return __atomic_and_fetch(&bktp->w1, v, __ATOMIC_ACQ_REL);
}
+static inline uint64_t
+cnxk_tim_bkt_fast_mod(uint64_t n, uint64_t d, struct rte_reciprocal_u64 R)
+{
+ return (n - (d * rte_reciprocal_divide_u64(n, &R)));
+}
+
+static __rte_always_inline void
+cnxk_tim_get_target_bucket(struct cnxk_tim_ring *const tim_ring,
+ const uint32_t rel_bkt, struct cnxk_tim_bkt **bkt,
+ struct cnxk_tim_bkt **mirr_bkt)
+{
+ const uint64_t bkt_cyc = cnxk_tim_cntvct() - tim_ring->ring_start_cyc;
+ uint64_t bucket =
+ rte_reciprocal_divide_u64(bkt_cyc, &tim_ring->fast_div) +
+ rel_bkt;
+ uint64_t mirr_bucket = 0;
+
+ bucket = cnxk_tim_bkt_fast_mod(bucket, tim_ring->nb_bkts,
+ tim_ring->fast_bkt);
+ mirr_bucket =
+ cnxk_tim_bkt_fast_mod(bucket + (tim_ring->nb_bkts >> 1),
+ tim_ring->nb_bkts, tim_ring->fast_bkt);
+ *bkt = &tim_ring->bkt[bucket];
+ *mirr_bkt = &tim_ring->bkt[mirr_bucket];
+}
+
+static struct cnxk_tim_ent *
+cnxk_tim_clr_bkt(struct cnxk_tim_ring *const tim_ring,
+ struct cnxk_tim_bkt *const bkt)
+{
+#define TIM_MAX_OUTSTANDING_OBJ 64
+ void *pend_chunks[TIM_MAX_OUTSTANDING_OBJ];
+ struct cnxk_tim_ent *chunk;
+ struct cnxk_tim_ent *pnext;
+ uint8_t objs = 0;
+
+ chunk = ((struct cnxk_tim_ent *)(uintptr_t)bkt->first_chunk);
+ chunk = (struct cnxk_tim_ent *)(uintptr_t)(chunk +
+ tim_ring->nb_chunk_slots)
+ ->w0;
+ while (chunk) {
+ pnext = (struct cnxk_tim_ent *)(uintptr_t)(
+ (chunk + tim_ring->nb_chunk_slots)->w0);
+ if (objs == TIM_MAX_OUTSTANDING_OBJ) {
+ rte_mempool_put_bulk(tim_ring->chunk_pool, pend_chunks,
+ objs);
+ objs = 0;
+ }
+ pend_chunks[objs++] = chunk;
+ chunk = pnext;
+ }
+
+ if (objs)
+ rte_mempool_put_bulk(tim_ring->chunk_pool, pend_chunks, objs);
+
+ return (struct cnxk_tim_ent *)(uintptr_t)bkt->first_chunk;
+}
+
+static struct cnxk_tim_ent *
+cnxk_tim_refill_chunk(struct cnxk_tim_bkt *const bkt,
+ struct cnxk_tim_bkt *const mirr_bkt,
+ struct cnxk_tim_ring *const tim_ring)
+{
+ struct cnxk_tim_ent *chunk;
+
+ if (bkt->nb_entry || !bkt->first_chunk) {
+ if (unlikely(rte_mempool_get(tim_ring->chunk_pool,
+ (void **)&chunk)))
+ return NULL;
+ if (bkt->nb_entry) {
+ *(uint64_t *)(((struct cnxk_tim_ent *)
+ mirr_bkt->current_chunk) +
+ tim_ring->nb_chunk_slots) =
+ (uintptr_t)chunk;
+ } else {
+ bkt->first_chunk = (uintptr_t)chunk;
+ }
+ } else {
+ chunk = cnxk_tim_clr_bkt(tim_ring, bkt);
+ bkt->first_chunk = (uintptr_t)chunk;
+ }
+ *(uint64_t *)(chunk + tim_ring->nb_chunk_slots) = 0;
+
+ return chunk;
+}
+
+static struct cnxk_tim_ent *
+cnxk_tim_insert_chunk(struct cnxk_tim_bkt *const bkt,
+ struct cnxk_tim_bkt *const mirr_bkt,
+ struct cnxk_tim_ring *const tim_ring)
+{
+ struct cnxk_tim_ent *chunk;
+
+ if (unlikely(rte_mempool_get(tim_ring->chunk_pool, (void **)&chunk)))
+ return NULL;
+
+ *(uint64_t *)(chunk + tim_ring->nb_chunk_slots) = 0;
+ if (bkt->nb_entry) {
+ *(uint64_t *)(((struct cnxk_tim_ent *)(uintptr_t)
+ mirr_bkt->current_chunk) +
+ tim_ring->nb_chunk_slots) = (uintptr_t)chunk;
+ } else {
+ bkt->first_chunk = (uintptr_t)chunk;
+ }
+ return chunk;
+}
+
+static __rte_always_inline int
+cnxk_tim_add_entry_sp(struct cnxk_tim_ring *const tim_ring,
+ const uint32_t rel_bkt, struct rte_event_timer *const tim,
+ const struct cnxk_tim_ent *const pent,
+ const uint8_t flags)
+{
+ struct cnxk_tim_bkt *mirr_bkt;
+ struct cnxk_tim_ent *chunk;
+ struct cnxk_tim_bkt *bkt;
+ uint64_t lock_sema;
+ int16_t rem;
+
+__retry:
+ cnxk_tim_get_target_bucket(tim_ring, rel_bkt, &bkt, &mirr_bkt);
+
+ /* Get Bucket sema*/
+ lock_sema = cnxk_tim_bkt_fetch_sema_lock(bkt);
+
+ /* Bucket related checks. */
+ if (unlikely(cnxk_tim_bkt_get_hbt(lock_sema))) {
+ if (cnxk_tim_bkt_get_nent(lock_sema) != 0) {
+ uint64_t hbt_state;
+#ifdef RTE_ARCH_ARM64
+ asm volatile(PLT_CPU_FEATURE_PREAMBLE
+ " ldxr %[hbt], [%[w1]] \n"
+ " tbz %[hbt], 33, dne%= \n"
+ " sevl \n"
+ "rty%=: wfe \n"
+ " ldxr %[hbt], [%[w1]] \n"
+ " tbnz %[hbt], 33, rty%= \n"
+ "dne%=: \n"
+ : [hbt] "=&r"(hbt_state)
+ : [w1] "r"((&bkt->w1))
+ : "memory");
+#else
+ do {
+ hbt_state = __atomic_load_n(&bkt->w1,
+ __ATOMIC_RELAXED);
+ } while (hbt_state & BIT_ULL(33));
+#endif
+
+ if (!(hbt_state & BIT_ULL(34))) {
+ cnxk_tim_bkt_dec_lock(bkt);
+ goto __retry;
+ }
+ }
+ }
+ /* Insert the work. */
+ rem = cnxk_tim_bkt_fetch_rem(lock_sema);
+
+ if (!rem) {
+ if (flags & CNXK_TIM_ENA_FB)
+ chunk = cnxk_tim_refill_chunk(bkt, mirr_bkt, tim_ring);
+ if (flags & CNXK_TIM_ENA_DFB)
+ chunk = cnxk_tim_insert_chunk(bkt, mirr_bkt, tim_ring);
+
+ if (unlikely(chunk == NULL)) {
+ bkt->chunk_remainder = 0;
+ tim->impl_opaque[0] = 0;
+ tim->impl_opaque[1] = 0;
+ tim->state = RTE_EVENT_TIMER_ERROR;
+ cnxk_tim_bkt_dec_lock(bkt);
+ return -ENOMEM;
+ }
+ mirr_bkt->current_chunk = (uintptr_t)chunk;
+ bkt->chunk_remainder = tim_ring->nb_chunk_slots - 1;
+ } else {
+ chunk = (struct cnxk_tim_ent *)mirr_bkt->current_chunk;
+ chunk += tim_ring->nb_chunk_slots - rem;
+ }
+
+ /* Copy work entry. */
+ *chunk = *pent;
+
+ tim->impl_opaque[0] = (uintptr_t)chunk;
+ tim->impl_opaque[1] = (uintptr_t)bkt;
+ __atomic_store_n(&tim->state, RTE_EVENT_TIMER_ARMED, __ATOMIC_RELEASE);
+ cnxk_tim_bkt_inc_nent(bkt);
+ cnxk_tim_bkt_dec_lock_relaxed(bkt);
+
+ return 0;
+}
+
+static __rte_always_inline int
+cnxk_tim_add_entry_mp(struct cnxk_tim_ring *const tim_ring,
+ const uint32_t rel_bkt, struct rte_event_timer *const tim,
+ const struct cnxk_tim_ent *const pent,
+ const uint8_t flags)
+{
+ struct cnxk_tim_bkt *mirr_bkt;
+ struct cnxk_tim_ent *chunk;
+ struct cnxk_tim_bkt *bkt;
+ uint64_t lock_sema;
+ int16_t rem;
+
+__retry:
+ cnxk_tim_get_target_bucket(tim_ring, rel_bkt, &bkt, &mirr_bkt);
+ /* Get Bucket sema*/
+ lock_sema = cnxk_tim_bkt_fetch_sema_lock(bkt);
+
+ /* Bucket related checks. */
+ if (unlikely(cnxk_tim_bkt_get_hbt(lock_sema))) {
+ if (cnxk_tim_bkt_get_nent(lock_sema) != 0) {
+ uint64_t hbt_state;
+#ifdef RTE_ARCH_ARM64
+ asm volatile(PLT_CPU_FEATURE_PREAMBLE
+ " ldxr %[hbt], [%[w1]] \n"
+ " tbz %[hbt], 33, dne%= \n"
+ " sevl \n"
+ "rty%=: wfe \n"
+ " ldxr %[hbt], [%[w1]] \n"
+ " tbnz %[hbt], 33, rty%= \n"
+ "dne%=: \n"
+ : [hbt] "=&r"(hbt_state)
+ : [w1] "r"((&bkt->w1))
+ : "memory");
+#else
+ do {
+ hbt_state = __atomic_load_n(&bkt->w1,
+ __ATOMIC_RELAXED);
+ } while (hbt_state & BIT_ULL(33));
+#endif
+
+ if (!(hbt_state & BIT_ULL(34))) {
+ cnxk_tim_bkt_dec_lock(bkt);
+ goto __retry;
+ }
+ }
+ }
+
+ rem = cnxk_tim_bkt_fetch_rem(lock_sema);
+ if (rem < 0) {
+ cnxk_tim_bkt_dec_lock(bkt);
+#ifdef RTE_ARCH_ARM64
+ asm volatile(PLT_CPU_FEATURE_PREAMBLE
+ " ldxr %[rem], [%[crem]] \n"
+ " tbz %[rem], 63, dne%= \n"
+ " sevl \n"
+ "rty%=: wfe \n"
+ " ldxr %[rem], [%[crem]] \n"
+ " tbnz %[rem], 63, rty%= \n"
+ "dne%=: \n"
+ : [rem] "=&r"(rem)
+ : [crem] "r"(&bkt->w1)
+ : "memory");
+#else
+ while (__atomic_load_n((int64_t *)&bkt->w1, __ATOMIC_RELAXED) <
+ 0)
+ ;
+#endif
+ goto __retry;
+ } else if (!rem) {
+ /* Only one thread can be here*/
+ if (flags & CNXK_TIM_ENA_FB)
+ chunk = cnxk_tim_refill_chunk(bkt, mirr_bkt, tim_ring);
+ if (flags & CNXK_TIM_ENA_DFB)
+ chunk = cnxk_tim_insert_chunk(bkt, mirr_bkt, tim_ring);
+
+ if (unlikely(chunk == NULL)) {
+ tim->impl_opaque[0] = 0;
+ tim->impl_opaque[1] = 0;
+ tim->state = RTE_EVENT_TIMER_ERROR;
+ cnxk_tim_bkt_set_rem(bkt, 0);
+ cnxk_tim_bkt_dec_lock(bkt);
+ return -ENOMEM;
+ }
+ *chunk = *pent;
+ if (cnxk_tim_bkt_fetch_lock(lock_sema)) {
+ do {
+ lock_sema = __atomic_load_n(&bkt->w1,
+ __ATOMIC_RELAXED);
+ } while (cnxk_tim_bkt_fetch_lock(lock_sema) - 1);
+ }
+ rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
+ mirr_bkt->current_chunk = (uintptr_t)chunk;
+ __atomic_store_n(&bkt->chunk_remainder,
+ tim_ring->nb_chunk_slots - 1,
+ __ATOMIC_RELEASE);
+ } else {
+ chunk = (struct cnxk_tim_ent *)mirr_bkt->current_chunk;
+ chunk += tim_ring->nb_chunk_slots - rem;
+ *chunk = *pent;
+ }
+
+ tim->impl_opaque[0] = (uintptr_t)chunk;
+ tim->impl_opaque[1] = (uintptr_t)bkt;
+ __atomic_store_n(&tim->state, RTE_EVENT_TIMER_ARMED, __ATOMIC_RELEASE);
+ cnxk_tim_bkt_inc_nent(bkt);
+ cnxk_tim_bkt_dec_lock_relaxed(bkt);
+
+ return 0;
+}
+
#endif /* __CNXK_TIM_WORKER_H__ */
git.droids-corp.org / dpdk.git / blobdiff