This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] nptl: Consolidate pthread_{timed,try}join{_np}
- From: Carlos O'Donell <carlos at redhat dot com>
- To: Adhemerval Zanella <adhemerval dot zanella at linaro dot org>, libc-alpha at sourceware dot org
- Date: Tue, 19 Dec 2017 21:44:00 -0800
- Subject: Re: [PATCH] nptl: Consolidate pthread_{timed,try}join{_np}
- Authentication-results: sourceware.org; auth=none
- References: <1513690827-11941-1-git-send-email-adhemerval.zanella@linaro.org>
On 12/19/2017 05:40 AM, Adhemerval Zanella wrote:
> This patch consolidates the pthread_join and gnu extensions to avoid
> code duplication. The function pthread_join, pthread_tryjoin_np, and
> pthread_timedjoin_np are now based on pthread_timedjoin_ex.
>
> It also fixes some inconsistencies on ESRCH, EINVAL, EDEADLK handling
> (where each implementation differs from each other) and also on
> clenup handler (which now always use a CAS).
High level:
Refactoring common code like this is a good idea.
Design:
Moving everything into __pthread_timedjoin_ex and using a bool for
blocking seems like a good refactoring, anything less would duplicate
code again.
I like the removal of the EINVAL checks in the lowlevellock.h code
which harmonizes this more with all the arches.
Implementation:
We have started using pthread_cond_common.c, pthread_rwlock_common.c,
and I'd like to see us use pthread_join_common.c for the common file
with all others including it. This might sound like bike-shedding, but
it helps one quickly determine where the common code lies. New developers
will thank us for the logical choices like this.
> Checked on i686-linux-gnu and x86_64-linux-gnu.
>
> * nptl/pthreadP.h (__pthread_timedjoin_np): Define.
> * nptl/pthread_join.c (pthread_join): Use __pthread_timedjoin_np.
> * nptl/pthread_tryjoin.c (pthread_tryjoin): Likewise.
> * nptl/pthread_timedjoin.c (cleanup): Use CAS on argument setting.
> (pthread_timedjoin_np): Define internal symbol and common code from
> pthread_join.
> * sysdeps/unix/sysv/linux/i386/lowlevellock.h (__lll_timedwait_tid):
> Remove superflous checks.
> * sysdeps/unix/sysv/linux/x86_64/lowlevellock.h (__lll_timedwait_tid):
> Likewise.
OK with rework of pthread_join_common.c.
Reviewed-by: Carlos O'Donell <carlos@redhat.com>
> ---
> ChangeLog | 13 ++++
> nptl/pthreadP.h | 3 +
> nptl/pthread_join.c | 94 +--------------------------
> nptl/pthread_timedjoin.c | 80 +++++++++++++++--------
> nptl/pthread_tryjoin.c | 43 ++----------
> sysdeps/unix/sysv/linux/i386/lowlevellock.h | 7 +-
> sysdeps/unix/sysv/linux/x86_64/lowlevellock.h | 7 +-
> 7 files changed, 75 insertions(+), 172 deletions(-)
>
> diff --git a/nptl/pthreadP.h b/nptl/pthreadP.h
> index 713000e..2646b8f 100644
> --- a/nptl/pthreadP.h
> +++ b/nptl/pthreadP.h
> @@ -506,6 +506,8 @@ extern int __pthread_setcanceltype (int type, int *oldtype);
> extern int __pthread_enable_asynccancel (void) attribute_hidden;
> extern void __pthread_disable_asynccancel (int oldtype) attribute_hidden;
> extern void __pthread_testcancel (void);
> +extern int __pthread_timedjoin_ex (pthread_t, void **, const struct timespec *,
> + bool);
OK.
>
> #if IS_IN (libpthread)
> hidden_proto (__pthread_mutex_init)
> @@ -524,6 +526,7 @@ hidden_proto (__pthread_setcancelstate)
> hidden_proto (__pthread_testcancel)
> hidden_proto (__pthread_mutexattr_init)
> hidden_proto (__pthread_mutexattr_settype)
> +hidden_proto (__pthread_timedjoin_ex)
OK.
> #endif
>
> extern int __pthread_cond_broadcast_2_0 (pthread_cond_2_0_t *cond);
> diff --git a/nptl/pthread_join.c b/nptl/pthread_join.c
> index afc8c37..47379a6 100644
> --- a/nptl/pthread_join.c
> +++ b/nptl/pthread_join.c
> @@ -16,103 +16,11 @@
> License along with the GNU C Library; if not, see
> <http://www.gnu.org/licenses/>. */
>
> -#include <errno.h>
> -#include <stdlib.h>
> -
> -#include <atomic.h>
OK.
> #include "pthreadP.h"
>
> -#include <stap-probe.h>
> -
> -
> -static void
> -cleanup (void *arg)
> -{
> - /* If we already changed the waiter ID, reset it. The call cannot
> - fail for any reason but the thread not having done that yet so
> - there is no reason for a loop. */
> - (void) atomic_compare_and_exchange_bool_acq ((struct pthread **) arg, NULL,
> - THREAD_SELF);
> -}
> -
> -
> int
> __pthread_join (pthread_t threadid, void **thread_return)
> {
> - struct pthread *pd = (struct pthread *) threadid;
> -
> - /* Make sure the descriptor is valid. */
> - if (INVALID_NOT_TERMINATED_TD_P (pd))
> - /* Not a valid thread handle. */
> - return ESRCH;
> -
> - /* Is the thread joinable?. */
> - if (IS_DETACHED (pd))
> - /* We cannot wait for the thread. */
> - return EINVAL;
> -
> - struct pthread *self = THREAD_SELF;
> - int result = 0;
> -
> - LIBC_PROBE (pthread_join, 1, threadid);
> -
> - /* During the wait we change to asynchronous cancellation. If we
> - are canceled the thread we are waiting for must be marked as
> - un-wait-ed for again. */
> - pthread_cleanup_push (cleanup, &pd->joinid);
> -
> - /* Switch to asynchronous cancellation. */
> - int oldtype = CANCEL_ASYNC ();
> -
> - if ((pd == self
> - || (self->joinid == pd
> - && (pd->cancelhandling
> - & (CANCELING_BITMASK | CANCELED_BITMASK | EXITING_BITMASK
> - | TERMINATED_BITMASK)) == 0))
> - && !CANCEL_ENABLED_AND_CANCELED (self->cancelhandling))
> - /* This is a deadlock situation. The threads are waiting for each
> - other to finish. Note that this is a "may" error. To be 100%
> - sure we catch this error we would have to lock the data
> - structures but it is not necessary. In the unlikely case that
> - two threads are really caught in this situation they will
> - deadlock. It is the programmer's problem to figure this
> - out. */
> - result = EDEADLK;
> - /* Wait for the thread to finish. If it is already locked something
> - is wrong. There can only be one waiter. */
> - else if (__builtin_expect (atomic_compare_and_exchange_bool_acq (&pd->joinid,
> - self,
> - NULL), 0))
> - /* There is already somebody waiting for the thread. */
> - result = EINVAL;
> - else
> - /* Wait for the child. */
> - lll_wait_tid (pd->tid);
> -
> -
> - /* Restore cancellation mode. */
> - CANCEL_RESET (oldtype);
> -
> - /* Remove the handler. */
> - pthread_cleanup_pop (0);
> -
> -
> - if (__glibc_likely (result == 0))
> - {
> - /* We mark the thread as terminated and as joined. */
> - pd->tid = -1;
> -
> - /* Store the return value if the caller is interested. */
> - if (thread_return != NULL)
> - *thread_return = pd->result;
> -
> -
> - /* Free the TCB. */
> - __free_tcb (pd);
> - }
> -
> - LIBC_PROBE (pthread_join_ret, 3, threadid, result, pd->result);
> -
> - return result;
> + return __pthread_timedjoin_ex (threadid, thread_return, NULL, true);
OK.
> }
> weak_alias (__pthread_join, pthread_join)
> diff --git a/nptl/pthread_timedjoin.c b/nptl/pthread_timedjoin.c
> index 567c171..6b894f2 100644
> --- a/nptl/pthread_timedjoin.c
> +++ b/nptl/pthread_timedjoin.c
> @@ -21,21 +21,25 @@
> #include <atomic.h>
> #include "pthreadP.h"
>
> +#include <stap-probe.h>
> +
>
> static void
> cleanup (void *arg)
> {
> - *(void **) arg = NULL;
> + /* If we already changed the waiter ID, reset it. The call cannot
> + fail for any reason but the thread not having done that yet so
> + there is no reason for a loop. */
> + struct pthread *self = THREAD_SELF;
> + atomic_compare_exchange_weak_acquire (&arg, &self, NULL);
> }
OK.
>
>
> int
> -pthread_timedjoin_np (pthread_t threadid, void **thread_return,
> - const struct timespec *abstime)
> +__pthread_timedjoin_ex (pthread_t threadid, void **thread_return,
> + const struct timespec *abstime, bool block)
> {
> - struct pthread *self;
> struct pthread *pd = (struct pthread *) threadid;
> - int result;
>
> /* Make sure the descriptor is valid. */
> if (INVALID_NOT_TERMINATED_TD_P (pd))
> @@ -47,8 +51,17 @@ pthread_timedjoin_np (pthread_t threadid, void **thread_return,
> /* We cannot wait for the thread. */
> return EINVAL;
>
> - self = THREAD_SELF;
> - if (pd == self || self->joinid == pd)
> + struct pthread *self = THREAD_SELF;
> + int result = 0;
> +
> + LIBC_PROBE (pthread_join, 1, threadid);
OK.
> +
> + if ((pd == self
> + || (self->joinid == pd
> + && (pd->cancelhandling
> + & (CANCELING_BITMASK | CANCELED_BITMASK | EXITING_BITMASK
> + | TERMINATED_BITMASK)) == 0))
> + && !CANCEL_ENABLED_AND_CANCELED (self->cancelhandling))
> /* This is a deadlock situation. The threads are waiting for each
> other to finish. Note that this is a "may" error. To be 100%
> sure we catch this error we would have to lock the data
> @@ -60,45 +73,56 @@ pthread_timedjoin_np (pthread_t threadid, void **thread_return,
>
> /* Wait for the thread to finish. If it is already locked something
> is wrong. There can only be one waiter. */
> - if (__builtin_expect (atomic_compare_and_exchange_bool_acq (&pd->joinid,
> - self, NULL), 0))
> + else if (__glibc_unlikely (atomic_compare_exchange_weak_acquire (&pd->joinid,
> + &self,
> + NULL)))
> /* There is already somebody waiting for the thread. */
> return EINVAL;
>
> + if (block)
> + {
> + /* During the wait we change to asynchronous cancellation. If we
> + are cancelled the thread we are waiting for must be marked as
> + un-wait-ed for again. */
> + pthread_cleanup_push (cleanup, &pd->joinid);
>
> - /* During the wait we change to asynchronous cancellation. If we
> - are cancelled the thread we are waiting for must be marked as
> - un-wait-ed for again. */
> - pthread_cleanup_push (cleanup, &pd->joinid);
> -
> - /* Switch to asynchronous cancellation. */
> - int oldtype = CANCEL_ASYNC ();
> -
> -
> - /* Wait for the child. */
> - result = lll_timedwait_tid (pd->tid, abstime);
> -
> + int oldtype = CANCEL_ASYNC ();
OK.
>
> - /* Restore cancellation mode. */
> - CANCEL_RESET (oldtype);
> + if (abstime != NULL)
> + result = lll_timedwait_tid (pd->tid, abstime);
> + else
> + lll_wait_tid (pd->tid);
>
> - /* Remove the handler. */
> - pthread_cleanup_pop (0);
> + CANCEL_RESET (oldtype);
>
> + pthread_cleanup_pop (0);
> + }
>
> - /* We might have timed out. */
> - if (result == 0)
> + if (__glibc_likely (result == 0))
> {
> + /* We mark the thread as terminated and as joined. */
> + pd->tid = -1;
OK.
> +
> /* Store the return value if the caller is interested. */
> if (thread_return != NULL)
> *thread_return = pd->result;
>
> -
> /* Free the TCB. */
> __free_tcb (pd);
> }
> else
> pd->joinid = NULL;
>
> + LIBC_PROBE (pthread_join_ret, 3, threadid, result, pd->result);
> +
> return result;
> }
> +hidden_def (__pthread_timedjoin_ex)
> +
> +int
> +__pthread_timedjoin_np (pthread_t threadid, void **thread_return,
> + const struct timespec *abstime)
> +{
> + return __pthread_timedjoin_ex (threadid, thread_return, abstime, true);
> +}
OK.
> +weak_alias (__pthread_timedjoin_np, pthread_timedjoin_np)
> diff --git a/nptl/pthread_tryjoin.c b/nptl/pthread_tryjoin.c
> index 6a3b62e..e7c7595 100644
> --- a/nptl/pthread_tryjoin.c
> +++ b/nptl/pthread_tryjoin.c
> @@ -26,47 +26,12 @@
> int
> pthread_tryjoin_np (pthread_t threadid, void **thread_return)
> {
> - struct pthread *self;
> - struct pthread *pd = (struct pthread *) threadid;
> -
> - /* Make sure the descriptor is valid. */
> - if (DEBUGGING_P && __find_in_stack_list (pd) == NULL)
> - /* Not a valid thread handle. */
> - return ESRCH;
> -
> - /* Is the thread joinable?. */
> - if (IS_DETACHED (pd))
> - /* We cannot wait for the thread. */
> - return EINVAL;
> -
> - self = THREAD_SELF;
> - if (pd == self || self->joinid == pd)
> - /* This is a deadlock situation. The threads are waiting for each
> - other to finish. Note that this is a "may" error. To be 100%
> - sure we catch this error we would have to lock the data
> - structures but it is not necessary. In the unlikely case that
> - two threads are really caught in this situation they will
> - deadlock. It is the programmer's problem to figure this
> - out. */
> - return EDEADLK;
> -
> /* Return right away if the thread hasn't terminated yet. */
> + struct pthread *pd = (struct pthread *) threadid;
> if (pd->tid != 0)
> return EBUSY;
>
OK.
> - /* Wait for the thread to finish. If it is already locked something
> - is wrong. There can only be one waiter. */
> - if (atomic_compare_and_exchange_bool_acq (&pd->joinid, self, NULL))
> - /* There is already somebody waiting for the thread. */
> - return EINVAL;
> -
> - /* Store the return value if the caller is interested. */
> - if (thread_return != NULL)
> - *thread_return = pd->result;
> -
> -
> - /* Free the TCB. */
> - __free_tcb (pd);
> -
> - return 0;
> + /* If pd->tid != 0 then lll_wait_tid will not block on futex
> + operation. */
> + return __pthread_timedjoin_ex (threadid, thread_return, NULL, false);
OK.
> }
> diff --git a/sysdeps/unix/sysv/linux/i386/lowlevellock.h b/sysdeps/unix/sysv/linux/i386/lowlevellock.h
> index 197bb1f..098c64b 100644
> --- a/sysdeps/unix/sysv/linux/i386/lowlevellock.h
> +++ b/sysdeps/unix/sysv/linux/i386/lowlevellock.h
> @@ -237,12 +237,7 @@ extern int __lll_timedwait_tid (int *tid, const struct timespec *abstime)
> ({ \
> int __result = 0; \
> if ((tid) != 0) \
> - { \
> - if ((abstime)->tv_nsec < 0 || (abstime)->tv_nsec >= 1000000000) \
> - __result = EINVAL; \
> - else \
> - __result = __lll_timedwait_tid (&(tid), (abstime)); \
> - } \
> + __result = __lll_timedwait_tid (&(tid), (abstime)); \
OK. Don't handle EINVAL in any special way and harmonize in the common code.
> __result; })
>
>
> diff --git a/sysdeps/unix/sysv/linux/x86_64/lowlevellock.h b/sysdeps/unix/sysv/linux/x86_64/lowlevellock.h
> index cbf6597..0b56a2e 100644
> --- a/sysdeps/unix/sysv/linux/x86_64/lowlevellock.h
> +++ b/sysdeps/unix/sysv/linux/x86_64/lowlevellock.h
> @@ -249,12 +249,7 @@ extern int __lll_timedwait_tid (int *, const struct timespec *)
> ({ \
> int __result = 0; \
> if ((tid) != 0) \
> - { \
> - if ((abstime)->tv_nsec < 0 || (abstime)->tv_nsec >= 1000000000) \
> - __result = EINVAL; \
> - else \
> - __result = __lll_timedwait_tid (&(tid), (abstime)); \
> - } \
> + __result = __lll_timedwait_tid (&(tid), (abstime)); \
OK.
> __result; })
>
> extern int __lll_lock_elision (int *futex, short *adapt_count, int private)
>
--
Cheers,
Carlos.