Home Phoronix Phoronix Forums X.Org Videos From FOSDEM 2008

Radeon IRC Logs For 2009-12-13

Search This Log:

DanaG: http://pastebin.com/f2ef0b9ae
DanaG: Is that that same initramfs-lacks-agpgart issue?
airlied: DanaG: looks like it
airlied: ah trolling phoronix is too easy
DanaG: also, would vga16fb claiming the console, break radeon?
DanaG: er, wait, I've blocked vga16fb.
DanaG: grr, agpgart and nvidia-agp are both in initramfs.
MostAwesomeDude: airlied: I'm partial to House, myself.
MostAwesomeDude: Makes me feel better about my sarcastic tendencies.
TBBle: Well, the HDMI audio support in drm-radeon-testing + patch to enable it in rv770.c seems to work fine on my Mobility HD4850 at 1080p, if that helps.
DanaG: argh, still null-pointer dereference,
DanaG: http://groups.google.com/group/linux.kernel/browse_thread/thread/33f5ff751a0872d3/c1241991de31a0f9?show_docid=c1241991de31a0f9
DanaG: hmm, though mine's not r600.
DanaG: mmmyeah, it's null-pointer-dereference'ing.
DanaG: http://pastebin.com/f2956135c
DanaG: hmm, should I change aperture-size setting?
DanaG: It's at 128M right now.
DanaG: And card has 128M video ram.
DanaG: hmm, setting agpmode to -1 makes it start without oopsing, at least.
DanaG: grr, failsafe-x broke KMS!
DanaG: Kernel driver in use: agpgart-nvidia
DanaG: Kernel modules: nvidia-agp
Mathiasdm: hugs radeon driver
Mathiasdm: finally, 3D acceleration :)
Mathiasdm: radeonhd 3650 btw
DanaG: anyway... my 9800 Pro works nicely, though no GLSL?
DanaG: yeah, is that hardware limitation, or software limitation?
soreau: DanaG: r300g will have support for it eventually
DanaG: That machine remains only as a curiosity, now. =þ
DanaG: It's my old, disused desktop.
DanaG: anyway, off to bed I go.
DanaG: ...maybe.
DanaG: [ 3254.162302] [drm:radeon_ib_schedule] *ERROR* radeon: couldn't schedule IB(7).
DanaG: [ 3254.162309] [drm:radeon_cs_ioctl] *ERROR* Faild to schedule IB !
AstralStorm: hey
AstralStorm: which branch of radeon should I pull into 2.6.32 for testing now?
soreau: hi osiris
osiris: soreau: hi
osiris: soreau: thanks for the fglrx info
AstralStorm: assume the kernel has nouveau added
AstralStorm: which one should I grab then?
soreau: osiris: I was just about to ask if you had received it :)
osiris: :)
dileX_: AstralStorm: take Linus-tree (2.6.32-git10) or extract commits from drm-2.6/drm-linus GIT branch against vanilla-2.6.32
AstralStorm: can't use drm-radeon-testing?
AstralStorm: it merged almost cleanly agains 2.6.32.
AstralStorm: remember, it has nouveau already
AstralStorm: (the conflicts were trivial)
AstralStorm: (one rename/rename - suprise suprise, git did conflict when both sides did the same rename)
dileX_: drm-radeon-testing should work (I tested w/ drm-linus)
dileX_: AstralStorm: sth like that
AstralStorm: I know how to merge branches
AstralStorm: thank you very much
dileX_: n.p.
AstralStorm: I suppose history base diverged a small bit, thus the conflicts
dileX_: might be
AstralStorm: someone should pull gregkh trivial fixes in
AstralStorm: from the stable branch
AstralStorm: so that people can merge it properly
AstralStorm: into stable branch
AstralStorm: right now it conflicts (trivially though)
dileX_: you mean the typo fixes?
AstralStorm: yep
AstralStorm: those cause the trivial conflicts
AstralStorm: and one missing include
AstralStorm: (in i915)
AstralStorm: some missing whitespace, such junk
dileX_: saw the trivial fixes in Linus-tree (hoped it were the drm-linus patches first, they came one commit later)
dileX_: AstralStorm: trivial fixes were in
AstralStorm: blah blah
AstralStorm: stable branch.
AstralStorm: I'm not using 2.6.32-almost-33
AstralStorm: at least until rc3
AstralStorm: it's begging for breakage
AstralStorm: using this during the merge window is crazy
dileX_: hehe
dileX_: so where you saw the trivial-fixes for stable?
AstralStorm: in 2.6.32.y
dileX_: dont see em in
AstralStorm: interesting, yet it conflicts
AstralStorm: hmm
AstralStorm: might be the nouveau part instead
AstralStorm: had that as well
AstralStorm: blames them
dileX_: isolated drm-radeon-testing.patch
osiris: airlied: here's a patch that allows for tiled textures http://pastebin.ca/1713060
mikkoc: dileX_: you have a drm-radeon-testing-patch that applies to vanilla 2.6.32?
dileX_: mikkoc: checking now
osiris: airlied: I'm not sure how to handle square micro tiling. it would be good to support it (in case you have attached depth texture to 16bit depth buffer, rendered into it and then want to sample from it - 16 bit depth buffer is always square microtiled)
dileX_: mikkoc:
mikkoc: nice
osiris: airlied: maybe we should add another flag like: RADEON_BO_FLAGS_MICRO_TILE_SQUARE?
mikkoc: i'll try it soon
papillon81: does Blender work reasonably for anyone? i have problems with the GUI elements showing up with big delay
AstralStorm: oh, my fail
AstralStorm: it seems someone rewound drm-radeon-testing
AstralStorm: dileX_: so, what should I merge into that nouveau-cursed tree?
AstralStorm: (cannot get rid of it)
AstralStorm: Renaming drivers/gpu/drm/radeon/radeon_pm.c => drivers/gpu/drm/nouveau/nouveau_i2c.h
AstralStorm: git fail!
AstralStorm: ok, again
AstralStorm: what should I pull into that stable tree + nouveau, to allow drm-radeon-testing to merge properly
AstralStorm: hmm
AstralStorm: what is drm-radeon-testing on top of
AstralStorm: definitely not 2.6.32
AstralStorm: it has some atombios functions that don't exist anymore there
ossman: airlied, ajax, problems with koji right now?
ossman: AstralStorm, I usually look at the top Makefile to get some idea of what the tree is based off
AstralStorm: ossman: ...
ossman: too obvious? :)
AstralStorm: says 2.6.32, but it's not
AstralStorm: it'd merge properly otherwise
ossman: 2.6.32 can be anything between 2.6.32 proper and 2.6.33-rc1
ossman: which is a very dynamic time in the kernel as all kinds of new things get merged
AstralStorm: exactly
AstralStorm: so, what is it based on?
AstralStorm: finding that base commit is not trivial
AstralStorm: also, why the hell is it not based on pure 2.6.32
ossman: AstralStorm, it is actually. "git merge-base linus master"
ossman: provided linus is Linus' upstream tree and master is drm-radeon-testing
AstralStorm: mhm
mikkoc: dileX_: thanks the patch worked fine
AstralStorm: heck, basing a tree on 2.6.32-git sounds like the dumbest idea ever
AstralStorm: ossman: there's no linux upstream tree tag in there ;p
mikkoc: ah i spoke too early... http://pastebin.com/d55ba1e51
ossman: AstralStorm, it's your git tree. I wouldn't know :)
AstralStorm: no
AstralStorm: I mean, in drm-2.6 tree
AstralStorm: someone forgot to fetch tags
ossman: if it's just a pure clone of airlied's, then I still wouldn't know. Haven't examined what branches he has
AstralStorm: so I get to run git merge-base vs 2.6.29-rc3 ;p
AstralStorm: he certainly doesn't have master as a branch
AstralStorm: and git merge-base doesn't want to work outside the repo
AstralStorm: ugh
ossman: AstralStorm, just pull down Linus' tree as well and you'll have a tree to compare with
AstralStorm: I HAVE the tree.
AstralStorm: what I don't have is it in the drm-2.6 repo
ossman: so pull it in :)
AstralStorm: no, I'd need it as a branch
AstralStorm: or it'd explode
ossman: Looking at the shortlog, it seems to be based on 2.6.32 proper
AstralStorm: yet it's not
ossman: AstralStorm, so fetch then. that will just modify a branch
AstralStorm: master branch
AstralStorm: how do I fetch that
AstralStorm: it's not in the remote list of drm-2.6
AstralStorm: airlied is doing something real funny when he merges
ossman: git fetch ../my-linus-tree master:linus
ossman: he probably just has linus' stuff on a remote, which means those branches doesn't get pushed to drm-2.6
AstralStorm: fb53f8621a3fab88776ae2450a1f3afc7920231b ha, ha
AstralStorm: that's definitely not 2.6.32
AstralStorm: unless linus pulled junk
AstralStorm: ehm.
AstralStorm: I'll better run it vs 2.6.32
ossman: Ah, right. you'll see whatever linus has already pulled in
ossman: you'll have to do some detective work
ossman: or start gitk and follow the pretty lines :)
AstralStorm: ugh. ok, time to run that qgit4
ossman: AstralStorm, what's the issue though? can't you work on top of that tree?
AstralStorm: no, can't
AstralStorm: I'd have to merge in a metric ton of other patches
AstralStorm: it would work, but I'd lose the other necessary things like firewire
ossman: breakage in upstream?
AstralStorm: still yes
AstralStorm: but not in the devel branches
AstralStorm: same with wifi
AstralStorm: really "loves" rt2860
AstralStorm: and merging these is not trivial, so I don't want to do it
AstralStorm: git somehow fails to automerge those
AstralStorm: for no real reason
AstralStorm: ok, seems like drm-radeon-testing diverged from drm-next a lot
AstralStorm: ugh
AstralStorm: so can't have nouveau and it in one tree w/o major changes
AstralStorm: so, I need to get rid of that nouveau
AstralStorm: meh
AstralStorm: fail was mine
AstralStorm: it is indeed on top of 2.6.32
papillon81: the blender problem is gone without KMS
dileX__: mikkoc: n.p.
AstralStorm: hmmhmmhmm
AstralStorm: I can't find the root
dileX: AstralStorm: query?
AstralStorm: dileX: yeah, go for it
AstralStorm: I need a rebase apparently
AstralStorm: ugh
maderat: [drm:radeon_gem_object_create] *ERROR* Failed to allocate GEM object (409600, 0, 4096) bug is known?
mikkoc: maderat: http://bugs.freedesktop.org/show_bug.cgi?id=25559
AstralStorm: great job with the driver
AstralStorm: now X with kms starts in split second
AstralStorm: :)
AstralStorm: oh, and UIs sped up as well
AstralStorm: heck, flash is faster too when scaled, which means the driver now must eat far less cpu
AstralStorm: can actually now keep up in my overclock mode
AstralStorm: but just barely
AstralStorm: I bet it's IRQ work, right?
_Groo_: hi/2 all
_Groo_: google earth is crashing with latest mesa/dri/ddx
_Groo_: http://pastebin.ca/1713263
kmays: AstralStorm: You should try running Scorched3D if you want to test the driver a bit more.
AstralStorm: ok
AstralStorm: but since I don't have that
AstralStorm: I'll run FreeSpace 2 Open
AstralStorm: btw
AstralStorm: I think I'm still seeing some very slight tearing in mplayer
AstralStorm: let me check again to be extra sure
kmays: Most games people can use are Urban Terror, Scorched3D, and Alien Arena 2009.
kmays: Mplayer..might be DMA
AstralStorm: -vo gl of course
AstralStorm: not xv
AstralStorm: so, likely vsync
AstralStorm: -vo xv doesn't tear and has never teared
kmays: What OS you using?
AstralStorm: linux ;P
AstralStorm: source-based
kmays: hehe..no..distro!
AstralStorm: let's say... gentoo
AstralStorm: (sctually not, but close)
kmays: Just seeing if youhave your own spin of Mesa or the distro default (i.e. Mesa 7.6.x)
AstralStorm: duh
AstralStorm: latest git
AstralStorm: of all of: pixman, libdrm, mesa, xf86-video-ati
AstralStorm: also kernel with drm-radeon-testing
AstralStorm: the card is RV690 I think (HD 3850)
kmays: Ok. Just wondering if apps like Mplayer and Quake4 would work well for you with all the latest enhancements..
adamk: Quake4 will definitely not work. It requires support for S3TC.
AstralStorm: why not
AstralStorm: it should, after unpacking the textures of course
AstralStorm: and repacking them with something more supported if possible
AstralStorm: pity S3TC is propietary still
AstralStorm: we could use an open alternative
spreeuw: there is a half working one
adamk: AstralStorm, It specifically checks for that extension. And having that extension screws up all textures in the game.
adamk: But if you think you can get it to work by unpacking the textures, go for it. Please let keep us informed of your progress :-)
AstralStorm: adamk: having or rather not having? :)
adamk: having.
AstralStorm: well... maybe we could write a fragment program that unpacks them on the fly ;p
AstralStorm: but that'd be almost as good as real support
AstralStorm: and I don't have the time to do this... yet
osiris: I've just pushed patches that let play Prey on KMS
osiris: r300-r500 cards
osiris: I've tested only on low details
spreeuw: http://icculus.org/prey/
osiris: oh, you need the libtxc_dtxn installed (game doesn't work when compressed textures aren't available)
Ghworg: s3tc is broken on r600+, there's a bug filed
spreeuw: osiris: ok will not work on r600 then
spreeuw: r300 should be ok
tlp: awesome osiris
kmays: AstralStorm: .the workaround would run into other issues that is better supported by the S3TC texture compression extension (and hardware essentially)... which resolves DXTC/BCn as well if done right.
spreeuw: osiris: could you make a patch for r600 too? ;p
spreeuw: or did you mean a game patch?
tlp: is the status even known on r6xx/r7xx?
tlp: I'll try the demo now
tlp: (RadeonProgram wiki is empty)
osiris: spreeuw: nope. I didn't work on r600 almost at all yet. it requires accelerated glCopyTexImage to run with decent speed
osiris: spreeuw: s/it/game/
tlp: did it work before, just very slowly?
osiris: spreeuw: and it isn't implemented on r600 yet
osiris: tlp: there were many graphical glitches
tlp: ah
tlp: so probably SILVER for the wiki
tlp: That thing was very useful for me when investigating the state of the free drivers, so I like to keep it updated.
kmays: tlp: It is nice to know if things like Blender3D work with the latest driver snapshots.
kmays: or even Celestia.
spreeuw: yep celestia runs smooth
spreeuw: in 1920x1200
spreeuw: only hq texture loading halts it sometimes
kmays: nice...no shading issues?
spreeuw: but that may be a disk issue
tlp: Celestia? Is that something like Stellarium?
spreeuw: kmays: not sure I have no reference
spreeuw: but it alls seems to be working
tlp: ah, cool
kmays: tlp: yes
spreeuw: using the gtk gui
spreeuw: tlp: yep but you can fly out into the universe
spreeuw: need to try stellarium sometime
spreeuw: once I get new qt built
osiris: tlp: etqw wil probably start working too once the #25588 is fixed
kmays: That will be good. A few games like World of Padman are being reviewed with the R600 driver (I think Phoronix use this game as a test app) so the more patches the better! ;)
mentor: I've got a 2.6.32 kernel, having a radeon kernel module version of 2.0.0. Unfortunately, the xf86-video-ati seems to require a major version of 1 for everything. Is there a correct branch to use with 2.6.32?
chithead: independent of kernel version, you need git master for kms. for non-kms, 6.12.4 should work
AstralStorm: I hope there will be a release soon of the xf86-video-ati
AstralStorm: one with KMS
AstralStorm: seems stable enough right now to me
AstralStorm: best to be sure though
spreeuw: AstralStorm: it requires a .32 kernel
spreeuw: or a patched 31
spreeuw: and for interrupt support you need an even newer kernel
spreeuw: will probably go into .33
spreeuw: most distros are stuck somewhere lower
chithead: I use xf86-video-ati from git master, and it works with both kms and ums
spreeuw: just ran prey demo
spreeuw: indeed the typical s3tc corruption
spreeuw: even with gamesetting turning it off
AstralStorm: spreeuw: the latter would be a game bug
AstralStorm: unused extension should be that.
RiotingPacifist: so i finally got glxgears to use dri2, but server glx vendor string is still SGI, im confused :S
RiotingPacifist: I have a the latest kernel (with modesetting on) the latest mesa (compiled from git with radeon drivers) and OpenGL renderer string: Mesa DRI R300 (RS400 5975) 20090101 NO-TCL DRI2 looks good, but the server says SGI/software rendering :( any ideas
MostAwesomeDude: SGI != software rendering.
MostAwesomeDude: The server's GLX *is* from SGI. Long story.
mentor: chithead: I am using git master
MostAwesomeDude: (Well, okay, maybe not that long. SGI open-sourced their GLX stuff, Mesa and Xserver pulled it in and based their GLX from that.)
MostAwesomeDude: RiotingPacifist: If your rendering string says R300 and DRI2, then you're good to go.
Nezmer: Hi, I grabbed a kernel snapshot from "drm-radeon-testing" head and built it.
Nezmer: KMS is working but it disables hardware acceleration "(II) RADEON(0): Acceleration disabled"
Nezmer: Is this normal?
MostAwesomeDude: Could you pastebin your Xorg.0.log?
adamk: Nezmer, Please pastebin your full log.
adamk: Heh.
olaf_: hello , i'm just trying last kernel git (10) , and i got an error : r600_cp: Failed to load firmware "radeon/R700_rlc.bin" , but i'v this file on /lib/firmware/radeon/
adamk: olaf_, Is radeon DRM built into the kernel?
adamk: olaf_, If so, you need to use the CONFIG_EXTRA_FIRMWARE (and CONFIG_EXTRA_FIRMWARE_DIR) kernel config option.
olaf_: adamk, CONFIG_DRM_RADEON=y ?
olaf_: ok
adamk: Right, so it's built directly in, which means you need that option.
mentor: I've got a 2.6.32 kernel, having a radeon kernel module version of 2.0.0. Unfortunately, the xf86-video-ati (master) seems to require a major version of 1 for everything. Is there a correct branch to use with 2.6.32?
olaf_: thanks adamk
adamk: mentor, You need to recompile libdrm with the experimental radeon api (ie. KMS support) and then recompile xf86-video-ati against that newer libdrm.
mentor: Yeah, done that
adamk: mentor, Check the wiki in the topic :-)
adamk: mentor, Apparently not :-0)
Nezmer: Xorg log http://pastebin.com/f37c3c8ee
adamk: mentor, Run ./configure in xf86-video-ati and see if it says that kernel modesetting is enabled.
mentor: aha
Nezmer: I have the same missing firmware as olaf_ in dmesg. Do I really need to include it in kernel even If I don't want to start KMS early?
MostAwesomeDude: Nezmer: I think that your xf86-video-ati doesn't know how to do r600 acceleration; you'll need go grab a newer one.
Nezmer: MostAwesomeDude: It works when I disable KMS. built from git/master today
adamk: Nezmer, If you've compiled radeon into the kernel, you definitely need that option to include the firmware.
adamk: Nezmer, I'm not sure if it's necessary to do that when you compile radeon DRM as a module.
olaf_: i'm trying with radeon drm as a module first
Ghworg: If compiled as a module you just need the firmware in /lib/firmware/radeon/
adamk: OK, good.
adamk: Nezmer, So you can have radeon DRM as a module, without compiling that firmware in.
mentor: adamk: "configure: X server has new mode code" is that the string that I should be looking for?
adamk: Nezmer, Did you download that firmware file and install it to /lib/firmware/radeon/ ?
adamk: mentor, I don't believe so. It should actually say "Kernel modesetting: yes" if it detects a libdrm built with KMS.
mentor: ok
adamk: mentor, Did you install the new libdrm anywhere special?
mentor: No
Nezmer: adamk: Should I look for it somewhere? I thought it would be available with kernel. Other firmware files are available but this one is not
Nezmer: adamk: http://pastebin.com/f260843e4
adamk: Nezmer, It's a very new firmware file and will not be included in the kernel.
adamk: Nezmer, http://www.mail-archive.com/dri-devel@lists.sourceforge.net/msg45004.html
Nezmer: adamk: I thought that was the point of "drm-radeon-testing" !
adamk: Nezmer, It will not be in the kernel because the kernel no longer accepts new firmware files.
adamk: Nezmer, It has nothing to do with the various development branches.
mentor: Aha, there we go
Nezmer: adamk: ok. thank you for the link.
adamk: No problem.
lordheavy: got this with TA spring with a RS880 (r600):
lordheavy: Mesa: User error: GL_INVALID_ENUM in glCompressedTexImage2D
lordheavy: spring: radeon_mipmap_tree.c:436: migrate_image_to_miptree: Assertion `mt->mesaFormat == image->base.TexFormat' failed.
MostAwesomeDude: User error -> App bug, not our bug.
chithead: lordheavy: there are more problems with r600 and spring, such as bug 25024
lordheavy: ok so it's not the better app for testing :-)
chithead: I think spring is not so much on the radar of the developers
lordheavy: MostAwesomeDude: ok
Nezmer: Alright, good news and bad news
Nezmer: good news, hardware acceleration is not disabled anymore when I add "R700_rlc.bin".
Nezmer: bad news, my desktop now looks like this http://omploader.org/vMnpoMA
Nezmer: It usually looks like this http://omploader.org/vMnpoMQ
kmays: ugh
kmays: chithead: I hope Blender3D stays on radar.
adamk: Nezmer, Hahah... I love that wallpaper.
adamk: Does the corruption exist without KMS?
Nezmer: adamk: let me try. I'll be back
Nezmer: adamk: the corruption I reported earlier doesn't happen with KMS disabled. But that's because "R700_rlc.bin" is not loaded during boot
airlied: osiris: oh yeah we need another square flag then I didn't add it though because I'm not sure we had the code in mesa to detile it
airlied: but maybe we do.
airlied: osiris: I wasn't sure what it was used for and I think it was different on r100/r200/r300
spreeuw: Nezmer: you need to download that
spreeuw: and make sure it loads
spreeuw: its required for irq in the newest kernel drm
Nezmer: spreeuw: I already have it. I meant It's not called with KMS disabled. It loads with KMS but causes this http://omploader.org/vMnpoMA
spreeuw: hey you tux blasphemer
airlied: osiris: btw I think under KMS we could get a good perf boots by just fast-tracking clears
airlied: and avoid the metaops again
airlied: though I always hated our old emit clear state stuff
dileX: "drm/radeon/kms: add dynamic engine reclocking": is that only working with r600(.c) or with all radeon hardware?
airlied: google fir gem.txt
airlied: oops
airlied: dileX: it might work on r500
dileX: how can I check if it works or not?
airlied: cat radeon_pm_info in debugfs in theory
mtkoan: anyone ever get cursor distortion when using radeon with multi-screen?
adamk: Yes, I have in the past.
Zajec: dileX: it's working on AtomBIOS devices
dileX: Zajec: how do I see if its working?
dileX: (just trying v3)
mtkoan: adamk: any clue of cause or fix?
Zajec: dileX: mount debugfs
Zajec: cat radeon_pm_info
Zajec: you should have engine downclocked by 50MHz
adamk: mtkoan, It went away after updating the driver quite a while back.
mtkoan: ah, i'm using the stock slackware 13, maybe an upgrade will fix it
dileX: Zajec: how do I downclock engine?
Zajec: dileX: it does itself when low load
eosie: soon enough we will be able to use pipe_context->surface_copy for untiling textures
dileX: Zajec: will report later (here: x1300 pci-e)
Zajec: dileX: would be nice :)
adamk: mtkoan, The first thing I did after installing Slackware 13 was to upgrade the kernel and the radeon driver :-)
lordheavy: damn, font corruption is back http://pastebin.ca/1713575
mikkoc_: lordheavy: http://bugs.freedesktop.org/show_bug.cgi?id=25559
cxo: Google maps keeps causing this error: [55095.035433] radeon 0000:02:00.0: object_init failed for (3276800, 0x00000004)
cxo: [55095.035443] [drm:radeon_gem_object_create] *ERROR* Failed to allocate GEM object (3276800, 4, 4096)
cxo: i'm running everything out of git from last night
dileX_: Zajec: I dont see any radeon_pm_info?
dileX_: mount -t debugfs none /sys/kernel/debug/ && ln -s /sys/kernel/debug /debug
dileX_: need to set any CONFIG_*_DEBUG?
lordheavy: got it here : /sys/kernel/debug/dri/0
lordheavy: [lordh@myhost 0]$ cat radeon_pm_info
dileX_: Zajec:
lordheavy: engine clock: 494040 kHz
lordheavy: memory clock: 00 kHz
dileX_: yeah
dileX_: root@seduxbox:~# cat /sys/kernel/debug/dri/0/radeon_pm_info
dileX_: engine clock: 391500 kHz
dileX_: memory clock: 297000 kHz
dileX_: Zajec: ^^
dileX_: lol
dileX_: find /debug -name radeon_pm_info vs. find /debug/ -name radeon_pm_info
Ghworg: Okay, here is a weird one. After running for several hours, firefox starts missing letters in webpages. All the Ts in a particular font for example
Ghworg: Other apps don't seem affected, but the desktop is KDE so most apps are QT
adamk: Any thoughts as to why explosions are so slow in quake2forge with the r600 driver?
lordheavy: known problem see http://bugs.freedesktop.org/show_bug.cgi?id=25559
Ghworg: lordheavy: That appears to be different. I get no error messages, and no other corruptions visible outside firefox
Zajec: dileX: what is your GPU?
Zajec: ah, moment, you wrote it i think
Zajec: x1300 :)
mentor: It's alive!
mentor: cackles
dileX: 01:00.0 VGA compatible controller [0300]: ATI Technologies Inc M52 [Mobility Radeon X1300] [1002:714a] (prog-if 00 [VGA controller])
dileX: rv515
Zajec: dileX: Wikipedia says your default engine clock is 450MHz... your logs shows 391MHz which should mean downclocking working :)
dileX: Zajec: any GPU-stress test?
Zajec: dileX: this PM upclocks really fast... even starting glxgears should make it upclock
spreeuw: good it clocks up fast ;p
spreeuw: I suggest also a slow decay
spreeuw: judging linux CPU scaling you're going to notice it
spreeuw: it directly affects user experience in all modes
spreeuw: if you pay attention
Zajec: spreeuw: it downclocks only if it finds GPU boring twice in a row
dileX: hmm
spreeuw: it's a trade off
Zajec: processing even 1 command won't cause downclocking
dileX: # cat /sys/kernel/debug/dri/0/radeon_pm_info
dileX: engine clock: 391500 kHz
dileX: memory clock: 297000 kHz
Zajec: dileX: i just noticed i don't see upclocking in pm_info...
Zajec: hm :)
Zajec: let me debug it :)
dileX: (running glxgears)
Zajec: yeah, sth is wrong...
Zajec: as me tomorrow please :)
Zajec: just bought Samsung R522 with 4650 :)
spreeuw: sweet
dileX: Zajec: n.p. (will continue watching kill bill vol.2)
spreeuw: Zajec: did you have other candidates too?
spreeuw: looks like it has its own video mem, nice
spreeuw: Zajec: the wifi works?
Zajec: spreeuw: reading about wifi right now :)
Zajec: spreeuw: i received this hour ago
Zajec: installing Linux right now :)
spreeuw: ok
spreeuw: did you have an xp one?
Zajec: win7
Zajec: got some candidates... some Asuses generally and some Dell 15 studio
Zajec: HP has nice notebooks but they blocks wireless cards other than whitelisted
Zajec: Toshiba has some nice L500, but generally every Toshiba lacks something from my want-list :)
spreeuw: I see theres also an Intel gpu version
Zajec: yup
Zajec: also 4330 is quite popular
spreeuw: yeah toshiba is making some lower budget models too now
spreeuw: ordered one for a colleague a while ago
Zajec: spreeuw: thing is a want all following in one: webcam, 802.11n, HDMI, ATI gpu... they always miss something :P
spreeuw: the matte steel finishing is nice
edt: would anyone know if this commit 'drm/radeon/kms: fix RS600 MC setup / 64bffd03756249e11b8651ccf33ac3a50a93ed4c' problem was to fix font corruption. If so its probably needed on rs780 too. drm-linus, drm-radeon-testing both have font corruption issues here.
RiotingPacifist: I'm still confused by this "server glx vendor string: SGI"+"client glx vendor string: Mesa Project and SGI"+"OpenGL vendor string: DRI R300 Project"=kwin won't do compositing is kwin wrong or do i not have dri running properly
spreeuw: edt: dunno try the latest code of everything
MostAwesomeDude: RiotingPacifist: Pastebin "LIBGL_DEBUG=verbose glxinfo" please.
RiotingPacifist: MostAwesomeDude: http://pastebin.com/d73014f6c thx
maro_: holy crap, just checked out the ati driver from git (using latest releases of everything else, except mesa 7.7-rc3) and I'm stunned at how well my R700 suddenly works
Terman: can someone please help me: I don't have acceleration and I see the following errors (kernel 2.6.32, git head of drm, xf86-video-ati and mesa (from 5 hours ago):
Terman: (EE) RADEON(0): [dri] RADEONDRIGetVersion failed because of a version mismatch.
Terman: [dri] radeon kernel module version is 2.0.0 but version 1.17.0 or newer is needed.
Terman: [dri] Disabling DRI.
MostAwesomeDude: RiotingPacifist: kwin needs to go stuff itself. Your setup is fine.
MostAwesomeDude: RiotingPacifist: But of course kwin *might* blacklist your chipset, since RS4xx are known to be unstable.
MostAwesomeDude: Either way, kwin's problem, not ours.
RiotingPacifist: why is it a mess of SGI and DRI?
maro_: compiz with wobbling 1080p videos no problems :-)
MostAwesomeDude: Terman: You need to rebuild your libdrm with --enable-radeon-experimental-api, then rebuild xf86-video-ati.
MostAwesomeDude: You should see Kernel Modesetting: yes, if you did it right.
MostAwesomeDude: RiotingPacifist: It's not really a mess.
Terman: MostAwesomeDude: I already do that
RiotingPacifist: MostAwesomeDude: perhaps but but i had it working on this chipset before I suspect kwin just checks for SGI and assumes software, thx for the help
MostAwesomeDude: Server GLX and client GLX might not match; the actual GLX version used is whatever's in common between them.
MostAwesomeDude: RiotingPacifist: You might be thinking of Flash.
MostAwesomeDude: RiotingPacifist: But yeah, kwin's being bad.
MostAwesomeDude: Terman: The xf86-video-ati being loaded in that log is not built with KMS support.
Terman: MostAwesomeDude: at least since oct 13th
Ghworg: RiotingPacifist: I've the same SGI strings in my vendor info and kwin compositing works fine, so it is not that. I'm running on different hardware to you though
RiotingPacifist: Ghworg: my opengl version is 1.2 could that be the problem
MostAwesomeDude: RiotingPacifist: Your GL version is 1.5.
MostAwesomeDude: GLX is just what the server and client use to talk to each other; unless you are a compiz dev, you should not need to care about it.
Ghworg: RiotingPacifist: OpenGL version or GLX version? My GLX is 1.2 (server 1.2, client 1.4), OpenGL is 1.5 Mesa 7.7-rc2
RiotingPacifist: hmm mines the same but mesa is 7.8-devel
Ghworg: git master, I'm running git 7_7 branch
Ghworg: Yours should be more capable than mine
dileX: Zajec: sent feedback mail to dri-devel ML
edgecase: ok Ubuntu xorg-edgers + Karmic, Radeon Mobility M7 LW [Radeon Mobility 7500], i have a dim green-ish display with KMS on
Terman: MostAwesomeDude: looking at the build logs, I find: Kernel modesetting: yes
MostAwesomeDude: Terman: And it's installed to the right spot?
edgecase: what knobs in xorg.conf can i twiddle? the cursor is bright and normal
MostAwesomeDude: Terman: Could you pastebin your entire Xorg.0.log?
Terman: MostAwesomeDude: I verified that the path is that of the newly built module
Terman: sure
edgecase: when i VT swictch back to tty1, it's also dim green
edgecase: Option "ColorTiling" ? is that used with KMS ?
MostAwesomeDude: edgecase: This is KMS-only? Doesn't happen with KMS off?
RiotingPacifist: ok it's a bug with the 4.4 kde better thanks for the help and thanks for the drivers
edgecase: hmm didn't try it, just installed the shiny pkgs ... now i feel dumb... i'll try that, i assume ommit cmdline radeon.modeset=1
MostAwesomeDude: edgecase: Yeah, or radeon.modeset=0
edgecase: it doesn't look like man pages are updated BTW. BBS
tormod: edgecase, for KMS you want a newer kernel also, try lucid kernel or drm-next
edgecase: ok
edgecase: good to know, i'll try kms off then new kernel
edgecase: also, should xrandr still work? it doesn't detect the external VGA being plugged in
edgecase: bbl
Terman: MostAwesomeDude: http://www-agrw.informatik.uni-kl.de/home/jmayer/radeon/
MostAwesomeDude: Terman: Yep, it's still as I said before. The driver that is being loaded, was not built with KMS support.
Terman: MostAwesomeDude: the driver that is loaded is in /usr/local/...
Terman: MostAwesomeDude: and the log for that one says modesetting on
Terman: during the build process
Terman: MostAwesomeDude: so I don't understand what I missed that you come to the opposite conclusion
lordheavy: firmware ?
Zajec: dileX: thx
Terman: the most irritating line is: [dri] radeon kernel module version is 2.0.0 but version 1.17.0 or newer is needed.
airlied: Terman: that message mans the driver has no KMS support
Terman: so 2.0.0 isn't newer than 1.17.0
airlied: or KMS isn't loaded before X is started
mentor: The major numbers have to match
mentor: But the message is wrong
DanaG: That string needs changing, then.
DanaG: It should say something more like, version is 2.x.x, but version 1.7.x is needed.
airlied: patches welcome
mentor: Is "drmWaitVBlank returned -1, IRQs don't seem to be working correctly." expected with r600?
MostAwesomeDude: That entire chunk of code with that message should probably say, "I was built for DRM version 1.x, not 2.x."
MostAwesomeDude: Because otherwise someday it might come back to bite us, if we ever get up to 2.18.0. :3
Ghworg: mentor: With 2.6.32 or older yes, irq code is only in drm-* git code
mentor: Ghworg: Tanks
Terman: weird, I start the kernel with radeon.modeset=1 and the resolution of the console changes from normal vga to something with a much higher resolution
MostAwesomeDude: Terman: That's supposed to happen. :3
lordheavy: :-)
Terman: MostAwesomeDude: yes, and that seems to indicate that kms is active
Terman: MostAwesomeDude: I've added the binary of my radeon driver, can you please verify that it really doesn't have kms built in?
Terman: oops: I've added it to the url above
MostAwesomeDude: Terman: Did things start working right when you loaded KMS early?
Terman: MostAwesomeDude: well, it seems I don't have 3d :-/
lordheavy: it depend on mesa
airlied: Terman: it'll print in the log before that point
airlied: if kms is detected or not
Terman: I've included dmesg
Terman: in the kernel/ directory
Terman: hmm, or maybe I'm telling you nonsense and the system froze with early kms and I rebooted
Terman: sigh, that was 5 hours ago
Terman: I
Terman: 'll reboot and let you know the result
Terman: result is, that airlied was right: It looks like I started X before turning on kms (race condition in my scripts between boot.local and xdm start)
Terman: thanks! and sorry for the noise :-(
benh: agd5f: ping
MostAwesomeDude: airlied: http://people.freedesktop.org/~csimpson/0001-Be-more-useful-if-disabling-DRI-due-to-KMS-UMS-misma.patch
MostAwesomeDude: I don't have commit privs to the DDX. Can you approve/disapprove?
airlied: MostAwesomeDude: looks good
MostAwesomeDude: We probably should have had that a few months ago, but it's never too late. :3
MostAwesomeDude: And besides, some of those other warnings are *way* older.
benh: airlied: ping
airlied: benh: pong
benh: airlied: remind me how you wanted me to handle the FW loading for that ATOM ROM for the mac X1900
benh: airlied: alex still hasn't sorted out the distribution of the file ... but I want to sort out the kernel patches so I can test KMS etc...
airlied: benh: just where we probe for the bios, check for that card ids and call the fw loader
airlied: for building radeon into the krenel you'll need to use CONFIG_EXTRA_FIRMWARE pointing at the bios image
benh: airlied: ok but you were also talking about some statement so it gets included in initrd or something like that too
benh: airlied: I can't remember what ;-)
airlied: add a MODULE_FIRMWARE line somewhere
benh: ok
airlied: might want to make it ppc oinly
benh: that or a CONFIG option
airlied: I'm not really caring for ppl who try and use a ppc card on an x86 machine
benh: (ie, one could use that card on an x86)
benh: or on sparc :-)
benh: ok
benh: will do CONFIG_PPC for now
airlied: benh: they could but that doesn't mean it should work ;-)
benh: hehe
spreeuw: lol damn spring AI got me
benh: airlied: is anybody moving over the PM code from radeonfb yet or it's something I could do while at it ?
edgecase: ok, Lucid kernel fixes my KMS, neat stuff
cxo: if i have a window with a lot of rapidly scrolling text (like something compiling for example) the system slows down like crazy. If i minimize the window, its all fast again
spreeuw: keep it minimized then ;p
spreeuw: is this in xorg or on console?
cxo: x, with compiz,kms,r700
lordheavy: cxo : kde ?
spreeuw: compiz is nr one suspect then
cxo: gnome (ubuntu 9.10)
spreeuw: and after that the font fuzzing engine
edgecase: BioTube, do you know about radeon power mgmt?
cxo: and this *ERROR* Failed to allocate GEM object business is really getting annoying
cxo: firefox seems to trigger it very easily, and it just keeps crashing
cxo: Is there a fix for that?
cxo: and the room goes silent
lordheavy: currently, no fixes
edgecase: is there a way to query the GPU memory allocations?
spreeuw: osiris: if you need a guinea pig for r600 s3tc send me patches I'll gladly test
edgecase: framebuffer size, etc
hbock: has anyone been working on switchable graphics support?
edgecase: hbock, what is "switchable" ?
hbock: i've been wondering what kind of infrastructure is necessary to implement something like that, and I'd be willing to try if no one else is interested
hbock: edgecase: newer lenovo laptops have both integrated (intel) and discrete (amd) graphics
airlied: benh: for non-powerpc we mostly resume properly using bios table
hbock: with proper drivers in windows vista+ you can switch cards on the fly
airlied: except some M10 cases and for those I think porting some of the radeonfb code could help
hbock: i imagine it involves copying video memory between the two and some format conversions, etc
airlied: just need to figure out what bits are modesetting related and get reset when we rewrite the mode
hbock: but i'm not terribly familiar with that
airlied: hbock: its kinda messy to do at runtime on Linus
airlied: Linux even
airlied: like you could do it by restarting X as a first step
edgecase: hbock, could you configure both drivers, and have dual-headed X?
airlied: and they is probablhy necessary
happycub1: edgecase - usually the intel is turned off, but if the bios is set up right it might be a GPIO switch somewhere
edgecase: happycub1, lspci doesn't show both at once?
edgecase: you can only have one card doing VGA io ports at once, but some drivers don't need that
hbock: edgecase: from what i recall if you enable switchable graphics in the bios, X won't even start
chithead: hbock: on lenovo (and iirc acer) that is the case. with sony it presumably works regardless of bios setting
chithead: well "works" as in uses one of the cards and not be able to switch
hbock: chithead: yeah, and both cards are probably consuming power too, right
hbock: airlied: since this switchable graphics business seems pretty non-standard at the moment, would it be in the ballpark to assume amd wouldn't have any real docs on it that could be cleaned up and released?
edgecase: hbock, there is a standard way, did you try lspci?
hbock: i haven't tried it, i should do it in a VT though
edgecase: i don't think that matters, but try both VT and X
hbock: maybe it is done through the standard init for both cards, with some additional driver code for the live switching
edgecase: well i can put 2 vid cards in a desktop, and it's pretty straight forward, maybe this is the same
hbock: that would make sense
hbock: perhaps with a userspace utility that controls which is on at any given time
hbock: since they share LVDS/DVI-D/VGA output lines
airlied: hbock: the problem is each laptop does it all differe nt
edgecase: oh that's the trick then, sharing outputs
hbock: that's a definite problem :/
airlied: though I think most of the ACPI calls are known to do the switch
airlied: I know nouveau has support to turn off the nvidia card in the sonys
airlied: when it detects both
airlied: so it doesn't chew poewr
hbock: is the handling of moving video ram buffers done via ACPI or would that have to be done manually?
airlied: like we have enough info to connect to the secondary card and post it
airlied: hbock: manually
airlied: i.e. thats step 2
airlied: step 1 is just doing it across X restart
airlied: so X picks the second video card instead of the first correctly
hbock: yeah, but then you lose all of your windows, no?
airlied: yes but you can't jump to fixing that problem
airlied: unless you fix the first problem
hbock: mmmm, yeah.
airlied: I think mjg59 knows a fair bit on the subject
airlied: like you could develop the second solution I suppose on a PC with 2 normal GPUs
airlied: X would require dynamic screen creation and a lot of xinerama/randr owrk
airlied: I think ajax has some ideas but its in the "shitloads of typing" category
edgecase: display hotplug eh?
airlied: hence why I'd solve the easy case first
airlied: edgecase: GPU hoyplug
airlied: display hotplug sort of mostly works
edgecase: right, i meant that
edgecase: i use display hotplug, it's awesome
hbock: i'd love to get involved in working on it
edgecase: hbock, look for ACPI specs on the subject
hbock: even though it's a huge hurdle getting involved with X.Org codebase :(
hbock: just need a lot of time i guess
edgecase: look at the .inf files for windows driver
edgecase: start with lspci :)
hbock: i think i will :)
edgecase: pastebin it now!
hbock: working with two discrete GPUs in a desktop might be a good way to get a grip on things
edgecase: well that's old news really
edgecase: unless you want to use linuxbios and don't bother mapping the legacy VGA registers and memory :)
benh: airlied: the ROM image alex sent is for a "52027-107" card while this is a "52031-109"
benh: but it seems to work fine so far in X, I'll try kms asap
lordheavy: nexuiz segfault :-/
edgecase: (II) RADEON(0): mem size init: gart size :4000000 vram size: s:1000000 visible:c00000 <-- can someone explain this a bit? vram is 16MB, that i understand, but is gart the ammount of system ram the GPU can access over AGP?
edgecase: also this one: RADEON(0): VRAM usage limit set to 8294K
edgecase: front buffer size is only for 3D, right?
benh: hrm
benh: airlied: radeon kms is blowing up :-)
benh: airlied: looks like I'll keep it as a module for the time being until I debug it
benh: airlied: looks like housekeeping bug ... it fails to init
benh: airlied: and unload_kms ->device_fini -> rv515_fini (hrm, it's a 515 ?) -> fence_driver_fini -> wakup_up -> kaboom
benh: probably hitting something not initialized yet :-)
benh: I'll dig later
edgecase: hmm should i expect KMS to do display hotplug? xrandr doesn't detect when i plug in VGA-0
edgecase: (II) RADEON(0): Allocate new frame buffer 1024x768 stride 1024
edgecase: (II) RADEON(0): VRAM usage limit set to 9676K
edgecase: so many interesting things
benh: edgecase: if you do xrandr --output VGA0 --auto ?
Ghworg: VGA hotplug takes too long to detect so it doesn't do it. I think digital outputs can do hotplug but not analog
benh: thinks that's the syntax
benh: Ghworg: yeah my experience is also that even with digital, it can be fairly unreliable
edgecase: well it used to detect VGA-0 each time i run xrandr or grandr etc
benh: especially with external TMDS
edgecase: well it's polling right, not event (interrupt) driven, so that's to be expected i guess
Ghworg: I could be wrong, this is just my recollection from conversations I've seen here
edgecase: yeah i should have tried that before asking, but i'm just so giddy right now with this new KMS toy thingy i just installed :)
edgecase: vt switch is so *fast* now!
edgecase: someone should make it rotate cube O.o
benh: airlied: does offb -> nouveau kms takeover work ?
edgecase: so is alt-fn-Fx handled by xserver or kernel? can i vt switch to console if X dies?
Wizzup: edgecase: yes, afaik
chithead: edgecase: you can ssh in and run "chvt 1"
edgecase: need a sys-rq key for vt switch
chithead: you can also use acpid to bind power button event to vtswitch
Ghworg: X can lock the input devices when it dies, in which case ctrl-alt-f1 on its own won't work
edgecase: chithead, good idea!
Ronis_BR: hi all
Ronis_BR: airlied: I have switched from tuxonice to default kernel hibernation and I'm still having the same problem
Ronis_BR: resume doesn't work... X becomes a mouse a grub splashscreen after resume
Ronis_BR: just like tuxonice
Ronis_BR: and I see tons of messages like:
Ronis_BR: [drm:radeon_ib_schedule] *ERROR* radeon: couldn't schedule IB(9).
Ronis_BR: [drm:radeon_cs_ioctl] *ERROR* Failed to schedule IB !
Ronis_BR: can anyone help me}
Ronis_BR: ?
edgecase: ok it seems xrandr expands the desktop, but doesn't enable the VGA-0 output
edgecase: VGA-0 disconnected
eosie: r300 can't render to a zbuffer only, can it?
eosie: with no colorbuffer set
edgecase: is there another kernel module needed for EDID to work ?
edgecase: Allocate new frame buffer 2048x768 stride 2048 <-- sweet!
Ronis_BR: man, can anyone make hibernation works with KMS+ radeon driver?
Ghworg: Worked fine last time I tried it
Ghworg: Can't do it right now 'cos I'm using 64-bit kernel with 32-bit userland, and hibernate doesn't support that combo (though suspend does strangely)
Ronis_BR: Ghworg: what are you using? uswsusp?
Ghworg: Yes
Ronis_BR: man, help me :)
Ronis_BR: Ghworg: I have tried tuxonice and uswsusp with the same results
Ronis_BR: [after resume, if KMS is enable, i got X with grub splash screen and a mouse
Ronis_BR: the computer is usable
Ronis_BR: but X isn't anymore, no matter if I restart it
Ronis_BR: suspend works fine
Ghworg: Ronis_BR: I really don't know what to suggest. I didn't do anything to make it work, it just works out of the box for me. Sorry
Ronis_BR: hum Ghworg what xorg version}
Ronis_BR: and what card
Ghworg: Ronis_BR: 1.6.5 and rv630
Ronis_BR: I'm using 1.6.5 also
Ronis_BR: Ghworg: I just don't have ideas what can be wrong....
Ronis_BR: Ghworg: can you pastebin your /etc/hibernate/common.conf
Ronis_BR: Ghworg: and /etc/hibernate/ususpend-disk.conf}
Ghworg: Ronis_BR: At least now you are not using a kernel patch you can file a bug
Ronis_BR: Ghworg: yes
Ghworg: Ronis_BR: I don't have a /etc/hibernate dir
Ronis_BR: hum
Ronis_BR: so you are probably using pm-utils
Ghworg: Yes
Ronis_BR: I'm affraid it is something with some library that I need to rebuilt here
Ghworg: Ronis_BR: The only setting in my pm config is SLEEP_MODULE="uswsusp", so not doing anything special there
Ronis_BR: hum I see
Ronis_BR: I'm using hibernate-script but I don't think it is the problem
Ronis_BR: since it is happening when I try s2disk
Ronis_BR: Ghworg: what distro are you using?
Ghworg: Ronis_BR: Debian testing/unstable
Ronis_BR: Ghworg: does your kernel have fbcondecor?
Ghworg: Ronis_BR: Not sure, how do I check? Is that a module or kernel opt or what?
Ronis_BR: it decorates your console (tty) :D
Ronis_BR: the kernel option is CONFIG_FB_CON_DECOR or something like that
Ghworg: Ronis_BR: Can't see anything like that in my kernel opts either way. I'm guessing no, since I just use plain text consoles no graphics or splash or anything like that
Ronis_BR: hum ok
Ronis_BR: Ghworg: but KMS is enable right
Ghworg: Ronis_BR: Yes
edgecase: ok with KMS, i have no way to use output VGA-0 (laptop uses LVDS by default)
Ronis_BR: ok, thanks Ghworg
edgecase: maybe I need extra Monitor sections?
edgecase: Output VGA-0 disconnected
Ronis_BR: Ghworg: do you know if i should run vbetool or something}
Ghworg: Ronis_BR: For hibernate you shouldn't have to I think, I've only seen that in reference to suspend. But I'm not an expert in this
Ronis_BR: ok thanks
airlied: benh: yup shipping it in F12, works fine on my g5
edgecase: in don't get EDID data for LVDS either
edgecase: am i missing firmware? i2c module?
Ronis_BR: airlied: man, have you read my problem above?
Ronis_BR: airlied: as you said, I remove tuxonice, but resume problem still hapenning
Ronis_BR: airlied: do you think I should open a bug?
cxo: ok this *ERROR* Failed to allocate GEM object error is driving me nuts. Does anyone have a commit id that is definitely free of this issue?
airlied: eosie: it sounds like something we might bar in the kernel
airlied: but might be possible, since you could just set the FP not to render anything to the CB
airlied: Ronis_BR: have you tried with an F12 LiveCD?
airlied: probably worth opening a bug
airlied: my laptop hibernates fine
airlied: cxo: go back before glisse TTM changes
edgecase: my drm module seems to find the connector table OK, no EDID output from kernel, and xorg can't do EDID
eosie: airlied: yeah I am getting "bad CS". Is it a kernel-only issue or a hw limitation?
Ronis_BR: airlied: no I don't
Ronis_BR: airlied: I'll take f12 livecd and i'll try
Ronis_BR: airlied: do you think vbetool can somewhat solve my problem
Ronis_BR: airlied: I saw many posts with problems very close to mine
airlied: eosie: I think the kernel expects you'll set a colorbuffer up
Ronis_BR: that should be already fixed....
airlied: eosie: though num_cb is setup accoridng RB3D_CCTL
airlied: so it should be okay
edgecase: seems R100 KMS doesn't handle multiple monitors yet?
airlied: eosie: should do works here at least a week or so ago
airlied: oops ^^ edgecase
edgecase: ok maybe i need fresher pkgs
airlied: eosie: actually num_cb is set to that value + 1
airlied: edgecase: its worked forever here
edgecase: airlied, what evidence would I see, kernel msgs, or xorg logs?
airlied: edgecase: dmesg and xrandr output
eosie: airlied: however RB3D_CCTL doesn't appear to have an option for 0 cbufs
edgecase: this is laptop M7
airlied: yeah it looks like we don't detect 0 cbs, how do you tell the hw there is 0?
airlied: edgecase: same as mine, but I haven't updated it in a week or two, too slow
eosie: I have no idea, but it's quite common to do a depth-only rendering in GL
edgecase: my dmesg shows for VGA: [drm] DDC: 0x60 0x60 0x60 0x60 0x60 0x60 0x60 0x60
edgecase: with monitor plugged in
airlied: edgecase: try booting with monoitor it and drm.debug=1
airlied: don't start X though
airlied: since you'll lose the important dmesg info
edgecase: ok, i was wondering where to enable debug output
edgecase: those don't look like valid EDID tho
eosie: airlied: shouldn't we ignore the cb setup if color_channel_mask is 0 and fastfill is disabled?
airlied: eosie: it sounds sane
airlied: do you know if CCM affects blending readback
eosie: airlied: in my opinion CCM comes after blending
Ronis_BR: [drm:r600_resume] *ERROR* r600 startup failed on resume
Ronis_BR: :(
eosie: airlied: we should probably check whether blending is disabled too
airlied: eosie: so probably need to make sure blending is off (READ_ENABLE bit in BLENDCNTL
Ronis_BR: airlied: what can cause r600_startup(rdev); to fail?
airlied: AGP failure maybe if its AGP it doesn't print anything else?
Ronis_BR: airlied: mine is PCIE
Ronis_BR: airlied: if I switch tty1 from X, it starts to show thousand of messages like
Ronis_BR: [drm:radeon_ib_schedule] *ERROR* radeon: couldn't schedule IB(9).
Ronis_BR: [drm:radeon_cs_ioctl] *ERROR* Failed to schedule IB !
Ronis_BR: no, it is different, if I go to X from tty1 is when these messages are shown
airlied: nothing about any other fialure before the starup failed on resume?
Ronis_BR: I just see
Ronis_BR: [drm:r600_ring_test] *ERROR* radeon: ring test failed
Ronis_BR: and some numbers
airlied: ah okay then the ring test failed
edt: airlied would you know if this commit 'drm/radeon/kms: fix RS600 MC setup / 64bffd03756249e11b8651ccf33ac3a50a93ed4c' problem was to fix font corruption. If so its probably needed on rs780 too. drm-linus, drm-radeon-testing both have font corruption issues here.
Ronis_BR: even if X is shutdown before hibernate this error continues
airlied: edt: nothing to do with any other chip
airlied: hence the rs600 setup
airlied: just building on my rs780, see if I can get it to happen
Ronis_BR: mine card is RS780
Ronis_BR: HD3200
edt: OK any idea what might be corrupting fonts on 2.6.32 + thursday's for-linus or drm-radeon-testing?
edt: I'm using a 790gx chipset with a builtin HD3300 (rs780)
Ronis_BR: actually RS780M/RS780MN don't know if it makes any difference
Ronis_BR: edt: I have no issues about rendering here
Ronis_BR: 3D, fonts, everything is veeeery fine
Ronis_BR: the only problem is the resume
evil_core: I got intercalcxing like effect in fullscreen XV mplayer in some movies
evil_core: no problem w/o xv or in window
evil_core: what can cause it?
evil_core: even vertical
evil_core: in code monkyeys
edt: its strange in that the konsole fonts are ok. Every other font on the screen is messaged up. Miss chars, lots of artifacts same color as the font should be where chars should be
Ronis_BR: edt: what xorg version/?
Ronis_BR: edt: I got wierd things when I upgraded to 1.7.1 sometime ago
edt: 1.7.3 with git ddx, libdrm, mesa
edt: Just rolling back the drm-next changes fix the font problems
Ronis_BR: hum
airlied: sounds like bisection time
edt: I was hoping you would not say that
edt: lop sided grin
Ronis_BR: edt: bisection time?!
edt: airlied which tree should I use for the bisection?
edgecase: airlied, i have dmesg output, nothing jumps out at me http://pastebin.ca/1713927
airlied: edt: drm-radeon-testing should be bisectable to the point
airlied: I'm just going to boot here in a few mins, have to rebuild kenrel
Ronis_BR: airlied: did you have any ideas how can I solve or help to solve my problem?
edt: ok expect something - probably tomorrow as its getting late here...
Ronis_BR: airlied: maybe try radeonhd?
airlied: Ronis_BR: with KMS it doesn't matter
airlied: rhd doesn't support kms
Ronis_BR: ah
Ronis_BR: ok
Ronis_BR: without KMS it is working
Ronis_BR: airlied: is r600_startup the right function for my card?
airlied: Ronis_BR: yes should be, sounds like resume isn't initialising something righgt
Ronis_BR: hum
Ronis_BR: airlied: I don't know if it helps, but suspend works fine
TCW: can one say that the ati gfx _hardware_ is of less quality than nvidias?
airlied: TCW: not really they both have thier shit points
airlied: Ronis_BR: so that failure was with F12 LiveC?
Ronis_BR: airlied: no, i'm downloading F12 livecd
Ronis_BR: airlied: it was with gentoo, using uswsusp and tuxonice, always the same thing
airlied: lets see if F12 works, hibernation isusually something that needs distro help
Ronis_BR: airlied: yes, I'll test.
Ronis_BR: airlied: but, does F12 livecd will let my hibernate? ?D
Ronis_BR: :)
airlied: it should do but thats a good q I've never test it
Ronis_BR: ok :) just need a swap right
Ronis_BR: airlied: ah, is it any command of procedure that need to be done at resume? I'm using hibernate-script right now
airlied: Ronis_BR: not sure, hence why I'd like to see if F12 works
Ronis_BR: ok
Ronis_BR: airlied: man, thanks! I'll get some sleep, tomorrow I'll tell you if it had worked
Ronis_BR: night
airlied: though you might be right in that the F12 livecd won't let you test
Ronis_BR: airlied: probably
Ronis_BR: and to install fedora here isn't a good option right now :)
airlied: I'll see if I can hibernate my rs780
edt: first of about 7 kernels building...
Ronis_BR: airlied: I can wait if you need any help
airlied: I have to open the machine, has a gpu in it so it'll a while
Ronis_BR: ok
edgecase: airlied, does drm put EDID info into /sys ?
airlied: edgecase: yes when it reads it
Ronis_BR: airlied: can you tell me if fedora has any scripts on /etc/pm/sleep.d
Ronis_BR: maybe there was a quirk or something that make it works
airlied: nope no scripts
Ronis_BR: ok
edgecase: airlied, sysfs shows no edid for VGA-1
Ronis_BR: airlied: does it can be a kernel option that I shouldn't have enable?
airlied: Ronis_BR: unlikely, I've just no idea how hibernate works on other distros
airlied: its a horrible mess thanks to crazy maintainers
Ronis_BR: airlied: :D
Ronis_BR: airlied: I thougth it is just a call to s2disk
Ronis_BR: at "Support for frame buffer devices" I have Enable firmware EDID and Enable Video mode Handling Helpers
Ronis_BR: is it necessary??
Ronis_BR: it cames from the time I was using uvesafb
airlied: Ronis_BR: shouldn't matter
Ronis_BR: ok
airlied: my rs780 failed to resume here, not sure why
airlied: so at least I can reproduce
Ronis_BR: airlied: same message?
Ronis_BR: ring test failed I think
airlied: doesn't come back up on the network
Ronis_BR: hum
Ronis_BR: I'll try some quirks of pm-utils to see If i can get something differente
airlied: thouhg hibernate should be easier since the GPU is already setup by the BIOS
Ronis_BR: yes
edgecase: airlied, do I need to load module fb_dcc.ko before radeon perhaps?
airlied: edgecase: no we don't use that stuff
edgecase: is this too old: xorg-server 2:1.6.5+git20091107+server-1.6-branch.2dbcb06a-0ubuntu0sarvatt~karmic ?
airlied: if the kernel can't find DD Cthe userspace won't
edgecase: radeon driver 6.12.99
airlied: no matter what it is
edgecase: ok so kernel then
edgecase: it's 2.6.32-7-generic #10-Ubuntu SMP Sun Dec 6 13:43:20 UTC 2009 i686
edgecase: no idea what patches they put in there
edgecase: airlied, what kernel branch has the latest radeon kms? i guess i should try that next
airlied: edgecase: drm-radeon-testing of my git tree
edgecase: k thx
airlied: anyone awake seeing the bo creation failure?
airlied: hmm fix diesn't work anyways
edt: I am not having any luck bisecting. What is the correct process to use when you want to bisect a tree that needs to be merged into another to get a working system?
edt: bisect seems to lose its place when you checkout a second tree
edt: and merge the second
airlied: edt: yeah that gets messy
edt: bisecting the merged tree does not work well either
edt: manual bisect I guess?
airlied: edt: any ideas when it worked, git log drivers/gpu/drm from then might provide a point to start
edt: it stopped working sometime after last weekend - I use kms, early start and last weekend IRQs was added and worked just fine. Did not try during the week at all. Friday evening it was broken along with the for-linus tree (with the same symptoms)
edt: it was 'fun' figuring out how to convert to ihex and build a patch for the firmware (grin)
edt: I know its bad at f2b115e69d46344ae7afcaad5823496d2a0d8650
airlied: might want to try before ca262a9998d46196750bb19a9dc4bd465b170ff7
edt: its good at aa1a750ecb3412f69fe34081b249aa978154f360
edt: the corruption is the cleanest I've had here
edt: just the fonts get missed up and even then sometimes they come out ok (5-10% ?)
edt: sleep time here. I will try around ca262a9998d46196750bb19a9dc4bd465b170ff7 tomorrow. Thanks.
edt: a thought. Is ca262a9998d46196750bb19a9dc4bd465b170ff7 in the for-linus stuff?
edt: if not then its not the problem
benh: airlied: ok so whet get_bios fails on r5xx we crash when unwinding
airlied: benh: probably not suprising nobody tested that
benh: airlied: right
benh: airlied: we're half uninitialized but we call fini() and the whole shebang
airlied: I fixed some issues on r100 with biod failing
benh: airlied: which blows up
airlied: benh: glisse has some patches recently that might fix that
airlied: probablyt the patch that caused it actually
airlied: cf0fe4566dcc0c5bd9b7da8c9a53e712593db118
benh: airlied: some stuff is really cdonfused
benh: airlied: such as using r520_init from r520.c but rv515_fini from rv515.c :-)
airlied: I had someone test it on an rn50 but it looks liek that patch will regress it
airlied: benh: yeah they shutdown the same
benh: airlied: so the thing is
benh: all the _fini functions more or less assume that everything is up
benh: it's going to be a mess to untangle :-)
airlied: previous to that "fix" it looks like we just left stuff to leak
benh: yes
benh: which is fine really
benh: better leak than crash :-)
airlied: probably should ping glisse, I'd consider reverting that patch
benh: yup
benh: Ill catch him tonight
benh: glisse: still asleep ? WAKE UP !
benh: kicks glisse between the ears :-)
cxo: kicks too for firefox crashes
airlied: I've already kicked him, hes causing too many regressions ;-)
benh: hehe
airlied: cxo: is that the allocation failures?
benh: non of the _fini code is called conditionally
benh: and I don't see much testing inside those function to see if requested stuff is there
benh: no wonder it blows up :-)
cxo: airlied, yead
cxo: yeah
airlied: cxo: trying to fix that now
benh: probably need a bitmask of parts of the driver that have been inited :-)
airlied: benh: or just make it safe to know which objects are created etc
rhodan: What will the xorg state tracker do for us?
airlied: damn, now I lose the network when it dies
airlied: rhodan: replace the DDX i ntheory
airlied: so far last I tried it just renders crap with radeon
rhodan: :(
rhodan: Ok, but when gallium replaces the DDX, will 2D be faster?
airlied: doesnt' have his magic 8 ball
rhodan: Oh, could've been that there was an obvious speed benefit.
airlied: in theory with state/shader caching it might be
airlied: but really history has proven measurement is the only truth
benh: airlied: yes, but there's so much crap in that init sequence
airlied: like I think it *should* be faster, but I can guess it won't be due to something else
benh: airlied: that a list of flags might be the easiest way our :-)
benh: out
benh: in that case, it blows up bcs the fence isn't initialized
benh: and fence_fini does wake_up on an uninitialized wq
airlied: the wq should be init first
benh: airlied: yes
benh: airlied: but we don't go through fence_init at all if we miss on the BIOS stuff :-)
benh: airlied: maybe we should split all the "init" from "setup" :-)
benh: or "init" from "start"
rhodan: By the way, how come 2D performance is a performance problem at all? If 3D runs well, shouldn't 2D compositing be blazing fast?
airlied: benh: ah yes
airlied: rhodan: who said it was aproblem?
rhodan: airlied: for me, it is :§
airlied: (apart from firefox scrolling I think is still crappy)
airlied: but thats because firefox is made of stupid assumptions
rhodan: Well, moving composited windows should have 60fps, always.
airlied: thats not 2D
airlied: or are you using Xrender compositng?
rhodan: No, OpenGL.
airlied: then its not 2D
rhodan: I see.
airlied: moving windosw with a compositor is purely 3D
rhodan: Is that good or bad?
rhodan: Or will *that* get sped up? :)
airlied: depends on the operation the compositor users
airlied: and whether mesa driver does it fast
rhodan: Like, software fallback?
benh: airlied: ok, it loads the ATOM bios now, and crashes later :-)
benh: in the MSI code ... probably not your fault
rhodan: Or is that something entirely different? I thought that when an OpenGL function isn't supported by the driver, it gets done in software. Is that true or am I mistaken?
airlied: rhodan: it could be doing a sw fallback
airlied: depends on what the operation is
rhodan: I wonder if kwin does that.
rhodan: Is there a way to find out?
airlied: ask a kwin dev how it moves windows under GL
airlied: then ask us what would be slow about that
rhodan: It's just a polygon move
airlied: yeah it should be fast really since it just re-renders
airlied: the content is static
rhodan: And some shadow drawing.
rhodan: Even if I do nothing at all, I only get 40fps.
rhodan: Moving windows, 20fps.