Re: [PATCH V3] firmware: google: Test spinlock on panic path to avoid lockups

[Evan Green]

On Thu, Sep 1, 2022 at 3:13 PM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Thu, 1 Sep 2022 16:02:08 -0300 "Guilherme G. Piccoli" <gpiccoli@xxxxxxxxxx> wrote:
>
> > On 01/09/2022 15:59, Greg KH wrote:
> > > [...]
> > > Ick, I don't know, this all feels odd.  I want someone else to review
> > > this and give their ack on the patch before I'll take it so someone else
> > > can share in the blame :)
> > >
> > > thanks,
> > >
> > > greg k-h
> >
> > LOL, that's OK for me! Evan seems to be fine with it BTW.
> >
> > Let's see if Petr can jump in, also adding Andrew here since he's
> > usually merging stuff for panic.
>
> Are the usual gsmi developers not operational?

I'm unsure who that is, I sort of Mr. Beaned my way in here having
touched the file recently. A lot of the people who historically
touched this file have gone.

>
> Patch seems sensible to me, although the deadlock sounds pretty
> theoretical.  A better code comment might be simply
>
>         /*
>          * Panic callbacks are executed with all other CPUs stopped, so we must
>          * not attempt to spin waiting for gsmi_dev.lock to be released.
>          */
>
> ?

I basically came to the same conclusion as Andrew. It seems like this
patch does fix a problem, which is a panic coming in on another CPU
and NMIing on top of a CPU doing a normal operation holding this lock.
The problem seems pretty theoretical, but I suppose I don't have
numbers one way or another since any attempt to gather numbers would
be reliant on this very mechanism. My Reviewed-by tag is already on
there.
-Evan