linux-kernel - Re: [PATCH 1/2] drm/vram: store dumb bo dimensions.

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <20190710152622.GQ15868@phenom.ffwll.local>
Date:   Wed, 10 Jul 2019 17:26:22 +0200
From:   Daniel Vetter <daniel@...ll.ch>
To:     Pekka Paalanen <ppaalanen@...il.com>
Cc:     Daniel Vetter <daniel@...ll.ch>, Sam Ravnborg <sam@...nborg.org>,
        Maxime Ripard <maxime.ripard@...tlin.com>,
        open list <linux-kernel@...r.kernel.org>,
        dri-devel@...ts.freedesktop.org, David Airlie <airlied@...ux.ie>,
        Gerd Hoffmann <kraxel@...hat.com>, tzimmermann@...e.de,
        Sean Paul <sean@...rly.run>
Subject: Re: [PATCH 1/2] drm/vram: store dumb bo dimensions.

On Mon, Jul 08, 2019 at 04:39:45PM +0300, Pekka Paalanen wrote:
> On Wed, 26 Jun 2019 18:27:54 +0200
> Daniel Vetter <daniel@...ll.ch> wrote:
> 
> > On Wed, Jun 26, 2019 at 04:40:13PM +0200, Sam Ravnborg wrote:
> > > Hi Gerd.
> > > 
> > > On Wed, Jun 26, 2019 at 08:55:50AM +0200, Gerd Hoffmann wrote:  
> > > > Store width and height of the bo.  Needed in case userspace
> > > > sets up a framebuffer with fb->width != bo->width..
> > > > 
> > > > Signed-off-by: Gerd Hoffmann <kraxel@...hat.com>
> > > > ---
> > > >  include/drm/drm_gem_vram_helper.h     | 1 +
> > > >  drivers/gpu/drm/drm_gem_vram_helper.c | 2 ++
> > > >  2 files changed, 3 insertions(+)
> > > > 
> > > > diff --git a/include/drm/drm_gem_vram_helper.h b/include/drm/drm_gem_vram_helper.h
> > > > index 1a0ea18e7a74..3692dba167df 100644
> > > > --- a/include/drm/drm_gem_vram_helper.h
> > > > +++ b/include/drm/drm_gem_vram_helper.h
> > > > @@ -39,6 +39,7 @@ struct drm_gem_vram_object {
> > > >  	struct drm_gem_object gem;
> > > >  	struct ttm_buffer_object bo;
> > > >  	struct ttm_bo_kmap_obj kmap;
> > > > +	unsigned int width, height;
> > > >  
> > > >  	/* Supported placements are %TTM_PL_VRAM and %TTM_PL_SYSTEM */
> > > >  	struct ttm_placement placement;
> > > > diff --git a/drivers/gpu/drm/drm_gem_vram_helper.c b/drivers/gpu/drm/drm_gem_vram_helper.c
> > > > index 4de782ca26b2..c02bf7694117 100644
> > > > --- a/drivers/gpu/drm/drm_gem_vram_helper.c
> > > > +++ b/drivers/gpu/drm/drm_gem_vram_helper.c
> > > > @@ -377,6 +377,8 @@ int drm_gem_vram_fill_create_dumb(struct drm_file *file,
> > > >  	gbo = drm_gem_vram_create(dev, bdev, size, pg_align, interruptible);
> > > >  	if (IS_ERR(gbo))
> > > >  		return PTR_ERR(gbo);
> > > > +	gbo->width = args->width;
> > > > +	gbo->height = args->height;
> > > >  
> > > >  	ret = drm_gem_handle_create(file, &gbo->gem, &handle);
> > > >  	if (ret)  
> > > 
> > > Be warned, I may have missed something in the bigger picture.
> > > 
> > > Your patch will set width and height only for dumb bo's
> > > But we have several users of drm_gem_vram_create() that will
> > > not set the width and height.
> > > 
> > > So only in some cases can we rely on them being set.
> > > Should this be refactored so we always set width, height.
> > > Or maybe say in a small comment that width,height are only
> > > set for dumb bo's?  
> > 
> > Also for dumb bo allocated buffers if userspace gets the dimensions wrong
> > between dumb_create and the addfb, something is wrong. Papering over that
> > by remembering the right dimensions doesn't look like a good solution.
> 
> Hi,
> 
> just a note irrelevant to this particular driver:
> 
> I have deliberately allocated a too high dumb buffer in userspace and
> created multiple smaller DRM FBs out of it with different offsets
> (i * 128 * stride). This has been a very useful hack to see that a
> GPU-less driver actually honours fences correctly, because if it
> doesn't, the whole image will be off by the offset delta, which is
> epileptically easy to see.
> 
> So I'm not getting the height wrong, I am deliberately overallocating
> and aliasing.

Yeah that's the other reason for why this patch is wrong: It would break
things like this trickery here :-) The separation between backing storage
and drm_fb is intentional, multiple fb in one overall fb is explicitly ok.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch