Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220819200928.401416-1-kherbst@redhat.com> <YymY+3+C2aI7T3GU@eldamar.lan>
In-Reply-To: <YymY+3+C2aI7T3GU@eldamar.lan>
From:   Karol Herbst <kherbst@redhat.com>
Date:   Tue, 20 Sep 2022 13:36:32 +0200
Message-ID: <CACO55ts7rpbyYv3ovWt1iCfkGsChCUVitmHqtzAwFpfbPEZGYQ@mail.gmail.com>
Subject: Re: [PATCH] nouveau: explicitly wait on the fence in nouveau_bo_move_m2mf
To:     Salvatore Bonaccorso <carnil@debian.org>
Cc:     linux-kernel@vger.kernel.org, Ben Skeggs <bskeggs@redhat.com>,
        Lyude Paul <lyude@redhat.com>, dri-devel@lists.freedesktop.org,
        nouveau@lists.freedesktop.org, stable@vger.kernel.org,
        Computer Enthusiastic <computer.enthusiastic@gmail.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Sep 20, 2022 at 12:42 PM Salvatore Bonaccorso <carnil@debian.org> wrote:
>
> Hi,
>
> On Fri, Aug 19, 2022 at 10:09:28PM +0200, Karol Herbst wrote:
> > It is a bit unlcear to us why that's helping, but it does and unbreaks
> > suspend/resume on a lot of GPUs without any known drawbacks.
> >
> > Cc: stable@vger.kernel.org # v5.15+
> > Closes: https://gitlab.freedesktop.org/drm/nouveau/-/issues/156
> > Signed-off-by: Karol Herbst <kherbst@redhat.com>
> > ---
> >  drivers/gpu/drm/nouveau/nouveau_bo.c | 9 +++++++++
> >  1 file changed, 9 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
> > index 35bb0bb3fe61..126b3c6e12f9 100644
> > --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> > +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> > @@ -822,6 +822,15 @@ nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict,
> >               if (ret == 0) {
> >                       ret = nouveau_fence_new(chan, false, &fence);
> >                       if (ret == 0) {
> > +                             /* TODO: figure out a better solution here
> > +                              *
> > +                              * wait on the fence here explicitly as going through
> > +                              * ttm_bo_move_accel_cleanup somehow doesn't seem to do it.
> > +                              *
> > +                              * Without this the operation can timeout and we'll fallback to a
> > +                              * software copy, which might take several minutes to finish.
> > +                              */
> > +                             nouveau_fence_wait(fence, false, false);
> >                               ret = ttm_bo_move_accel_cleanup(bo,
> >                                                               &fence->base,
> >                                                               evict, false,
> > --
> > 2.37.1
> >
> >
>
> While this is marked for 5.15+ only, a user in Debian was seeing the
> suspend issue as well on 5.10.y and did confirm the commit fixes the
> issue as well in the 5.10.y series:
>
> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=989705#69
>
> Karol, Lyude, should that as well be picked for 5.10.y?
>

mhh from the original report 5.10 was fine, but maybe something got
backported and it broke it? I'll try to do some testing on my machine
and see what I can figure out, but it could also be a debian only
issue at this point.

> Regards,
> Salvatore
>