hifi: soreau: 2.6.31 with ubuntu magic
hifi: it wasn't even a jump from rc to non-rc
soreau: hifi: Which card?
hifi: RV570, Radeon X1950 Pro
soreau: hifi: and how does kms fail?
hifi: so, console broke down but in X it works fine
hifi: when radeon module is loaded with modeset my screen losts signal
airlied: hifi: got fbcon?
hifi: and when I start X blindly I get picture back
hifi: but if I change to VT1 or similiar I see tha last frame from X
hifi: airlied: dunno, ubuntu kernel has been working fine with KMS until today
airlied: sounds like fbcon isn't loaded
hifi: hm, I think my desktop is online at home so I can check
hifi: no fbcon with lsmod, I think it's builded as module
airlied: Console: switching to colour frame buffer device
hifi: so thats probably it
airlied: look in dmesg for that
airlied: or something like it
hifi: [ 0.000879] Console: colour VGA+ 80x25
hifi: only line with Console
airlied: yup no fbcon then
hifi: failed kernel build probably
hifi: goes to poke ubuntu kernel team
airlied: you may need to modprove it
hifi: true, I wonder why it didn't load as it used to
hifi: [ 4661.125678] Console: switching to colour frame buffer device 160x64
hifi: and we have success
hifi: this raises a few questions though :)
hifi: 1. could KMS not kill the console without fbcon, 2. could radeon autoload fbcon
airlied: hifi: no and no
hifi: so it should be completely the distribution's job to load the correct modules?
airlied: yes, most distros build fbcon into the kernel
airlied: so vesafb works
hifi: goes to poke some more
mzz: there it went again, but this time I had netconsole going: "[drm:radeon_fence_wait] *ERROR* fence(e0c711e0:0x0006F34D) 504ms timeout going to reset GPU"
mzz: "[drm:r300_ga_reset] *ERROR* VAP & CP still busy (RBBM_STATUS=0x8400C100)[ 61.346377] netconsole: local port 6665"
mzz: and then it's pretty dead. Does that ring a bell with anyone?
mzz: err, the final half of that was a mispaste :(
mzz: it's dead after the RBBM_STATUS
airlied: mzz: looks like a gpu crash
airlied: glisse has some gpu reset code he should be publishing soon hopefully
airlied: what gpu is it?
airlied: and agp or pcie?
mzz: this is very intermittent, and it might (but I have zero evidence for this) happen more often if I have many images in video ram. One suspicious thing: it had started drawing one or more characters corrupted in firefox again.
airlied: try decreasing agp mode maybe
airlied: it sounds like what suokko might be looking for also
mzz: radeon 9600xt aka rv360
mzz: my xorg.conf is pretty sparse, although I just noticed I do have AccelDFS and EnablePageFlip on in there
airlied: well they won't do much in kms mode
airlied: what agp mode is it?
airlied: might be worth bootinh with radeon.agpmode=1
airlied: on the kernel command line
mzz: as far as agp goes: agpgart-nvidia (nforce 2) and "putting AGP V3 device into 8x mode"
mzz: putting that in grub.conf
mzz: if anyone wants full dmesg or Xorg.0.log just ask, but afaict it's pretty happy up until right before it dies, apart from the aforementioned corruption.
mzz: which I think I've only seen happen a bit before it dies, but I have seen it die without me noticing any corruption.
mzz: so far the only steps to reproduce I have are "use firefox", which doesn't really help.
airlied: actually radeon.agpmode=4
airlied: sinec its seems its a v3 card
mzz: oh, and 128M of VRAM, which it seems possible firefox is all eating
mzz: and just to state the obvious: the card seems to lock, but after failing to reset it the system's dead as far as both magic sysrq and ssh goes.
Nightwulf|work: hi all
mzz: so apart from the above two netconsole lines I have no idea how to debug.
mzz: even if I had a logic probe I wouldn't know where to stick it :)
airlied: mzz: I suspect the upcoming gpu reset code might help, or at least reet without locking up the whole box
mzz: airlied: a link to the relevant git branch that lands in around when it does would be appreciated then :)
mzz: also, if others don't see this happening at all and there's debugging code that provides further clues I'll happily run that
mzz: this is just one class of problem I really don't know how to debug (I'm ok with X locking as long as I can still ssh in)
soreau: attaches gdb to mzz
mzz: I hope you don't expect me to not move a muscle until you hit "c" in the debugger
soreau: gets frustrated and runs killall -9 mzz
mzz: decides he'd better pretend to play dead convincingly before soreau gets *really* frustrated and reboots the system he's running on
Lenin_Cat: how can I install the expermental 3D? Also radeon is practicly the same as radeonhd right?
soreau: Lenin_Cat: See the topic
Lenin_Cat: soreau, thats not for expreremental 3d
soreau: Lenin_Cat: What isn't?
Lenin_Cat: the link
soreau: coulda fooled me
soreau: Lenin_Cat: What are you trying to do exactly?
Lenin_Cat: have mesa
soreau: Then you didn't read the information in that link apparetnly
Lenin_Cat: soreau, I have radeon installed via apt-get
Lenin_Cat: so do I still have to build mesa?
soreau: I don't know what you're trying to do exactly
Lenin_Cat: soreau, I have already installed radeon via apt-get
Lenin_Cat: but I want the experement 3d support mesa, but sense I didnt build radeon do I have to build mesa?
airlied: you need to build mesa and probably a kernel
airlied: and libdrm
airlied: unless there are packages already for debian/ubuntu
boghog: well for ubuntu there is https://launchpad.net/~xorg-edgers/+archive/ppa
boghog: i use that on my intel laptop to get all the latest stuff
soreau: boghog: Does it have a kernel in there somewhere?
boghog: no I got that from somewhere else
boghog: though I don't remember where :p
EruditeHermit: soreau, yes it does
soreau: EruditeHermit: Can you confirm compiz water effect causes black screen there on your rv350?
EruditeHermit: soreau, kms or non kms?
soreau: Either way
EruditeHermit: is it going to crash me?
soreau: EruditeHermit: I did a git bisect on mesa today and found the problem commit. Just want to confirm I'm not the only one it affects. No, it should cause a black screen until the water has dissipated
EruditeHermit: hang on
soreau: EruditeHermit: Apparently nha cleaned up r300 fragment program code and caused this bug, but I kinda wanted to make sure it happens else where too
EruditeHermit: soreau, yes it does
soreau: EruditeHermit: Cool, thanks for checking
soreau: I sent him an email and memoserv, hopefully someone (nha) will fix it
EruditeHermit: soreau, water effect always causes issues
soreau: I am thinking about looking into the commit diff
soreau: EruditeHermit: Like what?
soreau: It's always worked fine here
EruditeHermit: it used to be really slow
EruditeHermit: really really slow and cause 100% cpu usage because it hit a software fallback
soreau: It's always worked ok here though I never look at cpu usage frivolous info ;)
EruditeHermit: well it slowed down all the rest of my computer
EruditeHermit: my computer isn't very powerful
soreau: mine is pretty quick
soreau: the ati box anyway
EruditeHermit: its a laptop 2GHz
EruditeHermit: why the heck do you have rv350 on a powerful machine
soreau: I have only p4 2.8ghz
EruditeHermit: pentuim M
soreau: EruditeHermit: Because that's what I have and can't afford more?
EruditeHermit: oh p4 is pretty old
EruditeHermit: i thought you had some new fangled CPU with that old card
soreau: But it's fast enough for what I do
soreau: EruditeHermit: Oh no, if I had that kind of money I'd have I better machine by now
soreau: EruditeHermit: Anyway, thanks for testing that
EruditeHermit: no probs
EruditeHermit: old hardware is perfectly good
soreau: I know what commit causes it ;)
EruditeHermit: apart from when it fails =p
EruditeHermit: which mine is doing right now
soreau: EruditeHermit: I have two other machines besides this one which is the fastest of the three ;)
soreau: one is the 'server' and the other irc box
EruditeHermit: my HDD and battery are almost dead
soreau: Oh yea, I have a really old stinkpad where the battery is dead and the machine in whole is very slow. I think it's a P2
soreau: but it still works ;)
soreau: Where can I go to get a list of mesa commits and see the diffs?
airlied: new drm-next tree pushed, please run it lots so I can send it to Linus on Monday
airlied: r600 suspend/resume under kms should work now
airlied: maybe some r500 suspend/resume kms fixes also
EruditeHermit: airlied, any chance you'll have time to look at my rv350 suspend issue?
EruditeHermit: soreau, btw, does your rv350 suspend/resume properly on kms?
mwu: pulls drm-next
soreau: EruditeHermit: I have no idea. I do not use resume/suspend
airlied: EruditeHermit: next week maybe.
airlied: oh its still mirroring should be there in 5 mins
airlied: EruditeHermit: we narrowed it down to just the LVDS didn't we?
EruditeHermit: well yes
EruditeHermit: I hit a different bug when the external monitor is plugged in
EruditeHermit: airlied, https://bugs.freedesktop.org/attachment.cgi?id=29408
EruditeHermit: this is what happens with an external monitor plugged in
EruditeHermit: but according to yangman, this is a problem that others have had
EruditeHermit: but LVDS doesn't even get that far
airlied: nice effect.
yangman: pretty sure it's not just LVDS
yangman: I haven't tested again in a while
yangman: if the CRTC is cloned, you get it on both output
EruditeHermit: yangman, right, LVDS is a different bug
EruditeHermit: yangman, so on my setup, I experience a black screen on LVDS at all times after resume
EruditeHermit: I hit a different bug when the external LCD is plugged in
EruditeHermit: the external LCD bug is shown in the picture
EruditeHermit: but they are 2 different problems
EruditeHermit: maybe they are related, but they may not be
yangman: not for me
yangman: same corruption all around :p
airlied: yangman: what gpu?
EruditeHermit: for me, LVDS doesn't even turn on
yangman: all the reports so far have been on Thinkpads, so I'm supsecting a bios quirk
airlied: yangman: the fixes I merged might help on r500
airlied: I also merged your register save/restore bits
airlied: both by thinkpads work :(
airlied: though I suspect the t60p is posting the bios without asking
EruditeHermit: mine is not a thinkpad
mzz: EruditeHermit: my desktop rv350 suspends and resumes quite happily
yangman: ok. I'll give it a go this weekend
mzz: EruditeHermit: but it's been doing so for quite some time now
airlied: I haven't tried my rv370 dell ina while but its got issues with staying powered on long enough to test suspend
mzz: (err, I actually mean the "to disk" kind of suspend here)
EruditeHermit: staying powered on long enough?
EruditeHermit: does it overheat?
mzz: so it suspending properly isn't that much of a trick
airlied: EruditeHermit: I think it either has a dodgy power button or dodogy power supply
EruditeHermit: kms is pretty usable apart from suspend/resume for me
EruditeHermit: its a little slower
EruditeHermit: but usable
airlied: I tihnk my rs480 desktop fails to resume, so that is taks for next week
mwu: it's taking forever to mirror.. just like I remember
mzz: I'm probably missing something, but this seems like something that should be easier with kms than without it
EruditeHermit: I thought so too
EruditeHermit: but the non kms path was well tested for years
mzz: yeah, there's that.
airlied: okay my desktop rv380 still resumes
airlied: I should update my r350 desktop next week also
EruditeHermit: where does a developer start when presented with such bugs?
EruditeHermit: just out of curiosity
airlied: EruditeHermit: try and reproduce locally first
airlied: EruditeHermit: failing that lots of radeontool and avivotool dumps
airlied: also dumping sections of VRAM to see if its messed up
airlied: in that case of the crazy lines, dumping the CRTC offset regs
airlied: and making sure they are similair to what went in before suspend
mzz: and I'm assuming the "failing that" steps also happen if you *can* reproduce locally?
airlied: pretty much
airlied: for an rv350 the amount of crtc related regs isn't huge
airlied: and it looks like the mode is set correctly
airlied: so its probably something in the scanout or memory controller setup
EruditeHermit: i'm impressed
EruditeHermit: I just think bugs that you can't reproduce locally are really hard to fix
mzz: I think that's generally true, and that kind of bug is just more common here (because to reproduce locally you might need to have very similar hardware, not just software)
airlied: EruditeHermit: the problem with them generally is that you have so many bugs
airlied: its easier to fix 5 locally than one rmeotely.
EruditeHermit: and also impressive that you remember that m10/11 has LVDS bugs
airlied: mzz: hence why I have lots of hw
mzz: yeah, I gathered that
airlied: I've bought motherboards so I could avoid remote debugging
EruditeHermit: me personally
airlied: home &
mzz: I'm mainly still completely out of my depth when either the kernel locks up or the actual hardware is really involved
mzz: I need to be able to run gdb or strace or the like, and that doesn't really work on a gpu or with a kernel that's locked
mwu: you usually need another computer if you want to run gdb
mwu: using something like firewire
EruditeHermit: ok bbl
mzz: yeah, but the lockup bug I'm currently seeing is at least thorough enough to break ssh over ethernet (and alt+sysrq)
mzz: does something like serial console often still work while ethernet does not?
mwu: any kernel backtraces?
mwu: yeah serial console is 2nd best
mzz: no backtrace, but once someone told me to use netconsole I got two lines out
mzz: they're mentioned up a bit
mwu: if you have a null modem cable you can usually get the logs you need
mwu: assuming you still have systems with serial ports
soreau: quarantines mzz
mzz: well, I'm happy netconsole caught some stuff, which means that if it turns out to be necessary I can probably get more out by sticking extra stacktraces and other prints into the kernel or its modules
soreau: uninstalls mzz
soreau: deletes mzz
mwu: I usually give up on the sophisticated kernel debugging stuff and stuff in a bunch of checks and printks
mwu: and then think/guess really hard
mzz: that sounds like what I'd do
mzz: except for me not knowing what checks and printks to stuff in, so fortunately not getting to the thinking stage at all :)
mzz: actually fixed a silly probing/initialization bug in an ancient driver on an ancient laptop pretty much like that
mwu: I usually read the code and make it through somehow
mwu: though not always
mzz: but that was just X crashing or displaying garbage, while still being killable through c-a-backspace. Easy!
mwu: the gfx stack is a lot more complicated looking than network
mwu: heh yeah
mwu: I remember hacking a firefox extension after doing a bunch of kernel work, and being amazed that mistakes didn't bring the system down
mwu: userspace people get it so easy
mzz: I've done a little bit of compiz-related hacking too, and already found it useful to have a second networked system there
mzz: compiz is pretty hard to kill if you send it into an infinite loop at an inconvenient time
mzz: I think I made a habit out of starting it together with a "killall compiz" kicking in after 10 seconds (unless aborted)
mwu: annoys me sometimes how easy it is to dos your own system
mzz: I've made it a little harder, through usually having available virtual memory pretty low
mwu: I turned off swap on this one
mzz: most default linux systems are willing to hand out way more virtual memory than really makes sense.
mzz: so yeah, either use very little swap or switch /proc/sys/vm/overcommit_memory
mwu: turning off overcommit is probably smarter than turning off swap, but I bought 8gb of ram so.. I'm set
mzz: well, I have a bunch of swap because I like to suspend to disk.
mwu: yeah I realized my mistake after
mzz: also, to do a few things I need to turn overcommit back on.
mzz: mainly java.
mzz: java apparently likes to map way more memory than it actually uses. I haven't tried to figure out why.
mzz: (it can't be using it, it runs without touching any swap if I turn overcommit back on)
mjr: yeah, it does like its mappings
mjr: you can control the memory usage with -Xms (for initial heap size)
mjr: -Xmx for maximum
taiu2: new drm-next does't lie to work for me
taiu2: starting X hangs at R600DownloadFromScreenCS bo_map bo_wait drmCommandWriteRead
glisse: airlied, agd5f: it's ok for me to push the query accel pathc to the ddx ?
glisse: what will happen is that without lastest drm-next people will have shadowfb with kms
airlied: glisse: yeah should be fine or if you get EINVAL just do kms accel
airlied: if it return EINVAL instead of 0 or 1
airlied: taiu2: damn not sure what cuold have broken that
airlied: care to bisect the drm-next tree its not that many commits
taiu2: i'll try
airlied: taiu2: got X log as well?
airlied: oh what gpu also
airlied: needs a nick->Gpu converter
taiu2: rv740, no log, i'v got 2 good commits already :)
airlied: I probaly shoudl have tested rv770 card before I went home
airlied: ah I think I see an issue
taiu2: somewhere inbetwenn I lose console also during boot
taiu2: i dont have fbcon built in btw
airlied: taiu2: patch mirroring out now
airlied: hopes it builds
taiu2: bc1a631e5104317cc8b4ef7d14adc597f2844003 is the first bad commit
airlied: taiu2: cool the patch I pushed should fix that
airlied: or at least part of it
airlied: ah it won't
airlied: glisse: I really don't like r600/rv770 split
airlied: forgetting to mirror stuff between them demonstrates pointless duplication
airlied: pushed a fix on the fixse
airlied: okay that one builds at least, let me know how itgoes.
suokko: mzz: Sounds like it is possible recursive locking issue in kernel. So it would help ifyou rn with lock debugging on
suokko: Or it just hangs in GPU reset because GPU didn't like reseting
mzz: suokko: should I try lockdep before or after trying the new resetting code mentioned earlier?
boghog: oh sweet, X doesn't lock up anymore after updating drm-next
suokko: mzz: With both it would be helpfull :)
stikonas: airlied: after upgrading to latest drm-next X locks up on RV730
mzz: suokko: ok, I'll do both at the same time then by the time someone tells me where to get that locking stuff (currently running a pretty boring .31 kernel)
glisse: airlied: i think they already share quite a lot
boghog: wow, quake3 plays flawlessly
glisse: new reset code is nowhere
boghog: well except for some decals showing up through walls
mzz: I'll do both at the same time by the time the reset code is in a place someone can point at, and someone has in fact done so.
mzz: better? :)
suokko: airlied: What about exposing blitting interface from drm for fbcon so it could be faster at rendering text?
glisse: i think we prefer having fbcon software, it's safer
nanonyme: Does this still allow eg fast fbdev over mplayer?
suokko: ok. It just takes too much cpu time in my system
glisse: for scrolling it might makes sense
suokko: Maybe the cpu usage could be optimized too
suokko: And maybe we could make the acceleration option so it can be disabled for panic handling or gpu hang easily
glisse: i just think that there is more urgent things than fbcon :)
suokko: sure :) But I jsut tough about it because I'm now working without X ;)
nanonyme: Loopity: isn't fbcon also capable of rendering graphics and wouldn't that be useful there?
suokko: And this feels very slow compared to terminal in X
airlied: stikonas: make sure you have latest latest drm-next ;-)
airlied: top two commits should be rv770 based
airlied: suokko: not really interested in making fbcon faster really, nouveau already turned off their accel due to races
stikonas: is fc30b8efbe1b271eb64e0d4f6cb2a91bb57ee5f3 the latest?
airlied: like I'd puat patches in but turn them off
airlied: stikonas: should be there in a few mins
airlied: has to mirror
stikonas: ah, ok
suokko: So problem is to do the acceleration without race conditions. Why there is no race conditions in other acceleration codes?
suokko: Looks like the problem in mixing dma and blit copies is only in timing when previous copy was done by blit and next copy is done by dma engine
suokko: So maybe I'm not flushing something for blit
suokko: and I couldn't get fglrx running yet because my CD burner is broken. It doesn't write anything or give any errors :/
MrCooper: airlied: could the 'VGA stolen memory' object be eliminated now that the AVIVO text mode rendering engine is disabled?
glisse: MrCooper: i think it could be killed
suokko: Maybe problem is that DMA engine has some caches which aren't flushed
glisse: suokko: does the blit depends on the result of the dma engine ?
suokko: glisse: not always but sometimes yes
suokko: But problem is always dma depending on blit result
glisse: does dma wait for blit before doing it's work ?
suokko: so dma engine seems to be reading old values from somewhere
MrCooper: 2D cache seems more likely than DMA cache
suokko: glisse: There is test for that s oyes
glisse: i guess your test framework is overenginered
glisse: and there is no dma cache from what i can gather from all source i have
glisse: but dma use different path
suokko: MrCooper: I hae now version that test if I do BLIT the copy, flush&wait idleclean, blit last 4 bytes back, flush&wait idleclean
suokko: And even after that dma engine is still reading old value for that memory location
suokko: I guess Ihave to add more values to be transfered so it would be easier to locate where it is getting the old value
MrCooper: yeah, that might be useful
MrCooper: it would be really weird if the DMA engine had caches, after all its purpose is just to move data from A to B
MrCooper: maybe there's another buffer / cache behind the 2D engine though, e.g. in the render backend or so
suokko: That could be too. that BLIT->DMA problem happens in GTT->VRAM and in VRAM->VRAM copies but never in VRAM->GTT
suokko: I mean that if blit has been making that copy I get next copy that is VRAM->VRAM or VRAM->GTT with wrong values from dma engine
glisse: suokko: so when you check vram memory do you flush hdp ?
suokko: glisse: no. But I also check the result from gtt first which is same as what looks to be in vram
airlied: MrCooper: no
airlied: MrCooper: the rendering engine uses that memory as a src and renders to framebuffer memory
airlied: its the other side of 0xa000
airlied: which is still mapped to the 256k
aliverius: OpenGL renderer string: Mesa DRI R300 20060815 x86/MMX/SSE2 TCL
aliverius: is the date too old or is it irrelevant?
nanonyme: Although the format shows that it's quite old.
aliverius: it is mesa 7.5 though
nanonyme: Modern Mesa puts in also your chipset's name there.
jcristau: yeah lots of radeon changes landed for 7.6
aliverius: i also heard exa is now default
aliverius: would this let me go back to hal, deleting xorg.conf?
aliverius: could i hope for better power saving with hal?
aliverius: then the hell with it
mzz: there it went again. Still agp 8. This time it dumped http://paste.pocoo.org/show/140328/ to netconsole.
aliverius: hi logari81, i see your name is kostas... well mine is giorgos ;)
aliverius: guys, aparts from - Option "DynamicClocks" "on"
aliverius: is there any other settign i can use to reduce consumption?
logari81: hi giorgos-aliverius ;)
chithead: dynamicclocks is so obsolete
chithead: basic power management you will find in git master, see the radeon manpage for details
aliverius: chithead: google doesnt offer much info
chithead: but the manpage does
aliverius: ok chithead
logari81: does anyone know why the tormodvolden driver packages break my Xserver on Ubuntu Jaunty? I have a Mobility X700.
nanonyme: aliverius: EXA and power management have pretty much nothing at all to do with Mesa being old.
nanonyme: And what did you want to do in the first place?
aliverius: yes yes, these where separate questions nanonyme
aliverius: i just want to save power and reduce heat
aliverius: (both are becoming serious problems and i am looking at the hw too)
aliverius: i still cant find anything more than dynamic clocks
chithead: logari81: do those include mesa from git? those need drm-next kernel or
chithead: logari81: commit 905b68223be18937159a29b354e6a332952ff952
nanonyme: aliverius: Your xf86-video-ati is probably too old.
nanonyme: man radeon tells you all available power saving options.
aliverius: local/xf86-video-ati 6.12.2-2 <---how does this look?
nanonyme: I have no idea in which versions but there should be DynamicPC, ForceLowPowerMode and ClockGating.
nanonyme: They were added somewhere.
aliverius: ok thanks very much :)
nanonyme: All three are listed in man radeon if you have new enough a driver.
ndim: In my multi display setup (internal laptop panel plus larger DVI TFT), when switching to a console and back to X11 (Fedora 11, kernel-PAE-18.104.22.168-43.fc11.i686, xorg-x11-drv-ati-6.12.2-14.fc11.i586), the external DVI TFT remains black.
ndim: Am I doing something wrong there?
ndim: Oh yes, using KMS.
ndim: Just running "xrandr --output LVDS --mode 1400x1050 --output DVI-0 --mode 2048x1152 --left-of LVDS" does not fix it.
ndim: But "xrandr --output LVDS --mode 1400x1050 --output DVI-0 --mode 2048x1152 --right-of LVDS; xrandr --output LVDS --mode 1400x1050 --output DVI-0 --mode 2048x1152 --left-of LVDS" does.
MrCooper: airlied: so why does the VGA memory matter with KMS but not with UMS?
AStorm: does cairo work faster with glitz on radeon?
AStorm: one person with RV730 is reporting that firefox scrolling works fast now
AStorm: while another with RV770 reports it's still slow
AStorm: both in KMS
AStorm: RV730 card is 512 MB, RV770 card is 1G
AStorm: hmm, it seems it is actually still eating CPU.
franjva: Hi. I read that suspend was working on r600, updated to the latest git...
franjva: and indeed, it works... the first time
franjva: if I suspend again the computer doesn't wake up
franjva: anyone else?
AStorm: seems that radeon drm is eating huge amounts of "kernel" cpu time
AStorm: why would that happen? it's not supposed to do any calculations
AStorm: memcpy eats time?
glisse: command checking and other related work likely guilty here, use sysprof or opprofile to find out
AStorm: yes, I will
AStorm: funny thing that the usage spreads across the cores
AStorm: drmCommandWriteRead eats ~18% of CPU time
AStorm: that's in ddx
thalunil: People keep me telling that DynamicClocks is largely outdated and i should use git master for xf86-video-ati to get some better power-saving features, right?
AStorm: check for yourself :)
AStorm: glisse: looks like the major CPU user is ioctl itself... wtf?
AStorm: but indeed, it sounds like the command parser is slow
zhasha: guise, is there an env for libdrm_radeon that will dump the raw cs the application sends?
MrCooper: AStorm: why wtf? the ioctl does work as a service for userspace
AStorm: MrCooper: it means that something that is inside kernel behind that ioctl is eating CPU
AStorm: and I can't pinpoint what
AStorm: probably ioctl sends too much data around?
MrCooper: you can with performance counters in 2.6.31
twnqx: hi, i have a small prob here...
AStorm: heck no, if sysprof can't do it..
dileX: can someone look at
twnqx: a) display detection on DisplayPort-0 doesn't work - it says disconnected even though something is connected
twnqx: b) manually setting a mode that works on an DVI connector shows "Con not display this mode" on the display :X
MrCooper: AStorm: my statement wasn't a guess or a question
thalunil: AStorm: how can i check for yourself, the wiki is largely unstructered for this kind of query
AStorm: thalunil: measure.
AStorm: also, driver tells you about unsupported options
AStorm: and I haven't heard about any power management improvements hitting the driver recently
thalunil: AStorm: yeah, sucks big time. DynamicClocks is set and still GPU is at 78°
AStorm: thalunil: although I did get some cooldown in an unrelated way
AStorm: by enabling PCI-e power saving
AStorm: (which is now supported by latest kernels)
dileX: AStorm: on the ML Zajec posted PM patchset for testing
thalunil: AStorm: latest means what? i am running 22.214.171.124
AStorm: thalunil: latest means latest git
MrCooper: AStorm: btw sysprof Git now also uses performance counters, though unfortunately it doesn't seem to support call traces from it yet
AStorm: and it also means I'm running on K10 mainboard
MrCooper: might still be good enough for finding a hotspot, otherwise use the perf command line tool
AStorm: MrCooper: could I use ftrace in some meaningful way there?
AStorm: perf? where do I get that?
hogbog: ah bummer, I was mistaken about latest drm-next fixing X lockup after ligin with gdm, forgot I had manually disabled blit :/
AStorm: MrCooper: btw, seems that mesa master is incompatible with xorg-server master
AStorm: some header rename?
AStorm: that is, xf86vidmodeproto 126.96.36.199
MrCooper: I think this has been fixed in xf86vidmodeproto Git
hogbog: argh now even machine locks up
AStorm: was xf86vmode.h, is xf86vm.h
hogbog: after I try to kill X
AStorm: MrCooper: blah, disregard then
dileX: AStorm: the proceedings PDF is worth reading
AStorm: dileX: thanks :) I wouldn't have found it otherwise
AStorm: dileX: so, latest sysprof can use ftrace?
AStorm: I wonder if I can measure CPU usage instead of call number
AStorm: with sysprof that is
kdekorte: Dang I thought the r600 KMS lockups were gone after todays code drop, but I just got two lockups in a row... back to DRI1
bpaterni: i must be lucky, i haven't experienced any yet, save for a couple "lost displays" at startup where I'd just get a blank screen but could still type commands to the console
kdekorte: I think I'm going to upgrade to rawhide and see if I have better luck with that
bpaterni: ...but then again, I'm not running anything fancy like compiz or games
kdekorte: me either
MrCooper: AStorm: again, latest sysprof no longer uses ftrace but performance counters; it's always measured CPU cycles though, never just number of calls
franjva: another try :): has anyone tried to suspend twice? (r600 + kms)
flo|linux: franjva, do you have problems with that?
flo|linux: with suspending to ram twice?
franjva: yeah, the first time resumes without problem
flo|linux: when you wake up the first time and halt then.
flo|linux: is the console black or white?
franjva: the second... it does not
flo|linux: (black bg or white bg)
franjva: monitor down
franjva: no led in keyboard, mouse off
flo|linux: because, i noticed that after the first wakeup the shutdown-console has a white BG
flo|linux: the first wakeup doesn't work for me, either.
flo|linux: at least at my fedora system
flo|linux: (fglrx, no KMS)
flo|linux: i haven't tried with arch yet
flo|linux: (opensource driver, no KMS)
franjva: I'm with radeon + kms.
franjva: until yesterday it only worked if X was off
franjva: today it works with X on, but only once
flo|linux: no, on my fedora 10 it _never_ worked
franjva: if you're with fglrx it's a different issue :)
AStorm: flo|linux: read the topic
flo|linux: AStorm, why?
flo|linux: i simply shared my experiences...
flo|linux: and, as i experience the same (or at least: similar) problem with different drivers, maybe it's not a driver-specific bug.
flo|linux: _possibly_ the kernel has to send some signal to the card or something like that
ymanton_: agd5f: is cooper on irc?
agd5f: thalunil: DynamicClocks got renamed to ClockGating. Also, it doesn't do anythign on r6xx+ yet
agd5f: ymanton_: sometimes, but not usually
nanonyme: agd5f: Is it even ever going to do anything for classic modesetting or only in KMS PM? (as in, does anyone consider it interesting enough to implement in classic modesetting)
agd5f: nanonyme: it works on older chips. the hw changed on r6xx+
nanonyme: Yeah, I meant r6xx+.
agd5f: nanonyme: I don't plan to do any more pm stuff for non-kms
nanonyme: Wondering if it'll get implemented for those only for KMS or also for classic modesetting.
nanonyme: Works for me.
tsamolotoff: Hi there)
tsamolotoff: Updated to the latest git, now there's this message - RADEON(0): KMS Color Tiling: disabled
tsamolotoff: Does it mean something CAN be enabled?
agd5f: tsamolotoff: yeah, at least for r1xx-r5xx
agd5f: tsamolotoff: I think it still a little buggy though
tsamolotoff: And if i'm running 2.6.31 ?
tsamolotoff: speaking as HAL-9000 Alex, mind you if I ask you a personal question? Are you on M56?
agd5f: tsamolotoff: should work. right now I'm on an rs690
tsamolotoff: agd5f: I'm asking this because of terrible performance with KMS and DRI2 on x1600...
tsamolotoff: After 3 month, it's still tenfold worse than it is without kms
fallenwizard: hi, I am running drm-next with my HD4650 now and after enabling KMS, I can't start any games in fullscreen because the drivers sets the wrong frequency
soreau: fallenwizard: Are you using the very latest drm-next?
fallenwizard: soreau: yes
fallenwizard: Maybe it's a bug in xf86-video-ati
fallenwizard: KMS and 3d is working without problems so far
fallenwizard: Only fullscreen doesn't work
soreau: You might need to speak with airlied about this
fallenwizard: soreau: ok, I will do that if he comes online
mwu: it seems like the driver checks for kms support before causing the radeon module to load
mwu: so libdrm ends up reporting no kms enabled even though it is
nanonyme: mwu: Try loading it in an initrd?
mwu: it works fine if I preload it of course
mwu: but it's still buggy behavior
nanonyme: Well, that's what everyone else is doing...
nanonyme: So patches welcome? :p
mwu: yeah but those familiar with the code can do a better job
nanonyme: Those familiar with the code load it in early boot anyway, me thinks.
mwu: that's not the point
mwu: stop blaming the victim
nanonyme: Since you need it for high-resolution framebuffer on boot.
mwu: I don't really need high res framebuffer on boot
mwu: and that's not the point anyway
nanonyme: As said, patches welcome.
mwu: can you merge/review them?
nanonyme: I can pretty much anything but no, I don't have authority to actually commit anything nor am I sure if I would want to.
saucerful: hi, i'm trying to rotate a screen and getting the error "xrandr: Configure crtc 0 failed"
saucerful: i'm trying 'xrandr --output DVI-0 --rotate left'
saucerful: its an x300 chip, in a t43 thinkpad
saucerful: anyone know if this is _supposed_ to work?
agd5f: saucerful: it works
saucerful: how can i diagnose the problem?
agd5f: pastebin your xorg log and output of xrandr
agd5f: saucerful: and xrandr --verbose
saucerful: okay, 1 sec
saucerful: Xorg.log - http://pastebin.com/m440cf2c9
saucerful: verbose - http://pastebin.com/m676af78a
saucerful: (i am using the external DVI-0 and the internal LVDS)
saucerful: i get the same error for rotating either one of them
saucerful: except crtc-0 instead of crtc-1 of course
dileX: (II) RADEON(0): KMS Color Tiling: disabled
dileX: how can I enable color-tiling?
saucerful: agd5f, i guess i should paste the whole xorg.log huh
saucerful: oops, 1 sec
saucerful: agd5f: http://pastebin.com/m6a5808c1
agd5f: saucerful: output of just xrandr --verbose
saucerful: agd5f, http://pastebin.com/m6a954b62
agd5f: saucerful: desktop isn't large enough
agd5f: maximum 2840 x 1440
agd5f: needs to be 2840 x 2840
saucerful: really? but my screens are 1400x1050 and 1440x900
agd5f: but it will be slow since the max texture size is 2048x2048
agd5f: saucerful: yes. we use one big surface for both screens
agd5f: limitation of the current xserver architexture
agd5f: architecture even
saucerful: right, but if i have the laptop monitor 1400x1050 on the left and the rotated external monitor 900x1440 on the right, shouldnt that fit in 2300x1440?
saucerful: cause i'm not rotating both screens
agd5f: saucerful: the entire surface is rotated, so you'd need 2300x2300
saucerful: agd5f, i see. that seems like a limitation indeed
saucerful: agd5f, what if i always want the second display rotated?
saucerful: agd5f, can i get away with a smaller texture size
nanonyme: agd5f: Should shatter enable much smaller surface there, btw?
agd5f: saucerful: what do you mean? it always works the same only difference would be if you had a cloned screen
agd5f: where you had both screens pointing at the same area, but one rotated
agd5f: nanonyme: yeah, that's the point of it
nanonyme: yays at MAD
saucerful: agd5f, no i still want separate screens, just i could always have the second one rotated so it would fit in a virtual screen size of 2300x1440
saucerful: would that work if i just initialized the second screen with '--rotate left'?
saucerful: agd5f, no that doesnt work either
agd5f: saucerful: the surface always has to be square
saucerful: agd5f, ohhhh now i see
saucerful: okay im going to try 2840x2840
kcodyjr: this is why i'm not entirely sold on the single-big-desktop model, it's bound to have huge areas that are always invisible and yet valid
agd5f: kcodyjr: that's why we want shatter
kcodyjr: methinks i should go read the last 6-12 months of updates about that
ajax: reminds me, i need to beat up corbin until he finishes that
saucerful: agd5f: same error...
agd5f: saucerful: pastebin new xrandr --verbose
saucerful: should i try a screen size of 1440x1440 and just see if rotating one screen works
agd5f: saucerful: yeah try that
kcodyjr: saucerful, bottom line, yes. not having shatter makes dual head at decent resolutions suck.
kcodyjr: just finished reading some stuff google threw at me.
saucerful: whats shatter?
agd5f: saucerful: xserver update to use a separate surface per head
saucerful: is it stable?
kcodyjr: isn't finished, by the looks of it
saucerful: agd5f, okay gonna try that brb
saucerful: agd5f, sure enough it works...
saucerful: hmmm so is there a way that i can just have my external screen @ 900x1440?
saucerful: i got this stupid widescreen display but it rotates and i'm always going to use it that way
saucerful: so i dont actually need to rotate on the fly
kdekorte: saucerful, yeah, just turn off the primary display and then rotate the external
kdekorte: xrandr --output [primary] --off
saucerful: and then i can turn primary back on?
saucerful: okay ill try that i need to enlarge the virtual size again
agd5f: saucerful: it should work, it will just be slow since the surface is larger than 2048x2048
kdekorte: start them in cloned mode, turn off the primary, rotate the seconddary... that will give you the best speed on the secondary
saucerful: well i was already using it with them side by side which was 2840x1050
saucerful: this won't be any slower will it?
agd5f: saucerful: yes
agd5f: the rotation will be done is software
agd5f: if the total desktop size is greater than 2048 x 2048
kcodyjr: seems he was already past that line with the side-by-side setup
saucerful: yeah thats what im saying
saucerful: anyway gonna try that brb
agd5f: side by side is fine for most things
kcodyjr: oh, right, 2K stride limit on the crtc
agd5f: since there's no transform involved
agd5f: kcodyjr: not the crtc, the texture sampler
agd5f: crtc limits on that chip were 4k IIRC
kcodyjr: does rotatation have to be done with a copy? i thought some crtc's could just read the memory sideways
agd5f: kcodyjr: yes. rotation uses a shadow buffer
saucerful: hmmm still not working
saucerful: looks like i need to switch to a tiling window manager
saucerful: any recommendations
agd5f: saucerful: what error is it giving you?
saucerful: same error
saucerful: but i was using screen size of 2840x1050
saucerful: im going to try again with 2840x2840
saucerful: though that was quite slow when i did it before
agd5f: still gives that error even with a virtual 2840 2840 line in your config?
saucerful: i tried 2840 2840 and rotating and it gave the same error, but i havent tried 2840 2840 and turning off one screen, rotating, and then turning back on
saucerful: going to try that now
saucerful: same error...
saucerful: agd5f, any ideas
saucerful: i tried placing the screen at 0x0 and at 1400x0, same error
agd5f: saucerful: oh, I suspect you are out of free vram
saucerful: i see
saucerful: i'm still wondering why there's no way to have the monitor initialize to a rotated position. then it would all fit in 2300x1440
agd5f: saucerful: as I said limitation of X. we need shatter to fix it
saucerful: oh well
saucerful: i've been meaning to try a tiling window manager anyway
agd5f: saucerful: you could try kms
agd5f: might have more luck there since it has a proper memory manager
saucerful: what is it
agd5f: kernel modesetting
saucerful: is it easy to enable
agd5f: saucerful: depends on what you consider easy :)
agd5f: requires 2.6.31 kernel and new ddx, mesa, libdrm bits
saucerful: im using 2.6.31
agd5f: saucerful: what distro?
saucerful: ubuntu, but i had to upgrade because of issues with ext4
agd5f: saucerful: karmic?
saucerful: no jaunty i manually upgraded the kernel
agd5f: saucerful: I think there may be ppas available to test kms
saucerful: even if it does work i'll need to use 2840x2840, right?
agd5f: saucerful: yes
saucerful: its too slow to bother i think
agd5f: saucerful: probably. until we get shatter done
saucerful: another question: can i do 1440x1950 with one screen on top of the other but still have windows/mouse move as if they are side by side
saucerful: or does that even make sense? would that be faster than 2440x1050?
saucerful: i would think it would be because it fits in 2048x2048 so i could still have DRI enabled right?
agd5f: saucerful: you'd still need 1950x1950
agd5f: and they will still be oriented as over/under
saucerful: agd5f, how come 2840x1050 feels faster than 2840x2840 if you say that it always uses a square
agd5f: 2840x1050 uses less vram than 2840x2840
saucerful: i see
agd5f: you only need a square if you want rotation
saucerful: yeah, im ditching rotation for now
saucerful: im just wondering for my side by side setup if 1440x1950 would be better than 2440x1050
kcodyjr: yes. keeping it under 2k on either axis lets textured copies work.
agd5f: saucerful: you'll have to use above/below for that however
saucerful: yeah, that would be confusing unless i give myself nice shortcuts for moving windows around
kcodyjr: what's so confusing? when you want to move a window right, drag it upward.
kcodyjr: sorry, couldn't resist, it's a boston thing.
saucerful: yeah i laughed
saucerful: agd5f, kcodyjr thanks a lot guys
agd5f: saucerful: np
saucerful: see ya (when shatter works?)
twnqx: kcodyjr: is that a hardware or a software problem?
twnqx: i mean, 2048... that's like, singlescreen
spreeuw: for windows you're btter off with 2 screens
spreeuw: for anything with decent windowmanagers 1 is more logical
spreeuw: I advice going for a 30" eizo
twnqx: currently has 2x 24" on his nvidia
spreeuw: vize too
twnqx: a single 30" would still be too small
twnqx: also, isn't that 2560 wide?
spreeuw: ok then I guess you need 2
spreeuw: nec makes sexy screens too
nanonyme: But yeah, I don't think anyone answered twnqx's question: what causes the 2048 limitation?
twnqx: i'm considering a second card and a third display :X
spreeuw: and for better price performance dell is king
spreeuw: but they look cheap
nanonyme: twnqx: Not a good idea for now.
twnqx: i don't buy ATI cards for a reason
twnqx: sadly it's hard to avoid them with laptops...
nanonyme: Because you don't want to use open drivers? :)
twnqx: because i want working drivers. i don't really care about open or not
twnqx: i love nvidia's vdpau, for example
twnqx: (and i don't consider fglrx "working")
nanonyme: We're going to eventually have VDPAU too as long as someone who can program can give it enough of his attention span. :3
nanonyme: Hasn't yet happened.
twnqx: there's hardware acceleration for video in ATI?
spreeuw: h264 decoding on gpu you mean?
twnqx: i could have bought a full HD tv then :X
spreeuw: thats deprecated when it will reach completion
spreeuw: 3 ghz quad core2's are getting common place
twnqx: for those who like to waste power maybe
twnqx: i'm pretty satisfied with my ion/atom combo
nanonyme: spreeuw: Yeah, I just got one. :)
spreeuw: if it's not free software it's moot
twnqx: except that it still uses too much power :P
twnqx: 40W is actually too much for this
nanonyme: Erm, not a quad core, just a dual core.
spreeuw: have you heard about atom / ion
nanonyme: But it's not like having two extra cores is going to help much there.
spreeuw: ion is some new nvidio gpu combined with atom
twnqx: why not, mplayer with ffmpeg-MT will use it :P
nanonyme: twnqx: Unnecessary.
nanonyme: One core can do it fine.
twnqx: no, can't
twnqx: been there, gave up on that
nanonyme: Can if those are 3 GHz Core 2 CPU's.
nanonyme: Works for me.
twnqx: not if you have material in 40mbit with higher-than-bluray compression settings
nanonyme: spreeuw: Yeah.
nanonyme: twnqx: I had that.
spreeuw: my amd x2 5600 cant yet quite
twnqx: had framedrops with 3ghz overclocked core 2 quad (q6600)
spreeuw: havent tried mt
spreeuw: just waiting for some more opts to leak into ffmpeg
nanonyme: Mine's E8500. *shrug*
spreeuw: for downloads I'm satisfied with 720p anyway
twnqx: of course, only a few files
twnqx: well, don't really care
spreeuw: and the consumer electronics standards well
nanonyme: Anyway, it works fine so that one core handles the 1080p and the other gives you fluid desktop usage.
spreeuw: optical media are deprecated ;p
spreeuw: pirate rips are the law
twnqx: those are causing more probs than blurays, believe me :P
spreeuw: twnqx: one tip, you should turn off cpu freq scaling
spreeuw: it kills hd
twnqx: of course, locked to max
twnqx: doesn't help
nanonyme: I think I have cpu freq scaling on. It's only noticeable at the beginning of the file though, it pulls juices up after that.
twnqx: right, just tried on this 2.7ghz i7
twnqx: can't play 1080p straight from bluray
agd5f: twnqx: 2048 is the max texture size on older radeons r1xx-r4xx
spreeuw: twnqx: what player?
nanonyme: agd5f: Was it 4096 for newer?
twnqx: mplayer standard
agd5f: r5xx was 4096 and r6xx/r7xx is 8192
nanonyme: Ooh, 8192 is plenty already.
twnqx: vdpau for background
spreeuw: you need mplayer svn
twnqx: that IS svn
spreeuw: havent tried 1080 in a while here
spreeuw: and I have never tried vbluray
twnqx: agd5f: why not use one texture per display, or even window?
spreeuw: I only watch x264 mkvs
twnqx: i have some of those that don't work either :P
spreeuw: bluray needs a monitor that supports it
agd5f: twnqx: X limitation
spreeuw: allkinds of sync bs
spreeuw: and tear prevention
agd5f: one that shatter will fix
spreeuw: and drm ports
twnqx: on linux?
kdekorte: twnqx, different drm
twnqx: and on windows there's coreavc which played 1080p from bluray on my old 4800+ x2 just fine
spreeuw: no I meant hdmi ;p drm drm ;p
twnqx: ah, direct renderin
twnqx: spreeuw: my display don't support it
twnqx: does displayport require drm?
nanonyme: agd5f: Do we need to do tiling to get optimal usage of the texture for multiple screens? :)
spreeuw: display port is a drm free spec
nanonyme: "Can use" and "requires" are two different things though.
twnqx: prolly need to get a git build of the driver then
agd5f: nanonyme: ?
twnqx: also, why doesthe virtual of my internal display change? isn't xrandr supposed to allow different virtuals per screen?
agd5f: nanonyme: tiling improves performance regardless of screen size
spreeuw: nanonyme: are you using mt ffmpeg?
nanonyme: agd5f: Okay. I'll assume I was stupid and just ask this: do we use one or more textures with shatter?
agd5f: twnqx: per X screen, not per head
twnqx: hm, but radeonhd manages per head :X
spreeuw: I still remmember when r300 got tiling
nanonyme: spreeuw: No.
spreeuw: super fast
agd5f: nanonyme: separate surfaces per head with shatter
spreeuw: for free software standards anyway ;p
nanonyme: agd5f: Okay, my idea is pretty much redundant then. :) I was just thinking of how to optimally fill a texture so that you get as many screens as possible to fit in.
spreeuw: hey btw are there any recommended video ati setting with this r600 3d stuff?
spreeuw: or should autoconfigure do the best
mwu: it seems to work out of the box pretty well
agd5f: spreeuw: aren't really any knobs for 3d
spreeuw: and for texture mem reservation?
spreeuw: r300 had that
spreeuw: with exa
spreeuw: you could slide it a bit or so?
agd5f: spreeuw: you could
nanonyme: Probably don't want to touch it on cards that have enough vram.
nanonyme: I got the impression from the man page that it's only really useful in low-memory environments where you have to make compromises since you can't get everything running at the same time.
spreeuw: disable low impact fallback still necessary?
spreeuw: just launched driconf for the first time
spreeuw: with the r600
spreeuw: has anybody tried vegastrike yet?
spreeuw: that and some other games keep giving me weird textures
spreeuw: often I see junk that was left in video memory by apps I launched the day before
nanonyme: spreeuw: With or without KMS and are you running everything latest from git?
airlied: MrCooper: its probably more of a race condition with how we startup in kms mode, before we have fbcon working, vgacon still does stuff until we get to fbcon
airlied: also we put the pcigart table at 0 instead of end of CPU visible VRAM
airlied: in KMS, so if vga area trashes something in UMS we don't see it as we overwrite the memory with framebuffer
lowkyalur: hi all.
spreeuw: nanonyme: dri1, it only happens in 3d
lowkyalur: i could use a hand with my x1400. it runs glxgears fine, however it does not give anything on glxinfo|grep irect
agd5f: lowkyalur: pastebin glxinfo
lowkyalur: glxinfo http://pastebin.com/d39236a54
airlied: lowkyalur: slackware?
lowkyalur: airlied: yes. 13.0, 64bit
airlied: heres 5 cents, get a real distro
airlied: or figure out where the hell glxinfo is coming from
airlied: that isn't the glxinfo we ship with mesa.
lowkyalur: airlied: thanks.
lowkyalur: airlied: why not?
airlied: lowkyalur: the glxinfo we ship with mesa prints things a lot differently
airlied: so Slcakware is packaging one from somewhere else
dileX: and uses lilo as default-boot-mngr
mzz: that's fine, unless they've renamed it to grub
lowkyalur: hey. i came here for help, not for getting bashed for my distro
dileX: mzz: its fine if you are on old hardware
agd5f: lowkyalur: 3d appears to be working. the reason you don't see "direct" is because they ship a different glxinfo
lowkyalur: agd5f: if that's about all... no worries
airlied: lowkyalur: I'm asking you to tell your distro its brkoen
lowkyalur: agd5f: thank you very much. and now i will go and ask my surreal distro channel why that is so...
dileX: lowkyalur: maybe its good to give feedback in this case to the maintainer(s)
airlied: lowkyalur: thanks
lowkyalur: it's strange as on 12.2 32bit everything seemed normal
adamk: On Slackware, glxinfo is a link to glinfo (from Mesa) for some strange reason.
airlied: adamk: ah clueless packaging for some reason
adamk: Yeah, pretty much.
lowkyalur: oh that's weird.
suokko: uname -r = 188.8.131.52 :) But how I can now spy fglrx?
legume: agd5f: The rv670 dce3 patch doesn't work for me - http://bugs.freedesktop.org/show_bug.cgi?id=24025
agd5f: legume: are you sure it built properly? I don't see why it would be loadin the wrong firmware
legume: agd5f: I'll try again, but I think it was OK.
agd5f: legume: are you building the firmware in?
suokko: already hates seamonkey
legume: agd5f: Not sure - I'll check config
legume: agd5f: CONFIG_FIRMWARE_IN_KERNEL=y if that's the one that affects radeon f/w?
agd5f: legume: not sure off hand
MrCooper: it is
suokko: and no git :/
lowkyalur: have a good night
adamk: So I e-mailed Patrick about the lack of glxinfo in Slackware 13 and the fact that he now symlinks glinfo to glxinfo. He says that upstream Mesa quit providing glxinfo...
adamk: Any idea what he's talking about?
legume: agd5f: Seems like it's building OK. Will reboot and try again.
airlied: adamk: its still there, I'm not sure it ever built by default
DanaG: hmm, I;ll be glad when the first 2.6.32 RC is released.
legume: glxinfo builds for me with mesa git but doesn't get installed with make install
legume: agd5f: Still the same modeset=1 loading wrong firmware with dce3 patch and no display.
soreau: Well, I narrowed down the compiz water causes black screen bug down to what nha did in r300_state.c and r300_fragprog_emit.c by hand reverting the patch which eliminates r300_fragprog.c as part of the problem. Just had to re-add 4 definitions in the header radeon_code.h
soreau: What nha did in r300_state.c and r300_fragprog_emit.c is prettier, but partly broken and was wondering if it had anything with the comment he left in there
adamk: airlied, Well I just grabbed MesaLib-7.5 and MesaDemos-7.5 (the version that comes with Slackware 13) ran './configure' and 'make' and glxinfo was built.
adamk: airlied, So all I can think is that Patrick is mistaken.
soreau: This is the comment he left /* Write the config register. * Note: The order in which the words for each node are written * is not correct here and needs to be fixed up once we're entirely done * Also note that the register specification from AMD is slightly * incorrect in its description of this register. */
soreau: I only wish I could make a patch for mesa master
soreau: but it seems a lot has changed around those areas
kkk: does kms works without mesa and libdrm?
MrCooper: soreau: pointing out the problem to nha and the mesa3d-dev list should be a good start
MrCooper: kkk: without Mesa sure, but no X without libdrm
soreau: MrCooper: I already sent nha an email and memoserv notice for him
soreau: Yet to file a bug report though.. think I should just make a comment on the ml?
kkk: MrCooper, a guy is using kms without mesa and libdrm
MrCooper: soreau: bug report is probably better
MrCooper: kkk: either he's only using console, or KMS-less X touching the hardware behind the kernel's back
dileX: or didnt get he is using only swrast
kkk: he is using X
m03sizlak: well i put rawhide on my r300-having laptop
m03sizlak: and the results are NOT good
m03sizlak: i cant use system for more than about 20 minutes before i experince a hard lock requiring a reboot
m03sizlak: i get a LOT of corruption, and im only using 2d gfx
m03sizlak: and i mean bad corruption, its not even fixed by min/maxxing the windows
m03sizlak: some font corruption too, like occasionally, all my lower case 'a''s look fscked up
agd5f: m03sizlak: if youa re on an agp card, try pci mode
m03sizlak: how to do that?
agd5f: load radeon with agpmode=-1
m03sizlak: i can put that on my kernel boot line?
m03sizlak: radeon.agpmode=-1 ?
m03sizlak: ill try that after i finish putting newest updates on
m03sizlak: what kind of performance impact will that have?
mzz: m03sizlak: I'm seeing that, but not nearly as quickly as you do, on my radeon 9600xt
mzz: m03sizlak: it's locking up once every several hours or so, sometimes giving me corruption on one or more characters (probably the same thing you describe)
mzz: I haven't tried agpmode -1 yet, agpmode 4 didn't really help
kkk: compiz doesn't start
kkk: WARNING: Application calling GLX 1.3 function "glXCreatePixmap" when GLX 1.3 is not supported! This is an application bug
kkk: using last mesa, libdrm, ati, dri2proto ecc git packages
kkk: and kms
soreau: kkk: How are you trying to start compiz?
kkk: compiz-manager --debug
soreau: Can you pastebin the complete output of that command?
kkk: soreau, http://dpaste.com/95438/
kkk: uh oh
soreau: Something's wrong with your compiz install, not your drivers. The warning you see wrt glXCreatePixmap is just a warning, not an error
soreau: I recommend checking you have all compiz components installed of matching versions
Batou: Is it still normal at this point for R600/700 KMS to have corruption with large windows in some cases? Chromium and plasma widgets both have issues for me if they get too large.
agd5f: Batou: does reverting a381287759b2b65e7de9fb35801c781cab016f10 fix it?
Batou: In 'radeon'? Let me try.
Batou: Hmm, not found. Is it in drm-next?
Batou: Ah, ok.
agd5f: Batou: yeah'
Batou: Let me reboot, I don't have DRM modulized. Probably should but haven't bothered with the firmware stuff yet.
Batou: Do I need to recompile radeon or anything else?
Batou: Nope, still there.
Batou: Entire Chromium window just fills with garbage and my framerate drops super low.
Batou: Last time I left it up too long Xorg locked up.
agd5f: what size does it happen at?
Batou: Hmm. Roughly half the monitor, which is 1920x1200.
Batou: Maybe a third.
Batou: In any shape, seems to happen regardless of which direction I expand it.
agd5f: Batou: file a bug. https://bugs.freedesktop.org
Batou: ok. I'm bad with git, what's the easiest way to reapply that commit?
Batou: Is there an unrevert?
Batou: That's not it.
airlied: Batou: git reset
Er1K: sorry if this is answered elsewhere, but I'm stumped. With a Radeon 7000 (PCI, RV100), is it possible to make the DVI and HD15 connectors come up as separate X displays?
Er1K: (as if one were intending to use xinerama, but I won't be using xinerama)
bridgman: Unless there's something wierd with your card, yes
bridgman: what's happening now - two displays both showing the same ?
Er1K: "(II) Screen 0 shares mem & io resources
Er1K: (II) Screen 1 shares mem & io resources"
Er1K: I have no problems making either come up, making them clone, making them span....just can't make my second display :1 without a segfault
bridgman: Hmm... I don't remember seeing that message
bridgman: would it be easy to pastebin your xorg log ?
Er1K: get out of my head :)
Er1K: if I uncomment the "Screen 1" line in "ServerLayout" kaploooey
Er1K: sorry if the file is a bit messy...I've been monkeying :)
Er1K: right now both displays are on hd15's, the 'desktop' is on the DVI port with a dongle
bridgman: ok, so you've got two device sections there... what they call Zaphod mode IIRC
bridgman: can you pastebin your xorg log ?
Er1K: yes, I know I had both resolutions in the metamodes...I just fixed that. same result
Er1K: yeah, momento
bridgman: disclaimer - I know squat about Zaphod mode
Er1K: great, we have something in common :)
bridgman: there's a warning about mergedfb; I know you're setting it to 0 but maybe comment out that line just to be safe
bridgman: I doubt that's the problem though
Er1K: same...also manually disabled dri and dri2, but it seems to have been doing that on it's own
bridgman: yeah, I don't remember the reason why zaphod and dri don't work together, but there's a message about it in the driver so there must be a reason ;)
bridgman: does everything work fine with a normal xorg conf, ie using randr ?
Er1K: yes, perfectly
bridgman: ok; airlied had a blog post about the Return of Zaphod a while back, will dig it up and see if there are any clues