ejs1920: what can i do to help debug the lockups i keep getting?
ejs1920: i tried drm from git which is supposed to have some lockup fixes, but it locked up after about 2 minutes of bzflag
ejs1920: mesa git seems to require and newer x server
ejs1920: s/and/a
ejs1920: at least rebooting with sysrq works
ghepeu: speaking of lockups, anyone here is using compiz and xserver/mesa/ati from git?
olegfink: me
ghepeu: Are you noticing random lockups?
olegfink: however, i have only ati from git
olegfink: nah
olegfink: after a long time a leave a macine it happens that the screen is frozen, but mouse cursor is till alive, so it's not quite a lockup
olegfink: but as the keyboard is chewed as well, the only way to fix it is to kill X remotely
ghepeu: I used xserver 1.4-1.4.0.90 and mesa 7.0.1-7.0.3_rc2 from september 2007 to march 19, and in 6 months I had 1 lockup
ghepeu: xf86-video-ati git
ghepeu: then I decided to try xserver/mesa from git and I had a number of lockups, two just yesterday
MrCooper: olegfink: that could be a GPU lockup - the mouse cursor is SIGIO driven
ghepeu: mouse keeps moving, X 100% cpu, sometime it can be killed, sometimes I must reboot
olegfink: aha, same here
ejs1920: my mouse usually freezes, but sysrq works
ejs1920: well bzflag doesn't show the mouse cursor
ghepeu: what's worrying for me is that the previous "iteration" of xserver/mesa worked really well, and random lockups now are a serious regression
ghepeu: not to mention that they're random, so I can't try to find what the problem is
MrCooper: everybody seeing lockups, try Option "CPusecTimeout" "100000"
MrCooper: ghepeu: did the 'previous iteration' have RENDER acceleration?
ghepeu: hw render?
MrCooper: (see http://bugs.freedesktop.org/show_bug.cgi?id=15203 about the option above)
ghepeu: I don't think render acceleration matters, I'm using a xf86-video-ati snapshot from before the r3xx-render branch
ghepeu: MrCooper, I patched my local copy yesterday, modified RADEON_DEFAULT_CP_TIMEOUT
MrCooper: does it still happen with current Git?
ghepeu: xf86-video-ati git? don't know, I'm not using it because exa is slow. I could test it, however
ejs1920: i'm using today's xf86-video-ati git, exa is fairly fast, render accel is nice
ejs1920: i restart x with that option
MrCooper: ghepeu: use XAA or Option "EXANoComposite"? *shrug*
ghepeu: eh, with XAA compiz is guaranteed to lockup in 10 minutes
ghepeu: I'm upgrading to latest git, I care for stability more than for speed
ghepeu: let's see if it helps
ejs1920: right, i'm trying audacious + extreme tuxracer, which was locking up yesterday
ejs1920: it locked up :(
ejs1920: can anyone else try that?
ejs1920: have an mp3 playing in audacious, play extreme tuxracer 0.4 in a window, try the "Doing" course, and watch your screen freeze
ejs1920: the mp3 keeps playing
ejs1920: i reboot with sysrq
MrCooper: sounds like a GPU lockup
ejs1920: could it be due to the exa render accel? i'm sure i would have noticed it earlier otherwise
ejs1920: i'll retry with xaa
MrCooper: ejs1920: you can also disable just the RENDER acceleration with Option "RenderAccel" "off" or Option "EXANoComposite"
ejs1920: there's also a minor problem with exa - etracer runs at 2fps for about 5 seconds of the Doing course, after that it runs normally at 15-50 fps
ejs1920: well i didn't lockup without the mp3s playing
ejs1920: now trying with mp3s playing
ejs1920: well i'm still here, so no lockup
ejs1920: let's try 20 minutes of bzflag
ejs1920: no lockup!
m3thos: that justifies a propper bug report!
ghepeu: ejs1920, are you using compiz?
ejs1920: no
ghepeu: did you check if the lockup occurs with exa but Option "EXANoComposite" in xorg.conf?
ejs1920: hang on, i'll try with "RenderAccel" "off"
ejs1920: whoops, didn't change it!
ejs1920: no lockup with "RenderAccel" "false"
ejs1920: but i'm still switching back to xaa
xen_: Hi there can someone help me with this problem all details explained here: http://ubuntuforums.org/showpost.php?p=4650305&postcount=1
TobiasTheCommie: xen_: wrong channel
TobiasTheCommie: join #ati
TobiasTheCommie: xen_: this is for the opensource driver
TobiasTheCommie: fglrx is the closed source driver by ati/amd
xen_: oh
xen_: sorry
sk1p_: hi there, i got a radeon 7000 with two vga outputs; it used to work just fine with mergedFB, but xrandr doesn't seem to detect the second head.
sk1p_: log is at http://alex.homelinux.net/~gu/Xorg.0.log
sk1p_: config @ http://alex.homelinux.net/~gu/xorg.conf
sk1p_: i noticed that the xrandr utility shows VGA-0 and DVI-0, even though the card doesn't have any DVI connector
agd5f: sk1p_: ddc doesn't seem to return an edid. you might try swapping the plugs
agd5f: sk1p_: you can also force the outputs on
sk1p: agd5f: ok, i'll try
agd5f: sk1p: I can add a quirk to have the DVI port show up as VGA if you send me your subsystem ids
agd5f: nevermind, they are in the log
sk1p: agd5f: ok, swapping the connectors didn't help
agd5f: sk1p: does it really have 2 VGA ports or are you using a DVI->VGA adapter
agd5f: sk1p: new log?
sk1p: yea, 2 vga ports (+ svideo out)
agd5f: cool. I'll fix that up
sk1p: http://alex.homelinux.net/~gu/Xorg.0.log_new
agd5f: to force on the second head: xrandr --output DVI-0 --set dvi_monitor_type analog; xrandr --addmode DVI-0 1280x1024; xrandr --output DVI-0 --mode 1280x1024
mikkoc: agd5f: hi, i've just closed this bug, as it seems to be only my distro's problem: http://bugs.freedesktop.org/show_bug.cgi?id=15249
mikkoc: any idea how to find out what's wrong?
agd5f: mikkoc: weird
sk1p: nice :) (though the second monitor does only 1024x768) it outputs something cloned, the mouse cursor seems to be broken
mikkoc: is it?
agd5f: sk1p: replace 1280x1024 with 1024x768 :)
sk1p: agd5f: yeah, did that ;)
agd5f: sk1p: for dualhead: xrandr --output DVI-0 --right-of VGA-0
sk1p: ok, but i guess i need to restart x for that because of fixed framebuffer size?
sk1p: can i force the monitor type in the xorg config?
agd5f: sk1p: yeah, you'll need to add a virtual line if the defaul isn't big enough
spstarr_work: agd5f: hmm
spstarr_work: agd5f: can we bump the default GART size up ?
spstarr_work: agd5f: also, with a 64MB card how come Scorched3D is awfully chuggy :/
spstarr_work: r350
sk1p: agd5f: ok, works :)
agd5f: sk1p: http://www.botchco.com/alex/xorg/sk1p.diff this patch should fix the connector to be reported as VGA
MrCooper: spstarr_work: grep textures /var/log/Xorg.0.log
spstarr_work: lemme check the bug I logged it has that info in the log
spstarr_work: (II) RADEON(0): Will use 35840 kb for textures at offset 0x1cf6000
spstarr_work: (II) RADEON(0): Using 110 MB for GART textures
spstarr_work: if given a GART of 128MB
MrCooper: 'GART textures' is misleading here, the Mesa r300 driver doesn't use GART memory for textures
spstarr_work: so then its only using 36MiB for textures ?
sk1p: agd5f: can't get the driver to compile.. ./configure: line 20655: syntax error near unexpected token `XINERAMA,' ./configure: line 20655: `XORG_DRIVER_CHECK_EXT(XINERAMA, xineramaproto)'
spstarr_work: sk1p: your missing Xinerama-devel or the macros for X server
agd5f: sk1p: you need to install the xserver-devel pacakges
agd5f: for you distro
sk1p: ah, i see :)
agd5f: or xorg-macros
spstarr_work: agd5f: by your estimation what remains left for the r300 series for full acceleration?
spstarr_work: i donno if AMD gave the specs for the r3xx yet or is planning to soon or NDA them
agd5f: spstarr_work: mostly 3D driver improvements
ejs1920: i can get a hard lockup when watching tv in mplayer, generally when seamonkey is rendering a webpage in the background. radeon or v4l problem?
ejs1920: it's analog tv, using yuy2 overlay
ejs1920: latest hg v4l drivers and latest git radeon driver
ejs1920: not using render accel
spstarr_work: agd5f: ok, then I'll be testing those as they filter into rawhide
agd5f: ejs1920: what chip?
ejs1920: 9550 / rv350
ejs1920: tv tuner is using em28xx driver, it's a usb hvr 900
ejs1920: the webpage tends to be phoronix forums
ejs1920: it could just be the fact that my computer is too old
agd5f: ejs1920: using textured video or overlay?
ejs1920: overlay
sk1p: agd5f: ok, the driver compiled, the workaround seems to be enabled but doesn't seem to work: http://alex.homelinux.net/~gu/Xorg.0.log_new_2
agd5f: sk1p: there's an error in the patch
agd5f: info->BiosConnector[index].ConnectorType = CONNECTOR_VGA;
agd5f: not info->BiosConnector[index].ConnectorType == CONNECTOR_VGA;
MrCooper: agd5f: *cough*-Werror*cough* ;)
agd5f: MrCooper: :)
sk1p: agd5f: ok, works :) though i still need to add the mode manually
agd5f: sk1p: yeah, it's not able to detect the monitor. you can force it on in your xorg log however
agd5f: sk1p: http://www.intellinuxgraphics.org/dualhead.html
ejs1920: i guess i'll have to try the vesa driver for a week and report back later
agd5f: ejs1920: weird. overlay should be fine.
agd5f: sk1p: add a monitor section, link it to VGA-1 and add monitor sync ranges, perferredmode, and Option "Enable" "true"
ejs1920: the weird thing is it locks up when a webpage has half loaded
ejs1920: still transferring data
ejs1920: half displayed
ejs1920: could it be related to dragging a window around freezing the xvideo output?
ejs1920: lets tempt fate with -vo x11 ...
spstarr_work: fate is tempted
MrCooper: ../../src/legacy_crtc.c: In function 'legacy_crtc_mode_set':
MrCooper: ../../src/legacy_crtc.c:1728: error: implicit declaration of function 'RADEONInitMemMapRegisters'
MrCooper: ../../src/legacy_crtc.c:1783: error: implicit declaration of function 'RADEONRestoreMemMapRegisters'
ejs1920: well it locked up, but the vesa driver displayed garbage
ejs1920: lets try noaccel and nodri
ejs1920: well now scrolling in nedit is painfull and i still have xvideo
ejs1920: and my cpu isn't fast enough for a deinterlacing filter :(
ejs1920: well now scrolling in seamonkey freezes the -vo x11 output
ejs1920: trying -vo x11 under NoAccel and NoDRI...
ejs1920: i mean -vo xv!
ejs1920: well NoAccel and NoDRI obviously didn't help
sk1p: agd5f: can't get it to work; log says (EE) RADEON(0): Output VGA-0 enabled but has no modes
sk1p: agd5f: http://alex.homelinux.net/~gu/xorg.conf
ejs1920_: DMAForXv false didn't help, i'll try one last crash with agpmode 2, then give up for the day
ejs1920: no luck with agpmode 2, guess i'll just have to be careful when watching tv
agd5f: sk1p: you'll need to add a modeline: Modeline "1024x768" 63.50 1024 1072 1176 1328 768 771 775 798 -hsync +vsync
agd5f: add it to your "Left" monitor
agd5f: ejs1920: try Option "BusType" "PCI"
agd5f: SiS AGP chipsets tend to be broken
ejs1920: i'll try it in 20 minutes
glisse: agd5f: i would just say: AGP tend to be broken ;)
agd5f: glisse: true that
mcgreg: is there a channel for mesa?
glisse: mcgreg: well dri-devel is a good place
glisse: but no channel dedicated to mesa
mcgreg: ok, thx
mcgreg: just built xserver git .. hope it'll work ;)
ejs1920: agd5f: BusType PCI didn't help
ejs1920: but i've thought of a few things to try
sk1p: agd5f: the modeline addition doesn't seem to change the situation.. http://alex.homelinux.net/~gu/Xorg.0.log_new_3 http://alex.homelinux.net/~gu/xorg.conf
agd5f: sk1p: try giving it another name like Modeline "1024x768_60" 63.50 1024 1072 1176 1328 768 771 775 798 -hsync +vsync
agd5f: Option "PreferredMode" "1024x768_60"
sk1p: still the same
ejs1920: since "sdlmame -yuvmode yuy2" doesn't seem to cause any lockups, that suggests the problem is in the em28xx or rt2x00 drivers and not radeon
agd5f: could be
PSYCHO___: i have one r500 now in my hands and it says it is unsuppoted
PSYCHO___: will paste relevant info when i boot up completely it
PSYCHO___: unknown chip id 0x791e, can't guess.
PSYCHO___: libGL warning: 3D driver returned no fbconfigs.
PSYCHO___: libGL error: InitDriver failed
PSYCHO___: 01:05.0 VGA compatible controller: ATI Technologies Inc Radeon X1200 Series
PSYCHO___: 01:05.0 0300: 1002:791e
airlied: PSYCHO___: thats an rs690
airlied: not an r500
airlied: I assume its a motherboard.
agd5f: PSYCHO___: need a newer mesa
PSYCHO___: ah
PSYCHO___: yes
PSYCHO___: i thought its rs500
PSYCHO___: damn
PSYCHO___: agd5f: forget to keyword it
PSYCHO___: :]
PSYCHO___: ok start with mesa
airlied: PSYCHO___: rs690 is nearly full 3D from r400, except vertex shaders
airlied: I've made two attempts to make compiz work and failed
PSYCHO___: i dont want compiz
PSYCHO___: i want "pan brambora" in english something kike "Mr. Potato"
PSYCHO___: s/kike/like/
PSYCHO___: its kde game
PSYCHO___: and it shows no image now
revellion1: anyone in here runned a ATI Radeon 9200SE 128MB PCI on X with EXA?
Revellion: anyone in here runned a ATI Radeon 9200SE 128MB PCI on X with EXA?
Revellion: oops
peterz: agd5f: any eta on r600 accel?
agd5f: peterz: I'm working on the r6xx documentation now. shouldn't be too long
peterz: agd5f: awesome !
PSYCHO___: ok same unknown chip text with mesa 7.0.2
PSYCHO___: i need newer?
PSYCHO___: s/newer/more recent/
agd5f: problem is it's WAY complex. all the cells run asynchronously and have their own ring buffers, but the GART has been replaced with a virtual memory system
agd5f: PSYCHO___: either add your id to 7.0.2 or upgrade to git
Magnade: PSYCHO___: if your looking at lspci output it would say unknown chip till you update the pciid database
Revellion: anyone knows if the rv280's with the current stable ATI driver is supposed to call memcpy so often in EXA mode?
PSYCHO___: Magnade: no lspci but glxinfo
Revellion: i smell some horrible software fallbacks
agd5f: Revellion: might be vram pressure
Revellion: agd5f: hmm
Revellion: agd5f: but isn't memcpy usually seen when copying from GPU-to-CPU ?
Revellion: unless i'm doing it against the wind here..
agd5f: Revellion: lack or free vram requires migration
Revellion: ah
agd5f: s/or/of/
Revellion: agd5f: but 128MB shouldn't supposedly run out too soon?, there's not a whole lot of windows open except a plain desktop and a terminal using vte
agd5f: Revellion: lots of offscreen pixmaps. plu if you have the dri enabled, much if that space is reserved for textures and back/depth buffers
agd5f: *plus
Revellion: ah
Revellion: ah. the allocations are still static?
Revellion: seeing as i'm not using 3D at all. setting FBTexPercent to 0 "could" possibly improve the situation, or am i dreaming?
agd5f: Revellion: yup
Revellion: did a test with that. seemingly it still likes to migrate/memcpy a lot
Magnade: Revellion: did it reduce the amount that it does it any tho?
Revellion: Magnade: hmm, i'd like to think i got atleast a placebo effect of better performance with the FBTexPercent at 0
Revellion: but in total CPU utilization it was about the same
Revellion: 50%
Magnade: dual core system?
Revellion: i wish :P
Revellion: no a old P3 1Ghz i'm experimenting with
Revellion: *an
Revellion: is there any changes related to EXA that might affect rv280's performance using it in post-6.8.0 (aka git master) ?
Magnade: was asking as i was wondering if its 50% of 1 cpu or 100% of a 2 cpu system
Revellion: ah nah 50% of 1
Revellion: i understand EXA is still in a bit of a flux, but i was hoping it had reached such a level that EXA would be better than XAA by now even on the older chips. or was those big improvements mostly centered around the git master?
Magnade: rv280... oh hey thats the same thing in my mac mini >_<
Magnade: i dont care for the speed i have ether but half my issue is having a nfs home
Revellion: Magnade: mhm, but it would be nice if moving windows didn't draw visible lag XD
Revellion: such as it jumping towards its destination
Revellion: when dragged :)
Magnade: well its not smooth moving windows around but it could be alot worse for me
Magnade: i have more issues with scrolling speed
arekm: too (scrolling)
leio: Revellion: I've got a ATI Technologies Inc RV280 [Radeon 9200 PRO]
Revellion: leio: roughly the same card as i have
leio: I knew I bought a 9250 though
leio: maybe that's the marketing name
Revellion: it is :S
Revellion: VGA compatible controller: ATI Technologies Inc RV280 [Radeon 9200 SE] (rev 01)
Revellion: my exact card
leio: so, EXA with default other settings is horrible
leio: worse than XAA. Using "MigrationHeuristic" "greedy" and/or "EXAOptimizeMigration" "true" make it somewhat match, supposedly greedy due to there being not much acceleration akin to XAA
leio: I have yet to reboot X for some EXA perf fix tries that was one of my current bottlenecks
airlied: Revellion: have you AccelDFS support ?
Revellion: leio: those options are only available in git builds?
Revellion: airlied: ill look at that hang on
airlied: Revellion: as without that you will get slow memcpys
Revellion: mhm
leio: EXAOptimizeMigration is new
leio: ah right, AccelDFS too. I believe git has it on by default
leio: (at least the option seems to be converted to a No version)
Revellion: airlied: indeed, it's enabled in config but not enabled in the driver itself
Revellion: "AccelDFS" "true"
Magnade: the exa migration bit i found works good with smart for everything but scrolling
Revellion: but (==) RADEON(0): Not using accelerated EXA DownloadFromScreen hook, for reason's i can't recognize
Revellion: no real answer is offered in the logs as to why it decided to not support/enable it
airlied: Revellion: its AGP its not on by default
Revellion: PCI in this case :\
Revellion: and i forced it on in the config
airlied: wierd I thought PCI turned it on by default.
Revellion: is there any special requirements for DFS to work on PCI?
Revellion: any limitations?
airlied: nothing I know off..
Revellion: airlied: well the docs states it's "off by default" on AGP
Revellion: wait... i just found something, might be a case of pebkac
Revellion: or a buggy module "nouveau" i had when testing the nvidia EXA support on the MX4000 card. seems it caused the radeon DRM module to fail to load, ill restart X and compare
Revellion: ok. that was the problem. but it instead openned up a new can of worms in the shape of graphical glitches
Revellion: borders being drawn as dashed lines would be the easiest way to describe it
Revellion: diffreent behaviour observed while using a compositing manager though
Magnade: different glitches?
leio: Do you have EXAOptimizeMigration on btw?
leio: oh, I suppose you are using a released version (though 1.4.99.901 has it too)
Revellion: leio: i'd have to get the git master i suppose for that?, since that option is not documented in the manpage (undocumented option?)
leio: man exa
Revellion: Magnade: no it looks fine using a compositor strangely
leio: I was asking because I've been told that EXAOptimizeMigration is not enabled by default only because some people see corruptions with it when starting a compositing manager, that isn't tracked down yet
Revellion: options for EXA is added in what section? "Device" ?
Revellion: leio: well my situation is reversed i suppose XD
leio: yeah, same section you have EXA chosen and driver told to be radeon
Revellion: leio: ah
Revellion: in this case of a release version i don't have the EXAOptimizeMigration option. for testing which Heuristic you reckon to try?, or just go through em all and test em?
Revellion: ah you meant the X-server version there earlier?, 1.4.0.90 here only
leio: "greedy" I understand it is just a way to stop most migration overhead by not accelerating much at all, but I am a newbie here, so I'm sure I'll be corrected :)
Revellion: well when it comes to the Xorg part of my system i'm not fully as educated in the inner works as i'd like to
Revellion: ill test greedy for now
Revellion: lets see
ghepeu: hi. Is anyone here using compiz and xf86-video-ati git?
airlied: agd5f: I wonder does R100/R200 need any cacheflush postEXA
Revellion: leio: greedy reduced performance noticably
agd5f: airlied: good question
Revellion: leio: hmm none of the migrationheuristics proved to improve the situation noticably. does EXAOptimizeMigration just detect and select the best migration scheme to use or does it do anything else?
Magnade: does r300 do cacheflushes?
Revellion: wow.. just tested git xf86-video-ati. scaling windows noticably a lot faster than release 6.8.0
agd5f: airlied: I think switching between TCL for 3D and non-TCL for Xv/EXA can result in issues. We probably need to flush the VAP better or something
airlied: agd5f: hmm...
agd5f: airlied: see bug 15356
rx__: oh
rx__: i see dynamicclocks got added in
rx__: tested on mobility yet?
mattmatteh: agd5f, ok, i was just messing with xattr and that does seem to improve the dvd
mattmatteh: agd5f, however finding the right values seems to be difficult, do you know of a way to calibrate that ?
eboettcher: agd5f: hmm.. I'm comming to you now from a box that has an RS400 and I'm looking into proposing some work for this and other r300 stuff for GSOC.
eboettcher: Are there any RS400 specific docs out there you suggest I look at?
eboettcher: I'm looking at some of the stuff on that R300 todo list now
agd5f: eboettcher: the R300 doc should cover it for 3D
spstarr_p4: hmmm
spstarr_p4: badness in rawhide ati
spstarr_p4: [drm:radeon_cp_start] *ERROR* radeon_cp_start called without lock held, held 0 owner f49bd840 f50df630
spstarr_p4: [drm:radeon_cp_idle] *ERROR* radeon_cp_idle called without lock held, held 0 owner f49bd840 f50df630
spstarr_p4: [drm:radeon_cp_reset] *ERROR* radeon_cp_reset called without lock held, held 0 owner f49bd840 f50df630
spstarr_p4: [drm:radeon_cp_start] *ERROR* radeon_cp_start called without lock held, held 0 owner f49bd840 f50df630
spstarr_p4: ....
spstarr_p4: its spinning with that in syslog over and over
spstarr_p4: this is with 3D (composite) being used
spstarr_p4: or in kwin when you switch to different modes of rendering
spstarr_p4: bilinear, etc..
spstarr_p4: known badness?
dli_: I am running git with Dynamic Clocks For Radeon R500+, gpu temperature is indeed higher than before
dli_: it's 78C now, compared with 71C before
spstarr: you can make me breakfast ;-)