Received-SPF: pass (google.com: domain of linux-kernel+bounces-36230-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Date: Wed, 24 Jan 2024 01:39:30 +0100
From: Oleg Nesterov <oleg@redhat.com>
To: Dylan Hatch <dylanbhatch@google.com>
Cc: Andrew Morton <akpm@linux-foundation.org>,
	"Eric W. Biederman" <ebiederm@xmission.com>,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2 0/2] getrusage: use sig->stats_lock
Message-ID: <20240124003930.GA26412@redhat.com>
References: <20240122155023.GA26169@redhat.com>
 <CADBMgpxNUoaermXsEj0Hs0KT=Q0xRpz5y+Px=oAGDP2Efg8yyw@mail.gmail.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CADBMgpxNUoaermXsEj0Hs0KT=Q0xRpz5y+Px=oAGDP2Efg8yyw@mail.gmail.com>
User-Agent: Mutt/1.5.24 (2015-08-30)

On 01/23, Dylan Hatch wrote:
>
> I have one last question -- is there possibly an edge case in which
> the hard lockup
> can still happen? How likely is it for many writers to force enough
> readers to do a
> retry on the seqlock, disabling irq and causing the lockup?

I don't know how likely is it, and I guess the repro should be more creative ;)

But yes. Please see the TODO: section in the changelog,

	- Turn sig->stats_lock into seqcount_rwlock_t, this way the
	  readers in the slow mode won't exclude each other.

and more importantly,

	- stats_lock has to disable irqs because ->siglock can be taken
	  in irq context, it would be very nice to change __exit_signal()
	  to avoid the siglock->stats_lock dependency.

There are other users which take stats_lock under siglock (and the
"fs/proc: do_task_stat" series changes 2 of them to not do this), but
__exit_signal() is most problematic.

If we remove this dependency, we can turn read_seqbegin_or_lock_irqsave()
into read_seqbegin_or_lock() which doesn't disable irqs.

Oleg.