[Xorg-driver-geode] last call for testing: X.org Geode2.11.9 release candidate

Mon Aug 16 03:58:37 PDT 2010

On E, 2010-08-16 at 18:28 +0800, Cui, Hunk wrote:
> Hi, Martin,
> 	Through some simple test, I found the root cause focus on the
> Xerver/dix/Pixmap.c -> FreeScratchPixmapHeader function -> Line
> "pScreen->pScratchPixmap = pPixmap;" when I commented out this line,
> also it seemed OK. So I think this issue doesn't associate with
> rotation patch, this is a new topic.

This associates with the patch just fine.
pScreen has a cache of one scratch pixmap, GetScratchPixmapHeader reuses
it, if it is set - by setting its headers with those given to
GetScratchPixmapHeader with a call to ModifyPixmapHeader. Then later a
FreeScratchPixmapHeader call simply puts that pixmap back into that one
item cache, if one isn't there yet, instead of destroying it. That's
all. This caching is supposed to be transparent to the API user.

ScratchPixmapHeader itself seems to be pretty much a wrapper pixmap,
with which you can give pixmaps to places that need one, but when you
have the actual pixmap image content somewhere else - so you just wrap
it with the pixmap headers, which contain the width, height, pitch and
that sort of things.

> 	My suggestion: rotation patch should be restored,

It hasn't been reverted, it's still there right now...

> it can release new version driver 2.11.9. At the same time, I will
> find the root reason, any new progress, I will sent to the mailing
> list and inform to you.

Meanwhile rotation is completely broken when user rotates and then
rotates back to normal. I'll take a look in the evening, and if can't
figure it out, we can do a workaround that just doesn't destroy the
shadow buffer once it has been created until the root reason is found.

Regards,
Mart Raudsepp

> Thanks,
> Hunk Cui
> 
> > -----Original Message-----
> > From: Huang, FrankR
> > Sent: Monday, August 16, 2010 3:43 PM
> > To: 'Mart Raudsepp'; q-funk at iki.fi; Cui, Hunk
> > Cc: Geode Mailing List; Bernardo Innocenti; Andres Salomon
> > Subject: RE: [Xorg-driver-geode] last call for testing: X.org Geode2.11.9
> > release candidate
> > 
> > We have reproduced this bug already.
> > It is caused by the commit:
> > 	http://cgit.freedesktop.org/xorg/driver/xf86-video-
> > geode/commit/?id=8a61ef8f9f6b9dac6804754572250e59d80bdf06
> > Hunk give research on this issue.
> > 
> > 
> > Martin-Éric,
> > 
> > 	If that is a plan to release new version driver today, rotation patch
> > above should be restored back before that is solved.
> > 
> > 
> > Thanks,
> > Frank
> > 
> > > -----Original Message-----
> > > From: xorg-driver-geode-bounces+frankr.huang=amd.com at lists.x.org
> > > [mailto:xorg-driver-geode-bounces+frankr.huang=amd.com at lists.x.org] On
> > > Behalf Of Mart Raudsepp
> > > Sent: 2010年8月16日 15:03
> > > To: q-funk at iki.fi
> > > Cc: Geode Mailing List; Bernardo Innocenti; Andres Salomon
> > > Subject: Re: [Xorg-driver-geode] last call for testing: X.org Geode2.11.9
> > > release candidate
> > >
> > > On E, 2010-08-16 at 09:51 +0300, Mart Raudsepp wrote:
> > > > On E, 2010-08-16 at 09:49 +0300, Martin-Éric Racine wrote:
> > > > > On Mon, Aug 16, 2010 at 9:36 AM, Mart Raudsepp <leio at gentoo.org> wrote:
> > > > > > On L, 2010-08-14 at 11:32 +0300, Martin-Éric Racine wrote:
> > > > > >> Greetings,
> > > > > >>
> > > > > >> As recently announced, we now have a release candidate for the
> > > X.org
> > > > > >> Geode 2.11.9 video driver. [1] Please test
> > > > > >> it extensively and immediately report any significant regression
> > > > > >> directly into the X.org Bugzilla. [2] If nothing major gets
> > > reported
> > > > > >> during this weekend, we'll release what we currently have in GIT by
> > > > > >> Monday morning.
> > > > > >
> > > > > > Looks like the rotation code isn't quite perfect yet.
> > > > > >
> > > > > > Before doing any rotation:
> > > > > >
> > > > > > http://dev.gentoo.org/~leio/xorg/geode-before-rotation-20100816.png
> > > > > >
> > > > > > After rotating to inverted mode, and then (importantly) back to
> > > normal
> > > > > > mode:
> > > > > >
> > > > > > http://dev.gentoo.org/~leio/xorg/geode-after-rotation-20100816.png
> > > > > >
> > > > > > Notice all icon images missing again, various glyphs turned to
> > > vertical
> > > > > > lines and a broken openbox panel black/grey gradient on the bottom.
> > > > > >
> > > > > > It seems that right after the rotation patch at commit 8a61ef8f, the
> > > > > > problem is already present.
> > > > > >
> > > > > > I would consider this a release blocker for a while. OTOH, rotation
> > > > > > didn't work at all with xserver-1.7+ before, so I guess it depends
> > > on
> > > > > > how quick we can find a solution for this, and if xserver-1.6 (where
> > > > > > rotation worked before) is affected by this.
> > > > >
> > > > > Thanks for reporting this.
> > > > >
> > > > > Since OLPC are the main user for this feature, I would have
> > > > > appreciated receiving feedback from them earlier than this.
> > > >
> > > > Bernie's feedback just now was that it works fine. But just asked to
> > > > test what happens after going back to normal mode, as that's where the
> > > > problem appears - after destroying of the scratch pixmap header that was
> > > > created upon entering rotated mode.
> > > >
> > > > Also on XO-1 things were working fine before, as the reserved memory
> > > > approach we had before didn't trigger any issues with xserver-1.6, that
> > > > is used on the XO-1 FC-11 images, but didn't work at all since
> > > > xserver-1.7+.
> > >
> > >
> > > Just to be more clear:
> > >
> > > Rotation works _fine_ with xf86-video-geode-2.11.8 + xserver-1.6(!), so
> > > if this change breaks it on xserver-1.6 too (which XO-1 uses), then it's
> > > a regression worth fixing.
> > >
> > > And I'm pretty sure a fix is not hard, after having debugged the issue a
> > > bit. I'm rather confident I can get it fixed within a couple hours, but
> > > I need to go to earn my salary now. Maybe Hunk can get it fixed today
> > > before me too.
> > >
> > > Mart
> > >
> > > >
> > > > > Personally, given how rotation has already been broken for quite some
> > > > > time, I don't consider this breakage as a good enough reason to avoid
> > > > > releasing as planned. Instead, I would hope that our AMD developers
> > > > > could be sent their own XO-1 to work with, along with a URL to
> > > > > instructions on how to build our driver for the OLPC and on how to
> > > > > install it onto the laptop's filesystem. I would also hope to see
> > > > > continuous feedback from the OLPC developer community towards ensuring
> > > > > that our driver keeps on working on the XO-1.
> > > > >
> > > > > Martin-Éric
> > > >
> > > >
> > > > _______________________________________________
> > > > Xorg-driver-geode mailing list
> > > > Xorg-driver-geode at lists.x.org
> > > > http://lists.x.org/mailman/listinfo/xorg-driver-geode
> > >
> > > _______________________________________________
> > > Xorg-driver-geode mailing list
> > > Xorg-driver-geode at lists.x.org
> > > http://lists.x.org/mailman/listinfo/xorg-driver-geode