[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <D3115315-12A9-43A9-9209-09553CF2C71C@lca.pw>
Date: Fri, 13 Mar 2020 17:00:41 -0400
From: Qian Cai <cai@....pw>
To: Peter Zijlstra <peterz@...radead.org>
Cc: Thomas Gleixner <tglx@...utronix.de>,
John Stultz <john.stultz@...aro.org>, sboyd@...nel.org,
rostedt@...dmis.org, hannes@...xchg.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH] timer: silenct a lockdep splat with debugobjects
> On Mar 13, 2020, at 4:13 PM, Peter Zijlstra <peterz@...radead.org> wrote:
>
> On Fri, Mar 13, 2020 at 03:32:52PM -0400, Qian Cai wrote:
>> Well, in mm/vmstat.c (init_mm_internals) case for example, it was initialized
>> as a static,
>>
>> static DECLARE_DEFERRABLE_WORK(shepherd, vmstat_shepherd);
>>
>> which will not call __debug_object_init().
>
> Then fix that.
>
> That is, all of:
>
> DECLARE_DEFERRABLE_WORK
> KTHREAD_DELAYED_WORK_INIT
> DEFINE_TIMER
>
> are broken and need to go.
This looks invasive with so many callsites...
>
> Unless of course, you can frob debugobjects such that we can provide the
> required storage through is_static_object() and modify the above static
> declarators to provide the storage.
I’ll explore this which looks more promising.
>
> Or, fix that random crud to do the wakeup outside of the lock.
That is likely to be difficult until we can find a creative way to not “uglifying" the
random code by dropping locks in the middle etc just because of debugojects.
>
> Plenty options..
Powered by blists - more mailing lists