4; Thu, 31 Oct 2024 04:32:59 +0800 (CST) Received: from localhost.localdomain (10.220.239.45) by frapeml500007.china.huawei.com (7.182.85.172) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.1.2507.39; Wed, 30 Oct 2024 21:32:58 +0100 From: Konstantin Ananyev To: CC: , , , , , , , , Subject: [PATCH v7 5/7] ring/soring: introduce Staged Ordered Ring Date: Wed, 30 Oct 2024 17:23:02 -0400 Message-ID: <20241030212304.104180-6-konstantin.ananyev@huawei.com> X-Mailer: git-send-email 2.35.3 In-Reply-To: <20241030212304.104180-1-konstantin.ananyev@huawei.com> References: <20241021174745.1843-1-konstantin.ananyev@huawei.com> <20241030212304.104180-1-konstantin.ananyev@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.220.239.45] X-ClientProxiedBy: frapeml100006.china.huawei.com (7.182.85.201) To frapeml500007.china.huawei.com (7.182.85.172) X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Staged-Ordered-Ring (SORING) provides a SW abstraction for 'ordered' queues with multiple processing 'stages'. It is based on conventional DPDK rte_ring, re-uses many of its concepts, and even substantial part of its code. It can be viewed as an 'extension' of rte_ring functionality. In particular, main SORING properties: - circular ring buffer with fixed size objects - producer, consumer plus multiple processing stages in the middle. - allows to split objects processing into multiple stages. - objects remain in the same ring while moving from one stage to the other, initial order is preserved, no extra copying needed. - preserves the ingress order of objects within the queue across multiple stages, i.e.: at the same stage multiple threads can process objects from the ring in any order, but for the next stage objects will always appear in the original order. - each stage (and producer/consumer) can be served by single and/or multiple threads. - number of stages, size and number of objects in the ring are configurable at ring initialization time. Data-path API provides four main operations: - enqueue/dequeue works in the same manner as for conventional rte_ring, all rte_ring synchronization types are supported. - acquire/release - for each stage there is an acquire (start) and release (finish) operation. after some objects are 'acquired' - given thread can safely assume that it has exclusive possession of these objects till 'release' for them is invoked. Note that right now user has to release exactly the same number of objects that was acquired before. After 'release', objects can be 'acquired' by next stage and/or dequeued by the consumer (in case of last stage). Expected use-case: applications that uses pipeline model (probably with multiple stages) for packet processing, when preserving incoming packet order is important. I.E.: IPsec processing, etc. Signed-off-by: Eimear Morrissey Signed-off-by: Konstantin Ananyev Acked-by: Morten Brørup --- devtools/build-dict.sh | 1 + doc/api/doxy-api-index.md | 1 + doc/guides/prog_guide/img/soring-pic1.svg | 635 ++++++++++++++++++++++ doc/guides/prog_guide/ring_lib.rst | 202 +++++++ doc/guides/rel_notes/release_24_11.rst | 8 + lib/ring/meson.build | 4 +- lib/ring/rte_soring.c | 198 +++++++ lib/ring/rte_soring.h | 556 +++++++++++++++++++ lib/ring/soring.c | 613 +++++++++++++++++++++ lib/ring/soring.h | 138 +++++ lib/ring/version.map | 19 + 11 files changed, 2373 insertions(+), 2 deletions(-) create mode 100644 doc/guides/prog_guide/img/soring-pic1.svg create mode 100644 lib/ring/rte_soring.c create mode 100644 lib/ring/rte_soring.h create mode 100644 lib/ring/soring.c create mode 100644 lib/ring/soring.h diff --git a/devtools/build-dict.sh b/devtools/build-dict.sh index a8cac49029..5eb592abfa 100755 --- a/devtools/build-dict.sh +++ b/devtools/build-dict.sh @@ -17,6 +17,7 @@ sed '/^..->/d' | sed '/^uint->/d' | sed "/^doesn'->/d" | sed '/^wasn->/d' | +sed '/^soring->/d' | # print to stdout cat diff --git a/doc/api/doxy-api-index.md b/doc/api/doxy-api-index.md index 266c8b90dc..ac7f3cd010 100644 --- a/doc/api/doxy-api-index.md +++ b/doc/api/doxy-api-index.md @@ -173,6 +173,7 @@ The public API headers are grouped by topics: [mbuf](@ref rte_mbuf.h), [mbuf pool ops](@ref rte_mbuf_pool_ops.h), [ring](@ref rte_ring.h), + [soring](@ref rte_soring.h), [stack](@ref rte_stack.h), [tailq](@ref rte_tailq.h), [bitset](@ref rte_bitset.h), diff --git a/doc/guides/prog_guide/img/soring-pic1.svg b/doc/guides/prog_guide/img/soring-pic1.svg new file mode 100644 index 0000000000..c97e66ca43 --- /dev/null +++ b/doc/guides/prog_guide/img/soring-pic1.svg @@ -0,0 +1,635 @@ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + image/svg+xml + + + + + + + + + + + + + + + + + obj5 + obj1 + obj2 + obj3 + + cons_head + cons_tail + prod_head + prod_tail + + lstages states + + producer and consumer states + + + stage[1].tail + stage[1].head + stage[0].tail + + + obj4 + stage[0].head + + diff --git a/doc/guides/prog_guide/ring_lib.rst b/doc/guides/prog_guide/ring_lib.rst index f7dbba0e4e..a023c2f697 100644 --- a/doc/guides/prog_guide/ring_lib.rst +++ b/doc/guides/prog_guide/ring_lib.rst @@ -491,6 +491,208 @@ Following is an example of usage: Note that between ``_start_`` and ``_finish_`` no other thread can proceed with enqueue(/dequeue) operation till ``_finish_`` completes. +Staged Ordered Ring API +----------------------- + +Staged-Ordered-Ring (SORING) API provides a SW abstraction for *ordered* queues +with multiple processing *stages*. It is based on conventional DPDK +``rte_ring`` API, re-uses many of its concepts, and even substantial part of +its code. It can be viewed as an 'extension' of ``rte_ring`` functionality. +In particular, main SORING properties: + +* circular ring buffer with fixed size objects and related metadata + +* producer, consumer plus multiple processing stages in between. + +* allows to split objects processing into multiple stages. + +* objects remain in the same ring while moving from one stage to the other, + initial order is preserved, no extra copying needed. + +* preserves the ingress order of objects within the queue across multiple + stages + +* each stage (and producer/consumer) can be served by single and/or + multiple threads. + +* number of stages, size and number of objects and their metadata in the + ring are configurable at ring initialization time. + +Data-Path API +~~~~~~~~~~~~~ + +SORING data-path API provided four main operations: + +* ``enqueue``/``dequeue`` works in the same manner as for conventional + ``rte_ring``, all rte_ring synchronization types are supported. + +* ``acquire``/``release`` - for each stage there is an ``acquire`` (start) + and ``release`` (finish) operation. + After some objects are ``acquired`` - given thread can safely assume that + it has exclusive possession of these objects till ``release`` for them is + invoked. + Note that right now user has to release exactly the same number of + objects that was acquired before. + After objects are ``released``, given thread loses its possession on them, + and they can be either acquired by next stage or dequeued + by the consumer (in case of last stage). + +A simplified representation of a SORING with two stages is shown below. +On that picture ``obj5`` and ``obj4`` elements are acquired by stage 0, +``obj2`` and ``obj3`` are acquired by stage 1, while ``obj11`` was already +released by stage 1 and is ready to be consumed. + +.. _figure_soring1: + +.. figure:: img/soring-pic1.* + +Along with traditional flavor there are enhanced versions for all these +data-path operations: ``enqueux``/``dequeux``/``acquirx``/``releasx``. +All enhanced versions take as extra parameter a pointer to an array of +metadata values. +At initialization user can request within the ``soring`` supplementary and +optional array of metadata associated with each object in the ``soring``. +While ``soring`` element size is configurable and user can specify it big +enough to hold both object and its metadata together, +for performance reasons it might be plausible to access them as separate arrays. +Note that users are free to mix and match both versions of data-path API in +a way they like. +As an example, possible usage scenario when such separation helps: + +.. code-block:: c + + /* + * use pointer to mbuf as soring element, while tx_state + * as a metadata. + * In this example we use a soring with just one stage. + */ + union tx_state { + /* negative values for error */ + int32_t rc; + /* otherwise contain valid TX port and queue IDs*/ + struct { + uint16_t port_id; + uint16_t queue_id; + } tx; + }; + struct rte_soring *soring; + + +producer/consumer part: + +.. code-block:: c + + struct rte_mbuf *pkts[MAX_PKT_BURST]; + union tx_state txst[MAX_PKT_BURST]; + ... + /* enqueue - writes to soring objects array no need to update metadata */ + uint32_t num = MAX_PKT_BURST; + num = rte_soring_enqueue_burst(soring, pkts, num, NULL); + .... + /* dequeux - reads both packets and related tx_state */ + uint32_t num = MAX_PKT_BURST; + num = rte_soring_dequeux_burst(soring, pkts, txst, num, NULL); + + /* + * TX packets out, or drop in case of error. + * Note that we don't need to dereference the soring objects itself + * to make a decision. + */ + uint32_t i, j, k, n; + struct rte_mbuf *dr[MAX_PKT_BURST]; + + k = 0; + for (i = 0; i != num; i++) { + /* packet processing reports an error */ + if (txst[i].rc < 0) + dr[k++] = pkts[i]; + /* valid packet, send it out */ + else { + /* group consequitive packets with the same port and queue IDs */ + for (j = i + 1; j < num; j++) + if (txst[j].rc != txst[i].rc) + break; + + n = rte_eth_tx_burst(txst[i].tx.port_id, txst[i].tx.queue_id, + pkts + i, j - i); + if (i + n != j) { + /* decide with unsent packets if any */ + } + } + } + /* drop errorneous packets */ + if (k != 0) + rte_pktmbuf_free_bulk(dr, k); + +acquire/release part: + +.. code-block:: c + + uint32_t ftoken; + struct rte_mbuf *pkts[MAX_PKT_BURST]; + union tx_state txst[MAX_PKT_BURST]; + ... + /* acquire - grab some packets to process */ + uint32_t num = MAX_PKT_BURST; + num = rte_soring_acquire_burst(soring, pkts, 0, num, &ftoken, NULL); + + /* process packets, fill txst[] for each */ + do_process_packets(pkts, txst, num); + + /* + * release - assuming that do_process_packets() didn't change + * contents of pkts[], we need to update soring metadata array only. + */ + rte_soring_releasx(soring, NULL, txst, 0, num, ftoken); + +Use Cases +~~~~~~~~~~ + +Expected use-cases include applications that use pipeline model +(probably with multiple stages) for packet processing, when preserving +incoming packet order is important. I.E.: IPsec processing, etc. + +SORING internals +~~~~~~~~~~~~~~~~ + +* In addition to accessible by the user array of objects (and metadata), + ``soirng`` also contains an internal array of states. Each ``state[]`` + corresponds to exactly one object within the soring. That ``state[]`` + array is used by ``acquire``/``release``/``dequeue`` operations to + store internal information and should not be accessed by the user directly. + +* At ``acquire``, soring moves stage's head (in a same way as ``rte_ring`` + ``move_head`` does), plus it saves in ``state[stage.old_head]`` + information about how many elements were acquired, acquired head position, + and special flag value to indicate that given elements are acquired + (``SORING_ST_START``). + Note that ``acquire`` returns an opaque ``ftoken`` value that user has + to provide for ``release`` function. + +* ``release`` extracts old head value from provided by user ``ftoken`` and + checks that corresponding ``state[]`` entry contains expected values + (mostly for sanity purposes). Then it marks this ``state[]`` entry with + ``SORING_ST_FINISH`` flag to indicate that given subset of objects was + released. After that, it checks does stage's old ``head`` value equals to + its current ``tail`` value. If so, then it performs ``finalize`` + operation, otherwise ``release`` just returns. + +* As ``state[]`` is shared by all threads, some other thread can perform + ``finalize`` operation for given stage. That allows ``release`` to avoid + excessive waits on the ``tail`` value. + Main purpose of ``finalize`` operation is to walk through ``state[]`` + array from current stage's ``tail`` position up to its ``head``, + check ``state[]`` and move stage ``tail`` through elements that already + are released (in ``SORING_ST_FINISH`` state). + Along with that, corresponding ``state[]`` entries are reset back to zero. + Note that ``finalize`` for given stage can be called from multiple places: + from ``release`` for that stage or from ``acquire`` for next stage, or + even from consumer's ``dequeue`` - in case given stage is the last one. + So ``finalize`` has to be MT-safe and inside it we have to guarantee that + at any given moment only one thread can update stage's ``tail`` and reset + corresponding ``state[]`` entries. + + References ---------- diff --git a/doc/guides/rel_notes/release_24_11.rst b/doc/guides/rel_notes/release_24_11.rst index fa4822d928..b8684d9f78 100644 --- a/doc/guides/rel_notes/release_24_11.rst +++ b/doc/guides/rel_notes/release_24_11.rst @@ -247,6 +247,14 @@ New Features Added ability for node to advertise and update multiple xstat counters, that can be retrieved using ``rte_graph_cluster_stats_get``. +* **Add Staged-Ordered-Ring (SORING) API to the rte_ring library.** + + New API to the ring library to provide a SW abstraction for + 'ordered' queues with multiple processing 'stages'. + It is based on conventional DPDK rte_ring, re-uses many of its concepts, + and even substantial part of its code. + It can be viewed as an 'extension' of rte_ring functionality. + Removed Items ------------- diff --git a/lib/ring/meson.build b/lib/ring/meson.build index 7fca958ed7..21f2c12989 100644 --- a/lib/ring/meson.build +++ b/lib/ring/meson.build @@ -1,8 +1,8 @@ # SPDX-License-Identifier: BSD-3-Clause # Copyright(c) 2017 Intel Corporation -sources = files('rte_ring.c') -headers = files('rte_ring.h') +sources = files('rte_ring.c', 'rte_soring.c', 'soring.c') +headers = files('rte_ring.h', 'rte_soring.h') # most sub-headers are not for direct inclusion indirect_headers += files ( 'rte_ring_core.h', diff --git a/lib/ring/rte_soring.c b/lib/ring/rte_soring.c new file mode 100644 index 0000000000..e297a2181d --- /dev/null +++ b/lib/ring/rte_soring.c @@ -0,0 +1,198 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2024 Huawei Technologies Co., Ltd + */ + +#include + +#include "soring.h" +#include + +RTE_LOG_REGISTER_DEFAULT(soring_logtype, INFO); + +static uint32_t +soring_calc_elem_num(uint32_t count) +{ + return rte_align32pow2(count + 1); +} + +static int +soring_check_param(uint32_t esize, uint32_t stsize, uint32_t count, + uint32_t stages) +{ + if (stages == 0) { + SORING_LOG(ERR, "invalid number of stages: %u", stages); + return -EINVAL; + } + + /* Check if element size is a multiple of 4B */ + if (esize == 0 || esize % 4 != 0) { + SORING_LOG(ERR, "invalid element size: %u", esize); + return -EINVAL; + } + + /* Check if ret-code size is a multiple of 4B */ + if (stsize % 4 != 0) { + SORING_LOG(ERR, "invalid retcode size: %u", stsize); + return -EINVAL; + } + + /* count must be a power of 2 */ + if (rte_is_power_of_2(count) == 0 || + (count > RTE_SORING_ELEM_MAX + 1)) { + SORING_LOG(ERR, "invalid number of elements: %u", count); + return -EINVAL; + } + + return 0; +} + +/* + * Calculate size offsets for SORING internal data layout. + */ +static size_t +soring_get_szofs(uint32_t esize, uint32_t stsize, uint32_t count, + uint32_t stages, size_t *elst_ofs, size_t *state_ofs, + size_t *stage_ofs) +{ + size_t sz; + const struct rte_soring * const r = NULL; + + sz = sizeof(r[0]) + (size_t)count * esize; + sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE); + + if (elst_ofs != NULL) + *elst_ofs = sz; + + sz = sz + (size_t)count * stsize; + sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE); + + if (state_ofs != NULL) + *state_ofs = sz; + + sz += sizeof(r->state[0]) * count; + sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE); + + if (stage_ofs != NULL) + *stage_ofs = sz; + + sz += sizeof(r->stage[0]) * stages; + sz = RTE_ALIGN(sz, RTE_CACHE_LINE_SIZE); + + return sz; +} + +static void +soring_dump_stage_headtail(FILE *f, const char *prefix, + struct soring_stage *st) +{ + fprintf(f, "%stail.pos=%"PRIu32"\n", prefix, st->sht.tail.pos); + fprintf(f, "%stail.sync=%"PRIu32"\n", prefix, st->sht.tail.sync); + fprintf(f, "%shead=%"PRIu32"\n", prefix, st->sht.head); +} + +void +rte_soring_dump(FILE *f, const struct rte_soring *r) +{ + uint32_t i; + char buf[32]; + + if (f == NULL || r == NULL) + return; + + fprintf(f, "soring <%s>@%p\n", r->name, r); + fprintf(f, " size=%"PRIu32"\n", r->size); + fprintf(f, " capacity=%"PRIu32"\n", r->capacity); + fprintf(f, " esize=%"PRIu32"\n", r->esize); + fprintf(f, " msize=%"PRIu32"\n", r->msize); + fprintf(f, " used=%u\n", rte_soring_count(r)); + fprintf(f, " avail=%u\n", rte_soring_free_count(r)); + + rte_ring_headtail_dump(f, " cons.", &(r->cons.ht)); + rte_ring_headtail_dump(f, " prod.", &(r->prod.ht)); + + fprintf(f, " nb_stage=%"PRIu32"\n", r->nb_stage); + for (i = 0; i < r->nb_stage; i++) { + snprintf(buf, sizeof(buf), " stage[%u].", i); + soring_dump_stage_headtail(f, buf, r->stage + i); + } +} + +ssize_t +rte_soring_get_memsize(const struct rte_soring_param *prm) +{ + int32_t rc; + uint32_t count; + + count = soring_calc_elem_num(prm->elems); + rc = soring_check_param(prm->elem_size, prm->meta_size, count, + prm->stages); + if (rc != 0) + return rc; + + return soring_get_szofs(prm->elem_size, prm->meta_size, count, + prm->stages, NULL, NULL, NULL); +} + +/* compilation-time checks */ +static void +soring_compilation_checks(void) +{ + RTE_BUILD_BUG_ON((sizeof(struct rte_soring) & + RTE_CACHE_LINE_MASK) != 0); + RTE_BUILD_BUG_ON((offsetof(struct rte_soring, cons) & + RTE_CACHE_LINE_MASK) != 0); + RTE_BUILD_BUG_ON((offsetof(struct rte_soring, prod) & + RTE_CACHE_LINE_MASK) != 0); + + RTE_BUILD_BUG_ON(offsetof(struct rte_ring_headtail, tail) != + offsetof(struct soring_stage_headtail, tail.pos)); + RTE_BUILD_BUG_ON(offsetof(struct rte_ring_headtail, sync_type) != + offsetof(struct soring_stage_headtail, unused)); +} + +int +rte_soring_init(struct rte_soring *r, const struct rte_soring_param *prm) +{ + int32_t rc; + uint32_t n; + size_t meta_ofs, stage_ofs, state_ofs; + + soring_compilation_checks(); + + if (r == NULL || prm == NULL) + return -EINVAL; + + n = soring_calc_elem_num(prm->elems); + rc = soring_check_param(prm->elem_size, prm->meta_size, n, prm->stages); + if (rc != 0) + return rc; + + soring_get_szofs(prm->elem_size, prm->meta_size, n, prm->stages, + &meta_ofs, &state_ofs, &stage_ofs); + + memset(r, 0, sizeof(*r)); + rc = strlcpy(r->name, prm->name, sizeof(r->name)); + if (rc < 0 || rc >= (int)sizeof(r->name)) + return -ENAMETOOLONG; + + r->size = n; + r->mask = r->size - 1; + r->capacity = prm->elems; + r->esize = prm->elem_size; + r->msize = prm->meta_size; + + r->prod.ht.sync_type = prm->prod_synt; + r->cons.ht.sync_type = prm->cons_synt; + + r->state = (union soring_state *)((uintptr_t)r + state_ofs); + memset(r->state, 0, sizeof(r->state[0]) * r->size); + + r->stage = (struct soring_stage *)((uintptr_t)r + stage_ofs); + r->nb_stage = prm->stages; + memset(r->stage, 0, r->nb_stage * sizeof(r->stage[0])); + + if (r->msize != 0) + r->meta = (void *)((uintptr_t)r + meta_ofs); + + return 0; +} diff --git a/lib/ring/rte_soring.h b/lib/ring/rte_soring.h new file mode 100644 index 0000000000..6d0976b1ac --- /dev/null +++ b/lib/ring/rte_soring.h @@ -0,0 +1,556 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2024 Huawei Technologies Co., Ltd + */ + +#ifndef _RTE_SORING_H_ +#define _RTE_SORING_H_ + +/** + * @file + * This file contains definition of RTE soring (Staged Ordered Ring) public API. + * Brief description: + * enqueue/dequeue works the same as for conventional rte_ring: + * any rte_ring sync types can be used, etc. + * Plus there could be multiple 'stages'. + * For each stage there is an acquire (start) and release (finish) operation. + * after some elems are 'acquired' - user can safely assume that he has + * exclusive possession of these elems till 'release' for them is done. + * Note that right now user has to release exactly the same number of elems + * he acquired before. + * After 'release', elems can be 'acquired' by next stage and/or dequeued + * (in case of last stage). + * Extra debugging might be enabled with RTE_SORING_DEBUG macro. + */ + +#include + +#ifdef __cplusplus +extern "C" { +#endif + +/** upper 2 bits are used for status */ +#define RTE_SORING_ST_BIT 30 + +/** max possible number of elements in the soring */ +#define RTE_SORING_ELEM_MAX (RTE_BIT32(RTE_SORING_ST_BIT) - 1) + +struct rte_soring_param { + /** expected name of the soring */ + const char *name; + /** number of elemnts in the soring */ + uint32_t elems; + /** size of elements in the soring, must be a multiple of 4 */ + uint32_t elem_size; + /** + * size of metadata for each elem, must be a multiple of 4. + * This parameter defines a size of supplementary and optional + * array of metadata associated with each object in the soring. + * While element size is configurable (see 'elem_size' parameter above), + * so user can specify it big enough to hold both object and its + * metadata together, for performance reasons it might be plausible + * to access them as separate arrays. + * Common usage scenario when such separation helps: + * enqueue() - writes to objects array + * acquire() - reads from objects array + * release() - writes to metadata array (as an example: return code) + * dequeue() - reads both objects and metadata array + */ + uint32_t meta_size; + /** number of stages in the soring */ + uint32_t stages; + /** sync type for producer */ + enum rte_ring_sync_type prod_synt; + /** sync type for consumer */ + enum rte_ring_sync_type cons_synt; +}; + +struct rte_soring; + +/** + * Calculate the memory size needed for a soring + * + * This function returns the number of bytes needed for a soring, given + * the expected parameters for it. This value is the sum of the size of + * the internal metadata and the size of the memory needed by the + * actual soring elements and their metadata. The value is aligned to a cache + * line size. + * + * @param prm + * Pointer to the structure that contains soring creation paramers. + * @return + * - The memory size needed for the soring on success. + * - -EINVAL if provided paramer values are invalid. + */ +__rte_experimental +ssize_t +rte_soring_get_memsize(const struct rte_soring_param *prm); + +/** + * Initialize a soring structure. + * + * Initialize a soring structure in memory pointed by "r". + * The size of the memory area must be large enough to store the soring + * internal structures plus the objects and ret-code tables. + * It is strongly advised to use @ref rte_soring_get_memsize() to get the + * appropriate size. + * + * @param r + * Pointer to the soring structure. + * @param prm + * Pointer to the structure that contains soring creation paramers. + * @return + * - 0 on success, or a negative error code. + */ +__rte_experimental +int +rte_soring_init(struct rte_soring *r, const struct rte_soring_param *prm); + +/** + * Return the total number of filled entries in a soring. + * + * @param r + * A pointer to the soring structure. + * @return + * The number of entries in the soring. + */ +__rte_experimental +unsigned int +rte_soring_count(const struct rte_soring *r); + +/** + * Return the total number of unfilled entries in a soring. + * + * @param r + * A pointer to the soring structure. + * @return + * The number of free entries in the soring. + */ +__rte_experimental +unsigned int +rte_soring_free_count(const struct rte_soring *r); + +/** + * Dump the status of the soring + * + * @param f + * A pointer to a file for output + * @param r + * Pointer to the soring structure. + */ +__rte_experimental +void +rte_soring_dump(FILE *f, const struct rte_soring *r); + +/** + * Enqueue several objects on the soring. + * Enqueues exactly requested number of objects or none. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to enqueue. + * Size of objects to enqueue must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param n + * The number of objects to add in the soring from the 'objs'. + * @param free_space + * if non-NULL, returns the amount of space in the soring after the + * enqueue operation has finished. + * @return + * - Actual number of objects enqueued, either 0 or n. + */ +__rte_experimental +uint32_t +rte_soring_enqueue_bulk(struct rte_soring *r, const void *objs, + uint32_t n, uint32_t *free_space); + +/** + * Enqueue several objects plus metadata on the soring. + * Enqueues exactly requested number of objects or none. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to enqueue. + * Size of objects to enqueue must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param meta + * A pointer to an array of metadata values for each object to enqueue. + * Note that if user not using object metadata values, then this parameter + * can be NULL. + * Size of elements in this array must be the same value as 'meta_size' + * parameter used while creating the soring. If user created the soring with + * 'meta_size' value equals zero, then 'meta' parameter should be NULL. + * Otherwise the results are undefined. + * @param n + * The number of objects to add in the soring from the 'objs'. + * @param free_space + * if non-NULL, returns the amount of space in the soring after the + * enqueue operation has finished. + * @return + * - Actual number of objects enqueued, either 0 or n. + */ +__rte_experimental +uint32_t +rte_soring_enqueux_bulk(struct rte_soring *r, const void *objs, + const void *meta, uint32_t n, uint32_t *free_space); + +/** + * Enqueue several objects on the soring. + * Enqueues up to requested number of objects. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to enqueue. + * Size of objects to enqueue must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param n + * The number of objects to add in the soring from the 'objs'. + * @param free_space + * if non-NULL, returns the amount of space in the soring after the + * enqueue operation has finished. + * @return + * - Actual number of objects enqueued. + */ +__rte_experimental +uint32_t +rte_soring_enqueue_burst(struct rte_soring *r, const void *objs, + uint32_t n, uint32_t *free_space); + +/** + * Enqueue several objects plus metadata on the soring. + * Enqueues up to requested number of objects. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to enqueue. + * Size of objects to enqueue must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param meta + * A pointer to an array of metadata values for each object to enqueue. + * Note that if user not using object metadata values, then this parameter + * can be NULL. + * Size of elements in this array must be the same value as 'meta_size' + * parameter used while creating the soring. If user created the soring with + * 'meta_size' value equals zero, then 'meta' parameter should be NULL. + * Otherwise the results are undefined. + * @param n + * The number of objects to add in the soring from the 'objs'. + * @param free_space + * if non-NULL, returns the amount of space in the soring after the + * enqueue operation has finished. + * @return + * - Actual number of objects enqueued. + */ +__rte_experimental +uint32_t +rte_soring_enqueux_burst(struct rte_soring *r, const void *objs, + const void *meta, uint32_t n, uint32_t *free_space); + +/** + * Dequeue several objects from the soring. + * Dequeues exactly requested number of objects or none. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to dequeue. + * Size of objects to enqueue must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param num + * The number of objects to dequeue from the soring into the objs. + * @param available + * If non-NULL, returns the number of remaining soring entries after the + * dequeue has finished. + * @return + * - Actual number of objects dequeued, either 0 or 'num'. + */ +__rte_experimental +uint32_t +rte_soring_dequeue_bulk(struct rte_soring *r, void *objs, + uint32_t num, uint32_t *available); + +/** + * Dequeue several objects plus metadata from the soring. + * Dequeues exactly requested number of objects or none. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to dequeue. + * Size of objects to enqueue must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param meta + * A pointer to array of metadata values for each object to dequeue. + * Note that if user not using object metadata values, then this parameter + * can be NULL. + * Size of elements in this array must be the same value as 'meta_size' + * parameter used while creating the soring. If user created the soring with + * 'meta_size' value equals zero, then 'meta' parameter should be NULL. + * Otherwise the results are undefined. + * @param num + * The number of objects to dequeue from the soring into the objs. + * @param available + * If non-NULL, returns the number of remaining soring entries after the + * dequeue has finished. + * @return + * - Actual number of objects dequeued, either 0 or 'num'. + */ +__rte_experimental +uint32_t +rte_soring_dequeux_bulk(struct rte_soring *r, void *objs, void *meta, + uint32_t num, uint32_t *available); + +/** + * Dequeue several objects from the soring. + * Dequeues up to requested number of objects. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to dequeue. + * Size of objects to enqueue must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param num + * The number of objects to dequeue from the soring into the objs. + * @param available + * If non-NULL, returns the number of remaining soring entries after the + * dequeue has finished. + * @return + * - Actual number of objects dequeued. + */ +__rte_experimental +uint32_t +rte_soring_dequeue_burst(struct rte_soring *r, void *objs, + uint32_t num, uint32_t *available); + +/** + * Dequeue several objects plus metadata from the soring. + * Dequeues up to requested number of objects. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to dequeue. + * Size of objects to enqueue must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param meta + * A pointer to array of metadata values for each object to dequeue. + * Note that if user not using object metadata values, then this parameter + * can be NULL. + * Size of elements in this array must be the same value as 'meta_size' + * parameter used while creating the soring. If user created the soring with + * 'meta_size' value equals zero, then 'meta' parameter should be NULL. + * Otherwise the results are undefined. + * @param num + * The number of objects to dequeue from the soring into the objs. + * @param available + * If non-NULL, returns the number of remaining soring entries after the + * dequeue has finished. + * @return + * - Actual number of objects dequeued. + */ +__rte_experimental +uint32_t +rte_soring_dequeux_burst(struct rte_soring *r, void *objs, void *meta, + uint32_t num, uint32_t *available); + +/** + * Acquire several objects from the soring for given stage. + * Acquires exactly requested number of objects or none. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to acquire. + * Size of objects must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param stage + * Stage to acquire objects for. + * @param num + * The number of objects to acquire. + * @param ftoken + * Pointer to the opaque 'token' value used by release() op. + * User has to store this value somewhere, and later provide to the + * release(). + * @param available + * If non-NULL, returns the number of remaining soring entries for given stage + * after the acquire has finished. + * @return + * - Actual number of objects acquired, either 0 or 'num'. + */ +__rte_experimental +uint32_t +rte_soring_acquire_bulk(struct rte_soring *r, void *objs, + uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available); + +/** + * Acquire several objects plus metadata from the soring for given stage. + * Acquires exactly requested number of objects or none. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to acquire. + * Size of objects must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param meta + * A pointer to an array of metadata values for each for each acquired object. + * Note that if user not using object metadata values, then this parameter + * can be NULL. + * Size of elements in this array must be the same value as 'meta_size' + * parameter used while creating the soring. If user created the soring with + * 'meta_size' value equals zero, then 'meta' parameter should be NULL. + * Otherwise the results are undefined. + * @param stage + * Stage to acquire objects for. + * @param num + * The number of objects to acquire. + * @param ftoken + * Pointer to the opaque 'token' value used by release() op. + * User has to store this value somewhere, and later provide to the + * release(). + * @param available + * If non-NULL, returns the number of remaining soring entries for given stage + * after the acquire has finished. + * @return + * - Actual number of objects acquired, either 0 or 'num'. + */ +__rte_experimental +uint32_t +rte_soring_acquirx_bulk(struct rte_soring *r, void *objs, void *meta, + uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available); + +/** + * Acquire several objects from the soring for given stage. + * Acquires up to requested number of objects. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to acquire. + * Size of objects must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param stage + * Stage to acquire objects for. + * @param num + * The number of objects to acquire. + * @param ftoken + * Pointer to the opaque 'token' value used by release() op. + * User has to store this value somewhere, and later provide to the + * release(). + * @param available + * If non-NULL, returns the number of remaining soring entries for given stage + * after the acquire has finished. + * @return + * - Actual number of objects acquired. + */ +__rte_experimental +uint32_t +rte_soring_acquire_burst(struct rte_soring *r, void *objs, + uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available); + +/** + * Acquire several objects plus metadata from the soring for given stage. + * Acquires up to requested number of objects. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to acquire. + * Size of objects must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param meta + * A pointer to an array of metadata values for each for each acquired object. + * Note that if user not using object metadata values, then this parameter + * can be NULL. + * Size of elements in this array must be the same value as 'meta_size' + * parameter used while creating the soring. If user created the soring with + * 'meta_size' value equals zero, then 'meta' parameter should be NULL. + * Otherwise the results are undefined. + * @param stage + * Stage to acquire objects for. + * @param num + * The number of objects to acquire. + * @param ftoken + * Pointer to the opaque 'token' value used by release() op. + * User has to store this value somewhere, and later provide to the + * release(). + * @param available + * If non-NULL, returns the number of remaining soring entries for given stage + * after the acquire has finished. + * @return + * - Actual number of objects acquired. + */ +__rte_experimental +uint32_t +rte_soring_acquirx_burst(struct rte_soring *r, void *objs, void *meta, + uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available); + +/** + * Release several objects for given stage back to the soring. + * Note that it means these objects become avaialble for next stage or + * dequeue. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to relase. + * Note that unless user needs to overwrite soring objects this parameter + * can be NULL. + * Size of objects must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param stage + * Current stage. + * @param n + * The number of objects to release. + * Has to be the same value as returned by acquire() op. + * @param ftoken + * Opaque 'token' value obtained from acquire() op. + */ +__rte_experimental +void +rte_soring_release(struct rte_soring *r, const void *objs, + uint32_t stage, uint32_t n, uint32_t ftoken); + +/** + * Release several objects plus metadata for given stage back to the soring. + * Note that it means these objects become avaialble for next stage or + * dequeue. + * + * @param r + * A pointer to the soring structure. + * @param objs + * A pointer to an array of objects to relase. + * Note that unless user needs to overwrite soring objects this parameter + * can be NULL. + * Size of objects must be the same value as 'elem_size' parameter + * used while creating the soring. Otherwise the results are undefined. + * @param meta + * A pointer to an array of metadata values for each object to release. + * Note that if user not using object metadata values, then this parameter + * can be NULL. + * Size of elements in this array must be the same value as 'meta_size' + * parameter used while creating the soring. If user created the soring with + * 'meta_size' value equals zero, then meta parameter should be NULL. + * Otherwise the results are undefined. + * @param stage + * Current stage. + * @param n + * The number of objects to release. + * Has to be the same value as returned by acquire() op. + * @param ftoken + * Opaque 'token' value obtained from acquire() op. + */ +__rte_experimental +void +rte_soring_releasx(struct rte_soring *r, const void *objs, + const void *meta, uint32_t stage, uint32_t n, uint32_t ftoken); + +#ifdef __cplusplus +} +#endif + +#endif /* _RTE_SORING_H_ */ diff --git a/lib/ring/soring.c b/lib/ring/soring.c new file mode 100644 index 0000000000..e8fe890597 --- /dev/null +++ b/lib/ring/soring.c @@ -0,0 +1,613 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2024 Huawei Technologies Co., Ltd + */ + +/** + * @file + * This file contains implementation of SORING 'datapath' functions. + * Brief description: + * ================== + * enqueue/dequeue works the same as for conventional rte_ring: + * any rte_ring sync types can be used, etc. + * Plus there could be multiple 'stages'. + * For each stage there is an acquire (start) and release (finish) operation. + * After some elems are 'acquired' - user can safely assume that he has + * exclusive possession of these elems till 'release' for them is done. + * Note that right now user has to release exactly the same number of elems + * he acquired before. + * After 'release', elems can be 'acquired' by next stage and/or dequeued + * (in case of last stage). + * Internal structure: + * =================== + * In addition to 'normal' ring of elems, it also has a ring of states of the + * same size. Each state[] corresponds to exactly one elem[]. + * state[] will be used by acquire/release/dequeue functions to store internal + * information and should not be accessed by the user directly. + * How it works: + * ============= + * 'acquire()' just moves stage's head (same as rte_ring move_head does), + * plus it saves in state[stage.cur_head] information about how many elems + * were acquired, current head position and special flag value to indicate + * that elems are acquired (SORING_ST_START). + * Note that 'acquire()' returns to the user a special 'ftoken' that user has + * to provide for 'release()' (in fact it is just a position for current head + * plus current stage index). + * 'release()' extracts old head value from provided ftoken and checks that + * corresponding 'state[]' contains expected values(mostly for sanity + * purposes). + * * Then it marks this state[] with 'SORING_ST_FINISH' flag to indicate + * that given subset of objects was released. + * After that, it checks does old head value equals to current tail value? + * If yes, then it performs 'finalize()' operation, otherwise 'release()' + * just returns (without spinning on stage tail value). + * As updated state[] is shared by all threads, some other thread can do + * 'finalize()' for given stage. + * That allows 'release()' to avoid excessive waits on the tail value. + * Main purpose of 'finalize()' operation is to walk through 'state[]' + * from current stage tail up to its head, check state[] and move stage tail + * through elements that already are in SORING_ST_FINISH state. + * Along with that, corresponding state[] values are reset to zero. + * Note that 'finalize()' for given stage can be done from multiple places: + * 'release()' for that stage or from 'acquire()' for next stage + * even from consumer's 'dequeue()' - in case given stage is the last one. + * So 'finalize()' has to be MT-safe and inside it we have to + * guarantee that only one thread will update state[] and stage's tail values. + */ + +#include "soring.h" + +/* + * Inline functions (fastpath) start here. + */ +static __rte_always_inline uint32_t +__rte_soring_stage_finalize(struct soring_stage_headtail *sht, uint32_t stage, + union soring_state *rstate, uint32_t rmask, uint32_t maxn) +{ + int32_t rc; + uint32_t ftkn, head, i, idx, k, n, tail; + union soring_stage_tail nt, ot; + union soring_state st; + + /* try to grab exclusive right to update tail value */ + ot.raw = rte_atomic_load_explicit(&sht->tail.raw, + rte_memory_order_acquire); + + /* other thread already finalizing it for us */ + if (ot.sync != 0) + return 0; + + nt.pos = ot.pos; + nt.sync = 1; + rc = rte_atomic_compare_exchange_strong_explicit(&sht->tail.raw, + (uint64_t *)(uintptr_t)&ot.raw, nt.raw, + rte_memory_order_release, rte_memory_order_relaxed); + + /* other thread won the race */ + if (rc == 0) + return 0; + + /* Ensure the head is read before rstate[] */ + head = rte_atomic_load_explicit(&sht->head, rte_memory_order_relaxed); + rte_atomic_thread_fence(rte_memory_order_acquire); + + /* + * start with current tail and walk through states that are + * already finished. + */ + + n = RTE_MIN(head - ot.pos, maxn); + for (i = 0, tail = ot.pos; i < n; i += k, tail += k) { + + idx = tail & rmask; + ftkn = SORING_FTKN_MAKE(tail, stage); + + st.raw = rte_atomic_load_explicit(&rstate[idx].raw, + rte_memory_order_relaxed); + if ((st.stnum & SORING_ST_MASK) != SORING_ST_FINISH || + st.ftoken != ftkn) + break; + + k = st.stnum & ~SORING_ST_MASK; + rte_atomic_store_explicit(&rstate[idx].raw, 0, + rte_memory_order_relaxed); + } + + + /* release exclusive right to update along with new tail value */ + ot.pos = tail; + rte_atomic_store_explicit(&sht->tail.raw, ot.raw, + rte_memory_order_release); + + return i; +} + +static __rte_always_inline uint32_t +__rte_soring_move_prod_head(struct rte_soring *r, uint32_t num, + enum rte_ring_queue_behavior behavior, enum rte_ring_sync_type st, + uint32_t *head, uint32_t *next, uint32_t *free) +{ + uint32_t n; + + switch (st) { + case RTE_RING_SYNC_ST: + case RTE_RING_SYNC_MT: + n = __rte_ring_headtail_move_head(&r->prod.ht, &r->cons.ht, + r->capacity, st, num, behavior, head, next, free); + break; + case RTE_RING_SYNC_MT_RTS: + n = __rte_ring_rts_move_head(&r->prod.rts, &r->cons.ht, + r->capacity, num, behavior, head, free); + *next = *head + n; + break; + case RTE_RING_SYNC_MT_HTS: + n = __rte_ring_hts_move_head(&r->prod.hts, &r->cons.ht, + r->capacity, num, behavior, head, free); + *next = *head + n; + break; + default: + /* unsupported mode, shouldn't be here */ + RTE_ASSERT(0); + *free = 0; + n = 0; + } + + return n; +} + +static __rte_always_inline uint32_t +__rte_soring_move_cons_head(struct rte_soring *r, uint32_t stage, uint32_t num, + enum rte_ring_queue_behavior behavior, enum rte_ring_sync_type st, + uint32_t *head, uint32_t *next, uint32_t *avail) +{ + uint32_t n; + + switch (st) { + case RTE_RING_SYNC_ST: + case RTE_RING_SYNC_MT: + n = __rte_ring_headtail_move_head(&r->cons.ht, + &r->stage[stage].ht, 0, st, num, behavior, + head, next, avail); + break; + case RTE_RING_SYNC_MT_RTS: + n = __rte_ring_rts_move_head(&r->cons.rts, &r->stage[stage].ht, + 0, num, behavior, head, avail); + *next = *head + n; + break; + case RTE_RING_SYNC_MT_HTS: + n = __rte_ring_hts_move_head(&r->cons.hts, &r->stage[stage].ht, + 0, num, behavior, head, avail); + *next = *head + n; + break; + default: + /* unsupported mode, shouldn't be here */ + RTE_ASSERT(0); + *avail = 0; + n = 0; + } + + return n; +} + +static __rte_always_inline void +__rte_soring_update_tail(struct __rte_ring_headtail *rht, + enum rte_ring_sync_type st, uint32_t head, uint32_t next, uint32_t enq) +{ + uint32_t n; + + switch (st) { + case RTE_RING_SYNC_ST: + case RTE_RING_SYNC_MT: + __rte_ring_update_tail(&rht->ht, head, next, st, enq); + break; + case RTE_RING_SYNC_MT_RTS: + __rte_ring_rts_update_tail(&rht->rts); + break; + case RTE_RING_SYNC_MT_HTS: + n = next - head; + __rte_ring_hts_update_tail(&rht->hts, head, n, enq); + break; + default: + /* unsupported mode, shouldn't be here */ + RTE_ASSERT(0); + } +} + +static __rte_always_inline uint32_t +__rte_soring_stage_move_head(struct soring_stage_headtail *d, + const struct rte_ring_headtail *s, uint32_t capacity, uint32_t num, + enum rte_ring_queue_behavior behavior, + uint32_t *old_head, uint32_t *new_head, uint32_t *avail) +{ + uint32_t n, tail; + + *old_head = rte_atomic_load_explicit(&d->head, + rte_memory_order_relaxed); + + do { + n = num; + + /* Ensure the head is read before tail */ + rte_atomic_thread_fence(rte_memory_order_acquire); + + tail = rte_atomic_load_explicit(&s->tail, + rte_memory_order_acquire); + *avail = capacity + tail - *old_head; + if (n > *avail) + n = (behavior == RTE_RING_QUEUE_FIXED) ? 0 : *avail; + if (n == 0) + return 0; + *new_head = *old_head + n; + } while (rte_atomic_compare_exchange_strong_explicit(&d->head, + old_head, *new_head, rte_memory_order_acq_rel, + rte_memory_order_relaxed) == 0); + + return n; +} + +static __rte_always_inline uint32_t +soring_enqueue(struct rte_soring *r, const void *objs, + const void *meta, uint32_t n, enum rte_ring_queue_behavior behavior, + uint32_t *free_space) +{ + enum rte_ring_sync_type st; + uint32_t nb_free, prod_head, prod_next; + + RTE_ASSERT(r != NULL && r->nb_stage > 0); + RTE_ASSERT(meta == NULL || r->meta != NULL); + + st = r->prod.ht.sync_type; + + n = __rte_soring_move_prod_head(r, n, behavior, st, + &prod_head, &prod_next, &nb_free); + if (n != 0) { + __rte_ring_do_enqueue_elems(&r[1], objs, r->size, + prod_head & r->mask, r->esize, n); + if (meta != NULL) + __rte_ring_do_enqueue_elems(r->meta, meta, r->size, + prod_head & r->mask, r->msize, n); + __rte_soring_update_tail(&r->prod, st, prod_head, prod_next, 1); + } + + if (free_space != NULL) + *free_space = nb_free - n; + return n; +} + +static __rte_always_inline uint32_t +soring_dequeue(struct rte_soring *r, void *objs, void *meta, + uint32_t num, enum rte_ring_queue_behavior behavior, + uint32_t *available) +{ + enum rte_ring_sync_type st; + uint32_t entries, cons_head, cons_next, n, ns, reqn; + + RTE_ASSERT(r != NULL && r->nb_stage > 0); + RTE_ASSERT(meta == NULL || r->meta != NULL); + + ns = r->nb_stage - 1; + st = r->cons.ht.sync_type; + + /* try to grab exactly @num elems first */ + n = __rte_soring_move_cons_head(r, ns, num, RTE_RING_QUEUE_FIXED, st, + &cons_head, &cons_next, &entries); + if (n == 0) { + /* try to finalize some elems from previous stage */ + n = __rte_soring_stage_finalize(&r->stage[ns].sht, ns, + r->state, r->mask, 2 * num); + entries += n; + + /* repeat attempt to grab elems */ + reqn = (behavior == RTE_RING_QUEUE_FIXED) ? num : 0; + if (entries >= reqn) + n = __rte_soring_move_cons_head(r, ns, num, behavior, + st, &cons_head, &cons_next, &entries); + else + n = 0; + } + + /* we have some elems to consume */ + if (n != 0) { + __rte_ring_do_dequeue_elems(objs, &r[1], r->size, + cons_head & r->mask, r->esize, n); + if (meta != NULL) + __rte_ring_do_dequeue_elems(meta, r->meta, r->size, + cons_head & r->mask, r->msize, n); + __rte_soring_update_tail(&r->cons, st, cons_head, cons_next, 0); + } + + if (available != NULL) + *available = entries - n; + return n; +} + +/* + * Verify internal SORING state. + * WARNING: if expected value is not equal to actual one, it means that for + * whatever reason SORING data constancy is broken. That is a very serious + * problem that most likely will cause race-conditions, memory corruption, + * program crash. + * To ease debugging it user might rebuild ring library with + * RTE_SORING_DEBUG enabled. + */ +static __rte_always_inline void +soring_verify_state(const struct rte_soring *r, uint32_t stage, uint32_t idx, + const char *msg, union soring_state val, union soring_state exp) +{ + if (val.raw != exp.raw) { +#ifdef RTE_SORING_DEBUG + rte_soring_dump(stderr, r); + rte_panic("line:%d from:%s: soring=%p, stage=%#x, idx=%#x, " + "expected={.stnum=%#x, .ftoken=%#x}, " + "actual={.stnum=%#x, .ftoken=%#x};\n", + __LINE__, msg, r, stage, idx, + exp.stnum, exp.ftoken, + val.stnum, val.ftoken); +#else + SORING_LOG(EMERG, "from:%s: soring=%p, stage=%#x, idx=%#x, " + "expected={.stnum=%#x, .ftoken=%#x}, " + "actual={.stnum=%#x, .ftoken=%#x};", + msg, r, stage, idx, + exp.stnum, exp.ftoken, + val.stnum, val.ftoken); +#endif + } +} + +/* check and update state ring at acquire op*/ +static __rte_always_inline void +acquire_state_update(const struct rte_soring *r, uint32_t stage, uint32_t idx, + uint32_t ftoken, uint32_t num) +{ + union soring_state st; + const union soring_state est = {.raw = 0}; + + st.raw = rte_atomic_load_explicit(&r->state[idx].raw, + rte_memory_order_relaxed); + soring_verify_state(r, stage, idx, __func__, st, est); + + st.ftoken = ftoken; + st.stnum = (SORING_ST_START | num); + + rte_atomic_store_explicit(&r->state[idx].raw, st.raw, + rte_memory_order_relaxed); +} + +static __rte_always_inline uint32_t +soring_acquire(struct rte_soring *r, void *objs, void *meta, + uint32_t stage, uint32_t num, enum rte_ring_queue_behavior behavior, + uint32_t *ftoken, uint32_t *available) +{ + uint32_t avail, head, idx, n, next, reqn; + struct soring_stage *pstg; + struct soring_stage_headtail *cons; + + RTE_ASSERT(r != NULL && stage < r->nb_stage); + RTE_ASSERT(meta == NULL || r->meta != NULL); + + cons = &r->stage[stage].sht; + + if (stage == 0) + n = __rte_soring_stage_move_head(cons, &r->prod.ht, 0, num, + behavior, &head, &next, &avail); + else { + pstg = r->stage + stage - 1; + + /* try to grab exactly @num elems */ + n = __rte_soring_stage_move_head(cons, &pstg->ht, 0, num, + RTE_RING_QUEUE_FIXED, &head, &next, &avail); + if (n == 0) { + /* try to finalize some elems from previous stage */ + n = __rte_soring_stage_finalize(&pstg->sht, stage - 1, + r->state, r->mask, 2 * num); + avail += n; + + /* repeat attempt to grab elems */ + reqn = (behavior == RTE_RING_QUEUE_FIXED) ? num : 0; + if (avail >= reqn) + n = __rte_soring_stage_move_head(cons, + &pstg->ht, 0, num, behavior, &head, + &next, &avail); + else + n = 0; + } + } + + if (n != 0) { + + idx = head & r->mask; + *ftoken = SORING_FTKN_MAKE(head, stage); + + /* check and update state value */ + acquire_state_update(r, stage, idx, *ftoken, n); + + /* copy elems that are ready for given stage */ + __rte_ring_do_dequeue_elems(objs, &r[1], r->size, idx, + r->esize, n); + if (meta != NULL) + __rte_ring_do_dequeue_elems(meta, r->meta, + r->size, idx, r->msize, n); + } + + if (available != NULL) + *available = avail - n; + return n; +} + +static __rte_always_inline void +soring_release(struct rte_soring *r, const void *objs, + const void *meta, uint32_t stage, uint32_t n, uint32_t ftoken) +{ + uint32_t idx, pos, tail; + struct soring_stage *stg; + union soring_state st; + + const union soring_state est = { + .stnum = (SORING_ST_START | n), + .ftoken = ftoken, + }; + + RTE_ASSERT(r != NULL && stage < r->nb_stage); + RTE_ASSERT(meta == NULL || r->meta != NULL); + + stg = r->stage + stage; + + pos = SORING_FTKN_POS(ftoken, stage); + idx = pos & r->mask; + st.raw = rte_atomic_load_explicit(&r->state[idx].raw, + rte_memory_order_relaxed); + + /* check state ring contents */ + soring_verify_state(r, stage, idx, __func__, st, est); + + /* update contents of the ring, if necessary */ + if (objs != NULL) + __rte_ring_do_enqueue_elems(&r[1], objs, r->size, idx, + r->esize, n); + if (meta != NULL) + __rte_ring_do_enqueue_elems(r->meta, meta, r->size, idx, + r->msize, n); + + /* set state to FINISH, make sure it is not reordered */ + rte_atomic_thread_fence(rte_memory_order_release); + + st.stnum = SORING_ST_FINISH | n; + rte_atomic_store_explicit(&r->state[idx].raw, st.raw, + rte_memory_order_relaxed); + + /* try to do finalize(), if appropriate */ + tail = rte_atomic_load_explicit(&stg->sht.tail.pos, + rte_memory_order_relaxed); + if (tail == pos) + __rte_soring_stage_finalize(&stg->sht, stage, r->state, r->mask, + r->capacity); +} + +/* + * Public functions (data-path) start here. + */ + +void +rte_soring_release(struct rte_soring *r, const void *objs, + uint32_t stage, uint32_t n, uint32_t ftoken) +{ + soring_release(r, objs, NULL, stage, n, ftoken); +} + + +void +rte_soring_releasx(struct rte_soring *r, const void *objs, + const void *meta, uint32_t stage, uint32_t n, uint32_t ftoken) +{ + soring_release(r, objs, meta, stage, n, ftoken); +} + +uint32_t +rte_soring_enqueue_bulk(struct rte_soring *r, const void *objs, uint32_t n, + uint32_t *free_space) +{ + return soring_enqueue(r, objs, NULL, n, RTE_RING_QUEUE_FIXED, + free_space); +} + +uint32_t +rte_soring_enqueux_bulk(struct rte_soring *r, const void *objs, + const void *meta, uint32_t n, uint32_t *free_space) +{ + return soring_enqueue(r, objs, meta, n, RTE_RING_QUEUE_FIXED, + free_space); +} + +uint32_t +rte_soring_enqueue_burst(struct rte_soring *r, const void *objs, uint32_t n, + uint32_t *free_space) +{ + return soring_enqueue(r, objs, NULL, n, RTE_RING_QUEUE_VARIABLE, + free_space); +} + +uint32_t +rte_soring_enqueux_burst(struct rte_soring *r, const void *objs, + const void *meta, uint32_t n, uint32_t *free_space) +{ + return soring_enqueue(r, objs, meta, n, RTE_RING_QUEUE_VARIABLE, + free_space); +} + +uint32_t +rte_soring_dequeue_bulk(struct rte_soring *r, void *objs, uint32_t num, + uint32_t *available) +{ + return soring_dequeue(r, objs, NULL, num, RTE_RING_QUEUE_FIXED, + available); +} + +uint32_t +rte_soring_dequeux_bulk(struct rte_soring *r, void *objs, void *meta, + uint32_t num, uint32_t *available) +{ + return soring_dequeue(r, objs, meta, num, RTE_RING_QUEUE_FIXED, + available); +} + +uint32_t +rte_soring_dequeue_burst(struct rte_soring *r, void *objs, uint32_t num, + uint32_t *available) +{ + return soring_dequeue(r, objs, NULL, num, RTE_RING_QUEUE_VARIABLE, + available); +} + +uint32_t +rte_soring_dequeux_burst(struct rte_soring *r, void *objs, void *meta, + uint32_t num, uint32_t *available) +{ + return soring_dequeue(r, objs, meta, num, RTE_RING_QUEUE_VARIABLE, + available); +} + +uint32_t +rte_soring_acquire_bulk(struct rte_soring *r, void *objs, + uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available) +{ + return soring_acquire(r, objs, NULL, stage, num, + RTE_RING_QUEUE_FIXED, ftoken, available); +} + +uint32_t +rte_soring_acquirx_bulk(struct rte_soring *r, void *objs, void *meta, + uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available) +{ + return soring_acquire(r, objs, meta, stage, num, + RTE_RING_QUEUE_FIXED, ftoken, available); +} + +uint32_t +rte_soring_acquire_burst(struct rte_soring *r, void *objs, + uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available) +{ + return soring_acquire(r, objs, NULL, stage, num, + RTE_RING_QUEUE_VARIABLE, ftoken, available); +} + +uint32_t +rte_soring_acquirx_burst(struct rte_soring *r, void *objs, void *meta, + uint32_t stage, uint32_t num, uint32_t *ftoken, uint32_t *available) +{ + return soring_acquire(r, objs, meta, stage, num, + RTE_RING_QUEUE_VARIABLE, ftoken, available); +} + +unsigned int +rte_soring_count(const struct rte_soring *r) +{ + uint32_t prod_tail = r->prod.ht.tail; + uint32_t cons_tail = r->cons.ht.tail; + uint32_t count = (prod_tail - cons_tail) & r->mask; + return (count > r->capacity) ? r->capacity : count; +} + +unsigned int +rte_soring_free_count(const struct rte_soring *r) +{ + return r->capacity - rte_soring_count(r); +} diff --git a/lib/ring/soring.h b/lib/ring/soring.h new file mode 100644 index 0000000000..b252ae4a61 --- /dev/null +++ b/lib/ring/soring.h @@ -0,0 +1,138 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2024 Huawei Technologies Co., Ltd + */ + +#ifndef _SORING_H_ +#define _SORING_H_ + +/** + * @file + * This file contains internal strctures of RTE soring: Staged Ordered Ring. + * Sort of extension of conventional RTE ring. + * Internal structure: + * In addition to 'normal' ring of elems, it also has a ring of states of the + * same size. Each state[] corresponds to exactly one elem[]. + * state[] will be used by acquire/release/dequeue functions to store internal + * information and should not be accessed by the user directly. + * For actual implementation details, please refer to soring.c + */ + +#include + +/* logging stuff, register our own tag for SORING */ +#include + +extern int soring_logtype; +#define RTE_LOGTYPE_SORING soring_logtype +#define SORING_LOG(level, ...) \ + RTE_LOG_LINE(level, SORING, "" __VA_ARGS__) + +/** + * SORING internal state for each element + */ +union soring_state { + alignas(sizeof(uint64_t)) RTE_ATOMIC(uint64_t) raw; + struct { + RTE_ATOMIC(uint32_t) ftoken; + RTE_ATOMIC(uint32_t) stnum; + }; +}; + +/* upper 2 bits are used for status */ +#define SORING_ST_START RTE_SHIFT_VAL32(1, RTE_SORING_ST_BIT) +#define SORING_ST_FINISH RTE_SHIFT_VAL32(2, RTE_SORING_ST_BIT) + +#define SORING_ST_MASK \ + RTE_GENMASK32(sizeof(uint32_t) * CHAR_BIT - 1, RTE_SORING_ST_BIT) + +#define SORING_FTKN_MAKE(pos, stg) ((pos) + (stg)) +#define SORING_FTKN_POS(ftk, stg) ((ftk) - (stg)) + +/** + * SORING re-uses rte_ring internal structures and implementation + * for enqueue/dequeue operations. + */ +struct __rte_ring_headtail { + + union __rte_cache_aligned { + struct rte_ring_headtail ht; + struct rte_ring_hts_headtail hts; + struct rte_ring_rts_headtail rts; + }; + + RTE_CACHE_GUARD; +}; + +union soring_stage_tail { + /** raw 8B value to read/write *sync* and *pos* as one atomic op */ + alignas(sizeof(uint64_t)) RTE_ATOMIC(uint64_t) raw; + struct { + RTE_ATOMIC(uint32_t) sync; + RTE_ATOMIC(uint32_t) pos; + }; +}; + +struct soring_stage_headtail { + volatile union soring_stage_tail tail; + enum rte_ring_sync_type unused; /**< unused */ + volatile RTE_ATOMIC(uint32_t) head; +}; + +/** + * SORING stage head_tail structure is 'compatible' with rte_ring ones. + * rte_ring internal functions for moving producer/consumer head + * can work transparently with stage head_tail and visa-versa + * (no modifications required). + */ +struct soring_stage { + + union __rte_cache_aligned { + struct rte_ring_headtail ht; + struct soring_stage_headtail sht; + }; + + RTE_CACHE_GUARD; +}; + +/** + * RTE soring internal structure. + * As with rte_ring actual elements array supposed to be located direclty + * after the rte_soring structure. + */ +struct __rte_cache_aligned rte_soring { + uint32_t size; /**< Size of ring. */ + uint32_t mask; /**< Mask (size-1) of ring. */ + uint32_t capacity; /**< Usable size of ring */ + uint32_t esize; + /**< size of elements in the ring, must be a multiple of 4*/ + uint32_t msize; + /**< size of metadata value for each elem, must be a multiple of 4 */ + + /** Ring stages */ + struct soring_stage *stage; + uint32_t nb_stage; + + /** Ring of states (one per element) */ + union soring_state *state; + + /** Pointer to the buffer where metadata values for each elements + * are stored. This is supplementary and optional information that + * user can attach to each element of the ring. + * While it is possible to incorporate this information inside + * user-defined element, in many cases it is plausible to maintain it + * as a separate array (mainly for performance reasons). + */ + void *meta; + + RTE_CACHE_GUARD; + + /** Ring producer status. */ + struct __rte_ring_headtail prod; + + /** Ring consumer status. */ + struct __rte_ring_headtail cons; + + char name[RTE_RING_NAMESIZE]; /**< Name of the ring. */ +}; + +#endif /* _SORING_H_ */ diff --git a/lib/ring/version.map b/lib/ring/version.map index 61f7464f5a..081034c30e 100644 --- a/lib/ring/version.map +++ b/lib/ring/version.map @@ -20,4 +20,23 @@ EXPERIMENTAL { # added in 24.11 rte_ring_headtail_dump; + rte_soring_acquire_bulk; + rte_soring_acquire_burst; + rte_soring_acquirx_bulk; + rte_soring_acquirx_burst; + rte_soring_count; + rte_soring_dequeue_bulk; + rte_soring_dequeue_burst; + rte_soring_dequeux_bulk; + rte_soring_dequeux_burst; + rte_soring_enqueue_bulk; + rte_soring_enqueue_burst; + rte_soring_enqueux_bulk; + rte_soring_enqueux_burst; + rte_soring_dump; + rte_soring_free_count; + rte_soring_get_memsize; + rte_soring_init; + rte_soring_release; + rte_soring_releasx; }; From patchwork Wed Oct 30 21:23:03 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Konstantin Ananyev X-Patchwork-Id: 147766 X-Patchwork-Delegate: thomas@monjalon.net Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 4B68045B82; Wed, 30 Oct 2024 21:33:25 +0100 (CET) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id A8AD7433D8; Wed, 30 Oct 2024 21:33:22 +0100 (CET) Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) by mails.dpdk.org (Postfix) with ESMTP id E4E884334B for ; Wed, 30 Oct 2024 21:33:02 +0100 (CET) Received: from mail.maildlp.com (unknown [172.18.186.216]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4XdzLR2FCtz6K5kN; Thu, 31 Oct 2024 04:31:43 +0800 (CST) Received: from frapeml500007.china.huawei.com (unknown [7.182.85.172]) by mail.maildlp.com (Postfix) with ESMTPS id 58C1F140B67; Thu, 31 Oct 2024 04:33:02 +0800 (CST) Received: from localhost.localdomain (10.220.239.45) by frapeml500007.china.huawei.com (7.182.85.172) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.1.2507.39; Wed, 30 Oct 2024 21:33:01 +0100 From: Konstantin Ananyev To: CC: , , , , , , , , Subject: [PATCH v7 6/7] app/test: add unit tests for soring API Date: Wed, 30 Oct 2024 17:23:03 -0400 Message-ID: <20241030212304.104180-7-konstantin.ananyev@huawei.com> X-Mailer: git-send-email 2.35.3 In-Reply-To: <20241030212304.104180-1-konstantin.ananyev@huawei.com> References: <20241021174745.1843-1-konstantin.ananyev@huawei.com> <20241030212304.104180-1-konstantin.ananyev@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.220.239.45] X-ClientProxiedBy: frapeml100006.china.huawei.com (7.182.85.201) To frapeml500007.china.huawei.com (7.182.85.172) X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Add both functional and stess test-cases for soring API. Stress test serves as both functional and performance test of soring enqueue/dequeue/acquire/release operations under high contention (for both over committed and non-over committed scenarios). Signed-off-by: Eimear Morrissey Signed-off-by: Konstantin Ananyev --- app/test/meson.build | 3 + app/test/test_soring.c | 442 +++++++++++++++ app/test/test_soring_mt_stress.c | 40 ++ app/test/test_soring_stress.c | 48 ++ app/test/test_soring_stress.h | 35 ++ app/test/test_soring_stress_impl.h | 834 +++++++++++++++++++++++++++++ 6 files changed, 1402 insertions(+) create mode 100644 app/test/test_soring.c create mode 100644 app/test/test_soring_mt_stress.c create mode 100644 app/test/test_soring_stress.c create mode 100644 app/test/test_soring_stress.h create mode 100644 app/test/test_soring_stress_impl.h diff --git a/app/test/meson.build b/app/test/meson.build index 0f7e11969a..e59f27fd51 100644 --- a/app/test/meson.build +++ b/app/test/meson.build @@ -177,6 +177,9 @@ source_file_deps = { 'test_security_proto.c' : ['cryptodev', 'security'], 'test_seqlock.c': [], 'test_service_cores.c': [], + 'test_soring.c': [], + 'test_soring_mt_stress.c': [], + 'test_soring_stress.c': [], 'test_spinlock.c': [], 'test_stack.c': ['stack'], 'test_stack_perf.c': ['stack'], diff --git a/app/test/test_soring.c b/app/test/test_soring.c new file mode 100644 index 0000000000..b2110305a7 --- /dev/null +++ b/app/test/test_soring.c @@ -0,0 +1,442 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2024 Huawei Technologies Co., Ltd + */ + +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include + +#include "test.h" + +#define MAX_ACQUIRED 20 + +#define SORING_TEST_ASSERT(val, expected) do { \ + RTE_TEST_ASSERT(expected == val, \ + "%s: expected %u got %u\n", #val, expected, val); \ +} while (0) + +static void +set_soring_init_param(struct rte_soring_param *prm, + const char *name, uint32_t esize, uint32_t elems, + uint32_t stages, uint32_t stsize, + enum rte_ring_sync_type rst_prod, + enum rte_ring_sync_type rst_cons) +{ + prm->name = name; + prm->elem_size = esize; + prm->elems = elems; + prm->stages = stages; + prm->meta_size = stsize; + prm->prod_synt = rst_prod; + prm->cons_synt = rst_cons; +} + +static int +move_forward_stage(struct rte_soring *sor, + uint32_t num_packets, uint32_t stage) +{ + uint32_t acquired; + uint32_t ftoken; + uint32_t *acquired_objs[MAX_ACQUIRED]; + + acquired = rte_soring_acquire_bulk(sor, acquired_objs, stage, + num_packets, &ftoken, NULL); + SORING_TEST_ASSERT(acquired, num_packets); + rte_soring_release(sor, NULL, stage, num_packets, + ftoken); + + return 0; +} + +/* + * struct rte_soring_param param checking. + */ +static int +test_soring_init(void) +{ + struct rte_soring *sor = NULL; + struct rte_soring_param prm; + int rc; + size_t sz; + memset(&prm, 0, sizeof(prm)); + +/*init memory*/ + set_soring_init_param(&prm, "alloc_memory", sizeof(uintptr_t), + 4, 1, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT); + sz = rte_soring_get_memsize(&prm); + sor = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE); + RTE_TEST_ASSERT_NOT_NULL(sor, "could not allocate memory for soring"); + + set_soring_init_param(&prm, "test_invalid_stages", sizeof(uintptr_t), + 4, 0, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT); + rc = rte_soring_init(sor, &prm); + RTE_TEST_ASSERT_FAIL(rc, "initted soring with invalid num stages"); + + set_soring_init_param(&prm, "test_invalid_esize", 0, + 4, 1, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT); + rc = rte_soring_init(sor, &prm); + RTE_TEST_ASSERT_FAIL(rc, "initted soring with 0 esize"); + + set_soring_init_param(&prm, "test_invalid_esize", 9, + 4, 1, 4, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT); + rc = rte_soring_init(sor, &prm); + RTE_TEST_ASSERT_FAIL(rc, "initted soring with esize not multiple of 4"); + + set_soring_init_param(&prm, "test_invalid_rsize", sizeof(uintptr_t), + 4, 1, 3, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT); + rc = rte_soring_init(sor, &prm); + RTE_TEST_ASSERT_FAIL(rc, "initted soring with rcsize not multiple of 4"); + + set_soring_init_param(&prm, "test_invalid_elems", sizeof(uintptr_t), + RTE_SORING_ELEM_MAX + 1, 1, 4, RTE_RING_SYNC_MT, + RTE_RING_SYNC_MT); + rc = rte_soring_init(sor, &prm); + RTE_TEST_ASSERT_FAIL(rc, "initted soring with invalid num elements"); + + rte_free(sor); + return 0; +} + +static int +test_soring_get_memsize(void) +{ + + struct rte_soring_param prm; + ssize_t sz; + + set_soring_init_param(&prm, "memsize", sizeof(uint32_t *), + 10, 1, 0, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT); + sz = rte_soring_get_memsize(&prm); + RTE_TEST_ASSERT(sz > 0, "failed to calculate size"); + + set_soring_init_param(&prm, "memsize", sizeof(uint8_t), + 16, UINT32_MAX, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT); + sz = rte_soring_get_memsize(&prm); + RTE_TEST_ASSERT_EQUAL(sz, -EINVAL, "calcuated size incorrect"); + + set_soring_init_param(&prm, "memsize", 0, + 16, UINT32_MAX, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT); + sz = rte_soring_get_memsize(&prm); + RTE_TEST_ASSERT_EQUAL(sz, -EINVAL, "calculated size incorrect"); + + return 0; + +} + +static int +test_soring_stages(void) +{ + struct rte_soring *sor = NULL; + struct rte_soring_param prm; + uint32_t objs[32]; + uint32_t rcs[32]; + uint32_t acquired_objs[32]; + uint32_t acquired_rcs[32]; + uint32_t dequeued_rcs[32]; + uint32_t dequeued_objs[32]; + size_t ssz; + uint32_t stage, enqueued, dequeued, acquired; + uint32_t i, ftoken; + + memset(&prm, 0, sizeof(prm)); + set_soring_init_param(&prm, "stages", sizeof(uint32_t), 32, + 10000, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT); + ssz = rte_soring_get_memsize(&prm); + RTE_TEST_ASSERT(ssz > 0, "parameter error calculating ring size"); + sor = rte_zmalloc(NULL, ssz, RTE_CACHE_LINE_SIZE); + RTE_TEST_ASSERT_NOT_NULL(sor, "couldn't allocate memory for soring"); + rte_soring_init(sor, &prm); + + for (i = 0; i < 32; i++) { + rcs[i] = i; + objs[i] = i + 32; + } + + enqueued = rte_soring_enqueux_burst(sor, objs, rcs, 32, NULL); + SORING_TEST_ASSERT(enqueued, 32); + + for (stage = 0; stage < 10000; stage++) { + int j; + dequeued = rte_soring_dequeue_bulk(sor, dequeued_objs, + 32, NULL); + /* check none at end stage */ + SORING_TEST_ASSERT(dequeued, 0); + + acquired = rte_soring_acquirx_bulk(sor, acquired_objs, + acquired_rcs, stage, 32, &ftoken, NULL); + SORING_TEST_ASSERT(acquired, 32); + + for (j = 0; j < 32; j++) { + SORING_TEST_ASSERT(acquired_rcs[j], j + stage); + SORING_TEST_ASSERT(acquired_objs[j], j + stage + 32); + /* modify both queue object and rc */ + acquired_objs[j]++; + acquired_rcs[j]++; + } + + rte_soring_releasx(sor, acquired_objs, + acquired_rcs, stage, 32, + ftoken); + } + + dequeued = rte_soring_dequeux_bulk(sor, dequeued_objs, dequeued_rcs, + 32, NULL); + SORING_TEST_ASSERT(dequeued, 32); + for (i = 0; i < 32; i++) { + /* ensure we ended up with the expected values in order*/ + SORING_TEST_ASSERT(dequeued_rcs[i], i + 10000); + SORING_TEST_ASSERT(dequeued_objs[i], i + 32 + 10000); + } + rte_free(sor); + return 0; +} + +static int +test_soring_enqueue_dequeue(void) +{ + struct rte_soring *sor = NULL; + struct rte_soring_param prm; + int rc; + uint32_t i; + size_t sz; + uint32_t queue_objs[10]; + uint32_t *queue_objs_p[10]; + uint32_t free_space; + uint32_t enqueued, dequeued; + uint32_t *dequeued_objs[10]; + + memset(&prm, 0, sizeof(prm)); + for (i = 0; i < 10; i++) { + queue_objs[i] = i + 1; + queue_objs_p[i] = &queue_objs[i]; + } + +/*init memory*/ + set_soring_init_param(&prm, "enqueue/dequeue", sizeof(uint32_t *), + 10, 1, 0, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT); + sz = rte_soring_get_memsize(&prm); + sor = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE); + RTE_TEST_ASSERT_NOT_NULL(sor, "alloc failed for soring"); + rc = rte_soring_init(sor, &prm); + RTE_TEST_ASSERT_SUCCESS(rc, "Failed to init soring"); + + free_space = 0; + + enqueued = rte_soring_enqueue_burst(sor, queue_objs_p, 5, &free_space); + + SORING_TEST_ASSERT(free_space, 5); + SORING_TEST_ASSERT(enqueued, 5); + + /* fixed amount enqueue */ + enqueued = rte_soring_enqueue_bulk(sor, queue_objs_p, 7, &free_space); + + SORING_TEST_ASSERT(free_space, 5); + SORING_TEST_ASSERT(enqueued, 0); + + /* variable amount enqueue */ + enqueued = rte_soring_enqueue_burst(sor, queue_objs_p + 5, 7, + &free_space); + SORING_TEST_ASSERT(free_space, 0); + SORING_TEST_ASSERT(enqueued, 5); + + /* test no dequeue while stage 0 has not completed */ + dequeued = rte_soring_dequeue_bulk(sor, dequeued_objs, 10, NULL); + SORING_TEST_ASSERT(dequeued, 0); + dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 10, NULL); + SORING_TEST_ASSERT(dequeued, 0); + + move_forward_stage(sor, 8, 0); + + /* can only dequeue as many as have completed stage*/ + dequeued = rte_soring_dequeue_bulk(sor, dequeued_objs, 10, NULL); + SORING_TEST_ASSERT(dequeued, 0); + dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 10, NULL); + SORING_TEST_ASSERT(dequeued, 8); + /* packets remain in order */ + for (i = 0; i < dequeued; i++) { + RTE_TEST_ASSERT_EQUAL(dequeued_objs[i], + queue_objs_p[i], "dequeued != enqueued"); + } + + dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 1, NULL); + SORING_TEST_ASSERT(dequeued, 0); + + move_forward_stage(sor, 2, 0); + dequeued = rte_soring_dequeue_burst(sor, dequeued_objs, 2, NULL); + SORING_TEST_ASSERT(dequeued, 2); + /* packets remain in order */ + for (i = 0; i < dequeued; i++) { + RTE_TEST_ASSERT_EQUAL(dequeued_objs[i], + queue_objs_p[i + 8], "dequeued != enqueued"); + } + + rte_soring_dump(stdout, sor); + rte_free(sor); + return 0; +} + +static int +test_soring_acquire_release(void) +{ + + struct rte_soring *sor = NULL; + struct rte_soring_param prm; + int rc, i; + size_t sz; + + memset(&prm, 0, sizeof(prm)); + uint32_t queue_objs[10]; + uint32_t rc_objs[10]; + uint32_t acquired_objs[10]; + uint32_t dequeued_objs[10]; + uint32_t dequeued_rcs[10]; + uint32_t s1_acquired_rcs[10]; + uint32_t free_space, enqueued, ftoken, acquired, dequeued; + + for (i = 0; i < 10; i++) { + queue_objs[i] = i + 5; + rc_objs[i] = i + 10; + } + +/*init memory*/ + set_soring_init_param(&prm, "test_acquire_release", sizeof(uint32_t), + 20, 2, sizeof(uint32_t), RTE_RING_SYNC_MT, RTE_RING_SYNC_MT); + sz = rte_soring_get_memsize(&prm); + sor = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE); + if (sor == NULL) { + printf("%s: alloc(%zu) for FIFO with %u elems failed", + __func__, sz, prm.elems); + return -ENOMEM; + } + + /* init ring */ + rc = rte_soring_init(sor, &prm); + RTE_TEST_ASSERT_SUCCESS(rc, "failed to init soring"); + + /* enqueue with associated rc */ + enqueued = rte_soring_enqueux_burst(sor, queue_objs, rc_objs, 5, + &free_space); + SORING_TEST_ASSERT(enqueued, 5); + /* enqueue without associated rc */ + enqueued = rte_soring_enqueue_burst(sor, queue_objs + 5, 5, + &free_space); + SORING_TEST_ASSERT(enqueued, 5); + + /* acquire the objects with rc's and ensure they are as expected */ + acquired = rte_soring_acquirx_burst(sor, acquired_objs, + s1_acquired_rcs, 0, 5, &ftoken, NULL); + SORING_TEST_ASSERT(acquired, 5); + for (i = 0; i < 5; i++) { + RTE_TEST_ASSERT_EQUAL(s1_acquired_rcs[i], rc_objs[i], + "acquired rc[%d]: %u != enqueued rc: %u", + i, s1_acquired_rcs[i], rc_objs[i]); + RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i], + "acquired obj[%d]: %u != enqueued obj %u", + i, acquired_objs[i], queue_objs[i]); + } + rte_soring_release(sor, NULL, 0, 5, ftoken); + + /* acquire the objects without rc's and ensure they are as expected */ + acquired = rte_soring_acquirx_burst(sor, acquired_objs, + s1_acquired_rcs, 0, 5, &ftoken, NULL); + SORING_TEST_ASSERT(acquired, 5); + for (i = 0; i < 5; i++) { + /* as the rc area of memory is zero'd at init this is true + * but this is a detail of implementation rather than + * a guarantee. + */ + RTE_TEST_ASSERT_EQUAL(s1_acquired_rcs[i], 0, + "acquired rc not empty"); + RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i + 5], + "acquired obj[%d]: %u != enqueued obj %u", + i, acquired_objs[i], queue_objs[i + 5]); + } + /*release the objects, adding rc's */ + rte_soring_releasx(sor, NULL, rc_objs + 5, 0, 5, + ftoken); + + acquired = rte_soring_acquirx_burst(sor, acquired_objs, + s1_acquired_rcs, 1, 10, &ftoken, NULL); + SORING_TEST_ASSERT(acquired, 10); + for (i = 0; i < 10; i++) { + /* ensure the associated rc's are the ones added at relase */ + RTE_TEST_ASSERT_EQUAL(s1_acquired_rcs[i], rc_objs[i], + "acquired rc[%d]: %u != enqueued rc: %u", + i, s1_acquired_rcs[i], rc_objs[i]); + RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i], + "acquired obj[%d]: %u != enqueued obj %u", + i, acquired_objs[i], queue_objs[i]); + } + /*release the objects, with rc's set to NULL */ + rte_soring_release(sor, NULL, 1, 10, ftoken); + + dequeued = rte_soring_dequeux_burst(sor, dequeued_objs, dequeued_rcs, + 10, NULL); + SORING_TEST_ASSERT(dequeued, 10); + for (i = 0; i < 10; i++) { + /* ensure the associated rc's are the ones added at relase */ + RTE_TEST_ASSERT_EQUAL(dequeued_rcs[i], rc_objs[i], + "dequeued rc[%d]: %u != enqueued rc: %u", + i, dequeued_rcs[i], rc_objs[i]); + RTE_TEST_ASSERT_EQUAL(acquired_objs[i], queue_objs[i], + "acquired obj[%d]: %u != enqueued obj %u", + i, acquired_objs[i], queue_objs[i]); + } + rte_soring_dump(stdout, sor); + rte_free(sor); + return 0; +} + +static int +test_soring(void) +{ + + /* Negative test cases */ + if (test_soring_init() < 0) + goto test_fail; + + /* Memory calculations */ + if (test_soring_get_memsize() < 0) + goto test_fail; + + /* Basic enqueue/dequeue operations */ + if (test_soring_enqueue_dequeue() < 0) + goto test_fail; + + /* Acquire/release */ + if (test_soring_acquire_release() < 0) + goto test_fail; + + /* Test large number of stages */ + if (test_soring_stages() < 0) + goto test_fail; + + return 0; + +test_fail: + return -1; +} + +REGISTER_FAST_TEST(soring_autotest, true, true, test_soring); + diff --git a/app/test/test_soring_mt_stress.c b/app/test/test_soring_mt_stress.c new file mode 100644 index 0000000000..2f90bb4598 --- /dev/null +++ b/app/test/test_soring_mt_stress.c @@ -0,0 +1,40 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2024 Huawei Technologies Co., Ltd + */ + +#include "test_soring_stress_impl.h" + +static inline uint32_t +_st_ring_dequeue_burst(struct rte_soring *r, void **obj, uint32_t n, + uint32_t *avail) +{ + return rte_soring_dequeue_burst(r, obj, n, avail); +} + +static inline uint32_t +_st_ring_enqueue_bulk(struct rte_soring *r, void * const *obj, uint32_t n, + uint32_t *free) +{ + return rte_soring_enqueue_bulk(r, obj, n, free); +} + +static inline uint32_t +_st_ring_acquire_burst(struct rte_soring *r, uint32_t stage, void **obj, + uint32_t num, uint32_t *token, uint32_t *avail) +{ + return rte_soring_acquire_burst(r, obj, stage, num, token, avail); +} + +static inline void +_st_ring_release(struct rte_soring *r, uint32_t stage, uint32_t token, + void * const *obj, uint32_t num) +{ + RTE_SET_USED(obj); + rte_soring_release(r, NULL, stage, num, token); +} + +const struct test test_soring_mt_stress = { + .name = "MT", + .nb_case = RTE_DIM(tests), + .cases = tests, +}; diff --git a/app/test/test_soring_stress.c b/app/test/test_soring_stress.c new file mode 100644 index 0000000000..334af6a29c --- /dev/null +++ b/app/test/test_soring_stress.c @@ -0,0 +1,48 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2024 Huawei Technologies Co., Ltd + */ + +#include "test_soring_stress.h" + +static int +run_test(const struct test *test) +{ + int32_t rc; + uint32_t i, k; + + for (i = 0, k = 0; i != test->nb_case; i++) { + + printf("TEST-CASE %s %s START\n", + test->name, test->cases[i].name); + + rc = test->cases[i].func(test->cases[i].wfunc); + k += (rc == 0); + + if (rc != 0) + printf("TEST-CASE %s %s FAILED\n", + test->name, test->cases[i].name); + else + printf("TEST-CASE %s %s OK\n", + test->name, test->cases[i].name); + } + + return k; +} + +static int +test_ring_stress(void) +{ + uint32_t n, k; + + n = 0; + k = 0; + + n += test_soring_mt_stress.nb_case; + k += run_test(&test_soring_mt_stress); + + printf("Number of tests:\t%u\nSuccess:\t%u\nFailed:\t%u\n", + n, k, n - k); + return (k != n); +} + +REGISTER_TEST_COMMAND(soring_stress_autotest, test_ring_stress); diff --git a/app/test/test_soring_stress.h b/app/test/test_soring_stress.h new file mode 100644 index 0000000000..6190e96117 --- /dev/null +++ b/app/test/test_soring_stress.h @@ -0,0 +1,35 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2024 Huawei Technologies Co., Ltd + */ + + +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include + +#include "test.h" + +struct test_case { + const char *name; + int (*func)(int (*)(void *)); + int (*wfunc)(void *arg); +}; + +struct test { + const char *name; + uint32_t nb_case; + const struct test_case *cases; +}; + +extern const struct test test_soring_mt_stress; diff --git a/app/test/test_soring_stress_impl.h b/app/test/test_soring_stress_impl.h new file mode 100644 index 0000000000..df4ca9564c --- /dev/null +++ b/app/test/test_soring_stress_impl.h @@ -0,0 +1,834 @@ +/* SPDX-License-Identifier: BSD-3-Clause + */ + +#include + +#include "test_soring_stress.h" + +/** + * Stress test for soring enqueue/dequeue/acquire/release operations. + * Depending on the role, performs at least one of the following patterns + * on each worker: + * - dequeue/read-write data from/to the dequeued objects/enqueue. + * - acquire/read-write data from/to the acquired objects/release. + * Serves as both functional and performance test of soring + * data-path API under high contention + * (for both over committed and non-over committed scenarios). + */ + +#define RING_NAME "SORING_STRESS" +#define BULK_NUM 32 +#define RING_SIZE (2 * BULK_NUM * RTE_MAX_LCORE) + +#define MAX_STAGES 16 + +enum { + WRK_CMD_STOP, + WRK_CMD_RUN, +}; + +static uint32_t wrk_cmd __rte_cache_aligned = WRK_CMD_STOP; + +/* test run-time in seconds */ +static const uint32_t run_time = 60; +static const uint32_t verbose; + +static rte_spinlock_t dump_lock; + +struct lcore_op_stat { + uint64_t nb_lcore; + uint64_t nb_call; + uint64_t nb_obj; + uint64_t nb_cycle; + uint64_t max_cycle; + uint64_t min_cycle; +}; + +struct lcore_stat { + uint64_t nb_cycle; + struct lcore_op_stat deqenq; + uint32_t role_mask; + uint32_t nb_stage; + struct lcore_op_stat stage[MAX_STAGES]; +}; + +#define ROLE_DEQENQ RTE_BIT32(0) +#define ROLE_STAGE(n) RTE_BIT32(n + 1) + +struct lcore_arg { + struct rte_soring *rng; + struct lcore_stat stats; +} __rte_cache_aligned; + +struct ring_elem { + uint32_t cnt[RTE_CACHE_LINE_SIZE / sizeof(uint32_t)]; +} __rte_cache_aligned; + +/* + * redefinable functions + */ + +static uint32_t +_st_ring_dequeue_burst(struct rte_soring *r, void **obj, uint32_t n, + uint32_t *avail); + +static uint32_t +_st_ring_enqueue_bulk(struct rte_soring *r, void * const *obj, uint32_t n, + uint32_t *free); + +static uint32_t +_st_ring_acquire_burst(struct rte_soring *r, uint32_t stage, void **obj, + uint32_t num, uint32_t *token, uint32_t *avail); + +static void +_st_ring_release(struct rte_soring *r, uint32_t stage, uint32_t token, + void * const *obj, uint32_t num); + +static void +lcore_op_stat_update(struct lcore_op_stat *ls, uint64_t call, uint64_t obj, + uint64_t tm, int32_t prcs) +{ + ls->nb_call += call; + ls->nb_obj += obj; + ls->nb_cycle += tm; + if (prcs) { + ls->max_cycle = RTE_MAX(ls->max_cycle, tm); + ls->min_cycle = RTE_MIN(ls->min_cycle, tm); + } +} + +static void +lcore_stat_update(struct lcore_stat *ls, uint64_t call, uint64_t obj, + uint64_t tm, int32_t prcs) +{ + uint32_t i; + + ls->nb_cycle += tm; + lcore_op_stat_update(&ls->deqenq, call, obj, tm, prcs); + for (i = 0; i != ls->nb_stage; i++) + lcore_op_stat_update(ls->stage + i, call, obj, tm, prcs); +} + +static void +lcore_op_stat_aggr(struct lcore_op_stat *ms, const struct lcore_op_stat *ls) +{ + ms->nb_call += ls->nb_call; + ms->nb_obj += ls->nb_obj; + ms->nb_cycle += ls->nb_cycle; + ms->max_cycle = RTE_MAX(ms->max_cycle, ls->max_cycle); + ms->min_cycle = RTE_MIN(ms->min_cycle, ls->min_cycle); +} + +static void +lcore_stat_aggr(struct lcore_stat *ms, const struct lcore_stat *ls) +{ + uint32_t i; + + ms->nb_cycle = RTE_MAX(ms->nb_cycle, ls->nb_cycle); + lcore_op_stat_aggr(&ms->deqenq, &ls->deqenq); + ms->deqenq.nb_lcore += ((ls->role_mask & ROLE_DEQENQ) != 0); + for (i = 0; i != ms->nb_stage; i++) { + lcore_op_stat_aggr(ms->stage + i, ls->stage + i); + ms->stage[i].nb_lcore += ((ls->role_mask & ROLE_STAGE(i)) != 0); + } +} + +static void +lcore_op_stat_dump(FILE *f, const struct lcore_op_stat *ls, const char *cap, + long double st) +{ + fprintf(f, "\t%s={\n", cap); + + fprintf(f, "\t\tnb_lcore=%" PRIu64 ",\n", ls->nb_lcore); + fprintf(f, "\t\tnb_call=%" PRIu64 ",\n", ls->nb_call); + fprintf(f, "\t\tnb_obj=%" PRIu64 ",\n", ls->nb_obj); + fprintf(f, "\t\tnb_cycle=%" PRIu64 ",\n", ls->nb_cycle); + fprintf(f, "\t\tobj/call(avg): %.2Lf\n", + (long double)ls->nb_obj / ls->nb_call); + fprintf(f, "\t\tcycles/obj(avg): %.2Lf\n", + (long double)ls->nb_cycle / ls->nb_obj); + fprintf(f, "\t\tcycles/call(avg): %.2Lf\n", + (long double)ls->nb_cycle / ls->nb_call); + + /* if min/max cycles per call stats was collected */ + if (ls->min_cycle != UINT64_MAX) { + fprintf(f, "\t\tmax cycles/call=%" PRIu64 "(%.2Lf usec),\n", + ls->max_cycle, + (long double)ls->max_cycle / st); + fprintf(f, "\t\tmin cycles/call=%" PRIu64 "(%.2Lf usec),\n", + ls->min_cycle, + (long double)ls->min_cycle / st); + } + + fprintf(f, "\t},\n"); +} + +static void +lcore_stat_dump(FILE *f, uint32_t lc, const struct lcore_stat *ls) +{ + uint32_t i; + long double st; + char cap[64]; + + st = (long double)rte_get_timer_hz() / US_PER_S; + + if (lc == UINT32_MAX) + fprintf(f, "%s(AGGREGATE)={\n", __func__); + else + fprintf(f, "%s(lcore=%u)={\n", __func__, lc); + + fprintf(f, "\tnb_cycle=%" PRIu64 "(%.2Lf usec),\n", + ls->nb_cycle, (long double)ls->nb_cycle / st); + + lcore_op_stat_dump(f, &ls->deqenq, "DEQ+ENQ", st); + for (i = 0; i != ls->nb_stage; i++) { + snprintf(cap, sizeof(cap), "%s#%u", "STAGE", i); + lcore_op_stat_dump(f, ls->stage + i, cap, st); + } + + fprintf(f, "};\n"); +} + +static void +fill_ring_elm(struct ring_elem *elm, uint32_t fill) +{ + uint32_t i; + + for (i = 0; i != RTE_DIM(elm->cnt); i++) + elm->cnt[i] = fill; +} + +static int32_t +check_updt_elem(struct ring_elem *elm[], uint32_t num, + const struct ring_elem *check, const struct ring_elem *fill, + const char *fname, const char *opname, const struct rte_soring *sor) +{ + uint32_t i; + + for (i = 0; i != num; i++) { + if (memcmp(check, elm[i], sizeof(*check)) != 0) { + rte_spinlock_lock(&dump_lock); + printf("%s:%s: %s(lc=%u, num=%u) failed at %u-th iter, " + "offending object: %p\n", + fname, opname, __func__, rte_lcore_id(), num, i, + elm[i]); + rte_memdump(stdout, "expected", check, sizeof(*check)); + rte_memdump(stdout, "result", elm[i], sizeof(*elm[i])); + rte_soring_dump(stdout, sor); + rte_spinlock_unlock(&dump_lock); + return -EINVAL; + } + memcpy(elm[i], fill, sizeof(*elm[i])); + } + + return 0; +} + +static int +check_ring_op(uint32_t exp, uint32_t res, uint32_t lc, + enum rte_ring_queue_behavior bhv, const char *fname, const char *opname, + const struct rte_soring *sor) +{ + if ((bhv == RTE_RING_QUEUE_FIXED && exp != res) || + (bhv == RTE_RING_QUEUE_VARIABLE && exp < res)) { + rte_spinlock_lock(&dump_lock); + printf("%s(lc=%u) failure: %s expected: %u, returned %u\n", + fname, lc, opname, exp, res); + rte_soring_dump(stdout, sor); + rte_spinlock_unlock(&dump_lock); + return -ENOSPC; + } + return 0; +} + +/* num in interval [7/8, 11/8] of BULK_NUM */ +static inline uint32_t +rand_elem_num(void) +{ + uint32_t num; + + num = 7 * BULK_NUM / 8 + rte_rand() % (BULK_NUM / 2); + return num; +} + +/* + * for each enabled stage do: + * acquire burst of objects + * read and check their contents + * update and check their contents + * release burst of objects + * done + */ +static int32_t +test_worker_stages(struct lcore_arg *la, uint32_t lc, const char *fname, + struct ring_elem *obj[2 * BULK_NUM], + const struct ring_elem *def_elm, const struct ring_elem *loc_elm, + const struct ring_elem stg_elm[MAX_STAGES], int32_t prcs) +{ + int32_t rc; + uint32_t i, n, num, tkn; + uint64_t tm0, tm1; + const struct ring_elem *celm, *pelm; + + num = rand_elem_num(); + + rc = 0; + tkn = 0; + for (i = 0, pelm = def_elm; i != la->stats.nb_stage; pelm = celm, i++) { + + celm = stg_elm + i; + + /* given stage is not enabled on that lcore */ + if ((la->stats.role_mask & ROLE_STAGE(i)) == 0) + continue; + + /* reset all pointer values */ + memset(obj, 0, sizeof(*obj) * num); + + /* acquire num elems */ + tm0 = (prcs != 0) ? rte_rdtsc_precise() : 0; + n = _st_ring_acquire_burst(la->rng, i, (void **)obj, num, + &tkn, NULL); + tm0 = (prcs != 0) ? rte_rdtsc_precise() - tm0 : 0; + + /* check return value and objects */ + rc = check_ring_op(num, n, lc, RTE_RING_QUEUE_VARIABLE, fname, + RTE_STR(_st_ring_stage_acquire), la->rng); + if (rc == 0) + rc = check_updt_elem(obj, n, pelm, loc_elm, fname, + RTE_STR(_st_ring_stage_acquire), la->rng); + if (rc != 0) + break; + + /* release num elems */ + rte_compiler_barrier(); + rc = check_updt_elem(obj, n, loc_elm, celm, fname, + RTE_STR(_st_ring_stage_release), la->rng); + if (rc != 0) + break; + + if (n == 0) + tm1 = 0; + else { + tm1 = (prcs != 0) ? rte_rdtsc_precise() : 0; + _st_ring_release(la->rng, i, tkn, + (void **)obj, n); + tm1 = (prcs != 0) ? rte_rdtsc_precise() - tm1 : 0; + } + lcore_op_stat_update(la->stats.stage + i, 1, n, tm0 + tm1, + prcs); + } + + return rc; +} + +static int32_t +test_worker_deqenq(struct lcore_arg *la, uint32_t lc, const char *fname, + struct ring_elem *obj[2 * BULK_NUM], + const struct ring_elem *def_elm, const struct ring_elem *loc_elm, + const struct ring_elem *pelm, int32_t prcs) +{ + int32_t rc; + uint32_t k, n, num; + uint64_t tm0, tm1; + + num = rand_elem_num(); + + /* reset all pointer values */ + memset(obj, 0, sizeof(*obj) * num); + + /* dequeue num elems */ + tm0 = (prcs != 0) ? rte_rdtsc_precise() : 0; + n = _st_ring_dequeue_burst(la->rng, (void **)obj, num, NULL); + + tm0 = (prcs != 0) ? rte_rdtsc_precise() - tm0 : 0; + + /* check return value and objects */ + rc = check_ring_op(num, n, lc, RTE_RING_QUEUE_VARIABLE, fname, + RTE_STR(_st_ring_dequeue_bulk), la->rng); + if (rc == 0) + rc = check_updt_elem(obj, n, pelm, loc_elm, fname, + RTE_STR(_st_ring_dequeue_bulk), la->rng); + if (rc != 0) + return rc; + + /* enqueue n elems */ + rte_compiler_barrier(); + rc = check_updt_elem(obj, n, loc_elm, def_elm, fname, + RTE_STR(_st_ring_enqueue_bulk), la->rng); + if (rc != 0) + return rc; + + tm1 = (prcs != 0) ? rte_rdtsc_precise() : 0; + k = _st_ring_enqueue_bulk(la->rng, (void **)obj, n, NULL); + tm1 = (prcs != 0) ? rte_rdtsc_precise() - tm1 : 0; + + /* check return value */ + rc = check_ring_op(n, k, lc, RTE_RING_QUEUE_FIXED, fname, + RTE_STR(_st_ring_enqueue_bulk), la->rng); + if (rc != 0) + return rc; + + lcore_op_stat_update(&la->stats.deqenq, 1, n, tm0 + tm1, prcs); + return 0; +} + +static int +test_worker(void *arg, const char *fname, int32_t prcs) +{ + int32_t rc; + uint32_t i, lc; + uint64_t cl; + struct lcore_arg *la; + struct ring_elem *obj[2 * BULK_NUM]; + struct ring_elem *pelm, def_elm, loc_elm, stg_elm[MAX_STAGES]; + + la = arg; + lc = rte_lcore_id(); + + fill_ring_elm(&def_elm, UINT32_MAX); + fill_ring_elm(&loc_elm, lc); + + for (i = 0; i != RTE_DIM(stg_elm); i++) + fill_ring_elm(stg_elm + i, (i + 1) << 24); + + pelm = stg_elm + la->stats.nb_stage - 1; + + /* Acquire ordering is not required as the main is not + * really releasing any data through 'wrk_cmd' to + * the worker. + */ + while (__atomic_load_n(&wrk_cmd, __ATOMIC_RELAXED) != WRK_CMD_RUN) + rte_pause(); + + cl = rte_rdtsc_precise(); + + do { + if ((la->stats.role_mask & ~ROLE_DEQENQ) != 0) { + rc = test_worker_stages(la, lc, fname, obj, + &def_elm, &loc_elm, stg_elm, prcs); + if (rc != 0) + break; + } + + if ((la->stats.role_mask & ROLE_DEQENQ) != 0) { + rc = test_worker_deqenq(la, lc, fname, obj, + &def_elm, &loc_elm, pelm, prcs); + if (rc != 0) + break; + } + + } while (__atomic_load_n(&wrk_cmd, __ATOMIC_RELAXED) == WRK_CMD_RUN); + + cl = rte_rdtsc_precise() - cl; + if (prcs == 0) + lcore_stat_update(&la->stats, 0, 0, cl, 0); + la->stats.nb_cycle = cl; + return rc; +} +static int +test_worker_prcs(void *arg) +{ + return test_worker(arg, __func__, 1); +} + +static int +test_worker_avg(void *arg) +{ + return test_worker(arg, __func__, 0); +} + +static void +mt1_fini(struct rte_soring *rng, void *data) +{ + rte_free(rng); + rte_free(data); +} + +static int +mt1_init(struct rte_soring **rng, void **data, uint32_t num, + enum rte_ring_sync_type prod_synt, enum rte_ring_sync_type cons_synt, + uint32_t nb_stages) +{ + int32_t rc; + size_t sz; + uint32_t i; + struct rte_soring *r; + struct ring_elem *elm; + void *p; + struct rte_soring_param prm; + + *rng = NULL; + *data = NULL; + + sz = num * sizeof(*elm); + elm = rte_zmalloc(NULL, sz, alignof(typeof(*elm))); + if (elm == NULL) { + printf("%s: alloc(%zu) for %u elems data failed", + __func__, sz, num); + return -ENOMEM; + } + + *data = elm; + + /* alloc soring */ + memset(&prm, 0, sizeof(prm)); + + prm.name = __func__; + prm.elems = num; + prm.elem_size = sizeof(uintptr_t); + prm.stages = nb_stages; + prm.prod_synt = prod_synt; + prm.cons_synt = cons_synt; + + sz = rte_soring_get_memsize(&prm); + r = rte_zmalloc(NULL, sz, RTE_CACHE_LINE_SIZE); + if (r == NULL) { + printf("%s: alloc(%zu) for FIFO with %u elems failed", + __func__, sz, prm.elems); + return -ENOMEM; + } + + *rng = r; + + rc = rte_soring_init(r, &prm); + if (rc != 0) { + printf("%s: rte_soring_init(r=%p,elems=%u,stages=%u) failed, " + "error: %d(%s)\n", + __func__, r, prm.elems, prm.stages, rc, strerror(-rc)); + return rc; + } + + for (i = 0; i != num; i++) { + fill_ring_elm(elm + i, UINT32_MAX); + p = elm + i; + if (_st_ring_enqueue_bulk(r, &p, 1, NULL) != 1) + break; + } + + if (i != num) { + printf("%s: _st_ring_enqueue(%p, %u) returned %u\n", + __func__, r, num, i); + return -ENOSPC; + } + + return 0; +} + +static int +test_mt(int (*test)(void *), enum rte_ring_sync_type prod_synt, + enum rte_ring_sync_type cons_synt, uint32_t nb_stage, + const uint32_t role_mask[RTE_MAX_LCORE]) +{ + int32_t rc; + uint32_t i, lc, mc; + struct rte_soring *r; + void *data; + struct lcore_arg arg[RTE_MAX_LCORE]; + + static const struct lcore_op_stat init_stat = { + .min_cycle = UINT64_MAX, + }; + + rc = mt1_init(&r, &data, RING_SIZE, prod_synt, cons_synt, nb_stage); + + if (rc != 0) { + mt1_fini(r, data); + return rc; + } + + memset(arg, 0, sizeof(arg)); + + /* launch on all workers */ + RTE_LCORE_FOREACH_WORKER(lc) { + arg[lc].rng = r; + arg[lc].stats.deqenq = init_stat; + arg[lc].stats.nb_stage = nb_stage; + arg[lc].stats.role_mask = role_mask[lc]; + for (i = 0; i != arg[lc].stats.nb_stage; i++) + arg[lc].stats.stage[i] = init_stat; + rte_eal_remote_launch(test, &arg[lc], lc); + } + + /* signal workers to start test */ + __atomic_store_n(&wrk_cmd, WRK_CMD_RUN, __ATOMIC_RELEASE); + + rte_delay_us(run_time * US_PER_S); + + /* signal workers to stop test */ + __atomic_store_n(&wrk_cmd, WRK_CMD_STOP, __ATOMIC_RELEASE); + + /* wait for workers and collect stats. */ + mc = rte_lcore_id(); + arg[mc].stats.deqenq = init_stat; + arg[mc].stats.nb_stage = nb_stage; + for (i = 0; i != arg[mc].stats.nb_stage; i++) + arg[mc].stats.stage[i] = init_stat; + + rc = 0; + RTE_LCORE_FOREACH_WORKER(lc) { + rc |= rte_eal_wait_lcore(lc); + lcore_stat_aggr(&arg[mc].stats, &arg[lc].stats); + if (verbose != 0) + lcore_stat_dump(stdout, lc, &arg[lc].stats); + } + + lcore_stat_dump(stdout, UINT32_MAX, &arg[mc].stats); + rte_soring_dump(stdout, r); + mt1_fini(r, data); + return rc; +} + +/* + * launch all stages and deq+enq on all worker lcores + */ +static void +role_mask_sym(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE]) +{ + uint32_t lc; + const uint32_t msk = RTE_BIT32(nb_stage + 2) - 1; + + memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE); + RTE_LCORE_FOREACH_WORKER(lc) + role_mask[lc] = msk; +} + +/* + * Divide all workers in two (nearly) equal groups: + * - workers from 'even' group do deque+enque + * - workers from 'odd' group do acquire/release (for all stages) + */ +static void +role_mask_even_odd(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE]) +{ + uint32_t i, lc; + const uint32_t msk[2] = { + [0] = ROLE_DEQENQ, + [1] = RTE_GENMASK32(nb_stage + 1, 1), + }; + + memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE); + + i = 0; + RTE_LCORE_FOREACH_WORKER(lc) { + role_mask[lc] = msk[i & 1]; + i++; + } + if (i == 1) { + lc = rte_get_next_lcore(-1, 1, 0); + role_mask[lc] |= msk[i & 1]; + } +} + +/* + * Divide all workers (nearly) evenly among all possible stages + */ +static void +role_mask_div(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE]) +{ + uint32_t i, lc; + uint32_t msk[nb_stage + 1]; + + memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE); + + for (i = 0; i != RTE_DIM(msk); i++) { + msk[i] = RTE_BIT32(i); + }; + + i = 0; + RTE_LCORE_FOREACH_WORKER(lc) { + role_mask[lc] = msk[i % RTE_DIM(msk)]; + i++; + } + if (i < RTE_DIM(msk)) { + lc = rte_get_next_lcore(-1, 1, 0); + for (; i != RTE_DIM(msk); i++) + role_mask[lc] |= msk[i % RTE_DIM(msk)]; + } +} + +/* + * one worker does ST enqueue+dequeue, while all others - stages processing. + */ +static void +role_mask_denq_st(uint32_t nb_stage, uint32_t role_mask[RTE_MAX_LCORE]) +{ + uint32_t i, lc; + const uint32_t msk[2] = { + [0] = ROLE_DEQENQ, + [1] = RTE_GENMASK32(nb_stage + 1, 1), + }; + + memset(role_mask, 0, sizeof(role_mask[0]) * RTE_MAX_LCORE); + + i = 0; + RTE_LCORE_FOREACH_WORKER(lc) { + if (i == 0) + role_mask[lc] = msk[0]; + else + role_mask[lc] = msk[1]; + i++; + } + if (i == 1) { + lc = rte_get_next_lcore(-1, 1, 0); + role_mask[lc] |= msk[1]; + } +} + + +static int +test_sym_mt1(int (*test)(void *)) +{ + uint32_t role_mask[RTE_MAX_LCORE]; + const uint32_t nb_stage = 1; + + role_mask_sym(nb_stage, role_mask); + return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT, + nb_stage, role_mask); +} + +static int +test_sym_mt4(int (*test)(void *)) +{ + uint32_t role_mask[RTE_MAX_LCORE]; + + const uint32_t nb_stage = 4; + + role_mask_sym(nb_stage, role_mask); + return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT, + nb_stage, role_mask); +} + +static int +test_sym_mt_rts4(int (*test)(void *)) +{ + uint32_t role_mask[RTE_MAX_LCORE]; + + const uint32_t nb_stage = 4; + + role_mask_sym(nb_stage, role_mask); + return test_mt(test, RTE_RING_SYNC_MT_RTS, RTE_RING_SYNC_MT_RTS, + nb_stage, role_mask); +} + +static int +test_sym_mt_hts4(int (*test)(void *)) +{ + uint32_t role_mask[RTE_MAX_LCORE]; + + const uint32_t nb_stage = 4; + + role_mask_sym(nb_stage, role_mask); + return test_mt(test, RTE_RING_SYNC_MT_HTS, RTE_RING_SYNC_MT_HTS, + nb_stage, role_mask); +} + +static int +test_stdenq_stage4(int (*test)(void *)) +{ + uint32_t role_mask[RTE_MAX_LCORE]; + + const uint32_t nb_stage = 4; + + role_mask_denq_st(nb_stage, role_mask); + return test_mt(test, RTE_RING_SYNC_ST, RTE_RING_SYNC_ST, + nb_stage, role_mask); +} + + +static int +test_even_odd_mt5(int (*test)(void *)) +{ + uint32_t role_mask[RTE_MAX_LCORE]; + + const uint32_t nb_stage = 5; + + role_mask_even_odd(nb_stage, role_mask); + return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT, + nb_stage, role_mask); +} + +static int +test_div_mt3(int (*test)(void *)) +{ + uint32_t role_mask[RTE_MAX_LCORE]; + + const uint32_t nb_stage = 3; + + role_mask_div(nb_stage, role_mask); + return test_mt(test, RTE_RING_SYNC_MT, RTE_RING_SYNC_MT, + nb_stage, role_mask); +} + +static const struct test_case tests[] = { + { + .name = "MT_DEQENQ-MT_STG1-PRCS", + .func = test_sym_mt1, + .wfunc = test_worker_prcs, + }, + { + .name = "MT_DEQENQ-MT_STG1-AVG", + .func = test_sym_mt1, + .wfunc = test_worker_avg, + }, + { + .name = "MT_DEQENQ-MT_STG4-PRCS", + .func = test_sym_mt4, + .wfunc = test_worker_prcs, + }, + { + .name = "MT_DEQENQ-MT_STG4-AVG", + .func = test_sym_mt4, + .wfunc = test_worker_avg, + }, + { + .name = "MTRTS_DEQENQ-MT_STG4-PRCS", + .func = test_sym_mt_rts4, + .wfunc = test_worker_prcs, + }, + { + .name = "MTRTS_DEQENQ-MT_STG4-AVG", + .func = test_sym_mt_rts4, + .wfunc = test_worker_avg, + }, + { + .name = "MTHTS_DEQENQ-MT_STG4-PRCS", + .func = test_sym_mt_hts4, + .wfunc = test_worker_prcs, + }, + { + .name = "MTHTS_DEQENQ-MT_STG4-AVG", + .func = test_sym_mt_hts4, + .wfunc = test_worker_avg, + }, + { + .name = "MT_DEQENQ-MT_STG5-1:1-PRCS", + .func = test_even_odd_mt5, + .wfunc = test_worker_prcs, + }, + { + .name = "MT_DEQENQ-MT_STG5-1:1-AVG", + .func = test_even_odd_mt5, + .wfunc = test_worker_avg, + }, + { + .name = "MT_DEQENQ-MT_STG3-1:3-PRCS", + .func = test_div_mt3, + .wfunc = test_worker_prcs, + }, + { + .name = "MT_DEQENQ_MT_STG3-1:3-AVG", + .func = test_div_mt3, + .wfunc = test_worker_avg, + }, + { + .name = "ST_DEQENQ-MT_STG4-PRCS", + .func = test_stdenq_stage4, + .wfunc = test_worker_prcs, + }, + { + .name = "ST_DEQENQ-MT_STG4-AVG", + .func = test_stdenq_stage4, + .wfunc = test_worker_avg, + }, +}; From patchwork Wed Oct 30 21:23:04 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Konstantin Ananyev X-Patchwork-Id: 147767 X-Patchwork-Delegate: thomas@monjalon.net Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id A94E045B82; Wed, 30 Oct 2024 21:33:32 +0100 (CET) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id D4DB6433ED; Wed, 30 Oct 2024 21:33:23 +0100 (CET) Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) by mails.dpdk.org (Postfix) with ESMTP id AC8BD433C4 for ; Wed, 30 Oct 2024 21:33:05 +0100 (CET) Received: from mail.maildlp.com (unknown [172.18.186.231]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4XdzKC1Hz7z6K6V2; Thu, 31 Oct 2024 04:30:39 +0800 (CST) Received: from frapeml500007.china.huawei.com (unknown [7.182.85.172]) by mail.maildlp.com (Postfix) with ESMTPS id 3F320140B38; Thu, 31 Oct 2024 04:33:05 +0800 (CST) Received: from localhost.localdomain (10.220.239.45) by frapeml500007.china.huawei.com (7.182.85.172) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.1.2507.39; Wed, 30 Oct 2024 21:33:04 +0100 From: Konstantin Ananyev To: CC: , , , , , , , , Subject: [PATCH v7 7/7] test: add stress test suite Date: Wed, 30 Oct 2024 17:23:04 -0400 Message-ID: <20241030212304.104180-8-konstantin.ananyev@huawei.com> X-Mailer: git-send-email 2.35.3 In-Reply-To: <20241030212304.104180-1-konstantin.ananyev@huawei.com> References: <20241021174745.1843-1-konstantin.ananyev@huawei.com> <20241030212304.104180-1-konstantin.ananyev@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.220.239.45] X-ClientProxiedBy: frapeml100006.china.huawei.com (7.182.85.201) To frapeml500007.china.huawei.com (7.182.85.172) X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Add a new test suite which purpose is to run 'stress' tests: main purpose is put a pressure to dpdk sync algorithms to flag their misbehaving/slowdown/etc. Right now it consists from just 2 test-cases: meson test --suite stress-tests --list DPDK:stress-tests / ring_stress_autotest DPDK:stress-tests / soring_stress_autotest These tests are quite time consuming (~15 mins each), that's another reason to put them into a separate test-suite. Signed-off-by: Konstantin Ananyev --- app/test/suites/meson.build | 10 ++++++++++ app/test/test.h | 1 + app/test/test_ring_stress.c | 2 +- app/test/test_soring_stress.c | 2 +- 4 files changed, 13 insertions(+), 2 deletions(-) diff --git a/app/test/suites/meson.build b/app/test/suites/meson.build index 191702cf76..e482373330 100644 --- a/app/test/suites/meson.build +++ b/app/test/suites/meson.build @@ -5,6 +5,7 @@ # to complete, so timeout to 10 minutes timeout_seconds = 600 timeout_seconds_fast = 10 +timeout_seconds_stress = 900 test_no_huge_args = ['--no-huge', '-m', '2048'] has_hugepage = run_command(has_hugepages_cmd, check: true).stdout().strip() != '0' @@ -21,6 +22,7 @@ endif # - fast_tests # - perf_tests # - driver_tests +# - stress_tests test_suites = run_command(get_test_suites_cmd, autotest_sources, check: true).stdout().strip().split() foreach suite:test_suites @@ -39,6 +41,14 @@ foreach suite:test_suites timeout: timeout_seconds, is_parallel: false) endforeach + elif suite_name == 'stress-tests' + foreach t: suite_tests + test(t, dpdk_test, + env: ['DPDK_TEST=' + t], + timeout: timeout_seconds_stress, + is_parallel: false, + suite: suite_name) + endforeach elif suite_name != 'fast-tests' # simple cases - tests without parameters or special handling foreach t: suite_tests diff --git a/app/test/test.h b/app/test/test.h index 15e23d297f..ebc4864bf8 100644 --- a/app/test/test.h +++ b/app/test/test.h @@ -208,5 +208,6 @@ void add_test_command(struct test_command *t); #define REGISTER_FAST_TEST(cmd, no_huge, ASan, func) REGISTER_TEST_COMMAND(cmd, func) #define REGISTER_PERF_TEST REGISTER_TEST_COMMAND #define REGISTER_DRIVER_TEST REGISTER_TEST_COMMAND +#define REGISTER_STRESS_TEST REGISTER_TEST_COMMAND #endif diff --git a/app/test/test_ring_stress.c b/app/test/test_ring_stress.c index 1af45e0fc8..82e19b02c3 100644 --- a/app/test/test_ring_stress.c +++ b/app/test/test_ring_stress.c @@ -63,4 +63,4 @@ test_ring_stress(void) return (k != n); } -REGISTER_TEST_COMMAND(ring_stress_autotest, test_ring_stress); +REGISTER_STRESS_TEST(ring_stress_autotest, test_ring_stress); diff --git a/app/test/test_soring_stress.c b/app/test/test_soring_stress.c index 334af6a29c..e5655d49cb 100644 --- a/app/test/test_soring_stress.c +++ b/app/test/test_soring_stress.c @@ -45,4 +45,4 @@ test_ring_stress(void) return (k != n); } -REGISTER_TEST_COMMAND(soring_stress_autotest, test_ring_stress); +REGISTER_STRESS_TEST(soring_stress_autotest, test_ring_stress);