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;
Date:   Wed, 30 Nov 2022 12:21:19 +0100
From:   Daniel Vetter <daniel@ffwll.ch>
To:     Rob Clark <robdclark@gmail.com>
Cc:     Guenter Roeck <linux@roeck-us.net>,
        Rob Clark <robdclark@chromium.org>,
        open list <linux-kernel@vger.kernel.org>,
        stable@vger.kernel.org, Eric Anholt <eric@anholt.net>,
        Noralf =?iso-8859-1?Q?Tr=F8nnes?= <noralf@tronnes.org>,
        dri-devel@lists.freedesktop.org,
        Thomas Zimmermann <tzimmermann@suse.de>
Subject: Re: [2/2] drm/shmem-helper: Avoid vm_open error paths
Message-ID: <Y4c8r0fxjI+WsvhM@phenom.ffwll.local>
Mail-Followup-To: Rob Clark <robdclark@gmail.com>,
        Guenter Roeck <linux@roeck-us.net>,
        Rob Clark <robdclark@chromium.org>,
        open list <linux-kernel@vger.kernel.org>, stable@vger.kernel.org,
        Eric Anholt <eric@anholt.net>,
        Noralf =?iso-8859-1?Q?Tr=F8nnes?= <noralf@tronnes.org>,
        dri-devel@lists.freedesktop.org,
        Thomas Zimmermann <tzimmermann@suse.de>
References: <20221129200242.298120-3-robdclark@gmail.com>
 <20221129203205.GA2132357@roeck-us.net>
 <CAF6AEGuK4jv25cQ4p-rrytx9Qn4JZdRRfkVJn9T3nf7vJmG5VQ@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CAF6AEGuK4jv25cQ4p-rrytx9Qn4JZdRRfkVJn9T3nf7vJmG5VQ@mail.gmail.com>
Precedence: bulk

On Tue, Nov 29, 2022 at 12:47:42PM -0800, Rob Clark wrote:
> On Tue, Nov 29, 2022 at 12:32 PM Guenter Roeck <linux@roeck-us.net> wrote:
> >
> > On Tue, Nov 29, 2022 at 12:02:42PM -0800, Rob Clark wrote:
> > > From: Rob Clark <robdclark@chromium.org>
> > >
> > > vm_open() is not allowed to fail.  Fortunately we are guaranteed that
> > > the pages are already pinned, and only need to increment the refcnt.  So
> > > just increment it directly.
> >
> > I don't know anything about drm or gem, but I am wondering _how_
> > this would be guaranteed. Would it be through the pin function ?
> > Just wondering, because that function does not seem to be mandatory.
> 
> We've pinned the pages already in mmap.. vm->open() is perhaps not the
> best name for the callback function, but it is called for copying an
> existing vma into a new process (and for some other cases which do not
> apply here because VM_DONTEXPAND).
> 
> (Other drivers pin pages in the fault handler, where there is actually
> potential to return an error, but that change was a bit more like
> re-writing shmem helper ;-))

Yhea vm_ops->open should really be called vm_ops->dupe or ->copy or
something like that ...
-Daniel

> 
> BR,
> -R
> 
> > >
> > > Fixes: 2194a63a818d ("drm: Add library for shmem backed GEM objects")
> > > Cc: stable@vger.kernel.org
> > > Signed-off-by: Rob Clark <robdclark@chromium.org>
> > > ---
> > >  drivers/gpu/drm/drm_gem_shmem_helper.c | 14 +++++++++++---
> > >  1 file changed, 11 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/drm_gem_shmem_helper.c b/drivers/gpu/drm/drm_gem_shmem_helper.c
> > > index 110a9eac2af8..9885ba64127f 100644
> > > --- a/drivers/gpu/drm/drm_gem_shmem_helper.c
> > > +++ b/drivers/gpu/drm/drm_gem_shmem_helper.c
> > > @@ -571,12 +571,20 @@ static void drm_gem_shmem_vm_open(struct vm_area_struct *vma)
> > >  {
> > >       struct drm_gem_object *obj = vma->vm_private_data;
> > >       struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
> > > -     int ret;
> > >
> > >       WARN_ON(shmem->base.import_attach);
> > >
> > > -     ret = drm_gem_shmem_get_pages(shmem);
> > > -     WARN_ON_ONCE(ret != 0);
> > > +     mutex_lock(&shmem->pages_lock);
> > > +
> > > +     /*
> > > +      * We should have already pinned the pages, vm_open() just grabs
> >
> > should or guaranteed ? This sounds a bit weaker than the commit
> > description.
> >
> > > +      * an additional reference for the new mm the vma is getting
> > > +      * copied into.
> > > +      */
> > > +     WARN_ON_ONCE(!shmem->pages_use_count);
> > > +
> > > +     shmem->pages_use_count++;
> > > +     mutex_unlock(&shmem->pages_lock);
> >
> > The previous code, in that situation, would not increment pages_use_count,
> > and it would not set not set shmem->pages. Hopefully, it would not try to
> > do anything with the pages it was unable to get. The new code assumes that
> > shmem->pages is valid even if pages_use_count is 0, while at the same time
> > taking into account that this can possibly happen (or the WARN_ON_ONCE
> > would not be needed).
> >
> > Again, I don't know anything about gem and drm, but it seems to me that
> > there might now be a severe problem later on if the WARN_ON_ONCE()
> > ever triggers.
> >
> > Thanks,
> > Guenter
> >
> > >
> > >       drm_gem_vm_open(vma);
> > >  }

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch