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

Message-ID: <20190708163945.1d3757b3@eldfell.localdomain>
Date:   Mon, 8 Jul 2019 16:39:45 +0300
From:   Pekka Paalanen <ppaalanen@...il.com>
To:     Daniel Vetter <daniel@...ll.ch>
Cc:     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 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.


Thanks,
pq

Content of type "application/pgp-signature" skipped