Home Phoronix Phoronix Forums X.Org Videos From FOSDEM 2008

Radeon IRC Logs For 2009-11-27

Search This Log:

MostAwesomeDude: yum update
MostAwesomeDude: Dammit.
airlied: agd5f: the whole VB wrapping code is messed up for kms,
airlied: realised he's reinvnting mesa inside the ddx ;-)
airlied: at least all the buffer handling
airlied: quits for weekend
ossman: MrCooper, ping
MrCooper: yeah?
ossman: MrCooper, I'm wondering, why are those vblank pre/post functions called from the driver and not by the drm core when it disables/enables a crtc?
Fjodor: Yesterday, a very helpful fellow, whose nick I forget, told me that there might be work underway to make the radeon driver do textures on both pipelines for multi-head. Does anyone know of a timeline for that?
MrCooper: ossman: only the driver knows it needs to do it
hifi: Fjodor: for most parts there are not timelines and no promises
ossman: MrCooper, oh? I thought drm was the main controller for modesetting, and it called the driver when it wanted a CRTC turned off, which in turn calls back to do the vblank fiddling
ossman: but I guess I'm missing something in that chain?
Fjodor: hifi: Yeah, well, that's the way things are. Maybe what I'm looking for, then, is if anyone is actively working on it, or if it's just "A Nice Idea(TM)" ;-)
hifi: fair enough
MostAwesomeDude: Split-frame multi-GPU? It'll be a bit.
Fjodor: MostAwesomeDude: Was that an answer to me?
MostAwesomeDude: Fjodor: Yep.
ossman: MostAwesomeDude, your shatter stuff is a big initial part of that though, right?
Fjodor: MostAwesomeDude: Ok, thank you for the input. To clarify a bit, though, it's multi-head on one card with two outputs, and an old one at that. Seems that I can only do 2048x2048 textures, and I have a 3200x1200 desktop...
MostAwesomeDude: ossman: No.
MostAwesomeDude: Fjodor: Oooh, *that*.
MostAwesomeDude: No, not planning on fixing that on our end.
Fjodor: MostAwesomeDude: Well, I'm sorry for being confusing, but I'm not really a graphics guy :-$
Fjodor: MostAwesomeDude: And ok, that's unfortunate for me, but thank you for the answer
MrCooper: ossman: the DRM core vblank and KMS code are separate entities, neither knows if the driver is using the other
MostAwesomeDude: Fjodor: Honestly, it's a tough problem and the current solution of "get a newer chipset" is far easier than anything else we've discussed so far.
ossman: MrCooper, mhm... so looking at crtc objects is probably a bad idea then as those will only be present in the KMS case?
Fjodor: MostAwesomeDude: Ok, I'll ask our tech dudes to hand one out to me. Thank you for the straight answer :-)
hifi: Fjodor: may I ask why you need bigger textures?
MostAwesomeDude: Fjodor: No problem. r500, from ~5 years ago, can do 4kx4k, which is why I honestly am loathe to explore fixing this.
ossman: MostAwesomeDude, what are the limits for current gen tech though? How long until the problem is back :)
MrCooper: ossman: the driver code in question is KMS specific
MostAwesomeDude: That, and the only app with this problem is compiz. :3
MostAwesomeDude: ossman: 8kx8k for current chips.
ossman: MrCooper, yeah, but I imagine the hardware bugs don't care if we're doing KMS or UMS :)
ossman: so I was planning on fixing both cases
Fjodor: hifi: As MostAwesomeDude just said, the problem is with compiz (though it looked funny ^_^)
mikkoc: MrCooper: if you could provide some patches to try for http://bugs.freedesktop.org/show_bug.cgi?id=24300#c28 I can test them
ossman: MostAwesomeDude, that's a lot of monitors...
Fjodor: 's lecture just paused, so I'll go out for a smoke. Brb
ZeZu: hmm
Fjodor: *he'll
ZeZu: chatter in here ?
ZeZu: amazing :)
MostAwesomeDude: ossman: Yep, which is why I'm not worried.
MostAwesomeDude: That, and hopefully compiz will grow a slightly better solution.
ossman: MrCooper, what is the full purpose of the pre/post functions? Grabbing the lock I guess so that nothing else enabled vblank in the middle of the mode switch, and making sure you get the most recent vblank counter value before it resets?
ossman: Is there more?
ZeZu: anyone know how i can id what registers are definitely used on an obscure ati chipset? its the xenos in xbox360, and it uses similar regs as r5xx + but some of them are hopelessly wrong :|
ossman: MrCooper, I'm trying to understand the function completely so I know what I have to work with
ZeZu: MM_INDEX, PCIE_INDEX, etc don't want to cooperate at any offset from docs , they always come back 0,0
ZeZu: while the DC & CRTC stuff works out partially & normal addresses
MrCooper: mikkoc: ugh I'm pretty swamped, hopefully someone else can
mikkoc: ok np :)
MrCooper: ossman: mostly the counter value I think
ossman: MrCooper, I didn't see anything to explicitly deal with the fact that the counter will be reset, so I take it it's only to make sure the code doesn't miss too many frames if there was a long time since the counter was update last?
MrCooper: ossman: it can only compensate for the hardware counter going backwards if it knows the last value before and the first one afterwards
ossman: MrCooper, it's supposed to do that? I'll have to look at the code again then I guess
ossman: I only remember it handling wraps properly
MrCooper: yeah, that's what it's for. There's a hardware counter mask for normal wraparounds
MrCooper: which would cause spurious increases of the userspace counter if we didn't compensate for the hardware counter going backwards
ossman: MrCooper, makes sense. I just didn't see where it did the compensation afterwards
Fjodor: MostAwesomeDude, hifi: Thank you for your informative answers - I'll get back to my lecture. Have a nice day :-)
damo22: what is the best driver to use for a ATI 9800 Pro in ubuntu
damo22: radeon or fglrx?
damo22: its a r350 based card
hifi: fglrx does not support R350 anymore
hifi: "radeon" is the way to go
damo22: okay, thanks, ive been having issues with the radeon driver though, X sometimes freezes so i disabled DRI
damo22: works well like that at least
damo22: my dad doesnt want 3d accelleration for his work, i just need a stable driver to display stuff without crashing
damo22: lol
BioTube: anybody else notice the radeon_pm_info debugfs node gives the wrong figures?
BioTube: "engine clock: 297000 Hz" should be KHz
Zajec: BioTube: i know that
Zajec: BioTube: posted patch for that typo 2-3 weeks ago
BioTube: okay
Zajec: BioTube: airlied seems to focus on other things now
Zajec: BioTube: but really should be fixed for 33-rc1
petar: this is pretty weird: when booting linux 2.6.31 with vesafb, the xserver fails to get KMS right. when killing x and starting it again---it suddenly works ((II) [KMS] Kernel modesetting enabled.).. all that on an old thinkpad t42 (ATI Technologies Inc RV350 [Mobility Radeon 9600 M10]). any ideas? oh and i forgot.. when loading the radeon kernel module before X (modeset=1), it will garble my console. (http://bugzilla.kernel.org/show_bug.cg
BioTube: petar: you need to avoid using vesafb and load fbcon with KMS
petar: BioTube: what is fbcon? i've been trying to find that particular module.. but i don't have it anywhere on my disk..
BioTube: should come with KMS
petar: (and, yes, i tried to boot without vesafb the first time---like described by the manual)
petar: BioTube: with KMS? is that a package? i thought thats a function of a kernel module?
BioTube: petar: It is
tavl: everytime i setup xorg to use radeon driver (or just let it choose it), i get a black screen, that only goes away with a hard reset. Btw, there's no Xorg log at all... The only way i'm able to use the video is through vesa driver (using xorg.conf, to set it up)... Any clue? (same behavior when using Slackware 13 or Ubuntu 9.04... and my video card is an X1650 Pro)
petar: yes, I have that already in my config.. :|
Zajec: petar: then blacklight vesafb
Zajec: and load radeon with modeset=1 =
Zajec: pass parameter at boot time, like radeon.modeset=1
Zajec: or if it doesn't load, load it manually (before starting X) with modprobe radeon modeset=1
petar: it doesn't load vesafb if i don't supply a vga= parameter
Zajec: boot parameterm ay be easier
Zajec: petar: that's ok
Zajec: petar: don't pass vga= and load radeon with modeset=1
petar: but that's what i'm saying.. even without vga= radeon.ko will garble my console with modeset=1
petar: except when i start x, kill it, and start it again.. then everything works
petar: kms and (i think) dri too
Zajec: "garble"?
petar: yeah, it looks like a bad lsd-trip. :)
Zajec: ahh
Zajec: that's some bug in radeon
Zajec: is that .32?
petar: you mean the kernel-object or the xorg radeon driver?
petar: s/object/module/
Zajec: kernel radeon module bug
petar: ah, now i get it.
petar: its kernel .31
petar: i knew it wasn't xorg's fault
petar: how could it be.. xorg isn't even started then..
Zajec: petar: that's right
Zajec: petar: 31 is first kernel with KMS for older cards
Zajec: petar: i believe KMS become somehow more stable in .32
Zajec: and probably you will have to wait for .33 to have something really usable :)
petar: and (according to hearsay) they fixed the io scheduler.. .32 will be really nice to have.
Zajec: petar: yeah, .32 contains so many improvements to KMS I thinkk it's not even worth of tracking bug you hit
Zajec: there is big probabiblity it was fixed in .32
petar: yes, you're right. i'll just forget about it.
petar: its not that it's not useable.
Zajec: as I say, .33, it will be great one ;)
petar: thanks for the help, though! :)
Zajec: :)
petar: and btw: kms is really fresh---before kms, it did hurt every time you had to restart x.. because it seemed that the display suffered some severe pain when changing modes back and forth. :)
tavl: everytime i setup xorg to use radeon driver (or just let it choose it), i get a black screen, that only goes away with a hard reset. Btw, there's no Xorg log at all... The only way i'm able to use the video is through vesa driver (using xorg.conf, to set it up)... Any clue? (same behavior when using Slackware 13 or Ubuntu 9.04... and my video card is an X1650 Pro)
legume: tavl: Is it AGP
|chiz|: does anyone here know of a good set of options to use with the radeon driver on a laptop running kde 4.3?
legume: tavl: Got to go, but if it is AGP you could try - in bios setting Aperture to 256. If that doesn't work in xorg.conf device section Option "AGPMode" "4" If that doesn't help Option "BusType" "PCIE"
dman777: is there a channel for mesa?
tavl: well, legume is gone, but it's not AGP... it's PCI-E, anyway...
tormod: dman777, I just discovered there is a #dri channel, but if it's radeon specific try here
tormod: tavl, try to disable DRI in your xorg.conf
dman777: tormod: i have a ATI Technologies Inc Radeon XPRESS 200M. when i use opacity with compiz fusion there is lag when i type. only for typing though. i was told it was a mesa and radeon problem.
dman777: tormod: everything else is smooth and resources are good, only when i type is there lag with the above mentioned.
dman777: join #dri
tormod: dman777, sounds like you have discovered a bug, so please look for / file a bug report and describe what software stack you are using
dman777: tormod: i'm new at that fileing a bug for this. where would the bug report be and what would i execute to get software stack info on my sys?
tormod: bugs.freedesktop.org, product=Mesa, component=Drivers/DRI/r300 for a first guess, although it could be a more general component. use your package manager to look up versions of kernel, mesa, radeon driver
Zajec: agd5f: what is "[PATCHES] radeon kms drm dp updates" ?
Zajec: it seems only to fix detection and add debugging
Zajec: what about base support for DP?
Zajec: *basic?
agd5f: Zajec: already in git
agd5f: drm-radeon-dp branch
Zajec: ah, new branch
Zajec: ok
Zajec: agd5f: did you get any info from that lazy guys reviewing your IRQs? ;)
agd5f: Zajec: it's approved IP-wise, just going through some wrangling on the legal side
Zajec: ups
Zajec: :)
dman777: tormod: do i need to assign it to someone?
tormod: dman777, no, people assign themselves
tormod: and by default it will be assigned to a relevant mailing list
Zajec: agd5f: what do you think about commiting Christian's patch for HDMI?
Zajec: agd5f: i know you have some additiional HDMI code
Zajec: agd5f: could we commit Christian's patch and then eventually you would add your improvements in future?
agd5f: Zajec: it's fine.
Zajec: ok, i'll ping Christian after airlied makes some awaiting-patches-cleaning
agd5f: Zajec: my patches are on top of his anyway
Zajec: that's good
Zajec: agd5f: may I ask what did you improve? or is that already a /secret/ in meaning that has to be reviewed?
agd5f: Zajec: the block routing and dto clock setup is wrong
agd5f: sort of
Zajec: agd5f: dto? i guess you mean two pairs for mul/div registers (as for timing)... but what is DTO actually?
agd5f: the hdmi clock
dman777: tormod: thanks
agd5f: for the data stream
Zajec: agd5f: interesting... was working pretty well
Zajec: agd5f: ah, moment...
Zajec: agd5f: actually Christian reported he has RV770 working with some guy but only for small resolutions
Zajec: agd5f: maybe it's related
tormod: dman777, and drop a link here in case somebody reads your original question
Zajec: agd5f: k, thx for info-status :)
agd5f: np
dman777: https://bugs.freedesktop.org/show_bug.cgi?id=25318
tormod: dman777, maybe you can try latest git of -ati also
dman777: tormod: ya, i can try it but someone said in the compiz fusion forum that the opactiy uses mesa. it's actually been a problem for me for over about a year, i just recently was able to narrow it down.
dman777: it's crazy... i coldnt figure out what it was and then someone told me to turn off the opacity and that fixed the lag for typing completely. i was like wow!
MusicGenius: anyone get a radeon 7k working in puppy linux?
petar: how can i prevent the xserver from pulling radeon.ko (& drm.ko) when starting up? i tried to blacklist the modules but x still seems to pull them..
adamk: petar, In the modules section, add:
adamk: Disable "dri"
adamk: MusicGenius, I don't know if anyone here has tried, but it should certainly work. The radeon 7000 has been supported for years now.
tavl: i'm repeating myself for about 4 hours, but here we go: everytime i setup xorg to use radeon driver (or just let it choose it), i get a black screen, that only goes away with a hard reset. Btw, there's no Xorg log at all... The only way i'm able to use the video is through vesa driver (using xorg.conf, to set it up)... Any clue? (same behavior when using Slackware 13 or Ubuntu 9.04... and my video card is an X1650 Pro)
petar: adamk: thanks! what is the difference between dri and drm, btw?
adamk: petar, 'DRI' refers to the entire infrastructure necessary for direct rendering. DRM refers specifically to the kernel driver.
petar: aw, right, direct rendering *module*.. ok
adamk: petar, Actually, DRM is 'direct rendering manager' :-)
adamk: tavl, Any chance you can pull off a log file from when you tried to use the 'radeon' driver?
petar: oh
petar: lets reboot
MusicGenius: adamk when i try to do stuff like use pSX or a n64 emulator it just throws me to prompt and i have to restart x
adamk: MusicGenius, Sounds like a bug, then :-) I suggest making sure you are using the latest available versions of the X server and radeon driver and, if the problem persists, open up a bug report and include the /var/log/Xorg.0.log file that is generated in that X session.
tormod: tavl, you still did not answer what I asked you one hour ago
robinson: Hello, is there a known performance issue with ATI integrated radeon xpress 1250 ( chipset is RS960m ) ? I get very poor performance when using any 3D application. And although desktop effects function, most games that require acceleration spit out some sort of failure... does anyone have any suggestion I could follow?
Wizzup: How recent is the driver?
robinson: It's an installation of Fedora 12. Does this line help? returned from glxinfo... : "Mesa DRI R300 (RS690 791F) 20090101 NO-TCL"
airlied: robinson: boot with nomodeset it might make things faster for 3d
airlied: we have a few 3d speed regressions
robinson: Well, it's pretty bad here. If I minimize a window ( with desktop effects ) it is supposed to do some pretty animation, but instead the computer locks up until it's done minimizing.. :p. And I have appended nomodeset, however I am unsure if there was any change in performance
airlied: robinson: wierd sounds like a bug rather than a speed thing
robinson: airlied: aww... well I just like things to be optimal, and the capability to play a game if I really want to.. but in actuality it does not bother my work too much. Any suggestions? Should I just wait it out?
robinson: ATIs proprietary driver no longer works on my computer for some reason or another... ( I dislike utilizing it anyway )
soreau: robinson: Can you pastebin your x log?
airlied: robinson: nomodeset is as fast as we can do now ;-)
airlied: the problem with those chips is they do vertex shaders on the CPU
airlied: and fglrx has a really good cpu vertex shader, and we don't
soreau: airlied: I was thinking maybe he has xaa in his x conf for some reason
robinson: you want my xorg.conf or Xorg.0.log?
robinson: The log is still "pasting", it is rather long... O_O
soreau: Yes, the log
robinson: ok, http://www.fpaste.org/b9fe/
robinson: one sec
Digital_Pioneer: OK, I'm trying to use the radeon driver with an HD4850. I've just compiled xf86-video-ati and mesa from git, installed both, but I'm still stuck with the stupid software rasterizer.
adamk: Digital_Pioneer, Use a service like http://pastebin.com/ to show us your /var/log/Xorg.0.log file.
soreau: robinson: I dont see anything too out of the ordinary
Digital_Pioneer: http://pastebin.org/57922
Digital_Pioneer: That long section starting at line 351 looks interesting.
adamk: Oh, your kernel is too old for 3D acceleration on that GPU.
adamk: Digital_Pioneer, I suggest taking a look at the wiki in the topic here.
Digital_Pioneer: I did.
Digital_Pioneer: My kernel is too old.... Arch is not known for its outdated kernels. :P
adamk: Well then you still didn't update your kernel to a new enough version :-)
adamk: Perhaps not, but it's still too old :-)
gimzo: arch has 2.6.31, I think you need 2.6.32
Digital_Pioneer: checks kernel.org... Latest stable is
soreau: Digital_Pioneer: You might consider trying airlies drm-radeon-next branch
adamk: 3D acceleration for r6xx/r7xx has only been introduced post 2.6.31... You either need the latest/greatest 2.6.32 rc, or drm-next.
Digital_Pioneer: OK.
Digital_Pioneer: This will take a while. :P
speps: hey guys what about that? [drm:radeon_cs_ioctl] *ERROR* Failed to initialize parser ! Thanks
Digital_Pioneer: This will take a long while. :(
soreau: Digital_Pioneer: If you want to try drm-radeon-testing, git clone git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6.git; cd drm-2.6; git checkout --track -b drm-radeon-testing origin/drm-radeon-testing
stikonas: ir won't take a long while. 2.6.32 should be released in 10 days
soreau: configure, compile, enjoy ;)
Digital_Pioneer: soreau: Well, right now I'm building my distribution's package for kernel26-drm-next, which is "KMS-enabled DRM kernel modules from Dave Airlie's drm-next branch."
Digital_Pioneer: soreau: Sometime in the next hundred years I'll finish checking it out, and then it should compile in pretty short order. :P
soreau: heh
speps: hey guys i get "drmRadeonCmdBuffer: -12" with several opengl apps or games.What about that?
soreau: speps: You might consider updating mesa and possibly your kernel
speps: soreau, mesa 7.6 kernel here
soreau: what card do you have?
speps: ati mobility radeon x1600
tavl: adamk: about your question, no way to get the log... the thing just freeze and stop responding even to ssh... no logs at all...
speps: soreau, xf86-video-ati git 2009 10 14
tavl: tormod: i'll take a look at the DRI option
soreau: speps: You probably should update libdrm and mesa
adamk: tavl, Well it will be really hard to figure out what's going on without log files. All I can suggest is making sure you are using the latest possible version of the driver.
speps: soreau, 7.6 is the last mesa right?You mean last git?
tavl: adamk: i already downloaded the last one... same problem... someone told me, once, that my X stack was broken... but i tried ubuntu 9.04 livecd with no luck... same freezing...
soreau: speps: git is already at 7.8
speps: soreau, last git solves?
soreau: speps: I cannot say for sure, but it is likely
soreau: tavl: Can you try booting karmic with boot option radeon.modeset=1 ?
soreau: or 9.10 at all..
tormod: tavl, and remove the "splash" option if you try modeset
tormod: soreau, kms on karmic is a game of luck since nothing stops X from starting before drm is fully initialised
soreau: tormod: I never said it wasnt
soreau: But if it does work, all the better ;)
tormod: :) I made a workaround on https://wiki.ubuntu.com/X/RadeonKMS for anyone interested
soreau: I misread 9.04 as 9.10 anyway
soreau: so he might not even have that immediate option
tormod: he's on 9.04? he should upgrade anyway
tavl: tormod: wow!! it worked!! thanks!! btw, what is the deal with the radeon.modeset=1 option?
tormod: tavl, that will switch to KMS and DRI2 and quite different code paths will be used
tavl: tormod: any bad effect?
tormod: some things might be slower, but other things might work better
tormod: tavl, are you on 9.10?
tavl: tormod: man, i've been trying to fix this for almost 3 months (with my Slackware 13)
tormod: tavl, it is not fixed, now you have no 3d acceleration
tavl: tormod: yes, but now i'll try to fix it in my slack and opensuse install
tavl: tormod: ok, so what would be the real fix? any work on it?
tormod: try the newest drivers and kernels and then file a bug, attach the non-DRI log if there's no DRI log
tavl: tormod: ok, need your help here... "non-DRI log" is the Xorg log, with that "modeset=1" option turned on?
tormod: the easiest way is to try Fedore 12 (live CD) or Ubuntu 9.10 with xorg-edgers PPA
tormod: tavl you only added DRI off to your xorg.conf, right?
tavl: tormod: well, i tested that modeset stuff with an Ubuntu 9.10 livecd, without any extra changes... and it worked..
tormod: oh I see, great
tormod: then it is fixed then, I thought you tried DRI off like I had suggested
tavl: tormod: so, no drawbacks?
tormod: tavl, http://www.phoronix.com/scan.php?page=article&item=fedora_12_ati&num=1
tormod: but its better than a hosed machine
tavl: tormod: of course it is! i was stucked at 800x600.. any idea how bad it is? :P
tavl: tormod: hum... i see... What should i expect: a fix for my problem, using UMS or a improvement in KMS performance?
tormod: you should expect huge improvements in KMS performance and UMS dying
tavl: tormod: ok, then... :)
tormod: tavl the 9.10 live CD works fine with KMS because it boots so slowly. if you install it, it will boot so fast that you might have the issue described on that wiki page
tavl: tormod: i know a thing or two about kernel and device drivers... where i can read about UMS vs KMS?
tavl: tormod: what wiki?
tormod: the one I posted one hour ago
tormod: well reading that phoronix stuff should give you some leads and links
tavl: tormod: could you send the link again? (another pc + no logs)
tormod: http://www.phoronix.com/scan.php?page=article&item=fedora_12_ati&num=1 and https://wiki.ubuntu.com/X/RadeonKMS
tormod: remember the IRC logs on www.radeon.org
Telek1: Hey quick question: Anyone here know if the R600/R700 can support OpenCL 1.0?
MostAwesomeDude: Yes and no.
MostAwesomeDude: HW can do it, but there's no drivers for it yet.
Telek1: That's cool was just curious about the hardware since I remember 'stream support' being touted for the 3xxx stuff at least, only never seeing it materializing, and the 4xxx stuff being the min supported, even though it was supposedly a minor revision of the earlier chips.
Telek1: is trying to line up a final round of pci video card upgrades for his less powerful systems to benefit from gpgpu capabilities.
agd5f: Telek1: only rv670 and r7xx+ support double precision ops
Telek1: agd5f: Which is required for OpenCL?
agd5f: Telek1: not sure off hand.
Telek1: nodnods.
Telek1: Well with the exception of a 3650 everything else I'd have is 4xxx or above.
Telek1: I'm just not sure if any 5xxx parts will end up with pci cards available :)
kdekorte: airlied, if you have a moment, could you spin a new libdrm rpm for F12 that includes the new drmGetDeviceNameFromFd
Digital_Pioneer: Not sure if this is the right place to ask, but how do I see if KMS is working?
BioTube: glxinfo | grep DRI2
Digital_Pioneer: BioTube: That gives me "IRQ's not enabled, falling back to busy waits: 2 0" which I assume is bad.
Digital_Pioneer: Since DRI2 is not in that anywhere. :P
MostAwesomeDude: grep -i kms /var/log/Xorg.0.log
Digital_Pioneer: MostAwesomeDude: Nohting.
Digital_Pioneer: Nothing even.
BioTube: IRQs aren't supported yet
Digital_Pioneer: OK
BioTube: and it seems you've got regular old userland modesetting
Telek1: Hey is KMS working on radeons yet? I've tried it on both a IGP345 and an R600 and it just blanks the screen on me.
Digital_Pioneer: Telek1: Well, if it's not then I'm wasting my time. :P
stikonas: KMS is working on all radeon except 5000 series
Digital_Pioneer: So any ideas on how I make KMS work for my 4850?
stikonas: Digital_Pioneer, you need kernel 2.6.32, xf86-video-ati from git, latest released libdrm and mesa from git or from 7.7 branch
gimzo: Digital_Pioneer: you need 2.6.32 kernel, it should just work then
Digital_Pioneer: Also, I had to kill my framebuffer to make KMS work (which it still doesn't) so how can I have my framebuffer and KMS? Right now, my terminals have horrifylingly low-resolution.
Digital_Pioneer: Well, I have 2.6.31 with the drm-next stuff.
Digital_Pioneer: I was under the impression that that would do it.
gimzo: Digital_Pioneer: with kms terminals will be in high resolution
stikonas: you must not use vesafb with KMS, kms provides radeondrmfb
Digital_Pioneer: gimzo: Sounds good. :)
Digital_Pioneer: So can I make KMS work with my 2.6.31 w/ drm-next or do I have to go all the way to 2.6.32?
gimzo: and you need to remove any vga kernel parameters from bootloader, if there are any
Digital_Pioneer: I did.
gimzo: it should work with drm-next
gimzo: that's what went into 2.6.32
gimzo: it should just work with new kernel, if it doesn't try adding radeon.modeset=1 to kernel line
Digital_Pioneer: So are there any logs or something I can look at to see why it's not working?
gimzo: there may be something in dmesg, I'm not sure
Digital_Pioneer: OK, added that to kernel line, rebooting...
Digital_Pioneer: Great.
BioTube: it worked?
Digital_Pioneer: KMS still not working, in other news, that just killed my 3D.
Digital_Pioneer: Back to software rasterizer.
gimzo: Digital_Pioneer: does dmesg say anything about invalid parameter ?
Digital_Pioneer: radeon: Unknown parameter `modeset'
gimzo: Digital_Pioneer: try radeon.nomodeset=0 .. possibly modeset=1 doesn't work any more
gimzo: do a modinfo radeon and see what does it say
Zajec: nomodeset? err? is that distro-specific?
Zajec: standard radeon module has "modeset"
Zajec: radeon.modeset=0 or radeon.modeset=1
gimzo: for me it says modeset
Zajec: tha'ts right according to me
Digital_Pioneer: gimzo: http://pastebin.com/m3e68a15f
gimzo: but I think I've seen people using nomodeset around
gimzo: Digital_Pioneer: that's a wrong kernel
Digital_Pioneer: gimzo: I have the drm-next patches
gimzo: hmm, something is wrong here, this doesn't look like drm-next module
gimzo: have you properly installed new modules?
Digital_Pioneer: Perhaps not. I'll reinstall drm-next
gimzo: Digital_Pioneer: what does uname -r say ?
Digital_Pioneer: 2.6.31-ARCH
gimzo: that looks like arch stock kernel
Digital_Pioneer: Yeah, I haven't changed its version string.
gimzo: ok
gimzo: did you do make modules_install when building kernel ?
gimzo: it should have overwritten modules with new versions then
Digital_Pioneer: wanders through the Makefile and wonders about "NAME = Man-Eating Seals of Antiquity"
BioTube: lots of projects give their releases codenames
Digital_Pioneer: That's a unique codename. :P
gimzo: Digital_Pioneer: try looking for "append" in make *config
Digital_Pioneer: gimzo: My distro's package should handle these details, should it not? :P
gimzo: not sure
gimzo: when I need to build kernel I do it manually, no packages involved
Digital_Pioneer: Yeah, but the smart people who know what they're doing already made the package. :P
gimzo: you mean drm-next package in AUR ?
Digital_Pioneer: Yeah
gimzo: I had a problem with it, a month or so ago, can't remember what it was, so I just built my own
gimzo: but anyway, let me see
Digital_Pioneer: Hmm.
gimzo: well, this might be the problem:
gimzo: Thanks for the package. Question: how do I use the new modules? They are stored in a different directory and they do not replace the old drm modules. It seems the kernel will use always the default ones.
Digital_Pioneer: That would be a problem.
Digital_Pioneer: gimzo: So should I just copy the drm-next modules over the old drm ones?
gimzo: I guess so
factor: Do the HD radeons cards work best with the Radeon xorg driver or does a HD driver exist?
BioTube: factor: both
Digital_Pioneer: factor: There is a radeonhd driver, but I've heard (word of mouth, nothing official) that the radeon driver has better support
factor: is the HD driver is the default load of xorg?
factor: ok
factor: will do radeon first then
BioTube: was last I checked, though that might've changed
Digital_Pioneer: reboots
factor: I know I have that.
factor: Which one has GL mode in it?
factor: radeon does not seems to have GL enabled
BioTube: OGL is in Mesa
BioTube: you need 7.7 or newer(7.6 might work)
BioTube: plus a recent kernel(2.6.32)
factor: I have 2.6.31
factor: but my NVIDIA works in ti just fine
factor: I have a pci card and the non oss driver.
BioTube: the 3d support code for r600 and r700 cards isn't in 2.6.31
factor: oh
BioTube: and both proprietary drivers are outside the normal graphics stack anyway
factor: Does the proprietary driver exist for the HD class
BioTube: fglrx only supports the hd class
factor: ok
BioTube: but it's not the most stable
factor: thanks forgot about that one
BioTube: plus 2d sucks
factor: downloading the proprietary driver now to test out
factor: drivers are huge nowdays:)
factor: 83 megs
tlp: hehe. everything is it seems
BioTube: that's why I love Linux: a hundred megabytes gets you thousands of drivers
factor: :)
tlp: everything scales right on up with the amount of storage/bandwidth available
spreeuw: factor: http://www.phoronix.com/forums/showthread.php?t=20484 short guide if you want to build form source
spreeuw: theres some user experience shared there too
factor: thanks, just trying the prorietary out now
factor: got that one book marked
factor: broke down and got the agp hd radeon.. really wanted the streams so I could test cuda against ati streams.my system with the pcie slot the sound did not work..
factor: but needed two systems to test out second life on.
hnsr: where does one adjust the vblank_mode parameter?
hnsr: I get this when I run OpenGL applications: "do_wait: drmWaitVBlank returned -1, IRQs don't seem to be working correctly.Try adjusting the vblank_mode configuration parameter."
BioTube: IRQs aren't supported yet for r600 and r700 hardware
BioTube: the code for it should be cleared sometime next century
Digital_Pioneer: OK, enough of this idiocy, I'm going to install 2.6.32
factor: :)
agd5f: Telek1: fglrx supports OpenCL on r7xx+
agd5f: with the stream SDK
Telek1: agd5f: Neat, thanks for the info.
agd5f: r6xx parts support CAL/brook+. I think OpenCL had some shared memory requirements that wasn't possible on r6xx
Telek1: nodnods.
Digital_Pioneer: OK, I feel like I should be getting close... :P
Digital_Pioneer: My kernel is 2.6.32-rc8
Digital_Pioneer: When I booted, it sat for a long time on a line that said it was requesting firmware "radeon/RV770_pfp.bin"
stikonas: you have to compile firmware into kernel
Digital_Pioneer: I didn't catch what it said when it finally moved on, but I have no 3D acceleration and no framebuffer, so I'm guessing it failed.
stikonas: there is menuconfig option for that
Digital_Pioneer: stikonas: So the fact that that file is sitting in /lib/firmware doesn't matter?
BioTube: well, by default firmware's not built period
stikonas: Digital_Pioneer there is no /lib/firmware in initrd
Digital_Pioneer: Fair enough.
Digital_Pioneer: stikonas: So where in menuconfig do I find this option?
stikonas: Device drivers->General Driver Options
Digital_Pioneer: stikonas: I guess it's the external firmware blobs to build in... Does it expect an absolute path or a path relative to /lib/firmware?
BioTube: that's not it
stikonas: in-kernel firmware
Digital_Pioneer: ...That's already selected...
stikonas: than I don't understand why it fails to load
stikonas: you need to ask the dev
soreau: Try '/' to search then type FIRMWARE and see if you've missed anything
BioTube: Make sure "Prevent firmware from being built" is unselected
Digital_Pioneer: BioTube: Ahh, it was selected.
Telek1: Say are PCI 4xxx cards supported in the open source or fglrx drivers?
Digital_Pioneer: So I don't need to specify *which* firmwares to include, just to include firmwares?
stikonas: correct
BioTube: just have it build the firmware and do a make firmware_install
soreau: howdy bridgman
bridgman: hi
bridgman: Telek1; didnt' know we had PCI 4xxx cards but if they exist and the driver team knew about them in advance (which doesn't always happen) they should be supported
bridgman: have you seen a pci HD4xxx card in the wild somewhere ?
Digital_Pioneer: bridgman: Had a whole herd of 'em grazing in my backyard last week.
Telek1: bridgman: Yep, one sec and I'll provide you a link :)
Telek1: http://www.newegg.com/Product/Product.aspx?Item=N82E16814131325 <-- Powercolor, and there's also an HIS, but it's out of stock :(
Telek1: Plus a bunch of AGP 46xx series cards, and PCIe x1 versions of those 2 PCI cards.
bridgman: interesting... thanks
Telek1: Welcome. I've been drooling over them for 6 months or so, just haven't worked up the nerve to spend the premium on them until I know where the drivers lay :)
bridgman: wow... crossfire ready ;)
Digital_Pioneer: BioTube: When I do a make firmware_install, those should go to /lib/firmware right?
BioTube: Digital_Pioneer: yes
Digital_Pioneer: OK, good. :)
Telek1: bridgman: Where'd you see that?
Telek1: Oh lol in the text.
bridgman: yep
Telek1: No, they're not crossfire capable :)
bridgman: it seems like a bad fit with PCI
Telek1: I WISH though, daisy chain about 4 of those on the PCI bus and I could do some awesome things :)
bridgman: I guess you could make it work, but it seems wrong somehow...
BioTube: weren't the first crossfire boards powered by extrabus cables?
bridgman: like building a supercomputer out of those Sinclair computers
Telek1: Maybe, but if it's a single set of data you can just broadcast to all attached cards :)
bridgman: BioTube; yeah, the compositor (video switch) was external too in a big dongle
bridgman: tres elegante
bridgman: then the compositor moved on board, and starting with RV570 (I think) moved on chip
BioTube: all I'm saying is that's the only way you're going to get crossfire with PCI
BioTube: could probably use that to get a decent data rate period
bridgman: Telek1; there used to be a big functionality/performance hit with PCI boards back in the AGP days, presumably because they couldn't render out of system memory
bridgman: don't know if that's still the case, will ask... or agd5f might know
bridgman: I *think* the newer chips can render out of system memory since they have a GART on chip but I
bridgman: 'm not totally sure
Telek1: bridgman: There was, although I'm not sure of the specifics. And it may not help for old PCI busses anyhow :)
bridgman: isn't new PCI bus == old PCI bus ?
Telek1: Not that I can see :)
bridgman: they're still 33 MHz, aren't they ?
Telek1: I've got some old systems that don't register newer PCI cards.
Telek1: Notably a 486 motherboard that will run a Rage 128, but either not a Radeon (Orig) or Not a Radeon 9000.
Telek1: I'll need to dig one out and test in in my mid 90s pentium.
bridgman: don't ask me to find the docco on dual-GPU rage 128 please
Telek1: :D
bridgman: Crossfire Retro
Telek1: Lol mine is a single :) It was picked up about 8 years ago because it was cheap :)
Telek1: Although docs on Rage IIC might help :D
Telek1: God that was a horrible chip :D It was FASTER to software render everything :)
bridgman: that might be easier; one of the fglrx devs seemed to have a big pile of old manuals
Telek1: Was my first 3d graphics card, followed ASAP by a Voodoo3 :)
bridgman: The IIc was our first 3D chip, I think
Telek1: I *LOVED* that card, for a good 2 years :)
Telek1: Yeah, wasn't bad as far as video cards went, but the 3d was horrible :(
Telek1: My prior card to that was a S3 :D
Telek1: Ohhh. Come to think of it maybe the Rage 128 didn't work, and it was the S3 that did. I'll need to dig up 2 sticks of EDO to find out though. I really want to pimp out that 486 with PCI :)
bridgman: IIRC we had just acquired Kubota Graphics to get into 3D
bridgman: I had an S3... it was every bit as slow as they say
bridgman: Virge, I think
Telek1: Hahaha :)
Telek1: Mine was one of the old 2d cards I believe.
Digital_Pioneer: I used to have an S3 Virge.
Telek1: I just remember how much faster it was than my previous card, which I think was ISA.
Telek1: I believe I got a VESA card not long after along with a VLB 486 motherboard (Couldn't been early pentium though?)
bridgman: that was back when Voodoo ruled
Telek1: I need to dig out the case that is in and if the bugs and mice and such haven't gotten to it, refurb it up :)
Telek1: Yeah :)
Telek1: Amazing to think of just how short of a period that was, huh?
happycube: heh... i rember the ati "3d" logo. which went away just about when ati chips actually *could* do 3d ;)_
Telek1: :D
Telek1: Anybody actually get to use an Intel 740?
Telek1: I've got one collecting dust somewhere i got about 5 years ago, but I never actually got to use it for 3d.
Telek1: Was Paradise the old 2d graphics company?
happycube: not me... i hear the i752 which barely was out is very close to i810, but there are no specs for either
happycube: yeah it was
Telek1: I've got a VLB Paradise card.
Telek1: An ISA ATI EGA/VGA card.
bridgman: I think I inherited a 740 around the time I joined ATI
bridgman: don't remember ever plugging it in though
Telek1: Another VLB card I THINK says Western Digital maybe?
happycube: i used to have lots more old stuffs
Telek1: nodnods.
happycube: i've pretty much moved away from any desktop Telek1: I got lucky and had my stuff end up in storage for a few years, and the rest piled on a desk in the parents house :)
happycube: hehe
Telek1: happycube: I wish, I still haven't been able to find a laptop that's got enough battery time to be portable, and enough videocard to not suck :D
happycube: (small form factor + p4 = not as awful as one'd think)
happycube: heh
Telek1: What speed p4? :D
happycube: good luck!
Telek1: Those chips run hot :D
happycube: 2.4ish... their idle power isn't too bad
Telek1: Reason I converted everything to either a core 2, or a sub 2ghz p4 :)
happycube: i don't wanna see a sff prescott tho.
Telek1: I've got whatever stepping the 3.0A was.
Telek1: That actually got moved into a 945/478 motherboard with dualchannel DDR... the Hyperthreading is almost dual core speed on that :)
happycube: if it has 512k l2, northwood (it might be 3.06)
happycube: :)
Telek1: It's also got enough ventilation now to not constantly overheat.
happycube: a socket 478 945? that can't be right can it?
Telek1: newegg baby :)
happycube: i could see asrock making one for fun ;)
Telek1: Fry's actually was carrying them for a while MSI 945 motherboard for ~50-60 bucks (can't remember which when I bought it)
happycube: (they did a core2-775 865 board for instance)
Telek1: The 775/478 are the same bus-wise, it's just electrical compatibility that's an issue.
happycube: yeah
Telek1: Anyhow it's a rock solid motherboard, 3 pci slots, and a PCIe x16 slot.
happycube: nice... i bet those go for a bit on ebay
Telek1: Up to 4 gigs of ram, sans the inaddressable 512 meg.
happycube: (you'ld be suprised what intel's top 875 board went for just a yearish ago)
Telek1: Haven't checked, somebody else had one on newegg not long ago for 69.99 I think.
Telek1: Well given that retail on a lot of those boards never dropped, prolly not THAT surprised :)
happycube: (acually the d875pbz is still $70)
Telek1: I've been waiting on a good deal on an X48 personally.
happycube: anyhow, gotta go... have fun!
Telek1: That and the new cheap dual core celerons with VT-x
Telek1: Take care happy :)
k\t: cxo: you around?
k\t: ok; on an x86_64 arch, i can not get glx to work properly in hardware mode, but if i use software... everythins is ok. --any suggestions?
agd5f: Telek1, bridgman: the on board gart works on pcie or pci
Telek1: agd5f: Awesome :)
Telek1: What gen of cards? R6xx/R7xx or previous gens too?
agd5f: Telek1: all
agd5f: r1xx-r7xx
agd5f: Telek1: actually even rage128
Telek1: aged5f: Awesome :)
Telek1: I've got a bunch of old radeon 9xxx cards I was planning to dualhead with :)
Telek1: Something else I was wondering: The TruForm extension never made it into the R200 driver, right?
agd5f: Telek1: I'm not sure if there was a GL extension to expose it or not
agd5f: Telek1: I think there's enough info in the mesa drivers to use it if you were so inclined
agd5f: Telek1: it's bascially just render to vertex buffer IIRC
Telek1: nodnods.
Telek1: I've mostly just wanted to test it out on my Radeon 9100 with the Morrowind patch, and maybe in the future emulate it via the new tesselation support in the R6/R7 series chips.
Telek1: NWN actually had support too, although I don't know if it actually worked in the linux port.
Digital_Pioneer: OK, I've got KMS working!
soreau: Took you long enough ;)
Digital_Pioneer: I had to specify the firmwares in the kernel configs, just like they told me I didn't have to do.
Digital_Pioneer: And now, X doesn't work.
Digital_Pioneer: :\
BioTube: fscked up screen?
soreau: Explain 'doesnt work'
Digital_Pioneer: [dri] radeon kernel module version is 2.0.0 but version 1.17.0 or newer is needed.
BioTube: you need to rebuild libdrm, mesa and ddx from git(in that order)
Digital_Pioneer: It turns my screen black for a while, then off, and slaughters my system so that I can't do anything but alt-sysrq.
soreau: Digital_Pioneer: You need to 1) Make sure to compile libdrm with kms support (see link in topic) 2) compile X, mesa and ddx against that libdrm (see link in topic) 3) Make sure fbcon was loaded before radeon
Digital_Pioneer: soreau: radeon is loaded in my initrd, so I should put fbcon in there too?
soreau: You need to make sure fbcon exists and that it is loaded before radeon
soreau: idk how that works with your specific distro
BioTube: actually, I load it after and it works fine
Digital_Pioneer: module fbcon doesn't exist on my system...
soreau: modinfo fbcon
soreau: If that returns nothing, make sure you have FRAMEBUFFER_CONSOLE in kconfig
Digital_Pioneer: No module fbcon, but CONFIG_FRAMEBUFFER_CONSOLE=y
Digital_Pioneer: It's y, not m, so perhaps it's compiled in, not a module?
soreau: points at 1) and 2) then vanishes
Digital_Pioneer: OK, I rebuilt stuff. Now trying to start X...
Digital-Pioneer: Ahh, that's cute...
Digital-Pioneer: (EE) RADEON(0): You will probably crash now ...
Digital-Pioneer: A few hundred of those, each with an accompanying (EE) RADEON(0): Timeout trying to update memory controller settings !
Digital-Pioneer: And the system actually is still running (I can SSH in)
Digital-Pioneer: But its screen is gone. It turns it off and won't turn it back on.
Digital-Pioneer: In other news, I recompiled libdrm-git, mesa-git and xf86-video-ati-git, but it still gives that stupid error about the version mismatch.
Digital-Pioneer: I didn't recompile X... Am I really going to have to recompile X? :(
BioTube: you did pass --enable-radeon-experimental-api to libdrm's configure, right?
Digital-Pioneer: Yeah
Digital-Pioneer: So what now?...
BioTube: no clue
Digital-Pioneer: What provides DRI?
BioTube: that's the whoole kit and kaboodle
Digital-Pioneer: It seems to me that DRI is outdated, since it can't handle radeon 2.0.0
BioTube: DRI and DRI2 are different protocols
Digital-Pioneer: So if I'm using radeon 2.0.0 then I don't need DRI? Just DRI2?
BioTube: actually, it's more an issue of being on the same page
BioTube: if a component is using the wrong protocol, it can't talk to the others
Digital-Pioneer: So.... What's not using the right protocol?
BioTube: for some reason your ddx isn't built with KMS support
Digital-Pioneer: OK, I'll check its config options.
Digital-Pioneer: ddx == xf86-video-ati right?
BioTube: correct
soreau: Digital-Pioneer: btw, why didn't you go with drm-next or drm-radeon-testing?
Digital-Pioneer: soreau: I got sick of fighting drm-next, so I just updated to 2.6.32
soreau: With your newer chipset, I'd definitely recommend one or the other (preferably -testing)
Digital-Pioneer: soreau: What exactly is drm-radeon-testing?
soreau: Digital-Pioneer: It is airlie's drm-radeon-testing branch with latest kernel code
soreau: To download it, do 'git clone git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6.git; cd drm-2.6; git checkout --track -b drm-radeon-testing origin/drm-radeon-testing'
Digital-Pioneer: soreau: Well, I've decided to take what I can get and live without KMS.
Digital-Pioneer: At least I finally got mesa going.