|
Packit |
6c4009 |
/* POSIX reader--writer lock: core parts.
|
|
Packit |
6c4009 |
Copyright (C) 2016-2018 Free Software Foundation, Inc.
|
|
Packit |
6c4009 |
This file is part of the GNU C Library.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
The GNU C Library is free software; you can redistribute it and/or
|
|
Packit |
6c4009 |
modify it under the terms of the GNU Lesser General Public
|
|
Packit |
6c4009 |
License as published by the Free Software Foundation; either
|
|
Packit |
6c4009 |
version 2.1 of the License, or (at your option) any later version.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
The GNU C Library is distributed in the hope that it will be useful,
|
|
Packit |
6c4009 |
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
Packit |
6c4009 |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
Packit |
6c4009 |
Lesser General Public License for more details.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
You should have received a copy of the GNU Lesser General Public
|
|
Packit |
6c4009 |
License along with the GNU C Library; if not, see
|
|
Packit |
6c4009 |
<http://www.gnu.org/licenses/>. */
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
#include <errno.h>
|
|
Packit |
6c4009 |
#include <sysdep.h>
|
|
Packit |
6c4009 |
#include <pthread.h>
|
|
Packit |
6c4009 |
#include <pthreadP.h>
|
|
Packit |
6c4009 |
#include <sys/time.h>
|
|
Packit |
6c4009 |
#include <stap-probe.h>
|
|
Packit |
6c4009 |
#include <atomic.h>
|
|
Packit |
6c4009 |
#include <futex-internal.h>
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* A reader--writer lock that fulfills the POSIX requirements (but operations
|
|
Packit |
6c4009 |
on this lock are not necessarily full barriers, as one may interpret the
|
|
Packit |
6c4009 |
POSIX requirement about "synchronizing memory"). All critical sections are
|
|
Packit |
6c4009 |
in a total order, writers synchronize with prior writers and readers, and
|
|
Packit |
6c4009 |
readers synchronize with prior writers.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
A thread is allowed to acquire a read lock recursively (i.e., have rdlock
|
|
Packit |
6c4009 |
critical sections that overlap in sequenced-before) unless the kind of the
|
|
Packit Service |
3b0880 |
rwlock is set to PTHREAD_RWLOCK_PREFER_WRITERS_NONRECURSIVE_NP.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
This lock is built so that workloads of mostly readers can be executed with
|
|
Packit |
6c4009 |
low runtime overheads. This matches that the default kind of the lock is
|
|
Packit |
6c4009 |
PTHREAD_RWLOCK_PREFER_READER_NP. Acquiring a read lock requires a single
|
|
Packit |
6c4009 |
atomic addition if the lock is or was previously acquired by other
|
|
Packit |
6c4009 |
readers; releasing the lock is a single CAS if there are no concurrent
|
|
Packit |
6c4009 |
writers.
|
|
Packit |
6c4009 |
Workloads consisting of mostly writers are of secondary importance.
|
|
Packit |
6c4009 |
An uncontended write lock acquisition is as fast as for a normal
|
|
Packit |
6c4009 |
exclusive mutex but writer contention is somewhat more costly due to
|
|
Packit |
6c4009 |
keeping track of the exact number of writers. If the rwlock kind requests
|
|
Packit Service |
3b0880 |
writers to be preferred (i.e., PTHREAD_RWLOCK_PREFER_WRITERS_NP or the
|
|
Packit |
6c4009 |
no-recursive-readers variant of it), then writer--to--writer lock ownership
|
|
Packit |
6c4009 |
hand-over is fairly fast and bypasses lock acquisition attempts by readers.
|
|
Packit |
6c4009 |
The costs of lock ownership transfer between readers and writers vary. If
|
|
Packit |
6c4009 |
the program asserts that there are no recursive readers and writers are
|
|
Packit |
6c4009 |
preferred, then write lock acquisition attempts will block subsequent read
|
|
Packit |
6c4009 |
lock acquisition attempts, so that new incoming readers do not prolong a
|
|
Packit |
6c4009 |
phase in which readers have acquired the lock.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
The main components of the rwlock are a writer-only lock that allows only
|
|
Packit |
6c4009 |
one of the concurrent writers to be the primary writer, and a
|
|
Packit |
6c4009 |
single-writer-multiple-readers lock that decides between read phases, in
|
|
Packit |
6c4009 |
which readers have acquired the rwlock, and write phases in which a primary
|
|
Packit |
6c4009 |
writer or a sequence of different primary writers have acquired the rwlock.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
The single-writer-multiple-readers lock is the central piece of state
|
|
Packit |
6c4009 |
describing the rwlock and is encoded in the __readers field (see below for
|
|
Packit |
6c4009 |
a detailed explanation):
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
State WP WL R RW Notes
|
|
Packit |
6c4009 |
---------------------------
|
|
Packit |
6c4009 |
#1 0 0 0 0 Lock is idle (and in a read phase).
|
|
Packit |
6c4009 |
#2 0 0 >0 0 Readers have acquired the lock.
|
|
Packit |
6c4009 |
#3 0 1 0 0 Lock is not acquired; a writer will try to start a
|
|
Packit |
6c4009 |
write phase.
|
|
Packit |
6c4009 |
#4 0 1 >0 0 Readers have acquired the lock; a writer is waiting
|
|
Packit |
6c4009 |
and explicit hand-over to the writer is required.
|
|
Packit |
6c4009 |
#4a 0 1 >0 1 Same as #4 except that there are further readers
|
|
Packit |
6c4009 |
waiting because the writer is to be preferred.
|
|
Packit |
6c4009 |
#5 1 0 0 0 Lock is idle (and in a write phase).
|
|
Packit |
6c4009 |
#6 1 0 >0 0 Write phase; readers will try to start a read phase
|
|
Packit |
6c4009 |
(requires explicit hand-over to all readers that
|
|
Packit |
6c4009 |
do not start the read phase).
|
|
Packit |
6c4009 |
#7 1 1 0 0 Lock is acquired by a writer.
|
|
Packit |
6c4009 |
#8 1 1 >0 0 Lock acquired by a writer and readers are waiting;
|
|
Packit |
6c4009 |
explicit hand-over to the readers is required.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
WP (PTHREAD_RWLOCK_WRPHASE) is true if the lock is in a write phase, so
|
|
Packit |
6c4009 |
potentially acquired by a primary writer.
|
|
Packit |
6c4009 |
WL (PTHREAD_RWLOCK_WRLOCKED) is true if there is a primary writer (i.e.,
|
|
Packit |
6c4009 |
the thread that was able to set this bit from false to true).
|
|
Packit |
6c4009 |
R (all bits in __readers except the number of least-significant bits
|
|
Packit |
6c4009 |
denoted in PTHREAD_RWLOCK_READER_SHIFT) is the number of readers that have
|
|
Packit |
6c4009 |
or are trying to acquired the lock. There may be more readers waiting if
|
|
Packit |
6c4009 |
writers are preferred and there will be no recursive readers, in which
|
|
Packit |
6c4009 |
case RW (PTHREAD_RWLOCK_RWAITING) is true in state #4a.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
We want to block using futexes but using __readers as a futex word directly
|
|
Packit |
6c4009 |
is not a good solution. First, we want to wait on different conditions
|
|
Packit |
6c4009 |
such as waiting for a phase change vs. waiting for the primary writer to
|
|
Packit |
6c4009 |
release the writer-only lock. Second, the number of readers could change
|
|
Packit |
6c4009 |
frequently, which would make it likely that a writer's futex_wait fails
|
|
Packit |
6c4009 |
frequently too because the expected value does not match the value of
|
|
Packit |
6c4009 |
__readers anymore.
|
|
Packit |
6c4009 |
Therefore, we split out the futex words into the __wrphase_futex and
|
|
Packit |
6c4009 |
__writers_futex fields. The former tracks the value of the WP bit and is
|
|
Packit |
6c4009 |
changed after changing WP by the thread that changes WP. However, because
|
|
Packit |
6c4009 |
of the POSIX requirements regarding mutex/rwlock destruction (i.e., that
|
|
Packit |
6c4009 |
destroying a rwlock is allowed as soon as no thread has acquired or will
|
|
Packit |
6c4009 |
acquire the lock), we have to be careful and hand over lock ownership (via
|
|
Packit |
6c4009 |
a phase change) carefully to those threads waiting. Specifically, we must
|
|
Packit |
6c4009 |
prevent a situation in which we are not quite sure whether we still have
|
|
Packit |
6c4009 |
to unblock another thread through a change to memory (executing a
|
|
Packit |
6c4009 |
futex_wake on a former futex word that is now used for something else is
|
|
Packit |
6c4009 |
fine).
|
|
Packit |
6c4009 |
The scheme we use for __wrphase_futex is that waiting threads that may
|
|
Packit |
6c4009 |
use the futex word to block now all have to use the futex word to block; it
|
|
Packit |
6c4009 |
is not allowed to take the short-cut and spin-wait on __readers because
|
|
Packit |
6c4009 |
then the waking thread cannot just make one final change to memory to
|
|
Packit |
6c4009 |
unblock all potentially waiting threads. If, for example, a reader
|
|
Packit |
6c4009 |
increments R in states #7 or #8, it has to then block until __wrphase_futex
|
|
Packit |
6c4009 |
is 0 and it can confirm that the value of 0 was stored by the primary
|
|
Packit |
6c4009 |
writer; in turn, the primary writer has to change to a read phase too when
|
|
Packit |
6c4009 |
releasing WL (i.e., to state #2), and it must change __wrphase_futex to 0
|
|
Packit |
6c4009 |
as the next step. This ensures that the waiting reader will not be able to
|
|
Packit |
6c4009 |
acquire, release, and then destroy the lock concurrently with the pending
|
|
Packit |
6c4009 |
futex unblock operations by the former primary writer. This scheme is
|
|
Packit |
6c4009 |
called explicit hand-over in what follows.
|
|
Packit |
6c4009 |
Note that waiting threads can cancel waiting only if explicit hand-over has
|
|
Packit |
6c4009 |
not yet started (e.g., if __readers is still in states #7 or #8 in the
|
|
Packit |
6c4009 |
example above).
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
Writers determine the primary writer through WL. Blocking using futexes
|
|
Packit |
6c4009 |
is performed using __writers_futex as a futex word; primary writers will
|
|
Packit |
6c4009 |
enable waiting on this futex by setting it to 1 after they acquired the WL
|
|
Packit |
6c4009 |
bit and will disable waiting by setting it to 0 before they release WL.
|
|
Packit |
6c4009 |
This leaves small windows where blocking using futexes is not possible
|
|
Packit |
6c4009 |
although a primary writer exists, but in turn decreases complexity of the
|
|
Packit |
6c4009 |
writer--writer synchronization and does not affect correctness.
|
|
Packit |
6c4009 |
If writers are preferred, writers can hand over WL directly to other
|
|
Packit |
6c4009 |
waiting writers that registered by incrementing __writers: If the primary
|
|
Packit |
6c4009 |
writer can CAS __writers from a non-zero value to the same value with the
|
|
Packit |
6c4009 |
PTHREAD_RWLOCK_WRHANDOVER bit set, it effectively transfers WL ownership
|
|
Packit |
6c4009 |
to one of the registered waiting writers and does not reset WL; in turn,
|
|
Packit |
6c4009 |
a registered writer that can clear PTHREAD_RWLOCK_WRHANDOVER using a CAS
|
|
Packit |
6c4009 |
then takes over WL. Note that registered waiting writers can cancel
|
|
Packit |
6c4009 |
waiting by decrementing __writers, but the last writer to unregister must
|
|
Packit |
6c4009 |
become the primary writer if PTHREAD_RWLOCK_WRHANDOVER is set.
|
|
Packit |
6c4009 |
Also note that adding another state/bit to signal potential writer--writer
|
|
Packit |
6c4009 |
contention (e.g., as done in the normal mutex algorithm) would not be
|
|
Packit |
6c4009 |
helpful because we would have to conservatively assume that there is in
|
|
Packit |
6c4009 |
fact no other writer, and wake up readers too.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
To avoid having to call futex_wake when no thread uses __wrphase_futex or
|
|
Packit |
6c4009 |
__writers_futex, threads will set the PTHREAD_RWLOCK_FUTEX_USED bit in the
|
|
Packit |
6c4009 |
respective futex words before waiting on it (using a CAS so it will only be
|
|
Packit |
6c4009 |
set if in a state in which waiting would be possible). In the case of
|
|
Packit |
6c4009 |
__writers_futex, we wake only one thread but several threads may share
|
|
Packit |
6c4009 |
PTHREAD_RWLOCK_FUTEX_USED, so we must assume that there are still others.
|
|
Packit |
6c4009 |
This is similar to what we do in pthread_mutex_lock. We do not need to
|
|
Packit |
6c4009 |
do this for __wrphase_futex because there, we always wake all waiting
|
|
Packit |
6c4009 |
threads.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
Blocking in the state #4a simply uses __readers as futex word. This
|
|
Packit |
6c4009 |
simplifies the algorithm but suffers from some of the drawbacks discussed
|
|
Packit |
6c4009 |
before, though not to the same extent because R can only decrease in this
|
|
Packit |
6c4009 |
state, so the number of potentially failing futex_wait attempts will be
|
|
Packit |
6c4009 |
bounded. All threads moving from state #4a to another state must wake
|
|
Packit |
6c4009 |
up threads blocked on the __readers futex.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
The ordering invariants that we have to take care of in the implementation
|
|
Packit |
6c4009 |
are primarily those necessary for a reader--writer lock; this is rather
|
|
Packit |
6c4009 |
straightforward and happens during write/read phase switching (potentially
|
|
Packit |
6c4009 |
through explicit hand-over), and between writers through synchronization
|
|
Packit |
6c4009 |
involving the PTHREAD_RWLOCK_WRLOCKED or PTHREAD_RWLOCK_WRHANDOVER bits.
|
|
Packit |
6c4009 |
Additionally, we need to take care that modifications of __writers_futex
|
|
Packit |
6c4009 |
and __wrphase_futex (e.g., by otherwise unordered readers) take place in
|
|
Packit |
6c4009 |
the writer critical sections or read/write phases, respectively, and that
|
|
Packit |
6c4009 |
explicit hand-over observes stores from the previous phase. How this is
|
|
Packit |
6c4009 |
done is explained in more detail in comments in the code.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
Many of the accesses to the futex words just need relaxed MO. This is
|
|
Packit |
6c4009 |
possible because we essentially drive both the core rwlock synchronization
|
|
Packit |
6c4009 |
and the futex synchronization in parallel. For example, an unlock will
|
|
Packit |
6c4009 |
unlock the rwlock and take part in the futex synchronization (using
|
|
Packit |
6c4009 |
PTHREAD_RWLOCK_FUTEX_USED, see above); even if they are not tightly
|
|
Packit |
6c4009 |
ordered in some way, the futex synchronization ensures that there are no
|
|
Packit |
6c4009 |
lost wake-ups, and woken threads will then eventually see the most recent
|
|
Packit |
6c4009 |
state of the rwlock. IOW, waiting threads will always be woken up, while
|
|
Packit |
6c4009 |
not being able to wait using futexes (which can happen) is harmless; in
|
|
Packit |
6c4009 |
turn, this means that waiting threads don't need special ordering wrt.
|
|
Packit |
6c4009 |
waking threads.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
The futex synchronization consists of the three-state futex word:
|
|
Packit |
6c4009 |
(1) cannot block on it, (2) can block on it, and (3) there might be a
|
|
Packit |
6c4009 |
thread blocked on it (i.e., with PTHREAD_RWLOCK_FUTEX_USED set).
|
|
Packit |
6c4009 |
Relaxed-MO atomic read-modify-write operations are sufficient to maintain
|
|
Packit |
6c4009 |
this (e.g., using a CAS to go from (2) to (3) but not from (1) to (3)),
|
|
Packit |
6c4009 |
but we need ordering of the futex word modifications by the waking threads
|
|
Packit |
6c4009 |
so that they collectively make correct state changes between (1)-(3).
|
|
Packit |
6c4009 |
The futex-internal synchronization (i.e., the conceptual critical sections
|
|
Packit |
6c4009 |
around futex operations in the kernel) then ensures that even an
|
|
Packit |
6c4009 |
unconstrained load (i.e., relaxed MO) inside of futex_wait will not lead to
|
|
Packit |
6c4009 |
lost wake-ups because either the waiting thread will see the change from
|
|
Packit |
6c4009 |
(3) to (1) when a futex_wake came first, or this futex_wake will wake this
|
|
Packit |
6c4009 |
waiting thread because the waiting thread came first.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
POSIX allows but does not require rwlock acquisitions to be a cancellation
|
|
Packit |
6c4009 |
point. We do not support cancellation.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
TODO We do not try to elide any read or write lock acquisitions currently.
|
|
Packit |
6c4009 |
While this would be possible, it is unclear whether HTM performance is
|
|
Packit |
6c4009 |
currently predictable enough and our runtime tuning is good enough at
|
|
Packit |
6c4009 |
deciding when to use elision so that enabling it would lead to consistently
|
|
Packit |
6c4009 |
better performance. */
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
static int
|
|
Packit |
6c4009 |
__pthread_rwlock_get_private (pthread_rwlock_t *rwlock)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
return rwlock->__data.__shared != 0 ? FUTEX_SHARED : FUTEX_PRIVATE;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
static __always_inline void
|
|
Packit |
6c4009 |
__pthread_rwlock_rdunlock (pthread_rwlock_t *rwlock)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
int private = __pthread_rwlock_get_private (rwlock);
|
|
Packit |
6c4009 |
/* We decrease the number of readers, and if we are the last reader and
|
|
Packit |
6c4009 |
there is a primary writer, we start a write phase. We use a CAS to
|
|
Packit |
6c4009 |
make this atomic so that it is clear whether we must hand over ownership
|
|
Packit |
6c4009 |
explicitly. */
|
|
Packit |
6c4009 |
unsigned int r = atomic_load_relaxed (&rwlock->__data.__readers);
|
|
Packit |
6c4009 |
unsigned int rnew;
|
|
Packit |
6c4009 |
for (;;)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
rnew = r - (1 << PTHREAD_RWLOCK_READER_SHIFT);
|
|
Packit |
6c4009 |
/* If we are the last reader, we also need to unblock any readers
|
|
Packit |
6c4009 |
that are waiting for a writer to go first (PTHREAD_RWLOCK_RWAITING)
|
|
Packit |
6c4009 |
so that they can register while the writer is active. */
|
|
Packit |
6c4009 |
if ((rnew >> PTHREAD_RWLOCK_READER_SHIFT) == 0)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
if ((rnew & PTHREAD_RWLOCK_WRLOCKED) != 0)
|
|
Packit |
6c4009 |
rnew |= PTHREAD_RWLOCK_WRPHASE;
|
|
Packit |
6c4009 |
rnew &= ~(unsigned int) PTHREAD_RWLOCK_RWAITING;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* We need release MO here for three reasons. First, so that we
|
|
Packit |
6c4009 |
synchronize with subsequent writers. Second, we might have been the
|
|
Packit |
6c4009 |
first reader and set __wrphase_futex to 0, so we need to synchronize
|
|
Packit |
6c4009 |
with the last reader that will set it to 1 (note that we will always
|
|
Packit |
6c4009 |
change __readers before the last reader, or we are the last reader).
|
|
Packit |
6c4009 |
Third, a writer that takes part in explicit hand-over needs to see
|
|
Packit |
6c4009 |
the first reader's store to __wrphase_futex (or a later value) if
|
|
Packit |
6c4009 |
the writer observes that a write phase has been started. */
|
|
Packit |
6c4009 |
if (atomic_compare_exchange_weak_release (&rwlock->__data.__readers,
|
|
Packit Service |
3b0880 |
&r, rnew))
|
|
Packit |
6c4009 |
break;
|
|
Packit |
6c4009 |
/* TODO Back-off. */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
if ((rnew & PTHREAD_RWLOCK_WRPHASE) != 0)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* We need to do explicit hand-over. We need the acquire MO fence so
|
|
Packit |
6c4009 |
that our modification of _wrphase_futex happens after a store by
|
|
Packit |
6c4009 |
another reader that started a read phase. Relaxed MO is sufficient
|
|
Packit |
6c4009 |
for the modification of __wrphase_futex because it is just used
|
|
Packit |
6c4009 |
to delay acquisition by a writer until all threads are unblocked
|
|
Packit |
6c4009 |
irrespective of whether they are looking at __readers or
|
|
Packit |
6c4009 |
__wrphase_futex; any other synchronizes-with relations that are
|
|
Packit |
6c4009 |
necessary are established through __readers. */
|
|
Packit |
6c4009 |
atomic_thread_fence_acquire ();
|
|
Packit |
6c4009 |
if ((atomic_exchange_relaxed (&rwlock->__data.__wrphase_futex, 1)
|
|
Packit |
6c4009 |
& PTHREAD_RWLOCK_FUTEX_USED) != 0)
|
|
Packit |
6c4009 |
futex_wake (&rwlock->__data.__wrphase_futex, INT_MAX, private);
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* Also wake up waiting readers if we did reset the RWAITING flag. */
|
|
Packit |
6c4009 |
if ((r & PTHREAD_RWLOCK_RWAITING) != (rnew & PTHREAD_RWLOCK_RWAITING))
|
|
Packit |
6c4009 |
futex_wake (&rwlock->__data.__readers, INT_MAX, private);
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
static __always_inline int
|
|
Packit |
6c4009 |
__pthread_rwlock_rdlock_full (pthread_rwlock_t *rwlock,
|
|
Packit |
6c4009 |
const struct timespec *abstime)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
unsigned int r;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Make sure we are not holding the rwlock as a writer. This is a deadlock
|
|
Packit |
6c4009 |
situation we recognize and report. */
|
|
Packit |
6c4009 |
if (__glibc_unlikely (atomic_load_relaxed (&rwlock->__data.__cur_writer)
|
|
Packit Service |
3b0880 |
== THREAD_GETMEM (THREAD_SELF, tid)))
|
|
Packit |
6c4009 |
return EDEADLK;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* If we prefer writers, recursive rdlock is disallowed, we are in a read
|
|
Packit |
6c4009 |
phase, and there are other readers present, we try to wait without
|
|
Packit |
6c4009 |
extending the read phase. We will be unblocked by either one of the
|
|
Packit |
6c4009 |
other active readers, or if the writer gives up WRLOCKED (e.g., on
|
|
Packit |
6c4009 |
timeout).
|
|
Packit |
6c4009 |
If there are no other readers, we simply race with any existing primary
|
|
Packit |
6c4009 |
writer; it would have been a race anyway, and changing the odds slightly
|
|
Packit |
6c4009 |
will likely not make a big difference. */
|
|
Packit |
6c4009 |
if (rwlock->__data.__flags == PTHREAD_RWLOCK_PREFER_WRITER_NONRECURSIVE_NP)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
r = atomic_load_relaxed (&rwlock->__data.__readers);
|
|
Packit Service |
3b0880 |
while (((r & PTHREAD_RWLOCK_WRPHASE) == 0)
|
|
Packit Service |
3b0880 |
&& ((r & PTHREAD_RWLOCK_WRLOCKED) != 0)
|
|
Packit Service |
3b0880 |
&& ((r >> PTHREAD_RWLOCK_READER_SHIFT) > 0))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* TODO Spin first. */
|
|
Packit |
6c4009 |
/* Try setting the flag signaling that we are waiting without having
|
|
Packit |
6c4009 |
incremented the number of readers. Relaxed MO is fine because
|
|
Packit |
6c4009 |
this is just about waiting for a state change in __readers. */
|
|
Packit |
6c4009 |
if (atomic_compare_exchange_weak_relaxed
|
|
Packit |
6c4009 |
(&rwlock->__data.__readers, &r, r | PTHREAD_RWLOCK_RWAITING))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* Wait for as long as the flag is set. An ABA situation is
|
|
Packit |
6c4009 |
harmless because the flag is just about the state of
|
|
Packit |
6c4009 |
__readers, and all threads set the flag under the same
|
|
Packit |
6c4009 |
conditions. */
|
|
Packit Service |
3b0880 |
while ((atomic_load_relaxed (&rwlock->__data.__readers)
|
|
Packit Service |
3b0880 |
& PTHREAD_RWLOCK_RWAITING) != 0)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
int private = __pthread_rwlock_get_private (rwlock);
|
|
Packit |
6c4009 |
int err = futex_abstimed_wait (&rwlock->__data.__readers,
|
|
Packit Service |
3b0880 |
r, abstime, private);
|
|
Packit |
6c4009 |
/* We ignore EAGAIN and EINTR. On time-outs, we can just
|
|
Packit |
6c4009 |
return because we don't need to clean up anything. */
|
|
Packit |
6c4009 |
if (err == ETIMEDOUT)
|
|
Packit |
6c4009 |
return err;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* It makes sense to not break out of the outer loop here
|
|
Packit |
6c4009 |
because we might be in the same situation again. */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
else
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* TODO Back-off. */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* Register as a reader, using an add-and-fetch so that R can be used as
|
|
Packit |
6c4009 |
expected value for future operations. Acquire MO so we synchronize with
|
|
Packit |
6c4009 |
prior writers as well as the last reader of the previous read phase (see
|
|
Packit |
6c4009 |
below). */
|
|
Packit Service |
3b0880 |
r = atomic_fetch_add_acquire (&rwlock->__data.__readers,
|
|
Packit Service |
3b0880 |
(1 << PTHREAD_RWLOCK_READER_SHIFT)) + (1 << PTHREAD_RWLOCK_READER_SHIFT);
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Check whether there is an overflow in the number of readers. We assume
|
|
Packit |
6c4009 |
that the total number of threads is less than half the maximum number
|
|
Packit |
6c4009 |
of readers that we have bits for in __readers (i.e., with 32-bit int and
|
|
Packit |
6c4009 |
PTHREAD_RWLOCK_READER_SHIFT of 3, we assume there are less than
|
|
Packit |
6c4009 |
1 << (32-3-1) concurrent threads).
|
|
Packit |
6c4009 |
If there is an overflow, we use a CAS to try to decrement the number of
|
|
Packit |
6c4009 |
readers if there still is an overflow situation. If so, we return
|
|
Packit |
6c4009 |
EAGAIN; if not, we are not a thread causing an overflow situation, and so
|
|
Packit |
6c4009 |
we just continue. Using a fetch-add instead of the CAS isn't possible
|
|
Packit |
6c4009 |
because other readers might release the lock concurrently, which could
|
|
Packit |
6c4009 |
make us the last reader and thus responsible for handing ownership over
|
|
Packit |
6c4009 |
to writers (which requires a CAS too to make the decrement and ownership
|
|
Packit |
6c4009 |
transfer indivisible). */
|
|
Packit |
6c4009 |
while (__glibc_unlikely (r >= PTHREAD_RWLOCK_READER_OVERFLOW))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* Relaxed MO is okay because we just want to undo our registration and
|
|
Packit |
6c4009 |
cannot have changed the rwlock state substantially if the CAS
|
|
Packit |
6c4009 |
succeeds. */
|
|
Packit Service |
3b0880 |
if (atomic_compare_exchange_weak_relaxed (&rwlock->__data.__readers, &r,
|
|
Packit Service |
3b0880 |
r - (1 << PTHREAD_RWLOCK_READER_SHIFT)))
|
|
Packit |
6c4009 |
return EAGAIN;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* We have registered as a reader, so if we are in a read phase, we have
|
|
Packit |
6c4009 |
acquired a read lock. This is also the reader--reader fast-path.
|
|
Packit |
6c4009 |
Even if there is a primary writer, we just return. If writers are to
|
|
Packit |
6c4009 |
be preferred and we are the only active reader, we could try to enter a
|
|
Packit |
6c4009 |
write phase to let the writer proceed. This would be okay because we
|
|
Packit |
6c4009 |
cannot have acquired the lock previously as a reader (which could result
|
|
Packit |
6c4009 |
in deadlock if we would wait for the primary writer to run). However,
|
|
Packit |
6c4009 |
this seems to be a corner case and handling it specially not be worth the
|
|
Packit |
6c4009 |
complexity. */
|
|
Packit |
6c4009 |
if (__glibc_likely ((r & PTHREAD_RWLOCK_WRPHASE) == 0))
|
|
Packit |
6c4009 |
return 0;
|
|
Packit |
6c4009 |
/* Otherwise, if we were in a write phase (states #6 or #8), we must wait
|
|
Packit |
6c4009 |
for explicit hand-over of the read phase; the only exception is if we
|
|
Packit |
6c4009 |
can start a read phase if there is no primary writer currently. */
|
|
Packit Service |
3b0880 |
while (((r & PTHREAD_RWLOCK_WRPHASE) != 0)
|
|
Packit Service |
3b0880 |
&& ((r & PTHREAD_RWLOCK_WRLOCKED) == 0))
|
|
Packit |
6c4009 |
{
|
|
Packit Service |
3b0880 |
/* Try to enter a read phase: If the CAS below succeeds, we have
|
|
Packit |
6c4009 |
ownership; if it fails, we will simply retry and reassess the
|
|
Packit |
6c4009 |
situation.
|
|
Packit |
6c4009 |
Acquire MO so we synchronize with prior writers. */
|
|
Packit |
6c4009 |
if (atomic_compare_exchange_weak_acquire (&rwlock->__data.__readers, &r,
|
|
Packit Service |
3b0880 |
r ^ PTHREAD_RWLOCK_WRPHASE))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* We started the read phase, so we are also responsible for
|
|
Packit |
6c4009 |
updating the write-phase futex. Relaxed MO is sufficient.
|
|
Packit |
6c4009 |
We have to do the same steps as a writer would when handing
|
|
Packit |
6c4009 |
over the read phase to us because other readers cannot
|
|
Packit |
6c4009 |
distinguish between us and the writer; this includes
|
|
Packit |
6c4009 |
explicit hand-over and potentially having to wake other readers
|
|
Packit |
6c4009 |
(but we can pretend to do the setting and unsetting of WRLOCKED
|
|
Packit |
6c4009 |
atomically, and thus can skip this step). */
|
|
Packit |
6c4009 |
if ((atomic_exchange_relaxed (&rwlock->__data.__wrphase_futex, 0)
|
|
Packit Service |
3b0880 |
& PTHREAD_RWLOCK_FUTEX_USED) != 0)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
int private = __pthread_rwlock_get_private (rwlock);
|
|
Packit |
6c4009 |
futex_wake (&rwlock->__data.__wrphase_futex, INT_MAX, private);
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
return 0;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
else
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* TODO Back off before retrying. Also see above. */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* We were in a write phase but did not install the read phase. We cannot
|
|
Packit |
6c4009 |
distinguish between a writer and another reader starting the read phase,
|
|
Packit |
6c4009 |
so we must wait for explicit hand-over via __wrphase_futex.
|
|
Packit |
6c4009 |
However, __wrphase_futex might not have been set to 1 yet (either
|
|
Packit |
6c4009 |
because explicit hand-over to the writer is still ongoing, or because
|
|
Packit |
6c4009 |
the writer has started the write phase but has not yet updated
|
|
Packit |
6c4009 |
__wrphase_futex). The least recent value of __wrphase_futex we can
|
|
Packit |
6c4009 |
read from here is the modification of the last read phase (because
|
|
Packit |
6c4009 |
we synchronize with the last reader in this read phase through
|
|
Packit |
6c4009 |
__readers; see the use of acquire MO on the fetch_add above).
|
|
Packit |
6c4009 |
Therefore, if we observe a value of 0 for __wrphase_futex, we need
|
|
Packit |
6c4009 |
to subsequently check that __readers now indicates a read phase; we
|
|
Packit |
6c4009 |
need to use acquire MO for this so that if we observe a read phase,
|
|
Packit |
6c4009 |
we will also see the modification of __wrphase_futex by the previous
|
|
Packit |
6c4009 |
writer. We then need to load __wrphase_futex again and continue to
|
|
Packit |
6c4009 |
wait if it is not 0, so that we do not skip explicit hand-over.
|
|
Packit |
6c4009 |
Relaxed MO is sufficient for the load from __wrphase_futex because
|
|
Packit |
6c4009 |
we just use it as an indicator for when we can proceed; we use
|
|
Packit |
6c4009 |
__readers and the acquire MO accesses to it to eventually read from
|
|
Packit |
6c4009 |
the proper stores to __wrphase_futex. */
|
|
Packit |
6c4009 |
unsigned int wpf;
|
|
Packit |
6c4009 |
bool ready = false;
|
|
Packit |
6c4009 |
for (;;)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
while (((wpf = atomic_load_relaxed (&rwlock->__data.__wrphase_futex))
|
|
Packit Service |
3b0880 |
| PTHREAD_RWLOCK_FUTEX_USED) == (1 | PTHREAD_RWLOCK_FUTEX_USED))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
int private = __pthread_rwlock_get_private (rwlock);
|
|
Packit |
6c4009 |
if (((wpf & PTHREAD_RWLOCK_FUTEX_USED) == 0)
|
|
Packit Service |
3b0880 |
&& !atomic_compare_exchange_weak_relaxed
|
|
Packit |
6c4009 |
(&rwlock->__data.__wrphase_futex,
|
|
Packit Service |
3b0880 |
&wpf, wpf | PTHREAD_RWLOCK_FUTEX_USED))
|
|
Packit |
6c4009 |
continue;
|
|
Packit |
6c4009 |
int err = futex_abstimed_wait (&rwlock->__data.__wrphase_futex,
|
|
Packit Service |
3b0880 |
1 | PTHREAD_RWLOCK_FUTEX_USED, abstime, private);
|
|
Packit |
6c4009 |
if (err == ETIMEDOUT)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* If we timed out, we need to unregister. If no read phase
|
|
Packit |
6c4009 |
has been installed while we waited, we can just decrement
|
|
Packit |
6c4009 |
the number of readers. Otherwise, we just acquire the
|
|
Packit |
6c4009 |
lock, which is allowed because we give no precise timing
|
|
Packit |
6c4009 |
guarantees, and because the timeout is only required to
|
|
Packit |
6c4009 |
be in effect if we would have had to wait for other
|
|
Packit |
6c4009 |
threads (e.g., if futex_wait would time-out immediately
|
|
Packit |
6c4009 |
because the given absolute time is in the past). */
|
|
Packit |
6c4009 |
r = atomic_load_relaxed (&rwlock->__data.__readers);
|
|
Packit |
6c4009 |
while ((r & PTHREAD_RWLOCK_WRPHASE) != 0)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* We don't need to make anything else visible to
|
|
Packit |
6c4009 |
others besides unregistering, so relaxed MO is
|
|
Packit |
6c4009 |
sufficient. */
|
|
Packit |
6c4009 |
if (atomic_compare_exchange_weak_relaxed
|
|
Packit |
6c4009 |
(&rwlock->__data.__readers, &r,
|
|
Packit |
6c4009 |
r - (1 << PTHREAD_RWLOCK_READER_SHIFT)))
|
|
Packit |
6c4009 |
return ETIMEDOUT;
|
|
Packit |
6c4009 |
/* TODO Back-off. */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* Use the acquire MO fence to mirror the steps taken in the
|
|
Packit |
6c4009 |
non-timeout case. Note that the read can happen both
|
|
Packit |
6c4009 |
in the atomic_load above as well as in the failure case
|
|
Packit |
6c4009 |
of the CAS operation. */
|
|
Packit |
6c4009 |
atomic_thread_fence_acquire ();
|
|
Packit |
6c4009 |
/* We still need to wait for explicit hand-over, but we must
|
|
Packit |
6c4009 |
not use futex_wait anymore because we would just time out
|
|
Packit |
6c4009 |
in this case and thus make the spin-waiting we need
|
|
Packit |
6c4009 |
unnecessarily expensive. */
|
|
Packit |
6c4009 |
while ((atomic_load_relaxed (&rwlock->__data.__wrphase_futex)
|
|
Packit Service |
3b0880 |
| PTHREAD_RWLOCK_FUTEX_USED)
|
|
Packit Service |
3b0880 |
== (1 | PTHREAD_RWLOCK_FUTEX_USED))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* TODO Back-off? */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
ready = true;
|
|
Packit |
6c4009 |
break;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* If we got interrupted (EINTR) or the futex word does not have the
|
|
Packit |
6c4009 |
expected value (EAGAIN), retry. */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
if (ready)
|
|
Packit |
6c4009 |
/* See below. */
|
|
Packit |
6c4009 |
break;
|
|
Packit |
6c4009 |
/* We need acquire MO here so that we synchronize with the lock
|
|
Packit |
6c4009 |
release of the writer, and so that we observe a recent value of
|
|
Packit |
6c4009 |
__wrphase_futex (see below). */
|
|
Packit |
6c4009 |
if ((atomic_load_acquire (&rwlock->__data.__readers)
|
|
Packit Service |
3b0880 |
& PTHREAD_RWLOCK_WRPHASE) == 0)
|
|
Packit |
6c4009 |
/* We are in a read phase now, so the least recent modification of
|
|
Packit |
6c4009 |
__wrphase_futex we can read from is the store by the writer
|
|
Packit |
6c4009 |
with value 1. Thus, only now we can assume that if we observe
|
|
Packit |
6c4009 |
a value of 0, explicit hand-over is finished. Retry the loop
|
|
Packit |
6c4009 |
above one more time. */
|
|
Packit |
6c4009 |
ready = true;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
return 0;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
static __always_inline void
|
|
Packit |
6c4009 |
__pthread_rwlock_wrunlock (pthread_rwlock_t *rwlock)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
int private = __pthread_rwlock_get_private (rwlock);
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
atomic_store_relaxed (&rwlock->__data.__cur_writer, 0);
|
|
Packit |
6c4009 |
/* Disable waiting by writers. We will wake up after we decided how to
|
|
Packit |
6c4009 |
proceed. */
|
|
Packit Service |
3b0880 |
bool wake_writers = ((atomic_exchange_relaxed
|
|
Packit Service |
3b0880 |
(&rwlock->__data.__writers_futex, 0) & PTHREAD_RWLOCK_FUTEX_USED) != 0);
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
if (rwlock->__data.__flags != PTHREAD_RWLOCK_PREFER_READER_NP)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* First, try to hand over to another writer. */
|
|
Packit |
6c4009 |
unsigned int w = atomic_load_relaxed (&rwlock->__data.__writers);
|
|
Packit |
6c4009 |
while (w != 0)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* Release MO so that another writer that gets WRLOCKED from us will
|
|
Packit |
6c4009 |
synchronize with us and thus can take over our view of
|
|
Packit |
6c4009 |
__readers (including, for example, whether we are in a write
|
|
Packit |
6c4009 |
phase or not). */
|
|
Packit Service |
3b0880 |
if (atomic_compare_exchange_weak_release (&rwlock->__data.__writers,
|
|
Packit Service |
3b0880 |
&w, w | PTHREAD_RWLOCK_WRHANDOVER))
|
|
Packit |
6c4009 |
/* Another writer will take over. */
|
|
Packit |
6c4009 |
goto done;
|
|
Packit |
6c4009 |
/* TODO Back-off. */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* We have done everything we needed to do to prefer writers, so now we
|
|
Packit |
6c4009 |
either hand over explicitly to readers if there are any, or we simply
|
|
Packit |
6c4009 |
stay in a write phase. See pthread_rwlock_rdunlock for more details. */
|
|
Packit |
6c4009 |
unsigned int r = atomic_load_relaxed (&rwlock->__data.__readers);
|
|
Packit |
6c4009 |
/* Release MO so that subsequent readers or writers synchronize with us. */
|
|
Packit |
6c4009 |
while (!atomic_compare_exchange_weak_release
|
|
Packit Service |
3b0880 |
(&rwlock->__data.__readers, &r, (r ^ PTHREAD_RWLOCK_WRLOCKED)
|
|
Packit Service |
3b0880 |
^ ((r >> PTHREAD_RWLOCK_READER_SHIFT) == 0 ? 0
|
|
Packit Service |
3b0880 |
: PTHREAD_RWLOCK_WRPHASE)))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* TODO Back-off. */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
if ((r >> PTHREAD_RWLOCK_READER_SHIFT) != 0)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* We must hand over explicitly through __wrphase_futex. Relaxed MO is
|
|
Packit |
6c4009 |
sufficient because it is just used to delay acquisition by a writer;
|
|
Packit |
6c4009 |
any other synchronizes-with relations that are necessary are
|
|
Packit |
6c4009 |
established through __readers. */
|
|
Packit |
6c4009 |
if ((atomic_exchange_relaxed (&rwlock->__data.__wrphase_futex, 0)
|
|
Packit |
6c4009 |
& PTHREAD_RWLOCK_FUTEX_USED) != 0)
|
|
Packit |
6c4009 |
futex_wake (&rwlock->__data.__wrphase_futex, INT_MAX, private);
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
done:
|
|
Packit |
6c4009 |
/* We released WRLOCKED in some way, so wake a writer. */
|
|
Packit |
6c4009 |
if (wake_writers)
|
|
Packit |
6c4009 |
futex_wake (&rwlock->__data.__writers_futex, 1, private);
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
static __always_inline int
|
|
Packit |
6c4009 |
__pthread_rwlock_wrlock_full (pthread_rwlock_t *rwlock,
|
|
Packit |
6c4009 |
const struct timespec *abstime)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* Make sure we are not holding the rwlock as a writer. This is a deadlock
|
|
Packit |
6c4009 |
situation we recognize and report. */
|
|
Packit |
6c4009 |
if (__glibc_unlikely (atomic_load_relaxed (&rwlock->__data.__cur_writer)
|
|
Packit Service |
3b0880 |
== THREAD_GETMEM (THREAD_SELF, tid)))
|
|
Packit |
6c4009 |
return EDEADLK;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* First we try to acquire the role of primary writer by setting WRLOCKED;
|
|
Packit |
6c4009 |
if it was set before, there already is a primary writer. Acquire MO so
|
|
Packit |
6c4009 |
that we synchronize with previous primary writers.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
We do not try to change to a write phase right away using a fetch_or
|
|
Packit |
6c4009 |
because we would have to reset it again and wake readers if there are
|
|
Packit |
6c4009 |
readers present (some readers could try to acquire the lock more than
|
|
Packit |
6c4009 |
once, so setting a write phase in the middle of this could cause
|
|
Packit |
6c4009 |
deadlock). Changing to a write phase eagerly would only speed up the
|
|
Packit |
6c4009 |
transition from a read phase to a write phase in the uncontended case,
|
|
Packit |
6c4009 |
but it would slow down the contended case if readers are preferred (which
|
|
Packit |
6c4009 |
is the default).
|
|
Packit |
6c4009 |
We could try to CAS from a state with no readers to a write phase, but
|
|
Packit |
6c4009 |
this could be less scalable if readers arrive and leave frequently. */
|
|
Packit |
6c4009 |
bool may_share_futex_used_flag = false;
|
|
Packit |
6c4009 |
unsigned int r = atomic_fetch_or_acquire (&rwlock->__data.__readers,
|
|
Packit Service |
3b0880 |
PTHREAD_RWLOCK_WRLOCKED);
|
|
Packit |
6c4009 |
if (__glibc_unlikely ((r & PTHREAD_RWLOCK_WRLOCKED) != 0))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* There is another primary writer. */
|
|
Packit Service |
3b0880 |
bool prefer_writer =
|
|
Packit Service |
3b0880 |
(rwlock->__data.__flags != PTHREAD_RWLOCK_PREFER_READER_NP);
|
|
Packit |
6c4009 |
if (prefer_writer)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* We register as a waiting writer, so that we can make use of
|
|
Packit |
6c4009 |
writer--writer hand-over. Relaxed MO is fine because we just
|
|
Packit |
6c4009 |
want to register. We assume that the maximum number of threads
|
|
Packit |
6c4009 |
is less than the capacity in __writers. */
|
|
Packit |
6c4009 |
atomic_fetch_add_relaxed (&rwlock->__data.__writers, 1);
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
for (;;)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* TODO Spin until WRLOCKED is 0 before trying the CAS below.
|
|
Packit |
6c4009 |
But pay attention to not delay trying writer--writer hand-over
|
|
Packit |
6c4009 |
for too long (which we must try eventually anyway). */
|
|
Packit |
6c4009 |
if ((r & PTHREAD_RWLOCK_WRLOCKED) == 0)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* Try to become the primary writer or retry. Acquire MO as in
|
|
Packit |
6c4009 |
the fetch_or above. */
|
|
Packit |
6c4009 |
if (atomic_compare_exchange_weak_acquire
|
|
Packit Service |
3b0880 |
(&rwlock->__data.__readers, &r,
|
|
Packit Service |
3b0880 |
r | PTHREAD_RWLOCK_WRLOCKED))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
if (prefer_writer)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* Unregister as a waiting writer. Note that because we
|
|
Packit |
6c4009 |
acquired WRLOCKED, WRHANDOVER will not be set.
|
|
Packit |
6c4009 |
Acquire MO on the CAS above ensures that
|
|
Packit |
6c4009 |
unregistering happens after the previous writer;
|
|
Packit |
6c4009 |
this sorts the accesses to __writers by all
|
|
Packit |
6c4009 |
primary writers in a useful way (e.g., any other
|
|
Packit |
6c4009 |
primary writer acquiring after us or getting it from
|
|
Packit |
6c4009 |
us through WRHANDOVER will see both our changes to
|
|
Packit |
6c4009 |
__writers).
|
|
Packit |
6c4009 |
??? Perhaps this is not strictly necessary for
|
|
Packit |
6c4009 |
reasons we do not yet know of. */
|
|
Packit Service |
3b0880 |
atomic_fetch_add_relaxed (&rwlock->__data.__writers,
|
|
Packit Service |
3b0880 |
-1);
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
break;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* Retry if the CAS fails (r will have been updated). */
|
|
Packit |
6c4009 |
continue;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* If writer--writer hand-over is available, try to become the
|
|
Packit |
6c4009 |
primary writer this way by grabbing the WRHANDOVER token. If we
|
|
Packit |
6c4009 |
succeed, we own WRLOCKED. */
|
|
Packit |
6c4009 |
if (prefer_writer)
|
|
Packit |
6c4009 |
{
|
|
Packit Service |
3b0880 |
unsigned int w = atomic_load_relaxed
|
|
Packit Service |
3b0880 |
(&rwlock->__data.__writers);
|
|
Packit |
6c4009 |
if ((w & PTHREAD_RWLOCK_WRHANDOVER) != 0)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* Acquire MO is required here so that we synchronize with
|
|
Packit |
6c4009 |
the writer that handed over WRLOCKED. We also need this
|
|
Packit |
6c4009 |
for the reload of __readers below because our view of
|
|
Packit |
6c4009 |
__readers must be at least as recent as the view of the
|
|
Packit |
6c4009 |
writer that handed over WRLOCKED; we must avoid an ABA
|
|
Packit |
6c4009 |
through WRHANDOVER, which could, for example, lead to us
|
|
Packit |
6c4009 |
assuming we are still in a write phase when in fact we
|
|
Packit |
6c4009 |
are not. */
|
|
Packit |
6c4009 |
if (atomic_compare_exchange_weak_acquire
|
|
Packit |
6c4009 |
(&rwlock->__data.__writers,
|
|
Packit |
6c4009 |
&w, (w - PTHREAD_RWLOCK_WRHANDOVER - 1)))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* Reload so our view is consistent with the view of
|
|
Packit |
6c4009 |
the previous owner of WRLOCKED. See above. */
|
|
Packit |
6c4009 |
r = atomic_load_relaxed (&rwlock->__data.__readers);
|
|
Packit |
6c4009 |
break;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* We do not need to reload __readers here. We should try
|
|
Packit |
6c4009 |
to perform writer--writer hand-over if possible; if it
|
|
Packit |
6c4009 |
is not possible anymore, we will reload __readers
|
|
Packit |
6c4009 |
elsewhere in this loop. */
|
|
Packit |
6c4009 |
continue;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* We did not acquire WRLOCKED nor were able to use writer--writer
|
|
Packit |
6c4009 |
hand-over, so we block on __writers_futex. */
|
|
Packit |
6c4009 |
int private = __pthread_rwlock_get_private (rwlock);
|
|
Packit Service |
3b0880 |
unsigned int wf = atomic_load_relaxed
|
|
Packit Service |
3b0880 |
(&rwlock->__data.__writers_futex);
|
|
Packit |
6c4009 |
if (((wf & ~(unsigned int) PTHREAD_RWLOCK_FUTEX_USED) != 1)
|
|
Packit |
6c4009 |
|| ((wf != (1 | PTHREAD_RWLOCK_FUTEX_USED))
|
|
Packit Service |
3b0880 |
&& !atomic_compare_exchange_weak_relaxed
|
|
Packit |
6c4009 |
(&rwlock->__data.__writers_futex, &wf,
|
|
Packit Service |
3b0880 |
1 | PTHREAD_RWLOCK_FUTEX_USED)))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* If we cannot block on __writers_futex because there is no
|
|
Packit |
6c4009 |
primary writer, or we cannot set PTHREAD_RWLOCK_FUTEX_USED,
|
|
Packit |
6c4009 |
we retry. We must reload __readers here in case we cannot
|
|
Packit |
6c4009 |
block on __writers_futex so that we can become the primary
|
|
Packit |
6c4009 |
writer and are not stuck in a loop that just continuously
|
|
Packit |
6c4009 |
fails to block on __writers_futex. */
|
|
Packit |
6c4009 |
r = atomic_load_relaxed (&rwlock->__data.__readers);
|
|
Packit |
6c4009 |
continue;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* We set the flag that signals that the futex is used, or we could
|
|
Packit |
6c4009 |
have set it if we had been faster than other waiters. As a
|
|
Packit |
6c4009 |
result, we may share the flag with an unknown number of other
|
|
Packit |
6c4009 |
writers. Therefore, we must keep this flag set when we acquire
|
|
Packit |
6c4009 |
the lock. We do not need to do this when we do not reach this
|
|
Packit |
6c4009 |
point here because then we are not part of the group that may
|
|
Packit |
6c4009 |
share the flag, and another writer will wake one of the writers
|
|
Packit |
6c4009 |
in this group. */
|
|
Packit |
6c4009 |
may_share_futex_used_flag = true;
|
|
Packit |
6c4009 |
int err = futex_abstimed_wait (&rwlock->__data.__writers_futex,
|
|
Packit Service |
3b0880 |
1 | PTHREAD_RWLOCK_FUTEX_USED, abstime, private);
|
|
Packit |
6c4009 |
if (err == ETIMEDOUT)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
if (prefer_writer)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* We need to unregister as a waiting writer. If we are the
|
|
Packit |
6c4009 |
last writer and writer--writer hand-over is available,
|
|
Packit |
6c4009 |
we must make use of it because nobody else will reset
|
|
Packit |
6c4009 |
WRLOCKED otherwise. (If we use it, we simply pretend
|
|
Packit |
6c4009 |
that this happened before the timeout; see
|
|
Packit |
6c4009 |
pthread_rwlock_rdlock_full for the full reasoning.)
|
|
Packit |
6c4009 |
Also see the similar code above. */
|
|
Packit Service |
3b0880 |
unsigned int w = atomic_load_relaxed
|
|
Packit Service |
3b0880 |
(&rwlock->__data.__writers);
|
|
Packit |
6c4009 |
while (!atomic_compare_exchange_weak_acquire
|
|
Packit Service |
3b0880 |
(&rwlock->__data.__writers, &w,
|
|
Packit |
6c4009 |
(w == PTHREAD_RWLOCK_WRHANDOVER + 1 ? 0 : w - 1)))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* TODO Back-off. */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
if (w == PTHREAD_RWLOCK_WRHANDOVER + 1)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* We must continue as primary writer. See above. */
|
|
Packit |
6c4009 |
r = atomic_load_relaxed (&rwlock->__data.__readers);
|
|
Packit |
6c4009 |
break;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* We cleaned up and cannot have stolen another waiting writer's
|
|
Packit |
6c4009 |
futex wake-up, so just return. */
|
|
Packit |
6c4009 |
return ETIMEDOUT;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* If we got interrupted (EINTR) or the futex word does not have the
|
|
Packit |
6c4009 |
expected value (EAGAIN), retry after reloading __readers. */
|
|
Packit |
6c4009 |
r = atomic_load_relaxed (&rwlock->__data.__readers);
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* Our snapshot of __readers is up-to-date at this point because we
|
|
Packit |
6c4009 |
either set WRLOCKED using a CAS (and update r accordingly below,
|
|
Packit |
6c4009 |
which was used as expected value for the CAS) or got WRLOCKED from
|
|
Packit |
6c4009 |
another writer whose snapshot of __readers we inherit. */
|
|
Packit |
6c4009 |
r |= PTHREAD_RWLOCK_WRLOCKED;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* We are the primary writer; enable blocking on __writers_futex. Relaxed
|
|
Packit |
6c4009 |
MO is sufficient for futex words; acquire MO on the previous
|
|
Packit |
6c4009 |
modifications of __readers ensures that this store happens after the
|
|
Packit |
6c4009 |
store of value 0 by the previous primary writer. */
|
|
Packit |
6c4009 |
atomic_store_relaxed (&rwlock->__data.__writers_futex,
|
|
Packit Service |
3b0880 |
1 | (may_share_futex_used_flag ? PTHREAD_RWLOCK_FUTEX_USED : 0));
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* If we are in a write phase, we have acquired the lock. */
|
|
Packit |
6c4009 |
if ((r & PTHREAD_RWLOCK_WRPHASE) != 0)
|
|
Packit |
6c4009 |
goto done;
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* If we are in a read phase and there are no readers, try to start a write
|
|
Packit |
6c4009 |
phase. */
|
|
Packit Service |
3b0880 |
while (((r & PTHREAD_RWLOCK_WRPHASE) == 0)
|
|
Packit Service |
3b0880 |
&& ((r >> PTHREAD_RWLOCK_READER_SHIFT) == 0))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* Acquire MO so that we synchronize with prior writers and do
|
|
Packit |
6c4009 |
not interfere with their updates to __writers_futex, as well
|
|
Packit |
6c4009 |
as regarding prior readers and their updates to __wrphase_futex,
|
|
Packit |
6c4009 |
respectively. */
|
|
Packit |
6c4009 |
if (atomic_compare_exchange_weak_acquire (&rwlock->__data.__readers,
|
|
Packit Service |
3b0880 |
&r, r | PTHREAD_RWLOCK_WRPHASE))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* We have started a write phase, so need to enable readers to wait.
|
|
Packit |
6c4009 |
See the similar case in __pthread_rwlock_rdlock_full. Unlike in
|
|
Packit |
6c4009 |
that similar case, we are the (only) primary writer and so do
|
|
Packit |
6c4009 |
not need to wake another writer. */
|
|
Packit |
6c4009 |
atomic_store_relaxed (&rwlock->__data.__wrphase_futex, 1);
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
goto done;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* TODO Back-off. */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* We became the primary writer in a read phase and there were readers when
|
|
Packit |
6c4009 |
we did (because of the previous loop). Thus, we have to wait for
|
|
Packit |
6c4009 |
explicit hand-over from one of these readers.
|
|
Packit |
6c4009 |
We basically do the same steps as for the similar case in
|
|
Packit |
6c4009 |
__pthread_rwlock_rdlock_full, except that we additionally might try
|
|
Packit |
6c4009 |
to directly hand over to another writer and need to wake up
|
|
Packit |
6c4009 |
other writers or waiting readers (i.e., PTHREAD_RWLOCK_RWAITING). */
|
|
Packit |
6c4009 |
unsigned int wpf;
|
|
Packit |
6c4009 |
bool ready = false;
|
|
Packit |
6c4009 |
for (;;)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
while (((wpf = atomic_load_relaxed (&rwlock->__data.__wrphase_futex))
|
|
Packit Service |
3b0880 |
| PTHREAD_RWLOCK_FUTEX_USED) == PTHREAD_RWLOCK_FUTEX_USED)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
int private = __pthread_rwlock_get_private (rwlock);
|
|
Packit Service |
3b0880 |
if (((wpf & PTHREAD_RWLOCK_FUTEX_USED) == 0)
|
|
Packit Service |
3b0880 |
&& !atomic_compare_exchange_weak_relaxed
|
|
Packit |
6c4009 |
(&rwlock->__data.__wrphase_futex, &wpf,
|
|
Packit Service |
3b0880 |
PTHREAD_RWLOCK_FUTEX_USED))
|
|
Packit |
6c4009 |
continue;
|
|
Packit |
6c4009 |
int err = futex_abstimed_wait (&rwlock->__data.__wrphase_futex,
|
|
Packit Service |
3b0880 |
PTHREAD_RWLOCK_FUTEX_USED, abstime, private);
|
|
Packit |
6c4009 |
if (err == ETIMEDOUT)
|
|
Packit |
6c4009 |
{
|
|
Packit Service |
3b0880 |
if (rwlock->__data.__flags
|
|
Packit Service |
3b0880 |
!= PTHREAD_RWLOCK_PREFER_READER_NP)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* We try writer--writer hand-over. */
|
|
Packit Service |
3b0880 |
unsigned int w = atomic_load_relaxed
|
|
Packit Service |
3b0880 |
(&rwlock->__data.__writers);
|
|
Packit |
6c4009 |
if (w != 0)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* We are about to hand over WRLOCKED, so we must
|
|
Packit |
6c4009 |
release __writers_futex too; otherwise, we'd have
|
|
Packit |
6c4009 |
a pending store, which could at least prevent
|
|
Packit |
6c4009 |
other threads from waiting using the futex
|
|
Packit |
6c4009 |
because it could interleave with the stores
|
|
Packit |
6c4009 |
by subsequent writers. In turn, this means that
|
|
Packit |
6c4009 |
we have to clean up when we do not hand over
|
|
Packit |
6c4009 |
WRLOCKED.
|
|
Packit |
6c4009 |
Release MO so that another writer that gets
|
|
Packit |
6c4009 |
WRLOCKED from us can take over our view of
|
|
Packit |
6c4009 |
__readers. */
|
|
Packit Service |
3b0880 |
unsigned int wf = atomic_exchange_relaxed
|
|
Packit Service |
3b0880 |
(&rwlock->__data.__writers_futex, 0);
|
|
Packit |
6c4009 |
while (w != 0)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
if (atomic_compare_exchange_weak_release
|
|
Packit |
6c4009 |
(&rwlock->__data.__writers, &w,
|
|
Packit Service |
3b0880 |
w | PTHREAD_RWLOCK_WRHANDOVER))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* Wake other writers. */
|
|
Packit |
6c4009 |
if ((wf & PTHREAD_RWLOCK_FUTEX_USED) != 0)
|
|
Packit |
6c4009 |
futex_wake (&rwlock->__data.__writers_futex,
|
|
Packit |
6c4009 |
1, private);
|
|
Packit |
6c4009 |
return ETIMEDOUT;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* TODO Back-off. */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* We still own WRLOCKED and someone else might set
|
|
Packit |
6c4009 |
a write phase concurrently, so enable waiting
|
|
Packit |
6c4009 |
again. Make sure we don't loose the flag that
|
|
Packit |
6c4009 |
signals whether there are threads waiting on
|
|
Packit |
6c4009 |
this futex. */
|
|
Packit Service |
3b0880 |
atomic_store_relaxed
|
|
Packit Service |
3b0880 |
(&rwlock->__data.__writers_futex, wf);
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* If we timed out and we are not in a write phase, we can
|
|
Packit |
6c4009 |
just stop being a primary writer. Otherwise, we just
|
|
Packit |
6c4009 |
acquire the lock. */
|
|
Packit |
6c4009 |
r = atomic_load_relaxed (&rwlock->__data.__readers);
|
|
Packit |
6c4009 |
if ((r & PTHREAD_RWLOCK_WRPHASE) == 0)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* We are about to release WRLOCKED, so we must release
|
|
Packit |
6c4009 |
__writers_futex too; see the handling of
|
|
Packit |
6c4009 |
writer--writer hand-over above. */
|
|
Packit Service |
3b0880 |
unsigned int wf = atomic_exchange_relaxed
|
|
Packit Service |
3b0880 |
(&rwlock->__data.__writers_futex, 0);
|
|
Packit |
6c4009 |
while ((r & PTHREAD_RWLOCK_WRPHASE) == 0)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* While we don't need to make anything from a
|
|
Packit |
6c4009 |
caller's critical section visible to other
|
|
Packit |
6c4009 |
threads, we need to ensure that our changes to
|
|
Packit |
6c4009 |
__writers_futex are properly ordered.
|
|
Packit |
6c4009 |
Therefore, use release MO to synchronize with
|
|
Packit |
6c4009 |
subsequent primary writers. Also wake up any
|
|
Packit |
6c4009 |
waiting readers as they are waiting because of
|
|
Packit |
6c4009 |
us. */
|
|
Packit |
6c4009 |
if (atomic_compare_exchange_weak_release
|
|
Packit |
6c4009 |
(&rwlock->__data.__readers, &r,
|
|
Packit |
6c4009 |
(r ^ PTHREAD_RWLOCK_WRLOCKED)
|
|
Packit |
6c4009 |
& ~(unsigned int) PTHREAD_RWLOCK_RWAITING))
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* Wake other writers. */
|
|
Packit |
6c4009 |
if ((wf & PTHREAD_RWLOCK_FUTEX_USED) != 0)
|
|
Packit |
6c4009 |
futex_wake (&rwlock->__data.__writers_futex,
|
|
Packit Service |
3b0880 |
1, private);
|
|
Packit |
6c4009 |
/* Wake waiting readers. */
|
|
Packit |
6c4009 |
if ((r & PTHREAD_RWLOCK_RWAITING) != 0)
|
|
Packit |
6c4009 |
futex_wake (&rwlock->__data.__readers,
|
|
Packit Service |
3b0880 |
INT_MAX, private);
|
|
Packit |
6c4009 |
return ETIMEDOUT;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* We still own WRLOCKED and someone else might set a
|
|
Packit |
6c4009 |
write phase concurrently, so enable waiting again.
|
|
Packit |
6c4009 |
Make sure we don't loose the flag that signals
|
|
Packit |
6c4009 |
whether there are threads waiting on this futex. */
|
|
Packit |
6c4009 |
atomic_store_relaxed (&rwlock->__data.__writers_futex, wf);
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* Use the acquire MO fence to mirror the steps taken in the
|
|
Packit |
6c4009 |
non-timeout case. Note that the read can happen both
|
|
Packit |
6c4009 |
in the atomic_load above as well as in the failure case
|
|
Packit |
6c4009 |
of the CAS operation. */
|
|
Packit |
6c4009 |
atomic_thread_fence_acquire ();
|
|
Packit |
6c4009 |
/* We still need to wait for explicit hand-over, but we must
|
|
Packit |
6c4009 |
not use futex_wait anymore. */
|
|
Packit Service |
3b0880 |
while ((atomic_load_relaxed
|
|
Packit Service |
3b0880 |
(&rwlock->__data.__wrphase_futex)
|
|
Packit Service |
3b0880 |
| PTHREAD_RWLOCK_FUTEX_USED)
|
|
Packit Service |
3b0880 |
== PTHREAD_RWLOCK_FUTEX_USED)
|
|
Packit |
6c4009 |
{
|
|
Packit |
6c4009 |
/* TODO Back-off. */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
ready = true;
|
|
Packit |
6c4009 |
break;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* If we got interrupted (EINTR) or the futex word does not have
|
|
Packit |
6c4009 |
the expected value (EAGAIN), retry. */
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
/* See pthread_rwlock_rdlock_full. */
|
|
Packit |
6c4009 |
if (ready)
|
|
Packit |
6c4009 |
break;
|
|
Packit |
6c4009 |
if ((atomic_load_acquire (&rwlock->__data.__readers)
|
|
Packit Service |
3b0880 |
& PTHREAD_RWLOCK_WRPHASE) != 0)
|
|
Packit |
6c4009 |
ready = true;
|
|
Packit |
6c4009 |
}
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
done:
|
|
Packit |
6c4009 |
atomic_store_relaxed (&rwlock->__data.__cur_writer,
|
|
Packit Service |
3b0880 |
THREAD_GETMEM (THREAD_SELF, tid));
|
|
Packit |
6c4009 |
return 0;
|
|
Packit |
6c4009 |
}
|