Use std::atomic for CQ data.

There was a sub-optimality in the CAS operation. vjpai@ and
I decided to move to std::atomic.

This commit basically moves CQ data to C++ structures, and
makes grpc_cq_event_queue a proper c++ class called CQEventQueue.
pull/18067/head
Soheil Hassas Yeganeh 6 years ago
parent f16340db0b
commit 9dd94ea1fa
  1. 12
      src/core/lib/gprpp/atomic.h
  2. 280
      src/core/lib/surface/completion_queue.cc

@ -49,8 +49,9 @@ class Atomic {
bool CompareExchangeWeak(T* expected, T desired, MemoryOrder success, bool CompareExchangeWeak(T* expected, T desired, MemoryOrder success,
MemoryOrder failure) { MemoryOrder failure) {
return GPR_ATM_INC_CAS_THEN( return GPR_ATM_INC_CAS_THEN(storage_.compare_exchange_weak(
storage_.compare_exchange_weak(*expected, desired, success, failure)); *expected, desired, static_cast<std::memory_order>(success),
static_cast<std::memory_order>(failure)));
} }
bool CompareExchangeStrong(T* expected, T desired, MemoryOrder success, bool CompareExchangeStrong(T* expected, T desired, MemoryOrder success,
@ -74,7 +75,7 @@ class Atomic {
// Atomically increment a counter only if the counter value is not zero. // Atomically increment a counter only if the counter value is not zero.
// Returns true if increment took place; false if counter is zero. // Returns true if increment took place; false if counter is zero.
bool IncrementIfNonzero(MemoryOrder load_order = MemoryOrder::ACQ_REL) { bool IncrementIfNonzero(MemoryOrder load_order = MemoryOrder::ACQUIRE) {
T count = storage_.load(static_cast<std::memory_order>(load_order)); T count = storage_.load(static_cast<std::memory_order>(load_order));
do { do {
// If zero, we are done (without an increment). If not, we must do a CAS // If zero, we are done (without an increment). If not, we must do a CAS
@ -83,9 +84,8 @@ class Atomic {
if (count == 0) { if (count == 0) {
return false; return false;
} }
} while (!storage_.AtomicCompareExchangeWeak( } while (!CompareExchangeWeak(&count, count + 1, MemoryOrder::ACQ_REL,
&count, count + 1, static_cast<std::memory_order>(MemoryOrder::ACQ_REL), load_order));
static_cast<std::memory_order>(load_order)));
return true; return true;
} }

@ -33,6 +33,7 @@
#include "src/core/lib/gpr/spinlock.h" #include "src/core/lib/gpr/spinlock.h"
#include "src/core/lib/gpr/string.h" #include "src/core/lib/gpr/string.h"
#include "src/core/lib/gpr/tls.h" #include "src/core/lib/gpr/tls.h"
#include "src/core/lib/gprpp/atomic.h"
#include "src/core/lib/iomgr/pollset.h" #include "src/core/lib/iomgr/pollset.h"
#include "src/core/lib/iomgr/timer.h" #include "src/core/lib/iomgr/timer.h"
#include "src/core/lib/profiling/timers.h" #include "src/core/lib/profiling/timers.h"
@ -44,6 +45,8 @@ grpc_core::TraceFlag grpc_trace_operation_failures(false, "op_failure");
grpc_core::DebugOnlyTraceFlag grpc_trace_pending_tags(false, "pending_tags"); grpc_core::DebugOnlyTraceFlag grpc_trace_pending_tags(false, "pending_tags");
grpc_core::DebugOnlyTraceFlag grpc_trace_cq_refcount(false, "cq_refcount"); grpc_core::DebugOnlyTraceFlag grpc_trace_cq_refcount(false, "cq_refcount");
namespace {
// Specifies a cq thread local cache. // Specifies a cq thread local cache.
// The first event that occurs on a thread // The first event that occurs on a thread
// with a cq cache will go into that cache, and // with a cq cache will go into that cache, and
@ -84,24 +87,22 @@ typedef struct {
grpc_closure* shutdown; grpc_closure* shutdown;
} non_polling_poller; } non_polling_poller;
static size_t non_polling_poller_size(void) { size_t non_polling_poller_size(void) { return sizeof(non_polling_poller); }
return sizeof(non_polling_poller);
}
static void non_polling_poller_init(grpc_pollset* pollset, gpr_mu** mu) { void non_polling_poller_init(grpc_pollset* pollset, gpr_mu** mu) {
non_polling_poller* npp = reinterpret_cast<non_polling_poller*>(pollset); non_polling_poller* npp = reinterpret_cast<non_polling_poller*>(pollset);
gpr_mu_init(&npp->mu); gpr_mu_init(&npp->mu);
*mu = &npp->mu; *mu = &npp->mu;
} }
static void non_polling_poller_destroy(grpc_pollset* pollset) { void non_polling_poller_destroy(grpc_pollset* pollset) {
non_polling_poller* npp = reinterpret_cast<non_polling_poller*>(pollset); non_polling_poller* npp = reinterpret_cast<non_polling_poller*>(pollset);
gpr_mu_destroy(&npp->mu); gpr_mu_destroy(&npp->mu);
} }
static grpc_error* non_polling_poller_work(grpc_pollset* pollset, grpc_error* non_polling_poller_work(grpc_pollset* pollset,
grpc_pollset_worker** worker, grpc_pollset_worker** worker,
grpc_millis deadline) { grpc_millis deadline) {
non_polling_poller* npp = reinterpret_cast<non_polling_poller*>(pollset); non_polling_poller* npp = reinterpret_cast<non_polling_poller*>(pollset);
if (npp->shutdown) return GRPC_ERROR_NONE; if (npp->shutdown) return GRPC_ERROR_NONE;
if (npp->kicked_without_poller) { if (npp->kicked_without_poller) {
@ -141,8 +142,8 @@ static grpc_error* non_polling_poller_work(grpc_pollset* pollset,
return GRPC_ERROR_NONE; return GRPC_ERROR_NONE;
} }
static grpc_error* non_polling_poller_kick( grpc_error* non_polling_poller_kick(grpc_pollset* pollset,
grpc_pollset* pollset, grpc_pollset_worker* specific_worker) { grpc_pollset_worker* specific_worker) {
non_polling_poller* p = reinterpret_cast<non_polling_poller*>(pollset); non_polling_poller* p = reinterpret_cast<non_polling_poller*>(pollset);
if (specific_worker == nullptr) if (specific_worker == nullptr)
specific_worker = reinterpret_cast<grpc_pollset_worker*>(p->root); specific_worker = reinterpret_cast<grpc_pollset_worker*>(p->root);
@ -159,8 +160,7 @@ static grpc_error* non_polling_poller_kick(
return GRPC_ERROR_NONE; return GRPC_ERROR_NONE;
} }
static void non_polling_poller_shutdown(grpc_pollset* pollset, void non_polling_poller_shutdown(grpc_pollset* pollset, grpc_closure* closure) {
grpc_closure* closure) {
non_polling_poller* p = reinterpret_cast<non_polling_poller*>(pollset); non_polling_poller* p = reinterpret_cast<non_polling_poller*>(pollset);
GPR_ASSERT(closure != nullptr); GPR_ASSERT(closure != nullptr);
p->shutdown = closure; p->shutdown = closure;
@ -175,7 +175,7 @@ static void non_polling_poller_shutdown(grpc_pollset* pollset,
} }
} }
static const cq_poller_vtable g_poller_vtable_by_poller_type[] = { const cq_poller_vtable g_poller_vtable_by_poller_type[] = {
/* GRPC_CQ_DEFAULT_POLLING */ /* GRPC_CQ_DEFAULT_POLLING */
{true, true, grpc_pollset_size, grpc_pollset_init, grpc_pollset_kick, {true, true, grpc_pollset_size, grpc_pollset_init, grpc_pollset_kick,
grpc_pollset_work, grpc_pollset_shutdown, grpc_pollset_destroy}, grpc_pollset_work, grpc_pollset_shutdown, grpc_pollset_destroy},
@ -188,7 +188,9 @@ static const cq_poller_vtable g_poller_vtable_by_poller_type[] = {
non_polling_poller_shutdown, non_polling_poller_destroy}, non_polling_poller_shutdown, non_polling_poller_destroy},
}; };
typedef struct cq_vtable { } // namespace
struct cq_vtable {
grpc_cq_completion_type cq_completion_type; grpc_cq_completion_type cq_completion_type;
size_t data_size; size_t data_size;
void (*init)(void* data, void (*init)(void* data,
@ -203,80 +205,116 @@ typedef struct cq_vtable {
void* reserved); void* reserved);
grpc_event (*pluck)(grpc_completion_queue* cq, void* tag, grpc_event (*pluck)(grpc_completion_queue* cq, void* tag,
gpr_timespec deadline, void* reserved); gpr_timespec deadline, void* reserved);
} cq_vtable; };
namespace {
/* Queue that holds the cq_completion_events. Internally uses gpr_mpscq queue /* Queue that holds the cq_completion_events. Internally uses gpr_mpscq queue
* (a lockfree multiproducer single consumer queue). It uses a queue_lock * (a lockfree multiproducer single consumer queue). It uses a queue_lock
* to support multiple consumers. * to support multiple consumers.
* Only used in completion queues whose completion_type is GRPC_CQ_NEXT */ * Only used in completion queues whose completion_type is GRPC_CQ_NEXT */
typedef struct grpc_cq_event_queue { class CqEventQueue {
public:
CqEventQueue() { gpr_mpscq_init(&queue_); }
~CqEventQueue() { gpr_mpscq_destroy(&queue_); }
/* Note: The counter is not incremented/decremented atomically with push/pop.
* The count is only eventually consistent */
intptr_t num_items() const {
return num_queue_items_.Load(grpc_core::MemoryOrder::RELAXED);
}
bool Push(grpc_cq_completion* c);
grpc_cq_completion* Pop();
private:
/* Spinlock to serialize consumers i.e pop() operations */ /* Spinlock to serialize consumers i.e pop() operations */
gpr_spinlock queue_lock; gpr_spinlock queue_lock_ = GPR_SPINLOCK_INITIALIZER;
gpr_mpscq queue; gpr_mpscq queue_;
/* A lazy counter of number of items in the queue. This is NOT atomically /* A lazy counter of number of items in the queue. This is NOT atomically
incremented/decremented along with push/pop operations and hence is only incremented/decremented along with push/pop operations and hence is only
eventually consistent */ eventually consistent */
gpr_atm num_queue_items; grpc_core::Atomic<intptr_t> num_queue_items_{0};
} grpc_cq_event_queue; };
struct cq_next_data {
~cq_next_data() { GPR_ASSERT(queue.num_items() == 0); }
typedef struct cq_next_data {
/** Completed events for completion-queues of type GRPC_CQ_NEXT */ /** Completed events for completion-queues of type GRPC_CQ_NEXT */
grpc_cq_event_queue queue; CqEventQueue queue;
/** Counter of how many things have ever been queued on this completion queue /** Counter of how many things have ever been queued on this completion queue
useful for avoiding locks to check the queue */ useful for avoiding locks to check the queue */
gpr_atm things_queued_ever; grpc_core::Atomic<intptr_t> things_queued_ever{0};
/* Number of outstanding events (+1 if not shut down) */ /** Number of outstanding events (+1 if not shut down)
gpr_atm pending_events; Initial count is dropped by grpc_completion_queue_shutdown */
grpc_core::Atomic<intptr_t> pending_events{1};
/** 0 initially. 1 once we initiated shutdown */ /** 0 initially. 1 once we initiated shutdown */
bool shutdown_called; bool shutdown_called = false;
} cq_next_data; };
struct cq_pluck_data {
cq_pluck_data() {
completed_tail = &completed_head;
completed_head.next = reinterpret_cast<uintptr_t>(completed_tail);
}
~cq_pluck_data() {
GPR_ASSERT(completed_head.next ==
reinterpret_cast<uintptr_t>(&completed_head));
}
typedef struct cq_pluck_data {
/** Completed events for completion-queues of type GRPC_CQ_PLUCK */ /** Completed events for completion-queues of type GRPC_CQ_PLUCK */
grpc_cq_completion completed_head; grpc_cq_completion completed_head;
grpc_cq_completion* completed_tail; grpc_cq_completion* completed_tail;
/** Number of pending events (+1 if we're not shutdown) */ /** Number of pending events (+1 if we're not shutdown).
gpr_atm pending_events; Initial count is dropped by grpc_completion_queue_shutdown. */
grpc_core::Atomic<intptr_t> pending_events{1};
/** Counter of how many things have ever been queued on this completion queue /** Counter of how many things have ever been queued on this completion queue
useful for avoiding locks to check the queue */ useful for avoiding locks to check the queue */
gpr_atm things_queued_ever; grpc_core::Atomic<intptr_t> things_queued_ever{0};
/** 0 initially. 1 once we completed shutting */ /** 0 initially. 1 once we completed shutting */
/* TODO: (sreek) This is not needed since (shutdown == 1) if and only if /* TODO: (sreek) This is not needed since (shutdown == 1) if and only if
* (pending_events == 0). So consider removing this in future and use * (pending_events == 0). So consider removing this in future and use
* pending_events */ * pending_events */
gpr_atm shutdown; grpc_core::Atomic<bool> shutdown{false};
/** 0 initially. 1 once we initiated shutdown */ /** 0 initially. 1 once we initiated shutdown */
bool shutdown_called; bool shutdown_called = false;
int num_pluckers; int num_pluckers = 0;
plucker pluckers[GRPC_MAX_COMPLETION_QUEUE_PLUCKERS]; plucker pluckers[GRPC_MAX_COMPLETION_QUEUE_PLUCKERS];
} cq_pluck_data; };
typedef struct cq_callback_data { struct cq_callback_data {
cq_callback_data(
grpc_experimental_completion_queue_functor* shutdown_callback)
: shutdown_callback(shutdown_callback) {}
/** No actual completed events queue, unlike other types */ /** No actual completed events queue, unlike other types */
/** Number of pending events (+1 if we're not shutdown) */ /** Number of pending events (+1 if we're not shutdown).
gpr_atm pending_events; Initial count is dropped by grpc_completion_queue_shutdown. */
grpc_core::Atomic<intptr_t> pending_events{1};
/** Counter of how many things have ever been queued on this completion queue /** Counter of how many things have ever been queued on this completion queue
useful for avoiding locks to check the queue */ useful for avoiding locks to check the queue */
gpr_atm things_queued_ever; grpc_core::Atomic<intptr_t> things_queued_ever{0};
/** 0 initially. 1 once we initiated shutdown */ /** 0 initially. 1 once we initiated shutdown */
bool shutdown_called; bool shutdown_called = false;
/** A callback that gets invoked when the CQ completes shutdown */ /** A callback that gets invoked when the CQ completes shutdown */
grpc_experimental_completion_queue_functor* shutdown_callback; grpc_experimental_completion_queue_functor* shutdown_callback;
} cq_callback_data; };
} // namespace
/* Completion queue structure */ /* Completion queue structure */
struct grpc_completion_queue { struct grpc_completion_queue {
@ -408,7 +446,7 @@ int grpc_completion_queue_thread_local_cache_flush(grpc_completion_queue* cq,
storage->done(storage->done_arg, storage); storage->done(storage->done_arg, storage);
ret = 1; ret = 1;
cq_next_data* cqd = static_cast<cq_next_data*> DATA_FROM_CQ(cq); cq_next_data* cqd = static_cast<cq_next_data*> DATA_FROM_CQ(cq);
if (gpr_atm_full_fetch_add(&cqd->pending_events, -1) == 1) { if (cqd->pending_events.FetchSub(1, grpc_core::MemoryOrder::ACQ_REL) == 1) {
GRPC_CQ_INTERNAL_REF(cq, "shutting_down"); GRPC_CQ_INTERNAL_REF(cq, "shutting_down");
gpr_mu_lock(cq->mu); gpr_mu_lock(cq->mu);
cq_finish_shutdown_next(cq); cq_finish_shutdown_next(cq);
@ -422,31 +460,21 @@ int grpc_completion_queue_thread_local_cache_flush(grpc_completion_queue* cq,
return ret; return ret;
} }
static void cq_event_queue_init(grpc_cq_event_queue* q) { bool CqEventQueue::Push(grpc_cq_completion* c) {
gpr_mpscq_init(&q->queue); gpr_mpscq_push(&queue_, reinterpret_cast<gpr_mpscq_node*>(c));
q->queue_lock = GPR_SPINLOCK_INITIALIZER; return num_queue_items_.FetchAdd(1, grpc_core::MemoryOrder::RELAXED) == 0;
gpr_atm_no_barrier_store(&q->num_queue_items, 0);
} }
static void cq_event_queue_destroy(grpc_cq_event_queue* q) { grpc_cq_completion* CqEventQueue::Pop() {
gpr_mpscq_destroy(&q->queue);
}
static bool cq_event_queue_push(grpc_cq_event_queue* q, grpc_cq_completion* c) {
gpr_mpscq_push(&q->queue, reinterpret_cast<gpr_mpscq_node*>(c));
return gpr_atm_no_barrier_fetch_add(&q->num_queue_items, 1) == 0;
}
static grpc_cq_completion* cq_event_queue_pop(grpc_cq_event_queue* q) {
grpc_cq_completion* c = nullptr; grpc_cq_completion* c = nullptr;
if (gpr_spinlock_trylock(&q->queue_lock)) { if (gpr_spinlock_trylock(&queue_lock_)) {
GRPC_STATS_INC_CQ_EV_QUEUE_TRYLOCK_SUCCESSES(); GRPC_STATS_INC_CQ_EV_QUEUE_TRYLOCK_SUCCESSES();
bool is_empty = false; bool is_empty = false;
c = reinterpret_cast<grpc_cq_completion*>( c = reinterpret_cast<grpc_cq_completion*>(
gpr_mpscq_pop_and_check_end(&q->queue, &is_empty)); gpr_mpscq_pop_and_check_end(&queue_, &is_empty));
gpr_spinlock_unlock(&q->queue_lock); gpr_spinlock_unlock(&queue_lock_);
if (c == nullptr && !is_empty) { if (c == nullptr && !is_empty) {
GRPC_STATS_INC_CQ_EV_QUEUE_TRANSIENT_POP_FAILURES(); GRPC_STATS_INC_CQ_EV_QUEUE_TRANSIENT_POP_FAILURES();
@ -456,18 +484,12 @@ static grpc_cq_completion* cq_event_queue_pop(grpc_cq_event_queue* q) {
} }
if (c) { if (c) {
gpr_atm_no_barrier_fetch_add(&q->num_queue_items, -1); num_queue_items_.FetchSub(1, grpc_core::MemoryOrder::RELAXED);
} }
return c; return c;
} }
/* Note: The counter is not incremented/decremented atomically with push/pop.
* The count is only eventually consistent */
static long cq_event_queue_num_items(grpc_cq_event_queue* q) {
return static_cast<long>(gpr_atm_no_barrier_load(&q->num_queue_items));
}
grpc_completion_queue* grpc_completion_queue_create_internal( grpc_completion_queue* grpc_completion_queue_create_internal(
grpc_cq_completion_type completion_type, grpc_cq_polling_type polling_type, grpc_cq_completion_type completion_type, grpc_cq_polling_type polling_type,
grpc_experimental_completion_queue_functor* shutdown_callback) { grpc_experimental_completion_queue_functor* shutdown_callback) {
@ -507,49 +529,33 @@ grpc_completion_queue* grpc_completion_queue_create_internal(
static void cq_init_next( static void cq_init_next(
void* data, grpc_experimental_completion_queue_functor* shutdown_callback) { void* data, grpc_experimental_completion_queue_functor* shutdown_callback) {
cq_next_data* cqd = static_cast<cq_next_data*>(data); new (data) cq_next_data();
/* Initial count is dropped by grpc_completion_queue_shutdown */
gpr_atm_no_barrier_store(&cqd->pending_events, 1);
cqd->shutdown_called = false;
gpr_atm_no_barrier_store(&cqd->things_queued_ever, 0);
cq_event_queue_init(&cqd->queue);
} }
static void cq_destroy_next(void* data) { static void cq_destroy_next(void* data) {
cq_next_data* cqd = static_cast<cq_next_data*>(data); cq_next_data* cqd = static_cast<cq_next_data*>(data);
GPR_ASSERT(cq_event_queue_num_items(&cqd->queue) == 0); cqd->~cq_next_data();
cq_event_queue_destroy(&cqd->queue);
} }
static void cq_init_pluck( static void cq_init_pluck(
void* data, grpc_experimental_completion_queue_functor* shutdown_callback) { void* data, grpc_experimental_completion_queue_functor* shutdown_callback) {
cq_pluck_data* cqd = static_cast<cq_pluck_data*>(data); new (data) cq_pluck_data();
/* Initial count is dropped by grpc_completion_queue_shutdown */
gpr_atm_no_barrier_store(&cqd->pending_events, 1);
cqd->completed_tail = &cqd->completed_head;
cqd->completed_head.next = (uintptr_t)cqd->completed_tail;
gpr_atm_no_barrier_store(&cqd->shutdown, 0);
cqd->shutdown_called = false;
cqd->num_pluckers = 0;
gpr_atm_no_barrier_store(&cqd->things_queued_ever, 0);
} }
static void cq_destroy_pluck(void* data) { static void cq_destroy_pluck(void* data) {
cq_pluck_data* cqd = static_cast<cq_pluck_data*>(data); cq_pluck_data* cqd = static_cast<cq_pluck_data*>(data);
GPR_ASSERT(cqd->completed_head.next == (uintptr_t)&cqd->completed_head); cqd->~cq_pluck_data();
} }
static void cq_init_callback( static void cq_init_callback(
void* data, grpc_experimental_completion_queue_functor* shutdown_callback) { void* data, grpc_experimental_completion_queue_functor* shutdown_callback) {
cq_callback_data* cqd = static_cast<cq_callback_data*>(data); new (data) cq_callback_data(shutdown_callback);
/* Initial count is dropped by grpc_completion_queue_shutdown */
gpr_atm_no_barrier_store(&cqd->pending_events, 1);
cqd->shutdown_called = false;
gpr_atm_no_barrier_store(&cqd->things_queued_ever, 0);
cqd->shutdown_callback = shutdown_callback;
} }
static void cq_destroy_callback(void* data) {} static void cq_destroy_callback(void* data) {
cq_callback_data* cqd = static_cast<cq_callback_data*>(data);
cqd->~cq_callback_data();
}
grpc_cq_completion_type grpc_get_cq_completion_type(grpc_completion_queue* cq) { grpc_cq_completion_type grpc_get_cq_completion_type(grpc_completion_queue* cq) {
return cq->vtable->cq_completion_type; return cq->vtable->cq_completion_type;
@ -632,37 +638,19 @@ static void cq_check_tag(grpc_completion_queue* cq, void* tag, bool lock_cq) {
static void cq_check_tag(grpc_completion_queue* cq, void* tag, bool lock_cq) {} static void cq_check_tag(grpc_completion_queue* cq, void* tag, bool lock_cq) {}
#endif #endif
/* Atomically increments a counter only if the counter is not zero. Returns
* true if the increment was successful; false if the counter is zero */
static bool atm_inc_if_nonzero(gpr_atm* counter) {
while (true) {
gpr_atm count = gpr_atm_acq_load(counter);
/* If zero, we are done. If not, we must to a CAS (instead of an atomic
* increment) to maintain the contract: do not increment the counter if it
* is zero. */
if (count == 0) {
return false;
} else if (gpr_atm_full_cas(counter, count, count + 1)) {
break;
}
}
return true;
}
static bool cq_begin_op_for_next(grpc_completion_queue* cq, void* tag) { static bool cq_begin_op_for_next(grpc_completion_queue* cq, void* tag) {
cq_next_data* cqd = static_cast<cq_next_data*> DATA_FROM_CQ(cq); cq_next_data* cqd = static_cast<cq_next_data*> DATA_FROM_CQ(cq);
return atm_inc_if_nonzero(&cqd->pending_events); return cqd->pending_events.IncrementIfNonzero();
} }
static bool cq_begin_op_for_pluck(grpc_completion_queue* cq, void* tag) { static bool cq_begin_op_for_pluck(grpc_completion_queue* cq, void* tag) {
cq_pluck_data* cqd = static_cast<cq_pluck_data*> DATA_FROM_CQ(cq); cq_pluck_data* cqd = static_cast<cq_pluck_data*> DATA_FROM_CQ(cq);
return atm_inc_if_nonzero(&cqd->pending_events); return cqd->pending_events.IncrementIfNonzero();
} }
static bool cq_begin_op_for_callback(grpc_completion_queue* cq, void* tag) { static bool cq_begin_op_for_callback(grpc_completion_queue* cq, void* tag) {
cq_callback_data* cqd = static_cast<cq_callback_data*> DATA_FROM_CQ(cq); cq_callback_data* cqd = static_cast<cq_callback_data*> DATA_FROM_CQ(cq);
return atm_inc_if_nonzero(&cqd->pending_events); return cqd->pending_events.IncrementIfNonzero();
} }
bool grpc_cq_begin_op(grpc_completion_queue* cq, void* tag) { bool grpc_cq_begin_op(grpc_completion_queue* cq, void* tag) {
@ -716,17 +704,14 @@ static void cq_end_op_for_next(grpc_completion_queue* cq, void* tag,
gpr_tls_set(&g_cached_event, (intptr_t)storage); gpr_tls_set(&g_cached_event, (intptr_t)storage);
} else { } else {
/* Add the completion to the queue */ /* Add the completion to the queue */
bool is_first = cq_event_queue_push(&cqd->queue, storage); bool is_first = cqd->queue.Push(storage);
gpr_atm_no_barrier_fetch_add(&cqd->things_queued_ever, 1); cqd->things_queued_ever.FetchAdd(1, grpc_core::MemoryOrder::RELAXED);
/* Since we do not hold the cq lock here, it is important to do an 'acquire' /* Since we do not hold the cq lock here, it is important to do an 'acquire'
load here (instead of a 'no_barrier' load) to match with the release load here (instead of a 'no_barrier' load) to match with the release
store store
(done via gpr_atm_full_fetch_add(pending_events, -1)) in cq_shutdown_next (done via pending_events.FetchSub(1, ACQ_REL)) in cq_shutdown_next
*/ */
bool will_definitely_shutdown = gpr_atm_acq_load(&cqd->pending_events) == 1; if (cqd->pending_events.Load(grpc_core::MemoryOrder::ACQUIRE) != 1) {
if (!will_definitely_shutdown) {
/* Only kick if this is the first item queued */ /* Only kick if this is the first item queued */
if (is_first) { if (is_first) {
gpr_mu_lock(cq->mu); gpr_mu_lock(cq->mu);
@ -740,7 +725,8 @@ static void cq_end_op_for_next(grpc_completion_queue* cq, void* tag,
GRPC_ERROR_UNREF(kick_error); GRPC_ERROR_UNREF(kick_error);
} }
} }
if (gpr_atm_full_fetch_add(&cqd->pending_events, -1) == 1) { if (cqd->pending_events.FetchSub(1, grpc_core::MemoryOrder::ACQ_REL) ==
1) {
GRPC_CQ_INTERNAL_REF(cq, "shutting_down"); GRPC_CQ_INTERNAL_REF(cq, "shutting_down");
gpr_mu_lock(cq->mu); gpr_mu_lock(cq->mu);
cq_finish_shutdown_next(cq); cq_finish_shutdown_next(cq);
@ -749,7 +735,7 @@ static void cq_end_op_for_next(grpc_completion_queue* cq, void* tag,
} }
} else { } else {
GRPC_CQ_INTERNAL_REF(cq, "shutting_down"); GRPC_CQ_INTERNAL_REF(cq, "shutting_down");
gpr_atm_rel_store(&cqd->pending_events, 0); cqd->pending_events.Store(0, grpc_core::MemoryOrder::RELEASE);
gpr_mu_lock(cq->mu); gpr_mu_lock(cq->mu);
cq_finish_shutdown_next(cq); cq_finish_shutdown_next(cq);
gpr_mu_unlock(cq->mu); gpr_mu_unlock(cq->mu);
@ -795,12 +781,12 @@ static void cq_end_op_for_pluck(grpc_completion_queue* cq, void* tag,
cq_check_tag(cq, tag, false); /* Used in debug builds only */ cq_check_tag(cq, tag, false); /* Used in debug builds only */
/* Add to the list of completions */ /* Add to the list of completions */
gpr_atm_no_barrier_fetch_add(&cqd->things_queued_ever, 1); cqd->things_queued_ever.FetchAdd(1, grpc_core::MemoryOrder::RELAXED);
cqd->completed_tail->next = cqd->completed_tail->next =
((uintptr_t)storage) | (1u & cqd->completed_tail->next); ((uintptr_t)storage) | (1u & cqd->completed_tail->next);
cqd->completed_tail = storage; cqd->completed_tail = storage;
if (gpr_atm_full_fetch_add(&cqd->pending_events, -1) == 1) { if (cqd->pending_events.FetchSub(1, grpc_core::MemoryOrder::ACQ_REL) == 1) {
cq_finish_shutdown_pluck(cq); cq_finish_shutdown_pluck(cq);
gpr_mu_unlock(cq->mu); gpr_mu_unlock(cq->mu);
} else { } else {
@ -856,8 +842,8 @@ static void cq_end_op_for_callback(
cq_check_tag(cq, tag, true); /* Used in debug builds only */ cq_check_tag(cq, tag, true); /* Used in debug builds only */
gpr_atm_no_barrier_fetch_add(&cqd->things_queued_ever, 1); cqd->things_queued_ever.FetchAdd(1, grpc_core::MemoryOrder::RELAXED);
if (gpr_atm_full_fetch_add(&cqd->pending_events, -1) == 1) { if (cqd->pending_events.FetchSub(1, grpc_core::MemoryOrder::ACQ_REL) == 1) {
cq_finish_shutdown_callback(cq); cq_finish_shutdown_callback(cq);
} }
@ -893,20 +879,20 @@ class ExecCtxNext : public grpc_core::ExecCtx {
cq_next_data* cqd = static_cast<cq_next_data*> DATA_FROM_CQ(cq); cq_next_data* cqd = static_cast<cq_next_data*> DATA_FROM_CQ(cq);
GPR_ASSERT(a->stolen_completion == nullptr); GPR_ASSERT(a->stolen_completion == nullptr);
gpr_atm current_last_seen_things_queued_ever = intptr_t current_last_seen_things_queued_ever =
gpr_atm_no_barrier_load(&cqd->things_queued_ever); cqd->things_queued_ever.Load(grpc_core::MemoryOrder::RELAXED);
if (current_last_seen_things_queued_ever != if (current_last_seen_things_queued_ever !=
a->last_seen_things_queued_ever) { a->last_seen_things_queued_ever) {
a->last_seen_things_queued_ever = a->last_seen_things_queued_ever =
gpr_atm_no_barrier_load(&cqd->things_queued_ever); cqd->things_queued_ever.Load(grpc_core::MemoryOrder::RELAXED);
/* Pop a cq_completion from the queue. Returns NULL if the queue is empty /* Pop a cq_completion from the queue. Returns NULL if the queue is empty
* might return NULL in some cases even if the queue is not empty; but * might return NULL in some cases even if the queue is not empty; but
* that * that
* is ok and doesn't affect correctness. Might effect the tail latencies a * is ok and doesn't affect correctness. Might effect the tail latencies a
* bit) */ * bit) */
a->stolen_completion = cq_event_queue_pop(&cqd->queue); a->stolen_completion = cqd->queue.Pop();
if (a->stolen_completion != nullptr) { if (a->stolen_completion != nullptr) {
return true; return true;
} }
@ -965,7 +951,7 @@ static grpc_event cq_next(grpc_completion_queue* cq, gpr_timespec deadline,
grpc_millis deadline_millis = grpc_timespec_to_millis_round_up(deadline); grpc_millis deadline_millis = grpc_timespec_to_millis_round_up(deadline);
cq_is_finished_arg is_finished_arg = { cq_is_finished_arg is_finished_arg = {
gpr_atm_no_barrier_load(&cqd->things_queued_ever), cqd->things_queued_ever.Load(grpc_core::MemoryOrder::RELAXED),
cq, cq,
deadline_millis, deadline_millis,
nullptr, nullptr,
@ -985,7 +971,7 @@ static grpc_event cq_next(grpc_completion_queue* cq, gpr_timespec deadline,
break; break;
} }
grpc_cq_completion* c = cq_event_queue_pop(&cqd->queue); grpc_cq_completion* c = cqd->queue.Pop();
if (c != nullptr) { if (c != nullptr) {
ret.type = GRPC_OP_COMPLETE; ret.type = GRPC_OP_COMPLETE;
@ -999,16 +985,16 @@ static grpc_event cq_next(grpc_completion_queue* cq, gpr_timespec deadline,
so that the thread comes back quickly from poll to make a second so that the thread comes back quickly from poll to make a second
attempt at popping. Not doing this can potentially deadlock this attempt at popping. Not doing this can potentially deadlock this
thread forever (if the deadline is infinity) */ thread forever (if the deadline is infinity) */
if (cq_event_queue_num_items(&cqd->queue) > 0) { if (cqd->queue.num_items() > 0) {
iteration_deadline = 0; iteration_deadline = 0;
} }
} }
if (gpr_atm_acq_load(&cqd->pending_events) == 0) { if (cqd->pending_events.Load(grpc_core::MemoryOrder::ACQUIRE) == 0) {
/* Before returning, check if the queue has any items left over (since /* Before returning, check if the queue has any items left over (since
gpr_mpscq_pop() can sometimes return NULL even if the queue is not gpr_mpscq_pop() can sometimes return NULL even if the queue is not
empty. If so, keep retrying but do not return GRPC_QUEUE_SHUTDOWN */ empty. If so, keep retrying but do not return GRPC_QUEUE_SHUTDOWN */
if (cq_event_queue_num_items(&cqd->queue) > 0) { if (cqd->queue.num_items() > 0) {
/* Go to the beginning of the loop. No point doing a poll because /* Go to the beginning of the loop. No point doing a poll because
(cq->shutdown == true) is only possible when there is no pending (cq->shutdown == true) is only possible when there is no pending
work (i.e cq->pending_events == 0) and any outstanding completion work (i.e cq->pending_events == 0) and any outstanding completion
@ -1049,8 +1035,8 @@ static grpc_event cq_next(grpc_completion_queue* cq, gpr_timespec deadline,
is_finished_arg.first_loop = false; is_finished_arg.first_loop = false;
} }
if (cq_event_queue_num_items(&cqd->queue) > 0 && if (cqd->queue.num_items() > 0 &&
gpr_atm_acq_load(&cqd->pending_events) > 0) { cqd->pending_events.Load(grpc_core::MemoryOrder::ACQUIRE) > 0) {
gpr_mu_lock(cq->mu); gpr_mu_lock(cq->mu);
cq->poller_vtable->kick(POLLSET_FROM_CQ(cq), nullptr); cq->poller_vtable->kick(POLLSET_FROM_CQ(cq), nullptr);
gpr_mu_unlock(cq->mu); gpr_mu_unlock(cq->mu);
@ -1074,7 +1060,7 @@ static void cq_finish_shutdown_next(grpc_completion_queue* cq) {
cq_next_data* cqd = static_cast<cq_next_data*> DATA_FROM_CQ(cq); cq_next_data* cqd = static_cast<cq_next_data*> DATA_FROM_CQ(cq);
GPR_ASSERT(cqd->shutdown_called); GPR_ASSERT(cqd->shutdown_called);
GPR_ASSERT(gpr_atm_no_barrier_load(&cqd->pending_events) == 0); GPR_ASSERT(cqd->pending_events.Load(grpc_core::MemoryOrder::RELAXED) == 0);
cq->poller_vtable->shutdown(POLLSET_FROM_CQ(cq), &cq->pollset_shutdown_done); cq->poller_vtable->shutdown(POLLSET_FROM_CQ(cq), &cq->pollset_shutdown_done);
} }
@ -1096,10 +1082,10 @@ static void cq_shutdown_next(grpc_completion_queue* cq) {
return; return;
} }
cqd->shutdown_called = true; cqd->shutdown_called = true;
/* Doing a full_fetch_add (i.e acq/release) here to match with /* Doing acq/release FetchSub here to match with
* cq_begin_op_for_next and cq_end_op_for_next functions which read/write * cq_begin_op_for_next and cq_end_op_for_next functions which read/write
* on this counter without necessarily holding a lock on cq */ * on this counter without necessarily holding a lock on cq */
if (gpr_atm_full_fetch_add(&cqd->pending_events, -1) == 1) { if (cqd->pending_events.FetchSub(1, grpc_core::MemoryOrder::ACQ_REL) == 1) {
cq_finish_shutdown_next(cq); cq_finish_shutdown_next(cq);
} }
gpr_mu_unlock(cq->mu); gpr_mu_unlock(cq->mu);
@ -1148,12 +1134,12 @@ class ExecCtxPluck : public grpc_core::ExecCtx {
GPR_ASSERT(a->stolen_completion == nullptr); GPR_ASSERT(a->stolen_completion == nullptr);
gpr_atm current_last_seen_things_queued_ever = gpr_atm current_last_seen_things_queued_ever =
gpr_atm_no_barrier_load(&cqd->things_queued_ever); cqd->things_queued_ever.Load(grpc_core::MemoryOrder::RELAXED);
if (current_last_seen_things_queued_ever != if (current_last_seen_things_queued_ever !=
a->last_seen_things_queued_ever) { a->last_seen_things_queued_ever) {
gpr_mu_lock(cq->mu); gpr_mu_lock(cq->mu);
a->last_seen_things_queued_ever = a->last_seen_things_queued_ever =
gpr_atm_no_barrier_load(&cqd->things_queued_ever); cqd->things_queued_ever.Load(grpc_core::MemoryOrder::RELAXED);
grpc_cq_completion* c; grpc_cq_completion* c;
grpc_cq_completion* prev = &cqd->completed_head; grpc_cq_completion* prev = &cqd->completed_head;
while ((c = (grpc_cq_completion*)(prev->next & while ((c = (grpc_cq_completion*)(prev->next &
@ -1209,7 +1195,7 @@ static grpc_event cq_pluck(grpc_completion_queue* cq, void* tag,
gpr_mu_lock(cq->mu); gpr_mu_lock(cq->mu);
grpc_millis deadline_millis = grpc_timespec_to_millis_round_up(deadline); grpc_millis deadline_millis = grpc_timespec_to_millis_round_up(deadline);
cq_is_finished_arg is_finished_arg = { cq_is_finished_arg is_finished_arg = {
gpr_atm_no_barrier_load(&cqd->things_queued_ever), cqd->things_queued_ever.Load(grpc_core::MemoryOrder::RELAXED),
cq, cq,
deadline_millis, deadline_millis,
nullptr, nullptr,
@ -1246,7 +1232,7 @@ static grpc_event cq_pluck(grpc_completion_queue* cq, void* tag,
} }
prev = c; prev = c;
} }
if (gpr_atm_no_barrier_load(&cqd->shutdown)) { if (cqd->shutdown.Load(grpc_core::MemoryOrder::RELAXED)) {
gpr_mu_unlock(cq->mu); gpr_mu_unlock(cq->mu);
memset(&ret, 0, sizeof(ret)); memset(&ret, 0, sizeof(ret));
ret.type = GRPC_QUEUE_SHUTDOWN; ret.type = GRPC_QUEUE_SHUTDOWN;
@ -1309,8 +1295,8 @@ static void cq_finish_shutdown_pluck(grpc_completion_queue* cq) {
cq_pluck_data* cqd = static_cast<cq_pluck_data*> DATA_FROM_CQ(cq); cq_pluck_data* cqd = static_cast<cq_pluck_data*> DATA_FROM_CQ(cq);
GPR_ASSERT(cqd->shutdown_called); GPR_ASSERT(cqd->shutdown_called);
GPR_ASSERT(!gpr_atm_no_barrier_load(&cqd->shutdown)); GPR_ASSERT(!cqd->shutdown.Load(grpc_core::MemoryOrder::RELAXED));
gpr_atm_no_barrier_store(&cqd->shutdown, 1); cqd->shutdown.Store(1, grpc_core::MemoryOrder::RELAXED);
cq->poller_vtable->shutdown(POLLSET_FROM_CQ(cq), &cq->pollset_shutdown_done); cq->poller_vtable->shutdown(POLLSET_FROM_CQ(cq), &cq->pollset_shutdown_done);
} }
@ -1334,7 +1320,7 @@ static void cq_shutdown_pluck(grpc_completion_queue* cq) {
return; return;
} }
cqd->shutdown_called = true; cqd->shutdown_called = true;
if (gpr_atm_full_fetch_add(&cqd->pending_events, -1) == 1) { if (cqd->pending_events.FetchSub(1, grpc_core::MemoryOrder::ACQ_REL) == 1) {
cq_finish_shutdown_pluck(cq); cq_finish_shutdown_pluck(cq);
} }
gpr_mu_unlock(cq->mu); gpr_mu_unlock(cq->mu);
@ -1368,7 +1354,7 @@ static void cq_shutdown_callback(grpc_completion_queue* cq) {
return; return;
} }
cqd->shutdown_called = true; cqd->shutdown_called = true;
if (gpr_atm_full_fetch_add(&cqd->pending_events, -1) == 1) { if (cqd->pending_events.FetchSub(1, grpc_core::MemoryOrder::ACQ_REL) == 1) {
gpr_mu_unlock(cq->mu); gpr_mu_unlock(cq->mu);
cq_finish_shutdown_callback(cq); cq_finish_shutdown_callback(cq);
} else { } else {

Loading…
Cancel
Save