This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH v4 12/12] nptl: Remove futex_supports_exact_relative_timeouts
On 18/06/2019 13:33, Mike Crowe wrote:
> The only implementation of futex_supports_exact_relative_timeouts always
> returns true. Let's remove it and all its callers.
LGTM.
>
> * nptl/pthread_cond_wait.c: (__pthread_cond_clockwait): Remove code that is
> only useful if futex_supports_exact_relative_timeouts () returns false.
>
> * nptl/pthread_condattr_setclock.c: (pthread_condattr_setclock): Likewise.
>
> * sysdeps/nptl/futex-internal.h: Remove comment about relative timeouts
> potentially being imprecise since it's no longer true. Remove declaration
> of futex_supports_exact_relative_timeouts.
>
> * sysdeps/unix/sysv/linux/futex-internal.h: Remove implementation of
> futex_supports_exact_relative_timeouts.
Ok.
> ---
> ChangeLog | 13 +++++++++++++
> nptl/pthread_cond_wait.c | 5 -----
> nptl/pthread_condattr_setclock.c | 5 -----
> sysdeps/nptl/futex-internal.h | 9 +--------
> sysdeps/unix/sysv/linux/futex-internal.h | 8 --------
> 5 files changed, 14 insertions(+), 26 deletions(-)
>
> diff --git a/ChangeLog b/ChangeLog
> index 984275c..cf07e4d 100644
> --- a/ChangeLog
> +++ b/ChangeLog
> @@ -1,3 +1,16 @@
> +2019-06-18 Mike Crowe <mac@mcrowe.com>
> +
> + * nptl/pthread_cond_wait.c: (__pthread_cond_clockwait): Remove code
> + that is only useful if futex_supports_exact_relative_timeouts ()
> + returns false.
> + * nptl/pthread_condattr_setclock.c: (pthread_condattr_setclock):
> + Likewise.
> + * sysdeps/nptl/futex-internal.h: Remove comment about relative
> + timeouts potentially being imprecise since it's no longer true.
> + Remove declaration of futex_supports_exact_relative_timeouts.
> + * sysdeps/unix/sysv/linux/futex-internal.h: Remove implementation
> + of futex_supports_exact_relative_timeouts.
> +
> 2019-05-27 Mike Crowe <mac@mcrowe.com>
>
> * NEWS: Mention recently-added pthread_cond_clockwait,
> diff --git a/nptl/pthread_cond_wait.c b/nptl/pthread_cond_wait.c
> index 558f930..cd98f50 100644
> --- a/nptl/pthread_cond_wait.c
> +++ b/nptl/pthread_cond_wait.c
> @@ -670,11 +670,6 @@ __pthread_cond_clockwait (pthread_cond_t *cond, pthread_mutex_t *mutex,
> if (!futex_abstimed_supported_clockid (clockid))
> return EINVAL;
>
> - /* If we do not support waiting using CLOCK_MONOTONIC, return an error. */
> - if (clockid == CLOCK_MONOTONIC
> - && !futex_supports_exact_relative_timeouts ())
> - return EINVAL;
> -
> return __pthread_cond_wait_common (cond, mutex, clockid, abstime);
> }
>
Ok.
> diff --git a/nptl/pthread_condattr_setclock.c b/nptl/pthread_condattr_setclock.c
> index 641a041..ac91923 100644
> --- a/nptl/pthread_condattr_setclock.c
> +++ b/nptl/pthread_condattr_setclock.c
> @@ -33,11 +33,6 @@ pthread_condattr_setclock (pthread_condattr_t *attr, clockid_t clock_id)
> in the pthread_cond_t structure needs to be adjusted. */
> return EINVAL;
>
> - /* If we do not support waiting using CLOCK_MONOTONIC, return an error. */
> - if (clock_id == CLOCK_MONOTONIC
> - && !futex_supports_exact_relative_timeouts())
> - return ENOTSUP;
> -
> /* Make sure the value fits in the bits we reserved. */
> assert (clock_id < (1 << COND_CLOCK_BITS));
>
Ok.
> diff --git a/sysdeps/nptl/futex-internal.h b/sysdeps/nptl/futex-internal.h
> index bc29bdb..9baae38 100644
> --- a/sysdeps/nptl/futex-internal.h
> +++ b/sysdeps/nptl/futex-internal.h
> @@ -51,8 +51,7 @@
> Both absolute and relative timeouts can be used. An absolute timeout
> expires when the given specific point in time on the specified clock
> passes, or when it already has passed. A relative timeout expires when
> - the given duration of time on the CLOCK_MONOTONIC clock passes. Relative
> - timeouts may be imprecise (see futex_supports_exact_relative_timeouts).
> + the given duration of time on the CLOCK_MONOTONIC clock passes.
>
> Due to POSIX requirements on when synchronization data structures such
> as mutexes or semaphores can be destroyed and due to the futex design
> @@ -81,12 +80,6 @@
> static __always_inline int
> futex_supports_pshared (int pshared);
>
> -/* Returns true if relative timeouts are robust to concurrent changes to the
> - system clock. If this returns false, relative timeouts can still be used
> - but might be effectively longer or shorter than requested. */
> -static __always_inline bool
> -futex_supports_exact_relative_timeouts (void);
> -
> /* Atomically wrt other futex operations on the same futex, this blocks iff
> the value *FUTEX_WORD matches the expected value. This is
> semantically equivalent to:
Ok.
> diff --git a/sysdeps/unix/sysv/linux/futex-internal.h b/sysdeps/unix/sysv/linux/futex-internal.h
> index 03312d6..30707e7 100644
> --- a/sysdeps/unix/sysv/linux/futex-internal.h
> +++ b/sysdeps/unix/sysv/linux/futex-internal.h
> @@ -46,14 +46,6 @@ futex_supports_pshared (int pshared)
> return EINVAL;
> }
>
> -/* The Linux kernel supports relative timeouts measured against the
> - CLOCK_MONOTONIC clock. */
> -static __always_inline bool
> -futex_supports_exact_relative_timeouts (void)
> -{
> - return true;
> -}
> -
> /* See sysdeps/nptl/futex-internal.h for details. */
> static __always_inline int
> futex_wait (unsigned int *futex_word, unsigned int expected, int private)
>
Ok.