agd5f: dileX: you need to set dynpm=1 to enable pm
dileX: agd5f: OK. which CONFIG*I2C* do I need - especially for debugging?
dileX: (reboot)
agd5f: dileX: don't need i2c for the pm stuff
dileX: agd5f: I get again vertical stripes in VT - even w/ dynpm=1
agd5f: dileX: if you get them without dynpm=1, you are hitting some other bug
agd5f: or do you only get them with the patches applied?
agd5f: if you only get them with the patches, can you bisect?
dileX: agd5f: without dynpm=1 I get a black screen
agd5f: dileX: does radeon-testing work ok without the patches?
dileX: linus-tree w/ d-r-t on top is OK
agd5f: dileX: can you bisect?
dileX: agd5f: OK. I do
agd5f: thanks
Nightwulf|work: hi all
dileX: agd5f: good: 0005-drm-radeon-kms-add-gui_idle-callback.patch // bad: 0006-drm-radeon-kms-add-support-for-gui-idle-interrupts.patch
twnqx: airlied: alex already fixed my issue in another bug ;)
twnqx: not to get working 2D out of KMS on r600
twnqx: now*
BeteNoire: darn
BeteNoire: after another reboot i can't get my login screen
Tommeh: R600?
Tommeh: +KMS?
BeteNoire: it's all gray, shivering
Tommeh: heh, weird.
BeteNoire: i had to start in single mode and manually start services
BeteNoire: Tommeh: dunno whats r600 but its x200m
Tommeh: KMS + R600 + compositing still causes some pretty weird problems for me with dual monitors. Almost exclusively with the second monitors.
twnqx: KMS + R600 still causes gpu lockups for me - without compositing or dual monitors :P
twnqx: wonders what enlightenment 17 does different
Tommeh: Thinks are fine without the compositing. But I have to revert to UMS to get compositing + dual monitors working successfully.
Tommeh: s/Thinks/Things/
BeteNoire: ums?
BeteNoire: (WW) xf86CloseConsole: VT_WAITACTIVE failed: Interrupted system call
BeteNoire: what's this?
mjt: that message quite frequently pops up here, esp. after suspend/resume cycle
mjt: after its apperance video does not work anymore, only reboot fixes it.
BeteNoire: so, the compositing is the reason?
mjt: i don't use compositing
BeteNoire: i need my terminal transparent
spreeuw: no you dont
dileX: BeteNoire Tommeh: which kernel are you using? where are those GPU lockups - in VT or X? tried latest ddx? checked bug-tracking-system? looked for fixes on dri-devel? tried glisse's kernel-tree?
Tommeh: I see no lockups etc. it works, just some problems with the 2nd screen. And only when compositing. Flickering + overly high contrast on the 2nd screen.
Tommeh: It's fine with UMS + compositing and KMS w/o compositing. It's just KMS + compositing.
Tommeh: Which interestingly, was the only time I required R600_rlc.bin to be loaded.
Tommeh: I'm using xorg-edgers PPA which are basically git releases, pre-packaged for Ubuntu.
Tommeh: Releases probably isn't the right word. They're just periodic packages from the git code.
Tommeh: 2.6.34-rc1, too.
BeteNoire: dileX: 2.6.32, when starting x, what is ddx?
twnqx: ddx is the display driver
dileX: xf86-video-ati
BeteNoire: those names
BeteNoire: i have 6.12.192
dileX: BeteNoire: for r600 you better should use >=2.6.33 or drm-radeon-testing GIT branch
twnqx: aaand they don't help!
twnqx: :S
Tommeh: BeteNoire: don't forget to include R600_rlc.bin firmware, too.
dileX: wiki (see topic) has lots of informations, hints, troubleshooting
BeteNoire: dileX: lspci says x 200m is rs482
kjeldahl: I'm getting intermittent hangups which recover after a few seconds/half a minute several times a day. On Ubuntu Lucid alpha system (Kubuntu, kernel 2.6.32-16) using xorg-edgers with Radeon RV710/730 (HD 4650) with a dual screen setup. Any ideas on why or how to get around it?
kjeldahl: Apart from the hangups which recover, everything seems smooth, even composting and 3D effects.
Tommeh: kjeldahl: I would suggest using a newer kernel from the kernel PPA
kjeldahl: Tommeh: Any easy to add ppa available?
Tommeh: Mm, I've not found a specific line
Tommeh: But .deb's are available
Tommeh: And they work fine.
Tommeh: Just google 'kernel PPA', should be the top result.
Tommeh: I'm currently running 2.6.34-rc1 on Ubuntu Karmic with the xorg-edgers repos fully up to date.
Tommeh: R600 card mostly works. :)
kjeldahl: Tommeh: Thanks, I'll ses if I can give it a try. I hate installing packages manually though.
Tommeh: Note you'll need the R700_rlc.bin file in /lib/firmware/`uname -r'/radeon/
edwin: boo, git mesa fails to build
Tommeh: kjeldahl: it's only three packages per kernel. Headers for all, headers for your arch (i386/x86_64) and then a kernel image for your arch.
edwin: http://rss.slashdot.org/~r/Slashdot/slashdot/~3/B9kcrJYl-EU/The-Problems-With-Video-Game-Voice-Acting
BeteNoire: eerm, seems that my card is not r600 dileX
edwin: err wrong paste
edwin: state_tracker/st_cb_texture.c:903: error: ‘struct pipe_screen’ has no member named ‘tex_transfer_destroy’
edwin: ^this is the build error :D
suokko: BeteNoire: yo usohuld stil ltry 2.6.33 for x200m because that card has some special problems
dileX: BeteNoire: anyway, give a new kernel a try. radeon OSS development is still WIP.
suokko: edwin: Looks like gallium error
MostAwesomeDude: edwin: One second, fixing.
edwin: this commit broke it 63af29bfbe265318bcf5be69e420de361b900321
BeteNoire: i'm waiting till they pull 2.6.33 of tuxonice to portage
BeteNoire: or should i just use gentoo-sources
edwin: what do you need tuxonice for?
BeteNoire: hibernation
BeteNoire: once it worked on my laptop :P
edwin: doesn't it work with latest kernels w/o tuxonice?
edwin: I thought tuxonice is just an alternative implementation
MostAwesomeDude: edwin: Fixed and pushed.
edwin: MostAwesomeDude: thanks
kjeldahl: Got 2.6.34 working. Still the same temporary hangups where gui becomes unresponsive, then suddenly recovers again. No artifacts, things just "hang".
kjeldahl: dmesg
kjeldahl: sry, wrong window (it's hanging again)
suokko: kjeldahl: does dmesg have something there?
kjeldahl: Nope, everything looks fine. It confirms loading the firmware files etc.
Tommeh: kjeldahl: specifically R700_rlc.bin?
Tommeh: As there are some that are already included.
Tommeh: Also, are you running with radeon.modeset=1 on your boot line?
kjeldahl: Regarding firmware, yes:
kjeldahl: [ 10.971994] platform radeon_cp.0: firmware: requesting radeon/RV730_pfp.bin
kjeldahl: [ 11.162027] platform radeon_cp.0: firmware: requesting radeon/RV730_me.bin
kjeldahl: [ 11.235608] platform radeon_cp.0: firmware: requesting radeon/R700_rlc.bin
kjeldahl: Regarding modeset, probably not (not seeing it in grub).
Tommeh: I think the default is 0
Tommeh: Might be worth forcing it on your grub boot line.
kjeldahl: I'll give it a try. Thanks.
Tommeh: Alternatively - try with 0, as it might be KMS causing the problem :)
Tommeh: But either way, try forcing it.
Tommeh: I personally have two boot options setup for each kernel I try currently.
dileX: agd5f: Tried "[PATCH] drm/radeon/kms: add hw_i2c module option" on top of pm2-patchset. 'modprobe -v radeon modeset=1 hw_i2c=1 dynpm=1' gives a black screen. unfortunately, no message in /var/log/ according to hw_i2c module-param. with hw_i2c=0 vertical stripes in VT. shouldnt there be a msg in the log-files?
MrCooper: airlied, glisse: unfortunately netconsole doesn't seem to say anything about those deadlocks
suokko: MrCooper: Did you build kernel with lock debugging configs enabled?
dileX: (forgot to depmod?)
MrCooper: suokko: CONFIG_DEBUG_{SPINLOCK,MUTEXES,LOCK_ALLOC}=y, any others?
suokko: CONFIG_LOCKDEP_SUPPORT=y and CONFIG_LOCKDEP=y?
suokko: I have CONFIG_DEBUG_LOCKDEP disabled but it might be usefull for debugging purpose
MrCooper: same here
legume: suokko: Is it you that did the last two commits to libdrm?
suokko: legume: yes
legume: suokko: OK - --disable-intel doesn't work anymore, which causes autogen to fail for me.
suokko: legume: What is the error?
suokko: It works for me
legume: suokko: I could probably work around but enabling atomic ops on my old LFS/GCC actually cost perf last time I did it.
legume: suokko: configure: error: libdrm_intel depends upon atomic operations
suokko: legume: aha. error checking was not completely correct
suokko: legume: but radeon started to depend on atomic operations too
legume: suokko: Hmm, that's a shame - not that I really know what I am talking about/doing here. But I do know -march=native cost me a bit of perf last time I used it, gcc 4.2.3 + athlon xp 2500.
suokko: legume: What arch are you using now?
kjeldahl: Tommyh: Running with radeon.modeset=1 now. Just a few minutes in, but no slowdowns yet. Keeping my fingers crossed. Thanks.
suokko: I'm using athlon xp too
legume: suokko: i686 as far as I know :-)
MrCooper: legume: you can always install libatomic ;)
legume: MrCooper: thanks, I'll have to look into that.
MrCooper: libatomic-ops even
suokko: legume: For me i686 configures correctly
suokko: legume: i386 fails
kjeldahl: Hang/slowdown again.. dmesg says "[ 142.333082] [drm:radeon_cs_ioctl] *ERROR* Failed to parse relocation !"
suokko: legume: even i486 works
Tommeh: kjeldahl: bizarre. :(
legume: suokko: Maybe it's just the way LFS sets things up, or GCC version or something.
Tommeh: Beyond my comprehension to be fair.
kjeldahl: will try again with modeset=0
Tommeh: kjeldahl: are you using the drivers-only xorg-edgers repo?
Tommeh: If so, use the full one, not drivers-only.
kjeldahl: Ok, will try. Thanks.
suokko: legume: Might be. I have gcc 4.4.1
suokko: legume: Can you try to configure with CPPFLAGS=-march=i686?
Nezmer: hi, I built mesa,xorg-server and xf86-video-ati yesterday all from git/master on FreeBSD 8/stable AMD64. Unfortunately, X doesn't work and I have this in my log:
kjeldahl: Tommeh: Choices are "less fresh more stable" or "drivers-only". You're suggesting I use "less fresh"? https://launchpad.net/~xorg-edgers/+archive/ppa
Nezmer: (EE) AIGLX error: dlopen of /usr/local/lib/xorg/modules/dri/r600_dri.so failed (/usr/local/lib/xorg/modules/dri/r600_dri.so: Undefined symbol "_glapi_tls_Context")
Nezmer: Is this expected?
legume: suokko: I expect that will work - before --disable-intel was implemented I had to do that, (well -march=native IIRC) just that building with that seemed like it cost a bit of perf.
kjeldahl: Tommeh: Sorry, never mind. Saw "adding this ppa" later down on that page. I will try that.
Tommeh: kjeldahl: yes, 'less fresh'
Tommeh: It includes more than jut DDX updates.
dileX: MrCooper suokko: "drm module loaded with debug=15" - 15 is a higher verbosity-level (I was used to debug=1)?
suokko: legume: If it is costing performance it sounds like bad bug in gcc optimization. But I'm anyway fixing the error message if disabling intel or radeon
MrCooper: dileX: can't have DRM debugging output enabled all the time
adamk: Nezmer: Yes, it's fully expected that the drivers shouldn't work :-) That's a pretty silly question.
adamk: Nezmer: It seems likely that you have something incorrectly installed or configured.
adamk: Do you have fglrx installed on that machine or did you ever have it installed there?
Nezmer: adamk: I meant expected on FreeBSD
dileX: MrCooper: I know its writing lots of msgs into log. I just wanted to ask if I get more detailed information when increasing from debug=1 to debug=15
suokko: Nezmer: --enable-glx-tls when configureing mesa
adamk: Ahhhh, definitely not.
adamk: That option certainly isn't needed here on FreeBSD.
MrCooper: dileX: no idea, possibly
adamk: Nezmer: Using the versions in ports or something newer?
suokko: dileX: It is not more verbose but different channels
dileX: suokko: what do you mean by "channels"?
Nezmer: adamk: From git/master. Should I disable glx-tls ?
legume: suokko: just to confirm CPPFLAGS=-march=i686 does work OK both with and without --disable-intel
adamk: Nezmer: I built Mesa and libdrm from git yesterday on -CURRENT. I used: ./configure --with-expat=/usr/local --prefix=/usr/local --with-dri-driverdir=/usr/local/lib/dri/ --with-dri-drivers=r300,r600,swrast --disable-egl --disable-gallium
Tommeh: kjeldahl: In fact, thinking about it - I use the bleeding edge updates and I don't have a problem. You should probably try the latest (freshest) releases from xorg-edgers.
adamk: Nezmer: Well,rather, that was for Mesa.
adamk: Nezmer: I didn't do anything other than ./configure for libdrm.
suokko: dileX: 0x1 is drm core, 0x2 is driver, 0x8 is kms. You then sum the bit mask to enable debuggin that you want to see
Nezmer: adamk: Ok , I'll try that. I hope -CURRENT isn't needed though
suokko: I don't remmber if the numbers are realy correct without looknig source
adamk: Nezmer: -STABLE or -CURRENT
adamk: Nezmer: 8.0-RELEASE will not work.
suokko: and I don't remmeber what is the 4th channel
adamk: Nezmer: If you are using the latest ports tree, you can always build 7.6.1 from ports and that will provide 3D acceleration.
Nezmer: adamk: I have -STABLE from Feb25. I'll upgrade then try your settings
adamk: Nezmer: But mesa from git has some improvements.
adamk: k
dileX: suokko: hey thats worth to note in wiki :-) where can I look for that debug-modes (source-code) - maybe from memory?
suokko: dileX: include/drm/drmP.h in kernel source
suokko: dileX: and kms is 0x4 :)
dileX: #define DRM_UT_CORE 0x01
dileX: #define DRM_UT_DRIVER 0x02
dileX: #define DRM_UT_KMS 0x04
suokko: yes
dileX: suokko: ok thats 0x07. you said sum up with bit-mask to enable drm-debugging? value for this?
dileX: #define DRM_FLAG_DEBUG 0x01
dileX: #define DRM_FLAG_DEBUG 0x01
suokko: dileX: the drm.debug takes the DRM_UT values. So if yo uwant driver+kms thne it is 6
dileX: suokko: yupp, sometimes its worth reading comments
dileX: or in other words: DRM_DEBUG: 0x01 | DRM_DEBUG_DRIVER: 0x02 | DRM_DEBUG_KMS: 0x04
suokko: dileX: yes
dileX: suokko: so drm.debug=0x07 will catch all DRM debug-levels
suokko: dileX: yes. Unless there will be new aditions
dileX: suokko: gotcha :-) you are planning new additions, haha
kjeldahl: Tommeh: Still running with modeset=1, but now with the "normal" (full) xorg-edgers. Hangs/X eating 100% CPU seem less frequent. Will also try modeset=0. Thanks again.
kjeldahl: Tommeh: Overall, system also feels more snappy, but that's based on "feel" alone.
dileX: suokko: what you think about a debug paragraph to wiki?
dileX: maybe you write a bit about investigating locking etc.
dileX: +can
dileX: (reboot)
suokko: dileX: I think we should have separate troubleshooting/debugging wiki pages
dileX: suokko: yeah - its long enough
kjeldahl: With radeon.modeset=0, X swapped my two screens. So obviously it's doing _something_. ;-)
AndrewR: suokko, hello!
Nezmer: adamk: FYI About FreeBSD. I think the issue was having "--enable-radeon" in drm config. Now It works without the radeon kms lib.
adamk: Ahhhh, interesting... If I just run ./configure it shows that it is building with the Radeon API, even if I don't specify --enable-radeon.
adamk: But since KMS isn't supported on FreeBSD anyway, I guess it doesn't really matter either way :-)
suokko: AndrewR: gi
suokko: hi
Tommeh: kjeldahl: sounds like you're having fun! :)
AndrewR: suokko, i found what OA (at least 0.7.0) CPU-limited on my system, always around 13-14 fps, in 640x480 and 1152x864 .... But turning on maximum texture quality (slider, not bi/tri linear ) always resulted in aborted anholt benchmark, any resolytion. less textures - and it is passed ....
dileX: suokko: ah, now I know where this "15" comes from
dileX: (I remember airlied told me so)
AndrewR: suokko, and wine-1.140 completely broken for me on r200/mesa 7.9, it simply stops in the middle of 3dmark2001 test (any "high quality" tests).....1.1.35 works .. i saw your wine bug, not sure if adding yet another one will help.....
suokko: AndrewR: OA for me at 640x480 at high quality settings runs 40-45 fps
suokko: AndrewR: If it is regression and you could bisect the wine commit that broke everything it would help a lot
mjg59: airlied: Solid window moves are *slow* on this es1000
mjg59: Everything else is fairly snappy
mjg59: But we're talking about a second to move a terminal from one side of the screen to the other
suokko: mjg59: without composite?
mjg59: suokko: Correct
glisse: mjg59: did we disable accel on those ?
glisse: xorg log should tell
mjg59: glisse: "Driver will not allow EXA pixmaps in VRAM", but otherwise it seems enabled
mjg59: To the extent that accelerated solid, copy, UTS and DFS is acceleration
glisse: i wonder if shadowfb would be preferable on those
ajax: almost certainly.
ajax: not least because i don't trust the CP on that thing further than i can throw it.
glisse: i might propose a patch to disable accel on those :)
ajax: it might be cute to shadowfb them with DMA for the upload
mjg59: ajax: Ohhey can I borrow a pile of random r500/600s?
ajax: mjg59: sure. back to poking at reclock i assume?
mjg59: ajax: Yeah
mjg59: Also, thermal
AndrewR: suokko, i have 256 mb system ram, 65 Mb VRAM, and OA set to uncompressed textures (compressed still have problems, at least here). May be some memory limitin will expose `anholt/max texturing early exit` bug for you?
AndrewR: *64 Mb vram.
glisse: agd5f: send a patch which should avoid the oops you had yesterday, thought here i wasn't able to reproduce it
agd5f: glisse: yeah, I can't reproduce it reliably here either
mjg59: agd5f: Hrm. With your latest patch set, if I pass radeon.dynpm=1 I get no display
mjg59: Caps lock still works, so I'll see if I can get anything out of it
agd5f: mjg59: I've fixed that
agd5f: bad guid idle interrupt setup on pre-r6xx chips
agd5f: new patch set about to post
mjg59: agd5f: Heh
mjg59: Ok, would explain that
mjg59: swaps for an r600
dileX: agd5f: you gotta a fix :-)?
mjg59: Oops. Maybe that's another r500.
dileX: hehe
agd5f: dileX: yup
mjg59: Huh. No, that's a 3650.
dileX: agd5f: great. ppl were annoyed of so much re|connects to IRC
agd5f: mjg59, dileX: http://people.freedesktop.org/~agd5f/pm2/
dileX: agd5f: thats on top of latest drm-radeon-testing (I miss two patches)?
agd5f: dileX: yes
mjg59: agd5f: Hrm. That's a rebase of the previous set?
agd5f: mjg59: 0004 was the only one that changed
mjg59: Ok
agd5f: mjg59: although it wasn't 0004 before
mjg59: agd5f: Ok!
mjg59: agd5f: 0001 doesn't apply
mjg59: (against drm-radeon-testing merged into linus master)
agd5f: mjg59: that's already in linus master
mjg59: Yeah, figured
agd5f: it came via the i2c tree
mjg59: agd5f: Nope, still failing for me with 3650
agd5f: mjg59: laptop?
mjg59: Desktop
agd5f: mjg59: can you pastebin your dmesg? I did most of my testing with a 3650
mjg59: agd5f: If I could get it...
dileX: (reboot)
mjg59: Let me try
Shuren: hello, one question: is there a video ram limit for actual r700 driver? i have a hd4670 1gb from sapphire, but only 256 mb are recognized... or maybe there aren't real 1024mb or vram?
glisse: Shuren: we only use the visible vram memory so far
glisse: visible == cpu mapable == pci bar size
agd5f: Shuren: the vram is real. it's a pci bar window limitation
Shuren: ok, thank's
agd5f: the gpu can access all of it, but the cpu can't directly
mattst88|laptop: why does the PCI BAR size exist? Is there a maximum of 256MiB or something defined by the spec?
mattst88|laptop: or is it there to not eat too much address space on a 32-bit system?
marvin24: testing the new pm code I got: http://pastie.org/873928 (gpu: rs785)
marvin24: "performance" has lowest clock rate
glisse: mattst88|laptop: avoid eating address space
mjg59: agd5f: Huh. Also failing with this other card. Let me identify it.
dileX: agd5f: here with rv515 still black screen
mjg59: Looks like an hd2600?
agd5f: dileX: can you pastebin the full dmesg?
dileX: agd5f: dmesg_full.txt: http://pastebin.ca/1843650
agd5f: dileX: looks like the radeon module never finishes loading
dileX: agd5f: thats from my running kernel
dileX: [ 50.186028] [drm] radeon: irq initialized.
dileX: IRQ seems not to be initialized
dileX: in the other log
dileX: Mar 17 16:37:25 seduxbox kernel: [ 164.763998] [drm] radeon: using MSI.
dileX: Mar 17 16:37:56 seduxbox shutdown[2697]: shutting down for system halt
agd5f: we probably need to init irq before pm or something like that. the pm idle work handler is probably getting called before the irq is inited
psychedelicious: is fan control implemented yet in Radeon or RadeonHD X drivers?
agd5f: psychedelicious: not yet
psychedelicious: thanks agd5f how good is the Evergreen support? I've seen several fanless Evergreen models for sale
agd5f: psychedelicious: modesetting only at the moment
psychedelicious: right, looks my long awaited switch from Nvidia to ATI will have to wait then, I only really need decent XV and 2D support for Evergreen ...
chithead: agd5f: does displayport work?
agd5f: chithead: not yet. it's close, but I haven't had time to finish it
chithead: ah, good
psychedelicious: so agd5f you're recommend sticking with r700 or lower for now then?
psychedelicious: *you'd
agd5f: psychedelicious: if you want acceleration
Thunderbird: of if you are using nvidia, consider using nouveau (though myself I still swear to nvidia drivers but I have different demands here)
psychedelicious: there's the 4350 range which are very reasonably priced so I could get one of those for now..
psychedelicious: Thunderbird, Nouveau has no power management or fan control and my 9600 makes a racket
Thunderbird: sure that's an issue; (you could easily hack nvclock for fan control though; right now I rely on some register values which the nvidia driver sets; I can't detect those from the bios or don't know how)
mjg59: agd5f: So, when there isn't a thermal monitoring chip, is fan control automatic?
agd5f: mjg59: if there's no thermal or fan chip, the fan is just on AFAIK
psychedelicious: Thunderbird, yeah I've already had that suggested on Nouveau's IRC channel, but I'd rather use products from a company that isn't hostile to freesoftware
Thunderbird: there are no AMD GPUs with a simple PWM output? (basic nvidia cards typically have that and then highend cards have an external controller)
agd5f: Thunderbird: r6xx+ have onboard thermal//fan control, but it's up to the oem. some use that, others use external chips
agd5f: pre-r6xx, it was all external chips
twnqx: agd5f: thanks for the quick pointer to the new pll code patches
agd5f: twnqx: np. I'm planning to fix that up properly today
dileX: agd5f: you mean in radeon_irq_kms.c?
agd5f: dileX: not sure what's going wrong yet
dileX: anyway, it would be cool to have a errmsg if IRQ is not initialized
dileX: I tried with drm.debug=0x0f and it said nothing
agd5f: dileX: it's hanging before it can print anything
agd5f: the module never finishes loading. I suspect it's waiting for the pm idle worker or something
dileX: I had also some flickering when rebooting and entering BIOS startup - sth like a pumping
agd5f: ok.. found the problem
agd5f: the gui idle interrupt is getting enabled early somehow and causing an interrupt storm
dileX: [ 50.849740] [drm] Initialized radeon 2.1.0 20080528 for 0000:01:00.0 on minor 0
dileX: there was a version bump to 2.1.0 recently but driver-date was not
dileX: commit 0de1a57bedce0899c98da4c0910d4f38c9b7c8f4 "drm/radeon/kms: bump the KMS version number for square tiling support."
dileX: http://git.kernel.org/?p=linux/kernel/git/airlied/drm-2.6.git;a=commit;h=0de1a57bedce0899c98da4c0910d4f38c9b7c8f4
dileX: maybe someone with commit rights can correct drv-date (just cosmetics)
evil_core: hi all
evil_core: /bin/sh ../../../../../../bin/minstall -m 755 radeong_drv.so //opt/xorg/lib/xorg/modules/drivers
evil_core: Unknown type of argument: radeong_drv.so
evil_core: gmake[5]: *** [install] Error 1
agd5f: mjg59, dileX: found the problem. the gui idle status bit is busted, but the interrupt still appears to work
mjg59: agd5f: Awesomes
agd5f: mjg59: not sure what's up with your r6xx cards yet
mjg59: My r400 appears to be failing to boot even without dynpm
agd5f: maybe they have similar issues. although it works ok on mine
mjg59: udev is sad about it
mjg59: "udevd: worker failed while handling '/devices/pci0000:00/0000:00:03.0"
agd5f: mjg59: with my patches none of the problematic cards will load properly
mjg59: agd5f: Even without dynpm?
agd5f: since the stat bit is permanently high
agd5f: yeah
mjg59: agd5f: My r600s will boot without dynpm
mjg59: But fail with it
agd5f: so the interrupt handler gets stuck in an infinite loop
mjg59: Whereas the r400 boots eventually, but dri fails
agd5f: on pre-r6xx
BeteNoire: boo :( my screen goes crap even on single boot up mode with 2.6.33 :|
mjg59: agd5f: Ok. Something interesting - the driver doesn't find a thermal chip on my r400. But if I force the creation of the i2c bus on gpio 144, I can see an lm63 chip on there
agd5f: mjg59: can you send me the bios?
mjg59: agd5f: Sure
dmb: agd5f: just curious, who is the one that makes the video bios for each card, is that the card maker, or ATI?
agd5f: dmb: we give them a sample bios, but then they give it the final "touches"
dmb: agd5f: so ati gives them the source code to the bios, and they play with it or whatever?
mjg59: dmb: There's often a plugin architecture
mjg59: With a tool to let vendors add tables or choose paramaters
glisse: like an sdk
agd5f: dmb: not exactly. they mostly fill in data tables. we have oem tools to do it
marvin24: BeteNoire: KMS?
dmb: oh
BeteNoire: marvin24: yes, i was advised to kms here, yesterday it worked like a charm, i didnt change anything till today and... its crap now
TuomasT: BeteNoire: KMS works for me without issues on HD3850. Using the in-kernel ATI Radeon driver and xf86-video-ati-6.12.192
suokko: BeteNoire: Can you dmesg from failed boot?
suokko: Can you ssh in?
BeteNoire: suokko: i can, its not totally lock up
BeteNoire: i can login and reboot without seing anything
suokko: BeteNoire: Does ssh work?
suokko: yes :)
suokko: So if you could pastebin the logs it might help solving the problem
BeteNoire: i dont think will be anything in logs but ok, i'll try next reboot
suokko: BeteNoire: Also can you get a picture of what the screen looks like when it displays crap.
Droste: BeteNoire: and you really didn't change _anything_? thats really weird. maybe you changed something from which you didn't thought it's related to KMS causing this.
BeteNoire: its usual "crap" - white=gray ;)
BeteNoire: white-gray
agd5f: mjg59: that bios doesn't have a powerplay table. not sure how we determine the chip on that unless we just add all the i2c busses and probe the all
mjg59: agd5f: Yeah...
mjg59: agd5f: Ok, I've found a card that actually has one in its pwoerplay table!
mjg59: "Possible LM64 thermal controller at 0x18 with fan control"
mjg59: Let me see if I can get that hooked up...
dileX: agd5f: how can I help? a vbios.rom I have here
BeteNoire: suokko: it's usually gray but sometimes gives another colors like this http://img705.imageshack.us/i/p1010304c.jpg/
BeteNoire: suokko: or it looks like this http://img267.imageshack.us/i/p1010305.jpg/
agd5f: dileX: new patches on the way
mjg59: agd5f: Oops!
mjg59: lm64 doesn't seem to be written yet :)
agd5f: bummer
mjg59: Shouldn't be too hard, though
agd5f: mjg59: maybe there's one in lmsensors that's not upstream yet
mjg59: agd5f: Yeah, I'll check that
mjg59: There was a patch sent to add it, but doesn't seem to have been merged
mjg59: Except their list server is down
mjg59: agd5f: Seems to be a trivial patch
mjg59: agd5f: Only difference I can see is that lm64 has a couple og gpio lines
mjg59: agd5f: What those gpio lines do, I have no idea, and nor am I sure how to expose them, so...
agd5f: probably not even used
mjg59: agd5f: Ok, that works nicely
agd5f: sweet
mjg59: agd5f: Sent the diff to you
mjg59: agd5f: HD2900 - I keep getting AER errors from the PCIe root bridge that the gpu is plugged into
mjg59: I'm also seeing occasional brief screen blankings, but they don't coincide too well
spstarr: hmmm
spstarr: airlied: is libdrm_radeon gone now?
spstarr: i just built git master and it doesn't exist anymore(!?)
spstarr: Radeon API yes
spstarr: odd it is there.. it didn't get compiled
mjg59: agd5f: Hey, what do you know. Temperatures go up and down as expected, too
agd5f: mjg59: cool!
spstarr: uh oh
spstarr: did someone break the Makefile in git master?
mjg59: agd5f: I needed to change your names to match the in-kernel naming (i2c binding gets done by name, sigh)
mjg59: agd5f: But I doubt that's really a problem
mjg59: agd5f: Oh, oops. Just seen a trivial error. I don't appear to request the module at any point.
mjg59: agd5f: On the other hand, it works, which kind of implies that the i2c core will ask for it in that case
mjg59: agd5f: So you can drop the module string stuff
mjg59: I'll send a new version
mjg59: agd5f: On the other hand, dynpm fails with every single r600+ I have
Jonimus: mjg59: heh it works for both of my r730s...
mjg59: Jonimus: Using Alex's latest patches?
Jonimus: mjg59: yeah, well the ones on freedesktop.org
Jonimus: one of the r730's is in my laptop the other is in my desktop
spstarr: glisse: you should sync your tree with airlied's it has conflicts do the asic header/struct changes
spstarr: since airlied has your gpu reset code merged i went back to his tree for now
spstarr: is starting to like git more
Dr_Jakob: yeah git grows on you...
spstarr: being able to rollback and revert commit changes
spstarr: so i can switch trees is neat
spstarr: for basic use it is easier to use now for me
spstarr: merging conflicts fixing them, commiting changes, cherrypicking
spstarr: and git resets
Dr_Jakob: yeah
Dr_Jakob: For me the killer feature is "git rebase -i" and "git add -i"
MostAwesomeDude: git add -p
airlied: ajax: we talked about making window pixmaps acceled only before, this proved a bit of a challenge with exa, but I could investigate again
airlied: the main reason on rn50 to avoid shadowfb is Xv support
airlied: but maybe that doesn't matter
airlied: since we lost overlay with kms
Dr_Jakob: Isn't that just a shortcut for git add -i "4" "
MostAwesomeDude: Something like that, yeah.
MostAwesomeDude: But it works great for my mindset.
MostAwesomeDude: "Awesome, it all works."
MostAwesomeDude: "Oh, that's too many changes. And I should probably split it up."
Obscene_CNN: Jonimus, ping
ajax: airlied: what kind of pervert needs xv on their server.
spstarr: Dr_Jakob: haven't gotten to git rebase yet :)
spstarr: ajax: tv tuners needed overlay, well my r1xx does, but this broke with KMS
agd5f: does the overlay even work on rn50?
agd5f: I suspect not
ajax: spstarr: focus. rn50.
spstarr: oh
Dr_Jakob: wtf is rn50?
ajax: rv100 mutant
spstarr: dell boxes used the rn50 alot
ajax: typically with a non-qualified 3d engine, cp that may or may not work...
Dr_Jakob: Let me guess its used in servers and you need to support it?
ajax: and bizarre outputs for things like always-on vnc
spstarr: even the new Dell 10G/11G boxes use the RN50 still *boggle*
ajax: they are, by far, the most competent server graphics chip.
spstarr: ya
Dr_Jakob: hehe
ajax: but there's absolutely no reason to try for awesome 3d support on them
twnqx: server == text mode, right?
ajax: or even glxgears
ajax: twnqx: shockingly not always
spstarr: ajax: the RN50 gave me grief though with RHEL5 eariler on
twnqx: i must admit i once bought a radeon add-in card
ajax: many people expect to be able to walk up to a server, plug in a monitor, and see gdm
spstarr: Dell had to give us updated dkms drivers due to bugs
twnqx: 128MB, first gen PCIe
twnqx: until to day i never switched it int gfx mode
spstarr: no INT10 support or something
twnqx: well, the bios might have put it to gfx mode to present the energy star logo :P
twnqx: but int10 gfx modes don't count.
ajax: you don't want to know how the energy star logo works
spstarr: twnqx: hahaha
twnqx: text mode with custom character set?
ajax: ding!
twnqx: \o/
twnqx: i'm too old not to know these options...
twnqx: still owns a book on "Programming the EGA/VGA Graphics Adapter"
spstarr: twnqx: the 'fix' was to build an xorg.conf and force video to VESA mode :(
edwin: wow, html5 ogg video playback is hogging my CPU
edwin: just looked at that video on wikipedia thing
edwin: I must say that flash works better than that :(
edwin: does anyone know what APIs firefox uses to play html5 video?
edwin: can acceleration be implemented for it?
MostAwesomeDude: Yes; yes.
suokko: MAD: ....
MostAwesomeDude: Dammit, just realized I answered on a technicality. Lemme bug my Moz pals.
edwin: I can do some profiling, but I assume its easily reproducible for you devs
edwin: example video: http://commons.wikimedia.org/wiki/File:13hpBoren.ogv
MostAwesomeDude: Ha, not really. It does better than Flash for me, and that's all I care about ATM.
edwin: which driver?
MostAwesomeDude: radeon, with r600 and r300.
edwin: r600 just hogs the CPU here, with the latest 2d driver...
edwin: the chips a rv730 actually
edwin: lets see if oprofile tells me anything useful
edwin: and if turning composite off helps
spreeuw: edwin: its normal to be slow
spreeuw: its a mozilla problem
edwin: samples % image name app name symbol name
edwin: 46066 57.3667 libpixman-1.so.0.16.4 libpixman-1.so.0.16.4 /usr/lib/libpixman-1.so.0.16.4
edwin: 8429 10.4968 libxul.so libxul.so /usr/lib/xulrunner-1.9.1/libxul.so
spreeuw: flash is indeed multiple times faster
spreeuw: and mplayer doesnt even register
MostAwesomeDude: Works fine over here, but I have mozplugger, so it's playing in mplayer.
edwin: let me try another html5 video site, I remember they worked quite well
edwin: ah no, this is firefox 3.5 with html5
edwin: no external plugins
spreeuw: MostAwesomeDude: thats not a html5 embedded video
spreeuw: edwin: would be nice if you can find something
spreeuw: but it seems to me its in mozillas engine
spreeuw: it draws css alpha just as slow
spreeuw: unusable on p4 in many cases
edwin: ok so dailymotion.com/openvideo is slow as hell too
edwin: like 1 fps
edwin: or 0.5 fps
spreeuw: and javascript popup junk they use for images now
spreeuw: just as slow
edwin: and with fglrx it used to be quite nice
spreeuw: its barely usable on phenom2 quad 3.4
spreeuw: dailymotion is flash here
spreeuw: fast
edwin: /openvideo is html5 there tooo
edwin: they have a demo video
Jonimus: yeah that wikipedia link def used html5 video for me
spreeuw: 0.2 load
suokko: edwin: http://fpaste.org/Prgg/
suokko: That is probably killing the performance
spreeuw: ah
spreeuw: Pagina niet gevonden
spreeuw: page not found localized
edwin: dailymotion.com/openvideodemo
spreeuw: unwatchable
spreeuw: stalling
edwin: ok, do you want me to file a bug, or can you handle that without me? (midnight here, going to bed)
spreeuw: .5 load
spreeuw: .7
spreeuw: I stopped filing bugs
spreeuw: its useless
edwin: 87534 72.0000 libpixman-1.so.0.16.4 libpixman-1.so.0.16.4 fetch_pixel_x8r8g8b8
edwin: ok its this function
spreeuw: I'll see if the current release helps
suokko: edwin: It would be nice if you could fill the bug report
edwin: ok will do that, let me install the dbg package in the 32-bit chroot
edwin: and try to get a callgraph
suokko: edwin: No need. That composite fallback is all that is required for bad performance
edwin: k
edwin: oh wait is it a mesa bug?
edwin: or a DDX EXA bug?
Jonimus: http://people.xiph.org/~maikmerten/demos/bigbuckbunny-videoonly.html there for you html5 video testing needs
suokko: edwin: ddx is missing some acceleration for exa
spreeuw: buck bunny runs ok here ~0.35 load
edwin: yeah that one works here too
suokko: Problem is that scaling of the video is not accelerated
spreeuw: hey thats cool
spreeuw: when you switch back to buckbunny page after leaving
spreeuw: it remmebers where you were
spreeuw: huh
spreeuw: now the load stays at 0.05
spreeuw: mplayer levels
spreeuw: maybe it has to do with a check on the downloader
spreeuw: (fake streamer)
suokko: You can even watch the wikipedia video with low cpu load
suokko: If you jsut right click the video area and select "view video"
spreeuw: yeah load dropped to 0 now entirely
Obscene_CNN: anyone with a r600 or higher want to test a couple of patches for me?
spreeuw: must have been some residulas of that transformers page
spreeuw: Obscene_CNN: dont tell me you added s3tc?
edwin: https://bugs.freedesktop.org/show_bug.cgi?id=27139
edwin: pls add more info to it if you have
Obscene_CNN: no, but I have gotten rid almost 16K of unnecessary cpu instructions in the r600_dri.so driver
edwin: Obscene_CNN: if its a 5m test I can do it
edwin: I have to go after that
Obscene_CNN: you'll have to build libdrm, mesa and the radeon driver
Obscene_CNN: well maybe not the radeon driver
edwin: 2.4.18 libdrm not good enough?
Obscene_CNN: it might be but you'll have to patch it
spreeuw: http://upload.wikimedia.org/wikipedia/commons/a/a7/6hpBoren.ogv
edwin: tomorrow then
Obscene_CNN: okay
edwin: Obscene_CNN: if you can mail me the instructions I'll test in the morning
spreeuw: standalone the vids work ok
Obscene_CNN: in, okay
edwin: good night!
suokko: spreeuw: yes. It is the scaling composite operation that is missing for the video playback
agd5f: dileX_, mjg59: new patches at http://people.freedesktop.org/~agd5f/pm2/
Obscene_CNN: great! someone removed the instructions on how to build libdrm on http://wiki.x.org/wiki/radeonBuildHowTo
Obscene_CNN: now I have to document things in an email to someone
suokko: Obscene_CNN: Or update the wiki pages
Obscene_CNN: suokko, I think I need an account for that
suokko: Obscene_CNN: sure but the account is only a few clicks away ;)
dileX_: agd5f: new patchset works here only in addition with 0001-drm-radeon-kms-add-hw_i2c-module-option.patch and 'modprobe -v radeon modeset=1 dynpm=1 hw_i2c=0'
dileX_: patch-series: http://paste.pocoo.org/show/190975/
dileX_: agd5f: dmesg.txt: http://pastebin.ca/1844099
dileX: agd5f: arghh. wrong patch-series. correct one is
dileX: now both work with hw_i2c=0,1
dileX: agd5f: http://paste.pocoo.org/show/190982/
agd5f: dileX: the hw_i2c stuff isn't directly related to pm.
dileX: agd5f: OK. new module-option hw_i2c works :-)
dileX: agd5f: might have a look into
agd5f: dileX: that's a kernel i2c algo config
agd5f: radeon doesn't use it
dileX: thx for clarification
Obscene_CNN: In case anyone else would like to try my libdrm, mesa, and radeon patches they can be found here http://www.phoronix.com/forums/showthread.php?s=90da7cf5c8254d972c16fb2146a5b359&t=22606
Obscene_CNN: Let the flame wars begin!!!!
airlied: Obscene_CNN: make a big difference?
airlied: unrolling memcpy seems a bit ott
Obscene_CNN: I went from a minimum 20 fps to 22 fps in torcs
airlied: did you try just makeing radeon_cd_write_x_dwords and a variable macro?
Obscene_CNN: I didn't try the variable macro.
Obscene_CNN: It does shave almost 16K of the r600_dri.so binary
Obscene_CNN: all of it redundant checks
Obscene_CNN: It generates nice and tight assembly too
Obscene_CNN: well compared to what it was :P
airlied: Obscene_CNN: it would be nice though to have some useable results
Obscene_CNN: I'll work on getting some of those
airlied: I could get behind a variable arg function
Obscene_CNN: well in most of the code it doesn't hamper readability more than whats already there. (Infact I left a lot of the macros commented out)
Obscene_CNN: I did get better performance from my mesa 7.7-r1 though
airlied: a lot of the changes just seem dirty for no reason
airlied: like inlining functions by hand you could just move to a header file for the compiler to inline
Obscene_CNN: well in a way I did with the radeon_cs.h file
Obscene_CNN: I was hoping to preserve most of the code look and feel to sort of explain what was going on
Obscene_CNN: It is a lot better than the stuff I wrote for the kernel :)
Obscene_CNN: well better loking
Obscene_CNN: looking even
Obscene_CNN: well its that time of year again I must protect the environment from the food dye in green beer.
Obscene_CNN: good night all
FIReun: hmm, mesa 'make' seems to be wanting su priv, s'nobueno
airlied: not likely
FIReun: it prompted to overwrite a file?
PuffTheMagic: how is r500 3d support these days?
PuffTheMagic: could I play xplane on it?
FIReun: and couldnt write some others, ran fine with 'sudo make'
airlied: FIReun: make install?
FIReun: nope
airlied: FIReun: or you ran make as root before
airlied: and screwed up
FIReun: airlied: nope, fresh git pull
FIReun: airlied: but sure, anythings possible
FIReun: phew, 9fps ion ioquake is not an improvement ):
spersaud: does anyone know when 3d support is expected ?
airlied: for what?
spersaud: radeonhd
airlied: 3d support ships in mesa
airlied: its already there for r600->r700
spersaud: airlied: what about rv770 ?
airlied: yes already there
spersaud: airlied: 3d support in mesa or radeonhd ?
airlied: its in mesa, radeonhd is in #radeonhd, this channel is for radeon
spersaud: oh ok cool
lordraptor: wow i had no idea how bad that 9250 was
lordraptor: i got a 9600 pro and with the open source driver is damn good
dileX_: agd5f: FYI: I started openarena - while closing OA I get black screen in X and VT
lordraptor: ooo a game? openarena?
lordraptor: if this 9600 pro does nexuiz i know im set
agd5f: dileX_: anything in dmesg?
agd5f: is it locked up or just blank screen?
dileX_: agd5f: blank screen
dileX_: http://pastebin.ca/1844331
agd5f: that doesn't look good
dileX_: BTW, I re-compiled a new kernel and I need 0001-drm-radeon-kms-remove-lvds-quirks.patch in addition. otherwise loading radeon w/ kms,dynpm,hw_i2c fails.
dileX_: (blank screen in VT)
dileX_: agd5f: errmsg_oa-start-stop.txt: http://pastebin.ca/1844338
dileX_: -> go sleeping