From: Adhemerval Zanella <adhemerval.zanella@linaro.org>
To: kemi <kemi.wang@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>,
Tim Chen <tim.c.chen@intel.com>,
Andi Kleen <andi.kleen@intel.com>,
Ying Huang <ying.huang@intel.com>, Aaron Lu <aaron.lu@intel.com>,
Glibc alpha <libc-alpha@sourceware.org>,
aubrey <aubrey.li@intel.com>
Subject: Re: [PATCH 1/3] Tunables: Add tunables of spin count for adaptive spin mutex
Date: Wed, 04 Apr 2018 17:17:00 -0000 [thread overview]
Message-ID: <f65e64f5-7289-faa4-2314-71ce5e904bf0@linaro.org> (raw)
In-Reply-To: <03d60e93-a167-1401-b67e-22d7d2443486@intel.com>
On 04/04/2018 07:25, kemi wrote:
> Hi, Adhemerval
> Thanks for your review. Could you please help to review the other two patches in this series
> if available? Also, please keep guys in the cc list in case someone may not subscribe glibc
> mail list. Thanks:)
Yeah, I am checking the patch along with the referenced benchmark on a aarch64
machine as well.
>
> See my reply below.
>
> On 2018å¹´03æ30æ¥ 15:14, Kemi Wang wrote:
>> On 30/03/2018 04:14, Kemi Wang wrote:
>>> This patch does not have any functionality change, we only provide a spin
>>> count tunes for pthread adaptive spin mutex. The tunable
>>> glibc.mutex.spin_count tunes can be used by system adminstrator to squeeze
>>> system performance according to different hardware capability and workload
>>> model.
>>>
>>> This is the preparation work for the next patch, in which the way of
>>> adaptive spin would be changed from an expensive cmpxchg to read while
>>> spinning.
>>>
>>> * elf/dl-tunables.list: Add glibc.mutex.spin_count entry.
>>> * manual/tunables.texi: Add glibc.mutex.spin_count description.
>>> * nptl/Makefile: Add mutex-conf.c for compilation.
>>> * nptl/mutex-conf.h: New file.
>>> * nptl/mutex-conf.c: New file.
>>>
>>> Suggested-by: Andi Kleen <andi.kleen@intel.com>
>>> Signed-off-by: Kemi Wang <kemi.wang@intel.com>
>>
>>> ---
>>> ChangeLog | 10 ++++++-
>>> elf/dl-tunables.list | 10 +++++++
>>> manual/tunables.texi | 17 ++++++++++++
>>> nptl/Makefile | 3 +-
>>> nptl/mutex-conf.c | 78 ++++++++++++++++++++++++++++++++++++++++++++++++++++
>>> nptl/mutex-conf.h | 31 +++++++++++++++++++++
>>> 6 files changed, 147 insertions(+), 2 deletions(-)
>>> create mode 100644 nptl/mutex-conf.c
>>> create mode 100644 nptl/mutex-conf.h
>>>
>>> diff --git a/ChangeLog b/ChangeLog
>>> index 1f98425..472657c 100644
>>> --- a/ChangeLog
>>> +++ b/ChangeLog
>>> @@ -1,5 +1,13 @@
>>> -2018-03-29 Florian Weimer <fweimer@redhat.com>
>>> +2018-03-30 Kemi Wang <kemi.wang@intel.com>
>>> +
>>> + * elf/dl-tunables.list: Add glibc.mutex.spin_count entry.
>>> + * manual/tunables.texi: Add glibc.mutex.spin_count description.
>>> + * nptl/Makefile: Add mutex-conf.c for compilation.
>>> + * nptl/mutex-conf.h: New file.
>>> + * nptl/mutex-conf.c: New file.
>>> + * nptl/Makefile: Add new file compilation.
>>>
>>> +2018-03-29 Florian Weimer <fweimer@redhat.com>
>>> * sysdeps/unix/sysv/linux/i386/tst-bz21269.c (do_test): Also
>>> capture SIGBUS.
>>>
>>> diff --git a/elf/dl-tunables.list b/elf/dl-tunables.list
>>> index 1f8ecb8..0c27c14 100644
>>> --- a/elf/dl-tunables.list
>>> +++ b/elf/dl-tunables.list
>>> @@ -121,4 +121,14 @@ glibc {
>>> default: 3
>>> }
>>> }
>>> +
>>> + mutex {
>>> + spin_count {
>>> + type: INT_32
>>> + minval: 0
>>> + maxval: 30000
>>> + env_alias: LD_SPIN_COUNT
>>> + default: 1000
>>> + }
>>> + }
>>
>> Indentation seems off here, rest the file uses double space while your
>> patch uses tabs.
>>
>
> Thanks for catching it.
>
>> Also I not sure if it worth to add environment variable for this tunable,
>> I would rather avoid adding newer ones (and naming seems off, since afaik
>> LD_* meaning some parameters that affects the loader).
>>
>
> AFAIK, environment variable is what I can think of to give people the possibility of
> tunes in shell for that, maybe you have better idea?
>
> Yes, we probably use other name like MUTEX_SPIN_COUNT to avoid confusion, agree?
The tunables framework already provides a environment variable to this [1],
the 'env_alias' is mainly to provide compatibility and to use the same logic
internally.
So I think it is better to use the default tunable env var.
[1] https://www.gnu.org/software/libc/manual/html_node/Tunables.html
>
>>> }
>>> diff --git a/manual/tunables.texi b/manual/tunables.texi
>>> index be33c9f..9c6a9f1 100644
>>> --- a/manual/tunables.texi
>>> +++ b/manual/tunables.texi
>>> @@ -281,6 +281,23 @@ of try lock attempts.
>>> The default value of this tunable is @samp{3}.
>>> @end deftp
>>>
>>> +@node Pthread Mutex Tunables
>>> +@section Pthread Mutex Tunables
>>> +@cindex pthread mutex tunables
>>> +
>>> +@deftp {Tunable namespace} glibc.mutex
>>> +Behavior of ptherad mutex can be tuned to acquire performance improvement
>>> +according to specific hardware capablity and workload character by setting
>>> +the following tunables in the @code{mutex} namespace.
>>> +@end deftp
>>
>> There is a typo (s/ptherad/pthread)
>
> thanks for catching it.
>
>> and I think 'acquire' is not the best
>> word here, I would use 'increase' (I am not a native speaker, so someone
>> might suggest a better wording here).
>>
>
> All right, I will ask for some native speakers to help check it.
>
>>> +
>>> +@deftp Tunable glibc.mutex.spin_count
>>> +The @code{glibc.mutex.spin_count} tunable set the maximum times the thread
>>> +should spin on the lock before going to sleep.
>>> +
>>> +The default value of this tunable is @samp{1000}.
>>> +@end deftp
>>
>> I think we need to expand it to specify:
>>
>> * spinning is only used for the mutex initialized with PTHREAD_MUTEX_ADAPTIVE_NP
>> GNU extension(current wording is ambiguous).
>>> * It affects both pthread_mutex_lock and pthread_mutex_timedlock.
>>
>> * The spinning is done by first issuing an atomic operation similar to trylock
>> followed by a arch-specific no operation (not sure if we need to expand
>> how spinning is backoff value is incremented).
>>
> How about this?
> The spinning is done in case of either the maximum spin count
> is reached or lock is acquired during spinning.
It is better for this specific part.
>
>> * Not sure 'sleep' is right terminology here, since for Linux 'futex' can
>> return right away when maximum spin_count it reached.
>>
> 'block' here may be more precise, agree?
> e.g. use "before calling into the kernel to block" or something like that
>
>>> +
>>> @node Hardware Capability Tunables
>>> @section Hardware Capability Tunables
>>> @cindex hardware capability tunables
>>> diff --git a/nptl/Makefile b/nptl/Makefile
>>> index 94be92c..5edacea 100644
>>> --- a/nptl/Makefile
>>> +++ b/nptl/Makefile
>>> @@ -139,7 +139,8 @@ libpthread-routines = nptl-init vars events version pt-interp \
>>> pthread_mutex_getprioceiling \
>>> pthread_mutex_setprioceiling \
>>> pthread_setname pthread_getname \
>>> - pthread_setattr_default_np pthread_getattr_default_np
>>> + pthread_setattr_default_np pthread_getattr_default_np \
>>> + mutex-conf
>>> # pthread_setuid pthread_seteuid pthread_setreuid \
>>> # pthread_setresuid \
>>> # pthread_setgid pthread_setegid pthread_setregid \
>>
>> I think a better name for the file would be pthread_mutex_conf.c.
>>
>
> Agree, will send V2 to fix it.
>
>>> diff --git a/nptl/mutex-conf.c b/nptl/mutex-conf.c
>>> new file mode 100644
>>> index 0000000..f4ffd6d
>>> --- /dev/null
>>> +++ b/nptl/mutex-conf.c
>>> @@ -0,0 +1,78 @@
>>> +/* mutex-conf.c: Pthread mutex tunable parameters.
>>> + Copyright (C) 2013-2018 Free Software Foundation, Inc.
>>> + This file is part of the GNU C Library.
>>> +
>>
>> Copyright should start at 2018 for newer implementations.
>>
>
> I am not aware of that before, do you mean "Copyright (C) 2018-2023"ï¼
No, just "Copyright (C) 2018".
>
>>> + The GNU C Library is free software; you can redistribute it and/or
>>> + modify it under the terms of the GNU Lesser General Public
>>> + License as published by the Free Software Foundation; either
>>> + version 2.1 of the License, or (at your option) any later version.
>>> +
>>> + The GNU C Library is distributed in the hope that it will be useful,
>>> + but WITHOUT ANY WARRANTY; without even the implied warranty of
>>> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
>>> + Lesser General Public License for more details.
>>> +
>>> + You should have received a copy of the GNU Lesser General Public
>>> + License along with the GNU C Library; if not, see
>>> + <http://www.gnu.org/licenses/>. */
>>> +
>>> +#include "config.h"
>>> +#include <pthreadP.h>
>>> +#include <init-arch.h>
>>> +#include <mutex-conf.h>
>>> +#include <unistd.h>
>>> +
>>> +#if HAVE_TUNABLES
>>> +# define TUNABLE_NAMESPACE mutex
>>> +#endif
>>> +#include <elf/dl-tunables.h>
>>> +
>>> +
>>> +struct mutex_config __mutex_aconf =
>>> + {
>>> + /* The maximum times a thread spin on the lock before
>>> + * going to sleep */
>>> + .spin_count = 1000,
>>> + };
>>> +
>>
>> Indentation seems off here and in other places in this file.
>>
>
> Will fix it in V2, thanks for pointing it out.
>
>>> +#if HAVE_TUNABLES
>>> +#define TUNABLE_CALLBACK_FNDECL(__name, __type) \
>>
>> Indentation for nested preprocessor is to just one space each level:
>>
>> #if HAVE_TUNABLES
>> # define ...
>>
>>> +static inline void \
>>> +__always_inline \
>>> +do_set_mutex_ ## __name (__type value) \
>>> +{ \
>>> + __mutex_aconf.__name = value; \
>>> +} \
>>> +void \
>>> +TUNABLE_CALLBACK (set_mutex_ ## __name) (tunable_val_t *valp) \
>>> +{ \
>>> + __type value = (__type) (valp)->numval; \
>>> + do_set_mutex_ ## __name (value); \
>>> +}
>>> +
>>> +TUNABLE_CALLBACK_FNDECL (spin_count, int32_t);
>>> +#endif
>>> +
>>> +static void
>>> +mutex_tunables_init (int argc __attribute__ ((unused)),
>>> + char **argv __attribute__ ((unused)),
>>> + char **environ)
>>> +{
>>> +#if HAVE_TUNABLES
>>> +
>>> + TUNABLE_GET (spin_count, int32_t,
>>> + TUNABLE_CALLBACK (set_mutex_spin_count));
>>> +#endif
>>> +}
>>> +
>>> +#ifdef SHARED
>>> +# define INIT_SECTION ".init_array"
>>> +#else
>>> +# define INIT_SECTION ".preinit_array"
>>> +#endif
>>> +
>>> +void (*const __pthread_mutex_tunables_init_array []) (int, char **, char **)
>>> + __attribute__ ((section (INIT_SECTION), aligned (sizeof (void *)))) =
>>> +{
>>> + &mutex_tunables_init
>>> +};> diff --git a/nptl/mutex-conf.h b/nptl/mutex-conf.h
>>> new file mode 100644
>>> index 0000000..babefe3
>>> --- /dev/null
>>> +++ b/nptl/mutex-conf.h
>>> @@ -0,0 +1,31 @@
>>> +/* mutex-conf.h: Pthread mutex tunable parameters.
>>> + Copyright (C) 2013-2018 Free Software Foundation, Inc.
>>
>> Copyright should start at 2018 for newer implementations.
>>
>
> Sure.
>
>>> + This file is part of the GNU C Library.
>>> +
>>> + The GNU C Library is free software; you can redistribute it and/or
>>> + modify it under the terms of the GNU Lesser General Public
>>> + License as published by the Free Software Foundation; either
>>> + version 2.1 of the License, or (at your option) any later version.
>>> +
>>> + The GNU C Library is distributed in the hope that it will be useful,
>>> + but WITHOUT ANY WARRANTY; without even the implied warranty of
>>> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
>>> + Lesser General Public License for more details.
>>> +
>>> + You should have received a copy of the GNU Lesser General Public
>>> + License along with the GNU C Library; if not, see
>>> + <http://www.gnu.org/licenses/>. */
>>> +#ifndef _MUTEX_CONF_H
>>> +#define _MUTEX_CONF_H 1
>>> +
>>> +#include <pthread.h>
>>> +#include <time.h>
>>> +
>>> +struct mutex_config
>>> +{
>>> + int spin_count;
>>> +};
>>> +
>>> +extern struct mutex_config __mutex_aconf attribute_hidden;
>>> +
>>> +#endif
>>>
>>
next prev parent reply other threads:[~2018-04-04 17:17 UTC|newest]
Thread overview: 14+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-03-30 7:17 Kemi Wang
2018-03-30 7:17 ` [PATCH 3/3] Mutex: Avoid useless spinning Kemi Wang
2018-04-05 20:59 ` Adhemerval Zanella
2018-04-08 8:33 ` kemi
2018-03-30 7:17 ` [PATCH 2/3] Mutex: Only read while spinning Kemi Wang
2018-04-05 20:55 ` Adhemerval Zanella
2018-04-08 8:30 ` kemi
2018-04-09 20:52 ` Adhemerval Zanella
2018-04-10 1:49 ` kemi
2018-04-11 13:28 ` Adhemerval Zanella
2018-04-02 15:19 ` [PATCH 1/3] Tunables: Add tunables of spin count for adaptive spin mutex Adhemerval Zanella
2018-04-04 10:27 ` kemi
2018-04-04 17:17 ` Adhemerval Zanella [this message]
2018-04-05 1:11 ` Carlos O'Donell
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=f65e64f5-7289-faa4-2314-71ce5e904bf0@linaro.org \
--to=adhemerval.zanella@linaro.org \
--cc=aaron.lu@intel.com \
--cc=andi.kleen@intel.com \
--cc=aubrey.li@intel.com \
--cc=dave.hansen@linux.intel.com \
--cc=kemi.wang@intel.com \
--cc=libc-alpha@sourceware.org \
--cc=tim.c.chen@intel.com \
--cc=ying.huang@intel.com \
/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).