Live Patching
 help / color / mirror / Atom feed
From: Josh Poimboeuf <jpoimboe@kernel.org>
To: Petr Mladek <pmladek@suse.com>
Cc: Yafang Shao <laoar.shao@gmail.com>,
	jikos@kernel.org, mbenes@suse.cz, joe.lawrence@redhat.com,
	live-patching@vger.kernel.org
Subject: Re: [PATCH 2/3] livepatch: Avoid blocking tasklist_lock too long
Date: Fri, 14 Feb 2025 10:12:06 -0800	[thread overview]
Message-ID: <20250214181206.xkvxohoc4ft26uhf@jpoimboe> (raw)
In-Reply-To: <Z69Wuhve2vnsrtp_@pathway.suse.cz>

On Fri, Feb 14, 2025 at 03:44:10PM +0100, Petr Mladek wrote:
> I guess that we really could consider the new task as migrated
> and clear TIF_PATCH_PENDING.
> 
> But we can't set child->patch_state to KLP_TRANSITION_IDLE. It won't
> work when the transition gets reverted. [*]

Hm, why not?  I don't see any difference between patching or unpatching?

klp_init_transition() has a barrier to enforce the order of the
klp_target_state and func->transition writes, as read by the
klp_ftrace_handler().

So in the ftrace handler, if func->transition is set and the task is
KLP_TRANSITION_IDLE, it can use klp_target_state to decide which
function to use.  Its patch state would effectively be the same as any
other already-transitioned task, whether it's patching or unpatching.

Then in klp_complete_transition(), after func->transition gets set to
false, klp_synchronize_transition() flushes out any running ftrace
handlers.  From that point on, func->transition is false, so the ftrace
handler would no longer read klp_target_state.

> [*] I gave this few brain cycles but I did not find any elegant
>     way how to set this a safe way and allow using rcu_read_lock()
>     in klp_try_complete_transition().
> 
>     It might be because it is Friday evening and I am leaving for
>     a trip tomorrow. Also I not motivated enough to think about it
>     because Yafang saw the RCU stall even with that rcu_read_lock().
>     So I send this just for record.

Even if it doesn't fix the RCU stalls, I think we should still try to
avoid holding the tasklist_lock.  It's a global lock which can be
contended, and we want the livepatch transition to be as unobtrusive as
possible.

If the system is doing a lot of forking across many CPUs, holding the
lock could block all the forking tasks and trigger system-wide
scheduling latencies.  And that could be compounded by the unnecessary
transitioning of new tasks every time the delayed work runs.

-- 
Josh

  reply	other threads:[~2025-02-14 18:12 UTC|newest]

Thread overview: 30+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-02-11  6:24 [PATCH 0/3] livepatch: Some improvements Yafang Shao
2025-02-11  6:24 ` [PATCH 1/3] livepatch: Add comment to clarify klp_add_nops() Yafang Shao
2025-02-12 12:51   ` Petr Mladek
2025-02-13  5:49     ` Yafang Shao
2025-02-11  6:24 ` [PATCH 2/3] livepatch: Avoid blocking tasklist_lock too long Yafang Shao
2025-02-12  0:40   ` Josh Poimboeuf
2025-02-12  2:34     ` Yafang Shao
2025-02-12 11:54       ` Yafang Shao
2025-02-12 15:42         ` Petr Mladek
2025-02-13  1:36           ` Josh Poimboeuf
2025-02-13  5:53             ` Yafang Shao
2025-02-13  9:48             ` Petr Mladek
2025-02-13 17:32               ` Josh Poimboeuf
2025-02-14 14:44                 ` Petr Mladek
2025-02-14 18:12                   ` Josh Poimboeuf [this message]
2025-02-18  2:37                     ` Yafang Shao
2025-02-13  2:47         ` Josh Poimboeuf
2025-02-13 11:19   ` Find root of the stall: was: " Petr Mladek
2025-02-13 12:32     ` Yafang Shao
2025-02-13 12:39       ` Yafang Shao
2025-02-14  2:44         ` Yafang Shao
2025-02-14  8:36           ` Josh Poimboeuf
2025-02-14 11:37             ` Petr Mladek
2025-02-18  2:19               ` Yafang Shao
2025-02-14  9:46       ` Petr Mladek
2025-02-11  6:24 ` [PATCH 3/3] livepatch: Avoid potential RCU stalls in klp transition Yafang Shao
2025-02-12  0:52   ` Josh Poimboeuf
2025-02-12  2:42     ` Yafang Shao
2025-02-13  1:58       ` Josh Poimboeuf
2025-02-13  5:51         ` Yafang Shao

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=20250214181206.xkvxohoc4ft26uhf@jpoimboe \
    --to=jpoimboe@kernel.org \
    --cc=jikos@kernel.org \
    --cc=joe.lawrence@redhat.com \
    --cc=laoar.shao@gmail.com \
    --cc=live-patching@vger.kernel.org \
    --cc=mbenes@suse.cz \
    --cc=pmladek@suse.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