virglrenderer regression in commit ad4f0f1941677c

virglrenderer regression in commit ad4f0f1941677c

[Rob Herring]

Hi,

This commit in virglrenderer causes a regression in Android for me.
The parameters that get passed in are last_level = 8, width = 1. I'm
not really sure if this is valid (I'm guessing there should be some
min width?), or where I should be looking to fix this. Any ideas?

commit ad4f0f1941677c6cd78bcd14348cd99ae7dd7527
Author: Marc-André Lureau <marcandre.lureau@redhat.com>
Date:   Tue Jan 19 14:37:50 2016 +0100

    renderer: reject large LOD values

    Or we could sit for a very long time in some further loops.

    Fix found thanks to american fuzzy lop.

    Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>

Thanks,
Rob
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: virglrenderer regression in commit ad4f0f1941677c

[Marc-André Lureau]

Hi

----- Original Message -----
> Hi,
> 
> This commit in virglrenderer causes a regression in Android for me.
> The parameters that get passed in are last_level = 8, width = 1. I'm
> not really sure if this is valid (I'm guessing there should be some
> min width?), or where I should be looking to fix this. Any ideas?
> 
> commit ad4f0f1941677c6cd78bcd14348cd99ae7dd7527
> Author: Marc-André Lureau <marcandre.lureau@redhat.com>
> Date:   Tue Jan 19 14:37:50 2016 +0100
> 
>     renderer: reject large LOD values
> 
>     Or we could sit for a very long time in some further loops.
> 
>     Fix found thanks to american fuzzy lop.
> 
>     Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>


I don't know whether a texture of 1x1 with a LOD of 8 is valid. I remember that fix was related to invalid virgl command stream found by afl, that would be able to loop millions of iterations in some code, such as vrend_renderer_resource_create(). I wonder if you could trace it back to some client code. Also worth if there is a valid opengl reproducer, that could be a new piglit test.

Btw, there is a virglrenderer mailing list: virglrenderer-devel@lists.freedesktop.org
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: virglrenderer regression in commit ad4f0f1941677c

[Rob Clark]

On Fri, Jul 29, 2016 at 3:57 PM, Rob Herring <robh@kernel.org> wrote:
> Hi,
>
> This commit in virglrenderer causes a regression in Android for me.
> The parameters that get passed in are last_level = 8, width = 1. I'm
> not really sure if this is valid (I'm guessing there should be some
> min width?), or where I should be looking to fix this. Any ideas?

what is the height?  last mip-map level is basically the number of
times you could half the width+height (round up to 1) until it is 1x1.
So something like 1x512 would have last_level=8.  (I might be off by
one right-shift..)

Try this:

- if (args->last_level > (floor(log2(MAX2(args->width, args->width))) + 1))
+ if (args->last_level > (floor(log2(MAX2(args->width, args->height))) + 1))

BR,
-R

> commit ad4f0f1941677c6cd78bcd14348cd99ae7dd7527
> Author: Marc-André Lureau <marcandre.lureau@redhat.com>
> Date:   Tue Jan 19 14:37:50 2016 +0100
>
>     renderer: reject large LOD values
>
>     Or we could sit for a very long time in some further loops.
>
>     Fix found thanks to american fuzzy lop.
>
>     Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
>
> Thanks,
> Rob
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev

Re: virglrenderer regression in commit ad4f0f1941677c

[Marc-André Lureau]

Hi

----- Original Message -----
> On Fri, Jul 29, 2016 at 3:57 PM, Rob Herring <robh@kernel.org> wrote:
> > Hi,
> >
> > This commit in virglrenderer causes a regression in Android for me.
> > The parameters that get passed in are last_level = 8, width = 1. I'm
> > not really sure if this is valid (I'm guessing there should be some
> > min width?), or where I should be looking to fix this. Any ideas?
> 
> what is the height?  last mip-map level is basically the number of
> times you could half the width+height (round up to 1) until it is 1x1.
> So something like 1x512 would have last_level=8.  (I might be off by
> one right-shift..)
> 
> Try this:
> 
> - if (args->last_level > (floor(log2(MAX2(args->width, args->width))) + 1))
> + if (args->last_level > (floor(log2(MAX2(args->width, args->height))) + 1))

Oh, looks like what I wanted but I wouldn't see the typo even after re-reading...

please send a patch on the virgl mailing list.
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev

Re: virglrenderer regression in commit ad4f0f1941677c

[Rob Clark]

On Mon, Aug 1, 2016 at 1:30 AM, Marc-André Lureau <mlureau@redhat.com> wrote:
>
> Hi
>
> ----- Original Message -----
>> On Fri, Jul 29, 2016 at 3:57 PM, Rob Herring <robh@kernel.org> wrote:
>> > Hi,
>> >
>> > This commit in virglrenderer causes a regression in Android for me.
>> > The parameters that get passed in are last_level = 8, width = 1. I'm
>> > not really sure if this is valid (I'm guessing there should be some
>> > min width?), or where I should be looking to fix this. Any ideas?
>>
>> what is the height?  last mip-map level is basically the number of
>> times you could half the width+height (round up to 1) until it is 1x1.
>> So something like 1x512 would have last_level=8.  (I might be off by
>> one right-shift..)
>>
>> Try this:
>>
>> - if (args->last_level > (floor(log2(MAX2(args->width, args->width))) + 1))
>> + if (args->last_level > (floor(log2(MAX2(args->width, args->height))) + 1))
>
> Oh, looks like what I wanted but I wouldn't see the typo even after re-reading...
>
> please send a patch on the virgl mailing list.


tbh, I don't even have a virgl setup (psuedo-patch was just from
looking at cgit)..  so I'd be more than happy if someone else could
actually test that change and send a patch.

BR,
-R
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev