From: Adhemerval Zanella <adhemerval.zanella@linaro.org>
To: Kurt Kanzenbach <kurt@linutronix.de>, libc-alpha@sourceware.org
Cc: Florian Weimer <fweimer@redhat.com>,
Carlos O'Donell <carlos@redhat.com>,
Thomas Gleixner <tglx@linutronix.de>,
Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Subject: Re: [PATCH RFC 0/3] nptl: Introduce and use FUTEX_LOCK_PI2
Date: Mon, 21 Jun 2021 18:32:41 -0300 [thread overview]
Message-ID: <bf4e7c28-11ce-fa4f-65b4-b00196440792@linaro.org> (raw)
In-Reply-To: <20210621111650.1164689-1-kurt@linutronix.de>
On 21/06/2021 08:16, Kurt Kanzenbach wrote:
> Hi,
>
> Linux real time application often make use of mutexes with priority inheritance
> enabled due to problems such as unbounded priority inversion. In addition, some
> applications make use of timeouts e.g., by utilizing pthread_mutex_clocklock().
>
> However, the combination of priority inheritance enabled mutexes with timeouts
> based on CLOCK_MONOTONIC doesn't work. That is because the underlying Linux
> kernel futex interface doesn't support it, yet. Using CLOCK_REALTIME is
> possible, but it is subject to adjustments (NTP, PTP, ...). Therefore, Thomas
> proposed to add a new futex operation called FUTEX_LOCK_PI2 with support for
> selectable clocks [1]. That patch series is not merged, yet.
>
> Here is a proof of concept patch set adding FUTEX_LOCK_PI2 in order to support
> pthread_mutex_clocklock(MONOTONIC)/PI. The idea is to use futex_lock_pi2() by
> default, and fallback to futex_lock_pi() in case the kernel is too old. I think,
> you get the idea.
>
> There's also a bugreport for glibc with a test case:
>
> https://sourceware.org/bugzilla/show_bug.cgi?id=26801
>
> Thoughts?
>
> Thanks,
> Kurt
Currently we check for PI mutex support on pthread_mutex_init which basically
check for futex_cmpxchg_enabled within kernel (so it fails only on a handful
configurations).
For FUTEX_LOCK_PI2 I think we will need to rework it, since we are moving
the futex PI failure from pthread_mutex_init to pthread_mutex_{timed}lock.
It means that we will need to remove the prio_inherit_missing() test on
pthread_mutex_init and make the pthread_mutex_{timed}lock fail instead
(not sure if we should use ENOTSUP or keep with current EINVAL).
The proposed futex_lockpi2_supported() incurs in an extra syscall on every
mutex slow path, we should avoid it. I would like also to avoid the CRIU
issue as well, so I think it would be better to avoid any caching (as done
by prio_inherit_missing()), and optimize the FUTEX_LOCK_PI2 to be used only
when the timeout for clock different than CLOCK_REALTIME is required.
So instead it would be better to move the logic solely on futex_lock_pi()
(I am assuming FUTEX_LOCK_PI2 would be only added for futex_time64):
static __always_inline int
futex_lock_pi2_64 (int *futex_word, const struct __timespec64 *abtime,
int private)
{
# if __ASSUME_FUTEX_LOCK_PI2
return INTERNAL_SYSCALL_CALL (futex_time64, futex_word,
__lll_private_flag (FUTEX_LOCK_PI2, private), 0,
abstime);
# else
if (abstime != NULL && clockid != CLOCK_REALTIME)
return INTERNAL_SYSCALL_CALL (futex_time64, futex_word,
__lll_private_flag (FUTEX_LOCK_PI2, private), 0,
abstime);
else
{
int err = INTERNAL_SYSCALL_CALL (futex_time64, futex_word,
__lll_private_flag (FUTEX_LOCK_PI, private), 0,
abstime);
if (err == -ENOSYS)
err = -EOVERFLOW;
}
# endif /* __ASSUME_FUTEX_LOCK_PI2 */
}
static __always_inline int
futex_lock_pi64 (int *futex_word, const struct __timespec64 *abstime,
int private)
{
int err;
#ifdef __ASSUME_TIME64_SYSCALLS
err = futex_lock_pi2_64 (futex_word, abstime, private);
#else /* __ASSUME_TIME64_SYSCALLS */
bool need_time64 = abstime != NULL && !in_time_t_range (abstime->tv_sec)
if (need_time64)
{
err = futex_lock_pi2_64 (futex_word, abstime, private);
}
else
{
struct timespec ts32;
if (abstime != NULL)
ts32 = valid_timespec64_to_timespec (*abstime);
err = INTERNAL_SYSCALL_CALL (futex, futex_word, __lll_private_flag
(FUTEX_LOCK_PI, private), 0,
abstime != NULL ? &ts32 : NULL);
}
#endif /* __ASSUME_TIME64_SYSCALLS */
[...]
}
It would make the changes on pthread_mutex code minimal, it would be only to
remove the extra check for clockid and adjust the comment.
Also, as Joseph has noted the __ASSUME_FUTEX_LOCK_PI2 should be the first
patch. It also does not make sense to add the __ASSUME_FUTEX_LOCK_PI2 on
tests, they need to be kernel agnostic so you will need to handle a possible
EINVAL/ENOSUP failure instead.
>
> [1] - https://lkml.kernel.org/lkml/20210422194417.866740847@linutronix.de/
>
> Kurt Kanzenbach (3):
> nptl: Introduce futex_lock_pi2()
> nptl: Use futex_lock_pi2()
> nptl: Include CLOCK_MONOTONIC in mutex tests
>
> nptl/pthread_mutex_timedlock.c | 24 ++++--
> nptl/tst-mutexpi10.c | 8 ++
> sysdeps/nptl/futex-internal.h | 94 +++++++++++++++++++++++
> sysdeps/nptl/lowlevellock-futex.h | 1 +
> sysdeps/pthread/tst-mutex5.c | 3 +-
> sysdeps/pthread/tst-mutex9.c | 3 +-
> sysdeps/unix/sysv/linux/kernel-features.h | 8 ++
> 7 files changed, 133 insertions(+), 8 deletions(-)
>
next prev parent reply other threads:[~2021-06-21 21:32 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-06-21 11:16 Kurt Kanzenbach
2021-06-21 11:16 ` [PATCH RFC 1/3] nptl: Introduce futex_lock_pi2() Kurt Kanzenbach
2021-06-21 11:16 ` [PATCH RFC 2/3] nptl: Use futex_lock_pi2() Kurt Kanzenbach
2021-06-21 11:16 ` [PATCH RFC 3/3] nptl: Include CLOCK_MONOTONIC in mutex tests Kurt Kanzenbach
2021-06-21 20:07 ` Joseph Myers
2021-06-22 8:58 ` Kurt Kanzenbach
2021-06-21 21:32 ` Adhemerval Zanella [this message]
2021-06-22 7:26 ` [PATCH RFC 0/3] nptl: Introduce and use FUTEX_LOCK_PI2 Kurt Kanzenbach
2021-06-22 7:29 ` Florian Weimer
2021-06-22 8:55 ` Kurt Kanzenbach
2021-06-22 12:30 ` Adhemerval Zanella
2021-06-22 14:25 ` Kurt Kanzenbach
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=bf4e7c28-11ce-fa4f-65b4-b00196440792@linaro.org \
--to=adhemerval.zanella@linaro.org \
--cc=bigeasy@linutronix.de \
--cc=carlos@redhat.com \
--cc=fweimer@redhat.com \
--cc=kurt@linutronix.de \
--cc=libc-alpha@sourceware.org \
--cc=tglx@linutronix.de \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).