After upgrading from release 6.8.1 to release 6.8.2 for an IBM Thinkpad 600E notebook running FC3, artifacts appear in the upper part of the desktop which result from erroneous mirroring of objects in the lower part of the screen. The operating mode is 1280x1024 pixel ("externdisp") at 16 bpp which requires the total 2560 KB of video memory available on this notebook. The artifacts (screenshot attached) disappear when clicking the background. The phenomenon only becomes visible for resolutions higher than the standard 1024 x 768 pixels of the internal LCD. The ghost images can be by prevented by adding the "ShadowFB" option to the "xorg.conf" device section. It is also worth mentioning that this behaviour which had been absent from release 6.7.0 showed up for some 6.8 prerelease, but had gone away as the final 6.8 release went public.
Created attachment 1988 [details] X.org 6.8.2 Neomagic 256AV (NM2200) screenshot on TP 600E
Initially posted by myself as Red Hat Bugzilla Bug (#127574): https://bugzilla.redhat.com/beta/show_bug.cgi?id=127574 Accordingly, I have to correct myself: it got posted as a 6.7.0 bug in the first place. My previous comment that the artifacts only appeared for some 6.8 beta release was wrong.
Do the artifacts go away if you turn XAA off? If so, could you locate which XAA operation is causing the problem? You can disable XAA entirely using Option "NoAccel", and individual XAA operations with Option "XaaNo...", where the available operations are documented in xorg.conf. All of these options go in your xorg.conf's Screen section.
As I wrote in my original posting, the "ShadowFB" option cures the artifact flaw. Adding the plain "noaccel" option is an utterly worse choice because in this case, there is no acceleration at all, and X becomes extremely sluggish. I have localized the accelerated feature which is responsible of the reported behaviour: by no surprise at all, adding the "XaaNoSolidFillRect" option to the "xorg.conf" file solves the problem. The X desktop remains quite responsive in this case, comparable to adding the "ShadowFB" option itself.
I wasn't suggesting that NoAccel was a usable configuration, just that it would help us in localizing the problem.
OK, I'll take a look at this once I find time.
Created attachment 2386 [details] [review] Fix for several issues (hopefully including the one reported here) This fix - eliminates the doublescan modes above 1024 lines hight. - eliminates interlaced modes (as we don't know how to activate them) - fixes pixel clock calculation to prevent lockups.
Committed, please test. 2005-04-11 Egbert Eich <eich-at-freedesktop-dot-org> * programs/Xserver/hw/xfree86/drivers/neomagic/neo_driver.c: (NEOPreInit), (NEOValidMode), (neoRestore), (neoModeInit), (neoCalcVCLK): Fix several mode setting issues: - eliminates DoubleScan modes above height == 1024. - eliminates interlaced modes as we don't know how to turn on interlace. - Fixes clock calculation to prevent lockups.
This one has been committed some time ago. No further comments seem to have been added here. Therefore I'm closing it now.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.