Bug 518595

Summary: [KMS] RV370 slow 2D and 3D unless radeon.modeset=0 used
Product: [Fedora] Fedora Reporter: Tim Waugh <twaugh>
Component: xorg-x11-drv-atiAssignee: X/OpenGL Maintenance List <xgl-maint>
Status: CLOSED RAWHIDE QA Contact: Fedora Extras Quality Assurance <extras-qa>
Severity: medium Docs Contact:
Priority: low    
Version: rawhideCC: bugs, vedran, xgl-maint
Target Milestone: ---   
Target Release: ---   
Hardware: All   
OS: Linux   
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2009-10-20 15:00:39 UTC Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Attachments:
Description Flags
Xorg.0.log-drm.debug=1
none
dmesg
none
/var/log/messages
none
Xorg.0.log-radeon.modset=0 none

Description Tim Waugh 2009-08-21 08:10:51 UTC
Description of problem:
Redrawing windows is visibly faster when booting with radeon.modeset=0.  Scrolling text input areas in Firefox is particularly slow without that.

Version-Release number of selected component (if applicable):
xorg-x11-server-Xorg-1.6.99-36.20090814.fc12.x86_64
xorg-x11-drv-ati-6.12.2-21.fc12.x86_64
kernel-2.6.31-0.162.rc6.git2.fc12.x86_64

How reproducible:
100%

Steps to Reproduce:
1.Boot as normal (i.e. without radeon.modeset=0).
2.Drag windows around the screen.
3.Boot with radeon.modeset=0 on the kernel command line
4.Compare
  
Actual results:
Much faster with radeon.modeset=0.

Expected results:
Fast even with modesetting. :-)

Comment 1 Tim Waugh 2009-08-21 08:11:45 UTC
Created attachment 358197 [details]
Xorg.0.log-drm.debug=1

Xorg.0.log from a boot with drm.debug=1 (and not radeon.modeset=0).

Comment 2 Tim Waugh 2009-08-21 08:12:29 UTC
Created attachment 358198 [details]
dmesg

dmesg from booting with drm.debug=1 (and not radeon.modeset=0).

Comment 3 Tim Waugh 2009-08-21 08:13:35 UTC
Created attachment 358200 [details]
/var/log/messages

/var/log/messages from booting with drm.debug=1 (and not radeon.modeset=0).  Rsyslog is configured to store debug level messages there too.

Comment 4 Tim Waugh 2009-08-21 08:14:22 UTC
Created attachment 358201 [details]
Xorg.0.log-radeon.modset=0

Xorg.0.log from booting with radeon.modeset=0

Comment 5 Nigel Jones 2009-09-01 00:13:00 UTC
seeing similar behaviour here on T60p with V5200 graphics. 200-600 FPS with kernel modesetting, back to 1300-1500 when disabled

Comment 6 Nigel Jones 2009-09-02 05:49:10 UTC
Further "general use" with kms enabled results in extremely slow UI response. Window movements are very jerky (redraw), just minimizing takes > 1s to repaint. Effectively near unusable. glxgears (albeit artifical test) appears to get slower the longer the system up / more windows open. Currently mine's down to ~150.

Can provide additional logs if required, but sounds like the same issue.

Did note that with kms disabled (radeon.modeset=0) that after a suspend/resume cycle "vbetool" would be spinning cpu cycles, and if killed would just call another suspend. Will try to reconfigure around it.

Currently using
kernel-2.6.31-0.190.rc8.fc12.i686
xorg-x11-drv-ati-6.13.0-0.2.20090821gitb1b77a4d6.fc12.i686

tried backing driver off to xorg-x11-drv-ati-6.12.2-21.fc12.i686 (alpha level) but made no difference

System is F12 alpha, updated with rawhide updates with no packages held back due to dependency issues, ie updates running clean as of 2 Sep 0630 UTC+1

Comment 7 Nigel Jones 2009-09-02 13:19:58 UTC
Further observation.

Went back to having modeset enabled, and got vbetool spinning.

Moved /usr/sbin/vbetool to vbetool.bad
No longer any spinning. After resume could unlock X, although VT switching did not work
However even with modesetting enabled, prior to suspend/resume was getting 2000 fps in glxgears, now getting ~800 with nothing else significant running.

Comment 8 Nigel Jones 2009-10-20 14:04:28 UTC
Have been running back with KMS enabled for a few weeks now, and this performance problem has not re-appeared, so looks like it's now fixed.

Comment 9 Vedran Miletić 2009-10-20 15:00:39 UTC
Thank you for reporting back.