[Lucid] Poor scrolling performance in browser (Firefox and Chrome, possibly others)

[willjcroz]

Binary package hint: firefox

Poor scrolling in Firefox and Google-Chrome beta in Lucid. Using radeon driver with X1400 GPU on a Thinkpad T60 on latest Lucid Beta 2 (fully updated on 13/04/2010).

Probably needs to be assigned to xorg-video-ati or xorg-xserver or something like that, but for now I just used apport to report the problem with Firefox.

Note: This has been tested with KMS both enabled and disabled (using nomodeset kernel append) the problem is present in both circumstances.

ProblemType: Bug
DistroRelease: Ubuntu 10.04
Package: firefox 3.6.3+nobinonly-0ubuntu2
ProcVersionSignature: Ubuntu 2.6.32-20.30-generic 2.6.32.11+drm33.2
Uname: Linux 2.6.32-20-generic x86_64
Architecture: amd64
Date: Tue Apr 13 14:34:47 2010
FirefoxPackages:
 firefox 3.6.3+nobinonly-0ubuntu2
 firefox-gnome-support 3.6.3+nobinonly-0ubuntu2
 firefox-branding 3.6.3+nobinonly-0ubuntu2
 abroswer N/A
 abrowser-branding N/A
InstallationMedia: Ubuntu 10.04 "Lucid Lynx" - Beta amd64 (20100406.1)
ProcEnviron:
 LANG=en_GB.utf8
 SHELL=/bin/bash
SourcePackage: firefox

[willjcroz (willjcroz)]

Apologies, this *does* in fact seem to be radeon KMS related, I was using nomodeset as the kernel parameter when in fact this seems to do nothing now. After doing 'radeon.modeset=0 vga=771' I am now seeing good performance in both browsers. If anyone wants me to refile this bug under xorg-video-radeon, please let me know...

[John Doe (jodo-deactivatedaccount)]

Could be ati/radeon issue. I'm having the same trouble, with an ATI X600M and radeon driver.

[willjcroz (willjcroz)]

An additional clue: performance in google-chrome is good after booting up and logging in (using KMS) but then becomes slow after some time.

[Øyvind Stegard (oyvindstegard)]

Tried the Lucid RC in a live session from USB stick.. Scrolling performance was abysmal in Firefox. So bad that I scrapped my plan to actually install Lucid onto the hard-drive. ATI X1400 128MB graphics card, system has 3GiB RAM. Strangely enough, when monitoring top while scrolling, a process called "hd-audio0" seems to suck up excessive amounts of CPU time (up to 50% on dual-core) in proportion to amount of graphics activity. So looks like a freakin audio card driver is actually using CPU whenever there are graphical operations going on (same things happens when moving windows around). This is further confirmed by the fact that audio crackles ever so slightly if I scroll fast while playing some music. Note that I have not actually tested an installed version of Lucid, but I find it strange that running from USB would consistently affect the graphics performance in this way.

[Øyvind Stegard (oyvindstegard)]

The issue is more general for me. After a while, and a few open windows (four virtual desktops), graphics performance turns sluggish with KMS. It doesn't just affect Firefox, but also other apps like Evince (looking through a PDF document becomes painful), Gnome-terminal and movie playback (Xv needs composite redirection disabled to avoid dropping frames in fullscreen). The slowness is much more prominent with larger/maximized/ fullscreen windows, rather than for smaller ones. I don't think I'm affected by a GEM memory leakage, as it looks like GEM objects are in fact cleaned up when closing X apps (allocation decreases).

My PCIE ATI X1400 graphics card has 128MB VRAM, system has 3GB. Right now, things are slow as molasses, and current GEM status is:
1699 objects
150470656 object bytes
0 pinned
0 pin bytes
0 gtt bytes
0 gtt total

I seem to be getting some messages about hpet timer in kernel log, together with sluggish graphics, but I don't know if it's related:
[...]
[19313.601918] CE: hpet increasing min_delta_ns to 113904 nsec
[19362.762508] CE: hpet increasing min_delta_ns to 170856 nsec
[20965.902711] CE: hpet increasing min_delta_ns to 256284 nsec

Running Compiz, Ubuntu 10.04 installed from RC, fully up to date as of now. I haven't tried living on the xorg edge, yet, but have tried a radeon driver git snapshot. It didn't really do anything for the performance problems I'm seeing with Lucid and radeon-KMS.

[willjcroz (willjcroz)]

Seems to be better for me with 2.6.34-rc6 from http://kernel.ubuntu.com/~kernel-ppa/mainline/v2.6.34-rc6-lucid/ but it is early days so I will post again soon to confirm this.

[willjcroz (willjcroz)]

Forgot to say: yes, I was seeing those (occasional) hpet messages also, so might be worth trying hpet=disable appended to the kernel command line and seeing how this affects the performance under 2.6.32-21. hpet on some machines can apparently cause problems with dynticks due to buggy hardware.

[Øyvind Stegard (oyvindstegard)]

Much better radeon KMS performance with mainline >= 2.6.34-rc6. Seems it good a nice speed boost indeed. Unfortunately, due to it causing other problems, I'm stuck with UMS at the moment.