Re: HWA nouveau regression?

Eirik Byrkjeflot Anonsen <[email protected]> Fri, 20 Apr 2012 12:57:52 +0200
Newsgroups gmane.comp.web.opera.linux
Message-ID <[email protected]>
"Eugene Paskevich" <[email protected]> writes:

> On Fri, 20 Apr 2012 08:28:28 +0300, Eirik Byrkjeflot Anonsen
> <[email protected]> wrote:
>
>> "Eugene Paskevich" <[email protected]> writes:
>>
>>> The latest snapshot (1372) seems to be completely broken with regard
>>> to HWA on my nouveau.
>>> I can see the window manager borders, but absolutely nothing inside
>>> the window.
>>> The window controls are functional, though invisible. :)
>>
>> "Absolutely nothing?"  As in completely black?  Or a mess of something
>> that is probably triangles, possibly filled with odd patterns?
>
> I've seen two different behaviors.
> One is when the window contents is the frame snapshot of the
> underlying desktop
> when the window receives show event. Erm... I ignorantly call that
> 'false compositing method'. When window is moved, its contents isn't
> changed,
> so the window isn't transparent.

Ah, ok.  That's when we're not rendering anything at all, so whatever
was on screen before remains.  (When running with a composite manager,
the result is somewhat different in this case, as we get uninitialized
memory instead.)

There's lots of possible reasons for this.  But for the time being I'll
go with the guess that we're messing up the hardware rendering so the
graphics driver can't make much sense of what we ask for.

> The other is somewhat different. I couldn't reproduce it anymore.
> The contents is like black and white vertical lines were seriously
> distorted horizontally... Yes, lame description indeed. But I couldn't
> recognize
> any patterns in what I saw.

:)

Hmm, I'll go with the assumption that it is the hardware rendering that
is messed up.  It can certainly look really "creative".

>>> Snapshot 1360 works OK on the same hardware/drivers.
>>> glxinfo reports:
>>> OpenGL vendor string: nouveau
>>> OpenGL renderer string: Gallium 0.4 on NV4B
>>> OpenGL version string: 2.1 Mesa 7.11.2
>>
>> That would be a 7300 or 7600, I guess.  Do you know how much graphics
>> memory that card has?
>
> Somewhat contradictory... I think it has 256M.
>
> Xorg.log says: (II) NOUVEAU(0): GART: 512MiB available
>
> dmesg:
> [   10.451125] [drm] nouveau 0000:04:00.0: Detected 256MiB VRAM
> [   10.461390] [drm] nouveau 0000:04:00.0: 512 MiB GART (aperture)

I believe that means the card has 256 MiB of memory, and that 512 MiB of
system memory is set up as accessible from the graphics card.

I suspect we might be running out of graphics memory and don't notice
that things fall apart.  We should definitely do something about error
recovery.  But I think it is going to be a while still until we start
looking seriously at memory usage.

Still, it is disconcerting if we run out of graphics memory on only
starting up a blank opera.  Have you tried starting with only a white
page active (to ensure opera paints as little as possible)?

> lspci -vv -nn -s 04:00.0
> 04:00.0 VGA compatible controller [0300]: NVIDIA Corporation G73
> [GeForce 7600 GT] [10de:0391] (rev a1) (prog-if 00 [VGA controller])

I suspect that card may be sufficiently fast that opera could work
acceptably with hw acceleration.

eirik
--
Opera-Linux: https://list.opera.com/mailman/listinfo/opera-linux
More lists: https://list.opera.com/mailman/listinfo/
Unsubscribe: mailto:[email protected]?subject=unsubscribe