Bugzilla – Bug 13405
radeonhd: cursor: linebuffer problems.
Last modified: 2009-05-11 21:20:14 UTC
using randr, I get a flickering mousecursor in a few well-defined areas
of one of my screens.
There are 4 vertical "lines", about 20 pixes in width that go across the
screen from top to bottom. When I place the mouse cursor on top of these
"lines", I get a flickering mouse cursor.
The 4 lines are at about 235-255 pixels from the left edge of the
screen, at 490-510 pixels from the left edge, at at 940-960 pixels from
the left edge and at 1005-1025 pixels from the left edge.
On the other monitor, this does not happen.
The resolution the monitor on which it is happening is running is 1280x1024.
I'll attach a small video showing the flickering.
Hm, the file is too large for bugzilla.
Find it here: http://www.deckmaker.mynetcologne.de/SNV30609.AVI
This looks like a memory controller issue. Did you happen to use fglrx before? That driver is known to mess with the MC.
I did, but that's more than a month ago, and I'm not using it anymore.
Have you done a powercycle since then?
Which hardware is this, which driver version is this, can you provide a log?
I've shut-down and restarted my computer several times since then.
It's a Radeon X1900GT from Sapphire,
and I'm running current git.
Created attachment 12742 [details]
*** Bug 13468 has been marked as a duplicate of this bug. ***
*** Bug 13735 has been marked as a duplicate of this bug. ***
Following on from the duplicated bug #13735, I have discovered that if I have an external monitor attached to the vga port, the laptop panel works correctly, correct resolution, no flickers on the laptop panel at all. I am adding the Xorg log from this invocation so that a comparison can be made to that of #13735 attachments.
Created attachment 13478 [details]
Xorg log with panel and external monitor
(In reply to comment #9)
> Following on from the duplicated bug #13735, I have discovered that if I have
> an external monitor attached to the vga port, the laptop panel works correctly,
> correct resolution, no flickers on the laptop panel at all. I am adding the
> Xorg log from this invocation so that a comparison can be made to that of
> #13735 attachments.
I have an Thinkpad T60 here and it is exactly the other way around. If I do not add an external monitor, everything works fine. If I add a 1024x768er one the internal monitor starts to flicker in some small areas.
Please let me know if any log file or other data is needed.
Created attachment 13665 [details]
No external monitors attached, just my Laptop Panel.
RADEONHD: version 1.1.0, built from git branch master, commit 0f066c20
Seems to be working for me now.
*** Bug 14499 has been marked as a duplicate of this bug. ***
I have the same issue, with latest version (commit a9af866ae712a0048d374dc640e482d1f4ce8859)
Also confirm that it's still here, see the duplicate Bug 14999 for my info (sorry for the dupe). Also a Thinkpad T60p, but the opposite of comment #11 - external monitor has the problem, internal is happy.
Excerpt from Bug 14499 by Konstantin, who found out some interesting details:
Cursor corruption appears when the cursor is on particular parts of the screen.
The corruption appears as blinking lines next to, or on top of the cursor.
These distortions don't change in patter if the cursor is moved strictly up and
down (same horizontal pixel), but change if the cursor is moved to left or
right, even by 1 pixel. The affected areas of the screen are columns ~10px wide
around every 255px (x % 255 == 0, but not x === 0).
The corruption seems to be limited to the cursor only - it doesn't stay around
when the cursor is moved away, and doesn't appear on screenshots.
Different cursors (arrow, text selection, etc) seem to be all affected by the
Same issue here with Fedora 8 and both a X1600XT and X1650
Same issue with an X1550, nearly latest GIT. Will have to compile the latest to check.
I have only the first two vertical lines, though. XAA acceleration. Desktop system, two screens, only the one connected to the DVI is affected. Resolution is 1280x1024.
Same issue on a X1950XT PCIe R580+. RadeonHD 1.1.0 on X.org server 126.96.36.199, DragonFly BSD 1.11 from january (thus, no acceleration). Cursor corruption in 4 vertical 'bands' perhaps the width of a curser (64 pixels), cursor corruption seems to be made out of just fore- and background cursor colors; not those of the underlying framebuffer. Can provide logs if necessary.
I have this problem, too. Dual-screen setup (only one screen affected, the external monitor, on VGA analog output). The mouse cursor flickers at the last 16 columns of 256 pixel boundaries, e.a. 240-255, 496-511 and so on.
My hardware: Mobility FireGL V5200 (Thinkpad T60p)
Operating system: FreeBSD 7 (STABLE branch)
I guess you can mark the bug platform-independent.
Created attachment 17829 [details]
PANEL only Xorg log file - Works fine
This is the Xorg log produced when there is nothing connected to DVI and VGA. This case doesn't seem to have the cursor issue.
Created attachment 17830 [details]
DVI+VGA (PANEL closed) Xorg log file - Facing cursor issue
This is the Xorg log file when both DVI and VGA are connected to external LCDs with PANEL closed. This case faces cursor issues.
I have this problem too.
Laptop model: Thinkpad T60
Chipset: ATI Mobility Radeon X1300 (R5xx)
Operating System: OpenSolaris 2008.11 snv_93
RadeonHD version: 1.2.1
Xorg version: 1.3
I face the cursor problem when I use dual screen setup (DVI+VGA) with PANEL closed. I don't see this issue when I a using PANEL only. I have posted the Xorg log files of working and issue cases. I am seeing the problem consistently over reboots. I will be more than happy to provide any additional information if required.
I'm facing the same problem on a Dell Optiplex 745 Workstation with an "ATI Technologies Inc RV516 [Radeon X1300 Pro]". I'm running a dual monitor setup with both monitors connected to dvi ports and running the same resolution (1280x1024). The cursor distortions only appear on the left (primary) monitor. Just as the aother folks here, i have 3 vertical lined areas, where mouse cursor gets distorted.
I just wanted to report, that i have the same issues when using the radeon driver. With fglrx it worked fine. Since the last usage of fglrx, i've had several powercycles.
Created attachment 20149 [details] [review]
Workaround: leave HW cursor enabled for both CRTCs
Please try if the attached patch solves the corruption problem. Note that it is a workaround, and may not be the correct solution.
There seems to be a bug in how rhdCrtc.Width and rhdCrtc.Height are set, resulting them being set to the full virtual size rather than individual CRTC dimensions. One side effect of this is that when the cursor is in the CRTC on the left, HW cursor for the CRTC to the right is disabled, but the reverse is not true.
With both CRTC enabled but HW cursor active only on one, the reported corruption is seen. If HW cursor for both CRTC is forced to stay on, then no corruption occurs. When the CRTC dimensions were hardcoded to more appropriate values, cursor corruption was seen on both screens.
Oddly enough, there doesn't seem to be any side effects of having HW cursor enabled for both CRTCs even when it is only visible on one.
I tried the path submitted by Yang Zhao, but the problem is still present on my ATI X1400.
The new randr cursor code now makes the buggy behaviour happen on both CRTCs. AFAICS, this is expected.
Does anyone have a working setup with fglrx? It would be useful to get some register dumps to see how it's handling the cursor on dual screen.
The following register ranges are relevant: 0x6400-0x6424 and 0x6C00-0x6C24.
Bug 19215 is a duplicate, only this time it's on the radeon driver.
Pushed Yang's fix in master. Please retest.
For those still having this issue - we could try to use the so called Icon instead of the mouse pointer. It's use has some ugly constraints, that's why we never bothered to implement it.
When starting X the radeonhd driver reports option "HWCursor not used" so I don't know if I'm really testing Yang's fix.
But the cursor corruption is still there with the latest git.
Only on the external monitor (not the laptop panel) if that matters at all.
Can you file another log? The message sounds strange...
Created attachment 23289 [details]
Xord.0.log showing HWCursor complaint
Here's the Xorg logfile, not done verbosely (can do so if you want) but it shows the complaint:
(WW) RADEONHD(0): Option "HWCursor" is not used
Ok, that's different. You have
in your xorg.conf. radeonhd doesn't have this option, but using hardware cursor is on by default.
as an interesting side note i saw this corruption on windows xp x64 after i was playing "supreme commander forged alliance" and my graphics card restarted itself (i assume... the screens locked up and i heard the fan do the thing it does at boot) crashing the game (it showed a grey screen) and causing all the graphics to "slow down" ie scrolling in ff/ie was painfully slow
i have a x1950xt
Alex found something hardware related and I pushed a fix today. It supersedes Yang's fix, so there *could* be regressions.
Can you all please retest?
(In reply to comment #37)
> Alex found something hardware related and I pushed a fix today. It supersedes
> Yang's fix, so there *could* be regressions.
> Can you all please retest?
I've built built a new RPM on Fedora 10, using the
xorg-x11-drv-radeonhd-snapshot.sh script. It says that it was built from
"git commit 8bb8cd80f08d094c8b451e005e57c495dc6e4d42".
I'm still seeing the problem.
(In reply to comment #38)
> I've built built a new RPM on Fedora 10, using the
> xorg-x11-drv-radeonhd-snapshot.sh script. It says that it was built from
> "git commit 8bb8cd80f08d094c8b451e005e57c495dc6e4d42".
That's still 2 commits behind.
Still reproducible by moving the cursor to the other screen, but flickering goes away again by causing the cursor image to change (ie, move to edge of window to activate "resize" cursor)
The places where corruption occur also seems to be different depending on whether the view ports are docked horizontally or vertically.
(In reply to comment #39)
> That's still 2 commits behind.
I just re-ran the script, and it came back with the same commit again.
The script appears to be doing
git clone http://yangman.ca/git/xf86-video-radeonhd.git/ xf86-video-radeonhd.src
What command should I be using to get the updated source?
(In reply to comment #41)
> (In reply to comment #39)
> > That's still 2 commits behind.
> The script appears to be doing
> git clone http://yangman.ca/git/xf86-video-radeonhd.git/
Not sure why your script is pulling from my personal repo instead of the fd.o one.
git://anongit.freedesktop.org/xorg/driver/xf86-video-radeonhd is the URL you want.
(In reply to comment #42)
> git://anongit.freedesktop.org/xorg/driver/xf86-video-radeonhd is the URL you
OK, pulling from that repo definitely gives me different behavior.
The bad news is that I now have a "lingering" mouse pointer on my 1280x1024
display when I move the pointer over to the 1680x1050 display. The reverse
does not happen. (I.e. there is no pointer "lingering" on the 1680x1050
display when I move the pointer to the 1280x1024 display.) I can some-
times get the pointer on the 1280x1024 display to disapper if I move the
mouse from display to display slowly.
The good news is that the mouse pointer "flickering" problem is now mostly
(but not completely) gone. I also believe that I've only seen it on the
Looks like Alex agrees that the first workaround is as proper of a fix as we'll get.
I'm working on refactoring the RandR cursor code right now, so I'll incorporate the fix into that and have it finalized and sent to the list in the next couple of days.
It seems you're right.
Yang, I removed your code because it seemed to be superfluous. Seems I was wrong.
Today's commit 6f378a0d fixes this issue for me. Great work guys!
(In reply to comment #0)
> using randr, I get a flickering mousecursor in a few well-defined areas
> of one of my screens.
> There are 4 vertical "lines", about 20 pixes in width that go across the
> screen from top to bottom. When I place the mouse cursor on top of these
> "lines", I get a flickering mouse cursor.
> The 4 lines are at about 235-255 pixels from the left edge of the
> screen, at 490-510 pixels from the left edge, at at 940-960 pixels from
> the left edge and at 1005-1025 pixels from the left edge.
> On the other monitor, this does not happen.
I have the flickering behaviour as well. Thinkpad t60, radeon mobility X1400, laptop's panel (1400x1050) + external (VGA connected) Dell 20' display (1680x1050), gentoo 32 bit, xorg 1.5.3, radeonhd 1.2.5.
Now, I see that I should get a git tree in order to get the fix (maybe a quick 1.2.6 follow-up, wink wink), but the funny thing is that I switched to xf86-video-ati (6.12.2) and I'm still seeing the same flickering mouse cursor at well defined areas on the display. So it looks like it's driver independent (or the problematic code is shared). And yeah, I don't remember seeing the flickering when only using the laptop's panel, but the mouse flickers on both panel and external monitor when they're connected.
(In reply to comment #47)
> I have the flickering behaviour as well. Thinkpad t60, radeon mobility X1400,
> laptop's panel (1400x1050) + external (VGA connected) Dell 20' display
> (1680x1050), gentoo 32 bit, xorg 1.5.3, radeonhd 1.2.5.
> Now, I see that I should get a git tree in order to get the fix (maybe a quick
> 1.2.6 follow-up, wink wink), but the funny thing is that I switched to
> xf86-video-ati (6.12.2) and I'm still seeing the same flickering mouse cursor
> at well defined areas on the display. So it looks like it's driver independent
> (or the problematic code is shared). And yeah, I don't remember seeing the
> flickering when only using the laptop's panel, but the mouse flickers on both
> panel and external monitor when they're connected.
6.12.2 is too old. the radeon fixes just went in over the last week.
> 6.12.2 is too old. the radeon fixes just went in over the last week.
understood. Is there a release of xf86-video-ati/radeon that has the fix?
I might stick with that for a while, radeonhd still has some issues for me (from time to time I get a NMI - "uh oh..dazed and confused.." and after that either the mouse cursor is severely damaged -a big thumb size rectangle-, or the system freezes hard - don't know if it's related to the bug we're discussing here).
(In reply to comment #49)
> > 6.12.2 is too old. the radeon fixes just went in over the last week.
> understood. Is there a release of xf86-video-ati/radeon that has the fix?
No, however, most distros provide git snapshots or you can build from source.
All issues should be fixed now with git commit 4be5f7152f71.
Please reopen if issues remain.
Also, please post if it works for you. This has been open so long, that I want to make sure we didn't miss something.
(In reply to comment #51)
> Also, please post if it works for you. This has been open so long, that I want
> to make sure we didn't miss something.
tried latest git (commit 4be5f715) and it doesn't flicker anymore, so thumbs up from me.
Commit 4be5f715 solves it for me as well. All cursor corruption is gone (same card as I mentioned earlier in this bug report, but under NetBSD/amd64 5.99.8 this time).
Working for me with 4be5f7152f71c292f16b6e30c59c07b282ea4772
Seems like the issue is gone for good. Phew!
Thanks for validation!
These changes made my dual monitor system worse. Not sure why I'm the only one seeing that.
(In reply to comment #55)
> Seems like the issue is gone for good. Phew!
> Thanks for validation!
(In reply to comment #56)
> These changes made my dual monitor system worse. Not sure why I'm the only one
> seeing that.
Can you send some more detailed information to the mailling list or, preferably, drop by the IRC channel? (#radeonhd on freenode). If it's a really obscure and isolated issue, it'll be easier to debug over a lower latency channel.