| Summary: | [r600g] Regression: EVE Online graphics borked (bisected) | ||
|---|---|---|---|
| Product: | Mesa | Reporter: | Luzipher <luziphermcleod> |
| Component: | Drivers/Gallium/r600 | Assignee: | Ian Romanick <idr> |
| Status: | RESOLVED FIXED | QA Contact: | |
| Severity: | normal | ||
| Priority: | medium | CC: | idr, tallica |
| Version: | git | ||
| Hardware: | x86-64 (AMD64) | ||
| OS: | Linux (All) | ||
| Whiteboard: | |||
| i915 platform: | i915 features: | ||
| Attachments: |
Screenshot of login screen (bad)
Screenshot of login screen (good, commit 015d4f61ef9116c9e844299ab9f2b15c653c0450) terrible hack |
||
|
Description
Luzipher
2011-10-05 18:33:53 UTC
Created attachment 52027 [details]
Screenshot of login screen (good, commit 015d4f61ef9116c9e844299ab9f2b15c653c0450)
Adding author of bisected commit to CC. *** Bug 41508 has been marked as a duplicate of this bug. *** Does the application produce any error logging or other debug output? No, there is no crash, so the application itself probably doesn't even notice something is wrong. I also checked /var/log/Xorg.0.log and /var/log/messages and couldn't see anything abnormal. Wine also didn't print anything unusual in the terminal. If I should look somewhere else or should try to run some debug mode, I'd be happy to do so. By the way ... the performance also seemed to suffer (framerate seems worse with the bug). Created attachment 52140 [details] [review] terrible hack The basic issue seems to be that in _mesa_BindAttribLocationARB() "name" is used for the hashtable key. This is unsafe because that memory is owned by the application and may go away / get overwritten / etc. The attached hack makes at least the Wine d3d tests pass again for me, but obviously introduces a memory leak instead. It's perhaps also worth pointing out that with the way hash_table_replace() works the key can be in use even after the (original) corresponding data is no longer in the table. This may be undesirable in cases where the lifetime of the key is tied to that of the data. Actually, ignore that. I just noticed Ian already fixed this at http://lists.freedesktop.org/archives/mesa-dev/2011-October/013075.html. Fixed on master by the commit below. The bug was a regression in master, so there's no need to cherry pick the patch to a stable branch. commit f3650b05cf4e37066d0f142a4c14fcc650de8d8d Author: Ian Romanick <ian.d.romanick@intel.com> Date: Fri Oct 7 14:29:51 2011 -0700 hash_table: Make string_to_uint_map make a copy of the name The hash table needs a copy of the key that it can keep for comparisons during searches. Signed-off-by: Ian Romanick <ian.d.romanick@intel.com> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=41499 Cc: Stéphane Marchesin <stephane.marchesin@gmail.com> Tested-by: Luzipher <luziphermcleod@yahoo.ie> Tested-by: Michał Lipski <tallica@o2.pl> Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> |
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.