loki_666: i'm having an issue since i switched from fglrx to radeon/radeonhd
hifi: what kind
loki_666: when listening to music, sometime it just like if the sound buffer fails
hifi: debian?
loki_666: ubuntu
loki_666: i tried radeon and radeonhd, same result
hifi: how the display driver is related to your music buffer
loki_666: well i dont know, but it is
hifi: is the rendering slow
loki_666: on radeon, it's ok exa and drm enabled
hifi: I see no reason why it would be related to the graphics driver
hifi: unless you use HDMI sound, but thats unlikely
loki_666: no i'm on a laptop
hifi: only reason why music buffering would stutter it that you have hard drive DMA disabled
loki_666: it's hapening even when listening to streaming from inet
loki_666: it's not stuttering, it stop
hifi: but nothing else stops?
loki_666: nop
hifi: you can like play a game while the music stops?
loki_666: yep, but videos stop (i guess because picture is synced on audio)
hifi: so it's not only music but movies too?
loki_666: yes but i guess it's the result of the audio stoping
hifi: tried without sound?
hifi: just to be sure
hifi: like mplayer -nosound movie.avi
loki_666: no, gonna try
loki_666: dmesg and Xorg.log say nothing when its happening
hifi: it'd be more related to radeon if the video stops without sound
loki_666: damn cant find a video on my drives
loki_666: found one
loki_666: w8
loki_666: seems to play fine
loki_666: and i can tell that sound is hanging even when playing mp3
hifi: I have problems to understand how this is related to the radeon driver
loki_666: yes I know, but it never happened when i was on fglrx, as soon as i switched i had this issue
hifi: even if you play it from a terminal with mplayer music.mp3?
loki_666: mmh, could be an issue with pulse
hifi: try switching to direct ALSA output
loki_666: i think we have a winner
loki_666: yep ty hifi
erjc: I have run into that "black screen" bug running drm-next-radeon and mesa radeon-rewrite as of yesterday. As soon as grub passes control to linux the screen goes black. I can reboot the machine with ctrl-alt-del but not ctrl-alt-sysrec-b. Nothing of this kernel is written to any log.
hifi: I get backtrace from radeon module when it's loaded with modesetting on
hifi: and if you have no framebuffer the screen stays black
erjc: now I run 2.6.30rc7 and X crashes if dri is called
dileX: hi hifi, could you solve your kernel-problems with kms and fb?
hifi: nope
hifi: I disabled all framebuffers
hifi: but no difference
dileX: which distro are you running?
hifi: ubuntu
hifi: shouldn't the kernel stuff be unrelated to which userland you are running?
glisse: hifi: iirc you problem is in ttm, for some reason it fails to register swap callback
hifi: can I do something to debug or solve the problem?
glisse: didn't you pasted a log yesterday ?
hifi: many logs :)
glisse: kernel one
hifi: http://hifi.iki.fi/dmesg_radeon_kms.txt
dileX: hi glisse
glisse: oh it seems that somehow radeon kernel modesetting module is loaded twice
glisse: don't know how such things is possible
glisse: hifi: did you compile the kernel with module builtin ?
glisse: than recompile the kernel without the module ?
hifi: I did make clean and make bzImage and make modules
hifi: radeon and ttm are modules
glisse: well somehow init is called twice
hifi: I'll get the .config up
hifi: http://hifi.iki.fi/drm_next_config
hifi: maybe ubuntu is doing something nasty in the init scripts
glisse: hifi: i am more worried about make modules_install && make install
hifi: I did install the modules and new kernel each time
glisse: also ubuntu could do somethings stupid
hifi: I'm guessing ubuntu
glisse: there initrd tools suck last time i heard of it
hifi: I'll blacklist the radeon module and see what happens
hifi: I don't use initrd with this kernel
hifi: glisse: does it matter I have drm compiled in rather than module?
glisse: hifi: compile as module better
hifi: ok, I'll recompile everything
glisse: thought it shouldn't matter
hifi: so, I'll now make clean && make bzImage && make modules && make install && make module_install
hifi: though I'll do the make install part manually
glisse: make modules_install before make install is better
dileX: glisse: can you look into
glisse: dunno about ACPI stuff
glisse: it's stuff about laptop backlight & co
hifi: mm, I'm doing remote reboot when I have compiled the kernel
glisse: but i don't know what we should do fir that
dileX: isnt "select FB" and "select FRAMEBUFFER_CONSOLE" missing? you told me not to use "CONFIG_FB_RADEON" (dont remember the reason, didnt make a note about that).
glisse: dileX: yes framebuffer console is missing
dileX: can you enlighten why not to use CONFIG_FB_RADEON? (out of curosity)
glisse: dileX: because it would mess the hw
glisse: radeonfb is different driver which will try to overwritte change made by radeon drm module
glisse: radeonfb will die
dileX: OK. is it possible to say dont select CONFIG_FB_RADEON in drm-next/drivers/gpu/drm/Kconfig? (not an expert on that field). would be helpful for avoiding user mis-configuration.
glisse: likely, but i don't know neither how to do that
glisse: likely documented somewhere
dileX: OK. no problem. I will search in kernel-docs.
dileX: glisse: I have a patch to satisfy 2.6.30-rcX on top. where shall I send it?
dileX: on top of drm-next-radeon
glisse: i am going to pull 2.6.30 soon
glisse: to make patch for 2.6.31
dileX: I know rc8 came last night :-)
glisse: only conflict is with drm crtc helper func iirc
dileX: yepp
hifi: glisse: http://hifi.iki.fi/dmesg.txt still
hifi: you can see I loaded the module by hand
hifi: modprobe radeon modeset=1
glisse: hifi: doesn't make sense once again initialization is called twice
dileX: glisse: here is my patch
hifi: glisse: what can I do?
glisse: hifi: i need to look at the code to understand why this could happen
glisse: hifi: you can try to find why radeon_device_init is called several time
glisse: and by who
hifi: at work currently
glisse: knowing who will likely tell you why
dileX: glisse: while you around
dileX: glisse: these patches work also fine
dileX: (reboot)
transplant: It is possible to use fbcon with the DRM drivers?
airlied: hifi: you got an lspci from that machine
airlied: glisse: there is two cards
airlied: glisse: have you tested multi-card?
hifi: http://hifi.iki.fi/lspci.txt provided
hifi: oh, I do have two cards
hifi: forgot to mention that
hifi: and the other one is R200
transplant: "Framebuffer console support [Y]" and "DRM: ATI Radeon [Y]", will do it?
hifi: can I prevent it from loading for my other card
airlied: well it should just work
hifi: just for debugging purposes
hifi: but true, the old drm stuff worked perfectly with both cards
transplant: that is without selecting any "Framebuffer devices"
transplant: I'm asking because i want to help testing
glisse: airlied: oh didn't think to that
glisse: well i need to detect tht and avoid reiniting ttm
Esmil_: Hi. Last night I wanted to try out kms. I followed this guide: http://jglisse.livejournal.com/1822.html pretty closely, but the kernel errors: http://pastebin.com/m48d72c0f
glisse: Esmil: disable all fb driver (vga,vesa,radeonfb,....)
Esmil: glisse: And have a text console?
glisse: Esmil: with kms you don't have text console, just active fbcon but nothings else in the fb section
Esmil: Ok I'll try that. Thank you
airlied: glisse: I think TH has some drm globals code for that
glisse: airlied: yeah likely
hifi: glisse: can you get a patch out soon to test it?
hifi: or a hack to disable TTM on the other board
glisse: i am testing agp if nothings big popup i will take a look at that while doing ttm patch
hifi: well, I can always pull the card out and test :)
hifi: I just hate doing that
sannes: If I use two screens with KMS I get this: http://www.sannes.org/misc/kms-dualscreen-corruption.jpg ..
bSON: hi, i am experiencing a bug with the current radeon driver (ubuntu karmic version = git from 2009/05/06, kernel 2.6.30) and was pointed here from #xorg, it might be an already well-known problem: the background of opengl windows seem to be almost completely transparent instead of solid as expected, for instance glxgears' background is slightly white and transparent instead of black
glisse: sannes: what are the resolution of screen ?
glisse: sannes: no corruption under X right ?
sannes: glisse: No corruption under X no
glisse: well it's fbcon stupidity problem
glisse: fbcon wasn't designed with multi-screen in mind
sannes: glisse: The resolution is different on the two screens, and they can't be equal either because one is a 4:3 and the other 16:9
sannes: unless they can and I'm wrong :P
sannes: glisse: Now that we "know" that 2.6.30 will be out this weekend, any chances of KMS/TTM/DRI2 kernel parts ending upstream in 2.6.31 ?
glisse: i am working on getting patch ready but i will be experimental
glisse: as i will break compat along the way for 2.6.32
max_r: glisse: and what about r6xx/r7xx users? are there any chances for us?
max_r: (chances for getting kms in 2.6.31)
glisse: likely no
max_r: and chances for going into 2.6.32 are quite good, right?
glisse: yeah 32 it will be ready at least it's the plan
glisse: but it will be in fedora12 too
sannes: glisse: What is known to be changed ABI-wise between 31-32 ?
glisse: current ddx likely won't work, as i will do more security check in kernel and forbid unsafe cmdstream
glisse: same for mesa
sannes: ah, nice :)
hifi: ok, I hacked the drm pci ids to skip my second head
hifi: now it kind of works :)
hifi: but yeah, sure, it was the second initialization that broke it
hifi: (II) GLX: Initialized DRISWRAST GL provider for screen 0
hifi: hm
hifi: (EE) AIGLX error: Calling driver entry point failed
hifi: ok, that was related to libdrm2
hifi: [dri] radeon.o kernel module version is 2.0.0 but version 1.17.0 or newer is needed.
hifi: now that defies logic :D
glisse: hifi: likely too old ddx ie ddx isn't the one i point to in my blog
hifi: hm, I'm sure I copied it
hifi: okay, I didn't
hifi: now the next problem is the screen is all garbled
hifi: and glxinfo gives
hifi: drm_radeon_getparam_t (RADEON_PARAM_DEVICE_ID): -13
glisse: wrong mesa dri driver
glisse: need one from radeon-rewrite
hifi: --with-gallium?
glisse: without
hifi: hm, that is built that way
hifi: it's configured with --with-dri-drivers=r300 --eith-demos= --disable-gallium
hifi: and branch is radeon-rewrite
hifi: oh, another failure by me
hifi: what a surprise
Esmil: glisse: Hmm.. I still don't get it. If I disable all framebuffer devices, but keep support for VGA text console and framebuffer console I get the same error when I load the radeon module http://pastebin.com/m48d72c0f
Esmil: If I disable the text console its really hard to load the radeon module ;) But if I compile it in my screen just flickers once and then my computer dies when loading the kernel
Esmil: Not even the 3 finger salute works
glisse: Esmil: what gpu ?
Esmil: Radeon Mobile X1400
Esmil: r500 right?
glisse: yeah should work
Esmil: Well, if I want to try out the new patches you just pushed. Should I compile in the text console and keep everything else as modules?
hifi: [ 45.355287] [drm:drm_mode_getfb] *ERROR* invalid framebuffer id
hifi: [ 46.051760] [drm:radeon_cp_start_kms] *ERROR* invalid ioctl with kms radeon_cp_start_kms
erjc: should be --with-dri-drivers=r300,radeon no?
hifi: glisse: does kms need vesa framebuffer or something like that?
hifi: I don't have any framebuffers enabled in the kernel
glisse: hifi: wrong ddx
glisse: or ddx not compiled against libdrm-radeon
hifi: ha, that might be
glisse: Esmil: well there is an issue wih reading your bios
glisse: so it won't work right now
nanonyme: glisse: Btw, should HAVE_LIBDRM_RADEON be undefined if not compiling Mesa against libdrm-radeon? (sounds like a silly question but it would explain a problem I've been having)
Esmil: Ahh.. well if I can do anything to help solving it, just say so
glisse: nanonyme: it's defined or not in config.h
nanonyme: Hmm.
glisse: Esmil: i need to ponder at code first to see what you can test
Esmil: glisse: Sure
nanonyme: glisse: So it's not detected on the fly?
glisse: nanonyme: no it's detected at ./configure time
nanonyme: That's kinda what I meant on the fly.
hifi: glisse: I still get it after I made sure my ddx is compiled against libdrm-radeon
nanonyme: glisse: configure mistakenly sets it on for me. Perhaps due to me using LIBDRM_CFLAGS and LIBDRM_LIBS, unsure.
glisse: hifi: pastebin output of ./configure of ddx
nanonyme: Oh! Now I think I got it.
glisse: nanonyme: yeah it might get confuse and not use what you provide with LIBDRM_LIBS
hifi: glisse: http://hifi.iki.fi/ddx.configure
nanonyme: glisse: I can find a PKG_CHECK_MODULES for HAVE_LIBDRM_RADEON=yes in configure.ac. Maybe it's asking pkg-config whether I have it?
nanonyme: That would explain it.
nanonyme: Since I have two libdrm's on my system, one is rawhide, one is r6xx-r7xx-3d.
nanonyme: And if configure asks pkg-config, it might be detecting the rawhide libdrm.
nanonyme: (Well, actually probably F11 libdrm at this point)
max_r: nanonyme: the problem is caused by rawhide libdrm
max_r: it adds file into /usr/lib/pkgconfig/<>.pc
nanonyme: max_r: Rawhide libdrm in itself is not a problem. The problem is that configure realizes it exists.
max_r: something like libdrm-radeon.pc
max_r: nanonyme: could you list all files installed by rawhide libdrm
max_r: only .pc ones
nanonyme: I'm having libdrm r6x-r7xx-3d under /usr/radeon so technically it should be asking pkg-config *under* /usr/radeon if it existed.
nanonyme: But the thing is that I have pkg-config installed only under /usr...
max_r: you could just rename bad .pc file before compilation
nanonyme: (Very likely overridable by setting LIBDRM_RADEON_CFLAGS and LIBDRM_RADEON_LIBS manually)
nanonyme: If the configure.ac is written properly.
glisse: hifi: do you have dri2 proto ?
glisse: doesn't seems so
hifi: yes
hifi: x11proto-dri2-dev is already the newest version.
hifi: if you mean that
nanonyme: (Err, no. The configure.ac isn't written to offer a sane failsafe. Apparently I seriously have to do that rename)
glisse: hifi: somethings is wrong with dri2 on your side
glisse: it shouldn't try to init dri1 with kms
hifi: mm, want to take a look at the xorg log
glisse: and it seems it doens't try to init dri2 but rather directly go to dri
nanonyme: Well, good to know.
hifi: something interesting is happening there
hifi: glisse: also remember I have two cards, I disabled DRI on the PCI one in xorg.conf
hifi: http://hifi.iki.fi/Xorg.0.log
nanonyme: max_r: I'm currently in Windows and can't check the exact name but you're very likely right. It's just that I'd prefer it to be workaroundable using a configure variable. :)
hifi: the RADEON(0) is the one that should have KMS/DRI2/whatever :)
nanonyme: I could always patch it to be so but I'm not sure if it makes much sense.
hifi: also I find it confusing AIGLX whine sthat screen 0 is not DRI capable and that screen 1 is not DRI or DRI2 capable
nanonyme: Depends on how much time time the r6xx-r7xx memory manager takes, I guess. :)
hifi: so it would mean screen 0 is DRI2 capable?
glisse: hifi: yeah well wait until i can get another pci radeon card to test that
glisse: it definitly get confuse with different card you have
hifi: I'll hack the pci id from ddx to try only on my screen 0 head
hifi: umm no
hifi: I can just disable the card in xorg conf, right?
nanonyme: max_r: Doable with a simple fix: just requires one if-clause that allows you to set HAVE_LIBDRM_RADEON on command line. ^^
nanonyme: (Or rather the skipping of pkg-config requires that)
hifi: glisse: uploaded xorg log with only the R500 card
hifi: same errors in dmesg and xorg log
max_r: max_r: lets better hope that libdrm_radeon will arrive soon and we all will be able to build with HAVE_LIBDRM_RADEON
max_r: *nanonyme:
erjc: drm-next $ git pull --> error: Entry 'drivers/gpu/drm/radeon/radeon.h' not uptodate. Cannot merge.
nanonyme: max_r: Which would be about immediately after r6xx/r7xx memory manager gets done? ;)
stikonas: erjc, drop local changes: git reset --hard
max_r: nanonyme: will see... :)
nanonyme: max_r: Sadly the kernel bits are untrivial. :(
nanonyme: (As far as I've understood)
hifi: I rebooted and now I'm getting this again: (EE) RADEON(0): [dri] RADEONDRIGetVersion failed because of a version mismatch.
glisse: hifi: doesn't seems there is any errir their
hifi: though now I have no errors in dmesg
glisse: hifi: load by hand the radeon module
hifi: oh
glisse: or xorg will mess up things
glisse: boot in init3
hifi: when xorg loads I get the errors in dmesg
hifi: [ 32.136387] [drm:drm_mode_getfb] *ERROR* invalid framebuffer id
hifi: though now that I rebooted and loaded the module myself I got EXA working again
agd5f: nanonyme: export PKG_CONFIG_PATH=/my/path:${PKG_CONFIG_PATH} should do what you want
nanonyme: Right, going to try that next time.
nanonyme: agd5f: I honestly didn't even think pkg-config could have caused my issue before looking at configure.ac today. :)
agd5f: glisse: you are calling r100_pci_gart_disable() in r300_gart_enable() in the AGP case. should be calling rv370_pcie_gart_disable() if the card is rv370 or higher
glisse: agd5f: i call pcie if it's a pcie card iirc
agd5f: glisse: but in the AGP case
glisse: oh yeah missed tha
agd5f: glisse: btw, I just confirmed that RS690_AIC_CTRL_SCRATCH is the same on rs4xx
glisse: ok thanks
agd5f: BTW, looks like rs4xx (but not rs690) has the old PCI gart regs (AIC_CTRL, etc.) as well. not sure if they work or not
nanonyme: agd5f: Anyway, now that I know where the mixup happens exactly, it's kinda trivial for me to fix it in a local build. :) Should cause more reliable test results.
glisse: sight lastest ttm bits broke things
glisse: need to go back at debuging ttm
hifi: glisse: was my problems related to that?
glisse: hifi: no it's local code
awaters1: on branch radeon-gem-cs3 of xf86-video-ati in glisse's repository
awaters1: the textured video takes up the whole window, but it only shows the top left of the video
awaters1: http://pastebin.com/m5b6f9541 seemed to fix the issue for me, on a r300
MostAwesomeDude: I was under the impression that it was already fixed?
awaters1: it was sort of
awaters1: before the fix the video would show the top left portion of the video in the top left of the window
awaters1: after the fix the video would show the left left portion of the video in the whole window
awaters1: let me see if i can make shift a screenshot
awaters1: is there a good site for uploading screenshots?
awaters1: here are screenshots of the issue http://www.cs.rpi.edu/~watera2/radeon_issue.html
dileX: for images: http://imagebin.org/index.php?page=add
stikonas: awaters1: I and _Groo_ experience the same problem
awaters1: thanks, screenshot before the fix http://imagebin.org/51329 screenshot after the fix http://imagebin.org/51328 screenshot of expected video
awaters1: http://imagebin.org/51330
awaters1: stikonas: does the video change when windows are dragged over it? after the fix if i drag a terminal over the video it seems to change the clipping in a weird way
stikonas: you have to apply aditional patch to fix clipping
stikonas: s/aditional/additional/
awaters1: with latest git and http://pastebin.com/m5b6f9541 it seems to work for me
stikonas: probably that patch reached cs3 branch
awaters1: the patch takes away fixes for the rs690 listed here http://cgit.freedesktop.org/~glisse/xf86-video-ati/commit/?h=radeon-gem-cs3&id=8338385cd94b35f54b25cf2576890159f8dd9f71
stikonas: earlier I fixed that problem locally but after git pull it appeared again
stikonas: I know
agd5f: that patch is correct
agd5f: glisse only fixed exa in his branch
awaters1: does glisse know of the issue - i don't it to get lost
awaters1: don't want it to get lost
agd5f: awaters1: yeah
agd5f: awaters1: it works in master and that's set to be merged at some point
awaters1: excellent
agd5f: lots of little diffs along the way make it harder to merge later
agd5f: osiris_, airlied: looks like we are setting GA_ENHANCE in r300ShadeModel()
agd5f: probably shouldn't be
osiris_: agd5f: glisse has already disabled it when kms enabled, but yeah, we should probably just set this reg in r300ResetHwState and never touch it again
agd5f: ok. cool
tball: Hello everybody. How is 3d support for r6xx / r7xx evolving? :)
nanonyme: tball: Slowly? ;)
nanonyme: I personally just check the RSS feeds for the git. If there's a bigger push, I suspect it'd be apparently from the git messages.
tball: nanonyme: You are properly right :)
nanonyme: Apparent even.
JarenJP: Hello everyone.
JarenJP: MostAwesomeDude, did you think I'd disappeared?
JarenJP: Christian helped me figure out which registers were different with my card and got the hdmi audio working for me!
pkt: I just tested the ati agp suspend fix under KMS.
airlied: pkt: any help?
pkt: Still nothing unfortunately. Remains frozen on wakeup :(
airlied: :-(
spstarr: airlied: glad you had some time off, how is baby?
spstarr: airlied: i have an r6xx laptop (switchable graphics with Intel GMA)
airlied: spstarr: nice we have one of those here. baby still thinks she is on GMT :(
spstarr: uh oh on baby part ;/
spstarr: airlied: I can test switchable graphics if that ever becomes a feature
spstarr: I bought a Thinkpad W500 series
spstarr: the other one died while you went on vacation (the LCD screen is dead, but the laptop works fine)
spstarr: airlied: im in the GMA right now
spstarr: the r6xx works fine with EXA but i didnt stress it enough to see yet, just waiting
spstarr: airlied: I like seeing [ 0.000999] No AGP bridge found :D (even though the intel GMA is AGP but prior to that loading.. )
spstarr: it loads it up after