From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
To: linux-kernel@vger.kernel.org, linux-ia64@vger.kernel.org
Cc: Andy Lutomirski <luto@kernel.org>,
Ben Segall <bsegall@google.com>,
Daniel Bristot de Oliveira <bristot@redhat.com>,
Dietmar Eggemann <dietmar.eggemann@arm.com>,
Ingo Molnar <mingo@redhat.com>,
Juri Lelli <juri.lelli@redhat.com>,
Peter Zijlstra <peterz@infradead.org>,
Steven Rostedt <rostedt@goodmis.org>,
Thomas Gleixner <tglx@linutronix.de>,
Vincent Guittot <vincent.guittot@linaro.org>
Subject: Re: [PATCH REPOST 0/8] kernel/fork: Move thread stack free otu of the scheduler path.
Date: Tue, 8 Feb 2022 18:10:19 +0100 [thread overview]
Message-ID: <YgKj+363/qsJHRGA@linutronix.de> (raw)
In-Reply-To: <20220125152652.1963111-1-bigeasy@linutronix.de>
On 2022-01-25 16:26:44 [+0100], To linux-kernel@vger.kernel.org wrote:
> [ This is a repost of https://lkml.kernel.org/r/20211118143452.136421-1-bigeasy@linutronix.de ]
>
> This is a follup-up on the patch
> sched: Delay task stack freeing on RT
> https://lkml.kernel.org/r/20210928122411.593486363@linutronix.de
>
> It addresses the review feedback:
> - Decouple stack accounting from its free invocation. The accounting
> happens in do_exit(), the final free call happens later.
>
> - Add put_task_stack_sched() to finish_task_switch(). Here the VMAP
> stack is cached only. If it fails, or in the !VMAP case then the final
> free happens in delayed_put_task_struct(). This is also an oportunity
> to cache the stack.
>
> >From testing I observe the following:
>
> | bash-1715 [006] ..... 124.901510: copy_process: allocC ffffc90007e70000
> | sh-cmds.sh-1746 [007] ..... 124.907389: copy_process: allocC ffffc90007dc4000
> | <idle>-0 [019] ...1. 124.918126: free_thread_stack: cache ffffc90007dc4000
> | sh-cmds.sh-1746 [007] ..... 124.918279: copy_process: allocC ffffc90007de8000
> | <idle>-0 [004] ...1. 124.920121: free_thread_stack: delay ffffc90007de8001
> | <idle>-0 [007] ...1. 124.920299: free_thread_stack: cache ffffc90007e70000
> | <idle>-0 [007] ..s1. 124.945433: free_thread_stack: cache ffffc90007de8000
>
> TS 124.901510, bash started sh-cmds.sh, obtained stack from cache.
> TS 124.907389, script invokes its first command, obtained stacak from
> cache. As you can see bash was running on CPU6 but its child was moved
> CPU7.
> TS 124.918126, the first command is done, stack is ached on CPU19.
> TS 124.918279, script's second command, ache from stack.
> TS 124.920121, the command is done. The stack cache on CPU4 is full.
> TS 124.920299, the script is done, caches stack on CPU7.
> TS 124.945433, the RCU-callback of last command is now happening. On
> CPU7, which is where the command was invoked (but not running). Instead
> of freeing the stack, it was cached since CPU7 had an empty slot.
>
> If I pin the script to CPU5 and run it with multiple commands then it
> works as expected:
>
> | bash-1799 [005] ..... 993.608131: copy_process: allocC ffffc90007fa0000
> | sh-cmds.sh-1827 [005] ..... 993.608888: copy_process: allocC ffffc90007fa8000
> | sh-cmds.sh-1827 [005] ..... 993.610734: copy_process: allocV ffffc90007ff4000
> | sh-cmds.sh-1829 [005] ...1. 993.610757: free_thread_stack: cache ffffc90007fa8000
> | sh-cmds.sh-1827 [005] ..... 993.612401: copy_process: allocC ffffc90007fa8000
> | <...>-1830 [005] ...1. 993.612416: free_thread_stack: cache ffffc90007ff4000
> | sh-cmds.sh-1827 [005] ..... 993.613707: copy_process: allocC ffffc90007ff4000
> | sh-cmds.sh-1831 [005] ...1. 993.613723: free_thread_stack: cache ffffc90007fa8000
> | sh-cmds.sh-1827 [005] ..... 993.615024: copy_process: allocC ffffc90007fa8000
> | <...>-1832 [005] ...1. 993.615040: free_thread_stack: cache ffffc90007ff4000
> | sh-cmds.sh-1827 [005] ..... 993.616380: copy_process: allocC ffffc90007ff4000
> | <...>-1833 [005] ...1. 993.616397: free_thread_stack: cache ffffc90007fa8000
> | bash-1799 [005] ...1. 993.617759: free_thread_stack: cache ffffc90007fa0000
> | <idle>-0 [005] ...1. 993.617871: free_thread_stack: delay ffffc90007ff4001
> | <idle>-0 [005] ..s1. 993.638311: free_thread_stack: free ffffc90007ff4000
>
> and no new is allocated during its runtime and a cached stack is used.
ping
Sebastian
prev parent reply other threads:[~2022-02-08 17:10 UTC|newest]
Thread overview: 20+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-01-25 15:26 [PATCH REPOST 0/8] kernel/fork: Move thread stack free otu of the scheduler path Sebastian Andrzej Siewior
2022-01-25 15:26 ` [PATCH 1/8] kernel/fork: Redo ifdefs around task's handling Sebastian Andrzej Siewior
2022-01-25 15:26 ` [PATCH 2/8] kernel/fork: Duplicate task_struct before stack allocation Sebastian Andrzej Siewior
2022-02-11 23:42 ` Andy Lutomirski
2022-02-14 11:39 ` Sebastian Andrzej Siewior
2022-01-25 15:26 ` [PATCH 3/8] kernel/fork, IA64: Provide a alloc_thread_stack_node() for IA64 Sebastian Andrzej Siewior
2022-02-14 18:00 ` Sebastian Andrzej Siewior
2022-01-25 15:26 ` [PATCH 4/8] kernel/fork: Don't assign the stack pointer in dup_task_struct() Sebastian Andrzej Siewior
2022-01-25 15:26 ` [PATCH 5/8] kernel/fork: Move memcg_charge_kernel_stack() into CONFIG_VMAP_STACK Sebastian Andrzej Siewior
2022-01-25 15:26 ` [PATCH 6/8] kernel/fork: Move task stack account to do_exit() Sebastian Andrzej Siewior
2022-02-11 23:43 ` Andy Lutomirski
2022-01-25 15:26 ` [PATCH 7/8] kernel/fork: Only cache the VMAP stack in finish_task_switch() Sebastian Andrzej Siewior
2022-02-11 23:55 ` Andy Lutomirski
2022-02-14 12:10 ` Sebastian Andrzej Siewior
2022-02-14 12:24 ` Sebastian Andrzej Siewior
2022-02-14 16:54 ` Sebastian Andrzej Siewior
2022-02-14 17:48 ` Sebastian Andrzej Siewior
2022-02-14 18:15 ` [PATCH v2 " Sebastian Andrzej Siewior
2022-01-25 15:26 ` [PATCH 8/8] kernel/fork: Use IS_ENABLED() in account_kernel_stack() Sebastian Andrzej Siewior
2022-02-08 17:10 ` Sebastian Andrzej Siewior [this message]
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=YgKj+363/qsJHRGA@linutronix.de \
--to=bigeasy@linutronix.de \
--cc=bristot@redhat.com \
--cc=bsegall@google.com \
--cc=dietmar.eggemann@arm.com \
--cc=juri.lelli@redhat.com \
--cc=linux-ia64@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=luto@kernel.org \
--cc=mingo@redhat.com \
--cc=peterz@infradead.org \
--cc=rostedt@goodmis.org \
--cc=tglx@linutronix.de \
--cc=vincent.guittot@linaro.org \
/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 NNTP newsgroup(s).