_kinski_: hi all, i have a problem with kms and dual screen setup
_kinski_: i get a freeze with dualhead and kms, using only kms without works just fine
_kinski_: here is my dmesg output: http://pastebin.com/m2ca50428
_kinski_: and my xorg log : http://pastebin.com/d7baa5850
soreau: I think something is wrong with cgit.. it seems to be 'stuck'. Ont the main mesa page it shows "r600: fix fragment.position" 17mins old but on the log it shows latest is "r300: disable blit debugging info" as 18hrs old and it's been in this same state regardless of reload/refresh pages for.. a couple days now
airlied: soreau: cgit gets stuck sometimes
_kinski_: airlied: are there known issues with kms + dualhead?
airlied: _kinski_: not that I know off
soreau: airlied: Any way to unstick it? Or, anyone I can bug to fix it?
airlied: soreau: benjsc in #freedesktop can kick it usually
soreau: airlied: Ok, thanks.
_kinski_: as posted before i get a freeze with kms+dual, but no problem without the second screen
airlied: _kinski_: dmesg + xlog in a bug
airlied: on bugs.freedesktop.org
_kinski_: ok
glisse: airlied: erestart was the reason of some alloc failure message ?
airlied: glisse: erestartsys
airlied: glisse: if radeon_bo_create fails, it tests for ERESTARTSYS already
airlied: but the gem wrapper doesn't check
airlied: so reports an error when its just ERESTARTSYS
glisse: so it was the reasons for the error messages we were seing ?
glisse: or is their still a bug in eviction ?
airlied: I still have one bug left see the rambling email I sent to you and TH
airlied: if I scroll this webpage my machines locks up and dies
airlied: glisse: http://www.brisbanetimes.com.au/queensland/brisbanes-streamy-steamy-night-20091214-kqqv.html
airlied: is the page I can crash my machine with ;-)
airlied: glisse: last error I got made it look like we were validating a bo that was already destroyed
airlied: or on the way to being destroyed
glisse: yeah i read the mail
JohnDoe_71Rus: ATI Technologies Inc RS482 [Radeon Xpress 200M] Mesa DRI R300 (RS400 5975) 20090101 NO-TCL DRI2 alien-arena artefacts
glisse: weird
airlied: however I think we have an issue where the not-destroyed list could take up all the offset space
airlied: I think we need some directed reclaim in certain places
airlied: where we sync free objects when we run out of space and there is objects sitting in the list
glisse: yeah this is what i was suspecting yesterday but according to thomas eviction is synchronous
glisse: but i don't see the ttm_bo_wait in the evict path
airlied: eviction is fine
airlied: the problem is freeing
airlied: if we alloc/free 1000s of objects, the freed objects still take up drm fd space
airlied: until they are reaped by the eviction thread
airlied: (not eviction thread, delayed delete thread)
glisse: it should fails before validation in bo_init then
airlied: glisse: the rb tree collision is also strange
glisse: yeah i think the rb tree is properly protected by lock
airlied: its like two objects get the same offset in the drm fd allocator
glisse: maybe when an object is on the destroy list we can free the fd offset first
airlied: that'll just paper over it mostly
airlied: in theory it should just work
airlied: so I think we do something silly ;-0
glisse: likely, all those path are so intermingled :(
airlied: glisse: btw there are still some ERESTART checks in radeon_object.h
airlied: they are pointless now
glisse: yeah i will clean them up
glisse: that's strange i am pretty i did grep when converting
airlied: glisse: oh can you ack the two patches I sent today if they make sense ;-)
airlied: btw do you use mouse scroll wheel to scroll in firefox? I think that can create lots of ERESTARTSYS ;-)
glisse: oh, i need to buy back a mouse with scroll, the scroll on the current one is busted
glisse: stupid microsoft mouse
dileX: any alternative to kdiff3 as git-mergetool (tkdiff is not my thingie)?
airlied: glisse: we actualy do release the VM block before adding it to the delayed queue
spstarr: morning :)
spstarr: how is IRQ and KMS ?
Zajec: spstarr: how can it be?
Zajec: spstarr: generally working :)
Tommeh: Morning all :)
Tommeh: (If it is morning for you!)
spstarr: well i have to see
dileX: airlied: when are you planning to push recent dri-devel patches to one of your drm-2.6 GIT branches?
nha: Any of you guys happen to be in Austin, Texas around 19th of January and want to hang out?
dileX: nha: 12 oclock hi-noon in front of the saloon :-)?
Tommeh: Anyone running an R600 card on Karmic with the latest Edgers PPA packages?
nha: dileX: something like that :P
nha: of course, weapons will be rulebooks + d20
dileX: airlied: glisse: checking my pending patches from dri-devel. might have a look into it
airlied: dileX: not all of those are acked yet ;-)
dileX: airlied: last three one yes. drm-radeon-kms-fix-r100-r500-CS-checker-for-compressed-V2.patch is OK?
airlied: but I'll be pushing out drm-radeno-testing tomorrwo
airlied: but I've one bug left on my PC
airlied: dileX: yeah should be fine
dileX: OK
glisse: airlied: how can i know the number of tabs open ? :)
airlied: glisse: I normally just guess ;-) I also have vertical tabs plugin
airlied: so my tabs go down the screen
airlied: the webpage actually caused wierd stuff with only one tab
airlied: when scrolling
airlied: but only when I had some debug printfs
airlied: glisse: http://people.freedesktop.org/~airlied/scratch/curhacks
airlied: is all crap I've been hacking on trying to figure it out
glisse: that's weird, it seems ok here, i likely have 40tabs on a 24"screen but no error (i also added debug printk)
glisse: what i get is small stop when scrolling
airlied: yeah I'm not sure if that patch I have flash installed or not
airlied: page even
airlied: the WARN_ON for ddestroy were what I hit when I hadn't eaten all the ram
amarks: i'm using the Tab Kit FF extension for vertical tabs, seems to work and i like how it can do hierarchical
glisse: i will try limitting vram
Nightwulf|work: hi all
orly_owl: firmware? I thought the radeonhd driver loaded firmware, not the radeon driver.
TBBle: I dunno about radeonhd, but the radeon drm/kms driver loads a ton of firmware. Three or four blobs for an r600/r700.
MostAwesomeDude: Oh noes.
TBBle: I'd assume radeonhd would have to load at least some of the same firmware, to drive the card. But I haven't looked.
MostAwesomeDude: Nope.
MostAwesomeDude: It does require the kernel module to be loaded, in legacy mode, to do acceleration.
hifi: what, radeonhd doesn't need a firmware to do acceleration in non-legacy(?) mode?
MrCooper: the kernel module always needs firmware for acceleration, even in 'legacy' (non-KMS?) mode
MostAwesomeDude: radeonhd doesn't work without the radeon kernel loaded, in legacy non-UMS mode, and even then, if the kernel doesn't have the firmware loaded, radeonhd can't do acceleration.
MostAwesomeDude: I'd draw a chart, but honestly I don't care about how firmware-safe radeonhd is.
dileX: opening 100+ tabs in firefox f*** up network not radeon :-)
maximi89: hi guys, i'm experimenting some errors, when i resume from hibernate state, i get a X entire in black, the computer still works but the X can't works, i'm using Fedora 12, xorg-x11-drv-ati-6.13.0-0.11.20091119git437113124.fc12.i686
maximi89: the error i get from dmesg is [drm:radeon_cs_ioctl] *ERROR* Faild to schedule IB ! [drm:radeon_ib_schedule] *ERROR* radeon: couldn't schedule
maximi89: any idea?
glisse: maximi89: sounds like lockup
maximi89: 01:05.0 VGA compatible controller: ATI Technologies Inc RS780M/RS780MN [Radeon HD 3200 Graphics]
maximi89: hoops
maximi89: glisse, lockup?
glisse: gpu lockup
riri: glisse, someone on linuxfr told me to thank you for the radeon kms support, so thank you :)
glisse: i am not the only one guilty
maximi89: glisse, so i guess that could be a regression, because in the last version hibernate suspend and all functions works perfectly, but when i upgrade to Fedora 12 i get hibernate not working
glisse: maximi89: in last version you didn't have acceleration
maximi89: sorry, in previous version
evil_core: anybody is usingg r500 and gallium daily?
roysjosh: agd5f, have you ever heard of anything like this: http://bugzilla.kernel.org/show_bug.cgi?id=6058#c4 . I think it may be one of the issues keeping my laptop from resuming from RAM
adamk_: Why do opengl windows stop rendering when they are moved partially off screen ot the left or top, but not the bottom or right?
cxo: you can see windows outside your screen?
adamk_: "partially"... Take glxgears, move part of it off-screen to the top. The entire window stops rendering, even the part that's still visible.
adamk_: If you move it to the bottom, the visible part still renders.
cxo: still renders on any side for me - but i'm using compiz so maybe thats why
cxo: cos everything is just one big cube, i can put glxgears on any edge and it sill renders
MostAwesomeDude: adamk_: Compositing manager?
adamk_: MostAwesomeDude: none...
MostAwesomeDude: Hm.
adamk_: DRI1, if it matters. I'll reboot into slackware and test with DRI2 in a minute.
evil_core: radeon supports pbuffer on r500?
cxo: puff the magic buffer
adamk_: Bah... So DRI2 handles it properly.
MostAwesomeDude: evil_core: Not daily, but sometimes. Been having issues?
MostAwesomeDude: evil_core: Yeah, pbuffer works with r500.
edgecase1: ok you guys gave me some bad git advice
edgecase1: git checkout -b only works if you're already working on the branch or something
edgecase1: airlied, so anyhow, finally using drm-radeon-testing, Thinkpad T30 M7 R100 ... DDC for VGA is 8 bytes 0x60 even when monitor plugged in
MrCooper: osiris: does mesa/progs/demos/gearbox work correctly for you with accelerated blits?
osiris: MrCooper: nope, but I've got a patch for it. will push it in a few hours
agd5f: edgecase1: does ddc work on that monitor on another card?
Zajec2: agd5f: on what GPU did you experience HDMI lock up?
MrCooper: osiris: cool, what did you test the accelerated blits with?
osiris: MrCooper: I don't remember right now
osiris: MrCooper: I don't remember right now
cxo: what is with irc today... its like it had bad chinese food or something
edgecase: airlied, dmesg as follows:
edgecase: [ 0.623807] [drm] Radeon Display Connectors
edgecase: [ 0.623854] [drm] Connector 0:
edgecase: [ 0.623897] [drm] VGA
edgecase: [ 0.623941] [drm] DDC: 0x60 0x60 0x60 0x60 0x60 0x60 0x60 0x60
edgecase: [ 0.623991] [drm] Encoders:
edgecase: [ 0.624035] [drm] CRT1: INTERNAL_DAC1
edgecase: that's drm-radeon-testing booted with modesetting on and VGA connected to laptop
osiris: MrCooper: but I know, it was testing only glCopyTexImage, not glCopyTexSubImage
rindolf: Hi all.
rindolf: What should I use - radeon or radeonhd?
edgecase: so, is Preempt fail? Xorg kernel bug Scheduling while atomic...
Ghworg: rindolf: radeon gets more code love, I'd go with that
rindolf: Ghworg: ok.
AndrewR: hi all ... I don't have rs780 anymore .. but still have rv280 %)
evil_core: anybvody uses glide wrapper in wine?
airlied: christ freenode ddos hell
edgecase: airlied, i can use some help with this M7 DDC if you can get thru the netsplit storms
agd5f: edgecase: does ddc work with that monitor on another card?
edgecase: agd5f, it works with modeset off and with radeon_fb
airlied: edgecase: that dmesg bit isn't helpful
Zajec: agd5f: could you answer me please? on what card did you notice lock up with HDMI patch?
airlied: boot with drm.debug=1 and see if it prints anything later
Zajec: agd5f: finally we won't splited this time ;)
agd5f: Zajec: any card
Zajec: oh... wow
edgecase: airlied, ok i'll pastebin dmesg, bbl
Zajec: that's weird then I don't notice problems here :|
agd5f: Zajec: using interrupts rather than a timer fixes it though ;)
Zajec: agd5f: do we have audio interrupts? i don't think so
agd5f: Zajec: not yet ;)
Zajec: agd5f: you have initially?
airlied: we do get a hint on the audio irq ;-)
Zajec: nice
airlied: since we get one and it prints it out
Zajec: airlied: but that is not public code, right?
airlied: no that happens with my tree now
Zajec: ok
airlied: you get an unknown interrupt from the hdmi controller ;-)
Zajec: guys (agd5f?) I've posted my problem about IRQs in "Re: [PATCH] drm/radeon/kms: add dynamic engine reclocking (v3)", could you check that please, please, please?
Zajec: airlied: that's great, hope you will release it soon :)
agd5f: Zajec: but disabling the call to r600_hdmi_update_audio_settings() in the timer handler avoids the hang in the current cde
airlied: Zajec: I don't want to push the current code if its all hangy
agd5f: Zajec: Can you resend? I only see v2 of your patch
dileX: Zajec: interesting explanations in your mail
Zajec: agd5f: refresh mailbox, I've posted it 3mins ago
Zajec: :)
Zajec: maybe that's issue
Zajec: dri-devel@
agd5f: Zajec: I'll let you know when I get it
Zajec: agd5f: i understand disabling it fixes hang for you, but this makes audio unusable, doesn't it?
Zajec: agd5f: if we don't update HDMI status we won't get sound
Zajec: *sound
agd5f: Zajec: I see replies to it, but I never got the original patch
agd5f: Zajec: could be. I don't have an hdmi receiver
Zajec: agd5f: sec
Zajec: agd5f: ah, i posted v3 in v2 thread
Zajec: it was dileX that replied creating new thread, sorry
Zajec: agd5f: can you see my 2nd mail in "[PATCH] drm/radeon/kms: add dynamic engine reclocking (v2)" thread?
dileX: Zajec: I am not subscribed to ML
Zajec: agd5f: if you wish, i can forward it to you
agd5f: Zajec: vblank interrrupts are only on when requested
Zajec: dileX: ah :)
dileX:
Zajec: agd5f: how can I request?
agd5f: Zajec: generally 3D apps, for example, request it
Zajec: agd5f: is that possible to request it from kernel side (as I need from PM part of radeon driver)
agd5f: Zajec: you can turn it on/off by setting rdev->irq.vblank[x] to true and calling radeon_set_irq()
dileX: Zajec: replies are all in one thread
Zajec: agd5f: i don't really see some request enabling in r600.c
Zajec: hm, ok
agd5f: Zajec: but you don't want to leave it on all the time or it keeps the CPU from going into lower power c-states
agd5f: Zajec: the driver doesn't turn it on by default. only when an app requests it
Zajec: agd5f: hm, i think i got some response already on this on ML
Zajec: will check it once more
Zajec: agd5f: thanks for tip
Zajec: big tip :)
agd5f: Zajec: np
evil_core: what TCL mode should I use in driconf for r500?
seb_: (i built everything from gits repository)
dileX: seb_: missing firmware
seb_: radeon/R700_rlc.bin is unknown in debian
seb_: apt-file search 'radeon/R700_rlc.bin' ----> no results
Zajec: hopes to fix no vblank issue :)
agd5f: seb_: http://people.freedesktop.org/~agd5f/radeon_ucode/
Zajec: agd5f: only thanks to your fast help :)
dileX: seb_: other files
seb_: ^^ thx
Zajec: it works! :)
dileX: Zajec: push v4?
seb_: hope it gonna work for me as well
seb_: see you after my reboot
Zajec: dileX: will do soon
Zajec: dileX: i've to consider case of old GPUs since we call PM functions on every GPU
Zajec: dileX: also I hacked a little and request IRQ on every CRTC, that's not really nice :)
seb_: not that good i got a blank screen and computer freezes
seb_: some kind of 'black screen of the death'
seb_: the only thing i tweaked is /Etc/modules
seb_: with radeon modeset=1
agd5f: Zajec: set_engine_clock is implemened on all gpus
agd5f: Zajec: although changing the engine clock is generally kind of tweaky on desktop r1xx cards IIRC
seb_: dunno if framebuffer could interfere
seb_: or do you think gpm could mess the whole display?
Zajec: agd5f: so should we avoid that?
agd5f: Zajec: have to test it out on older cards
Zajec: airlied: does audio IRQ work with currect microcode?
agd5f: Zajec: yes. it's just another interrupt source
kdekorte: I get a lot of corruption in when using compiz after running Windows 7 and then rebooting into Linux. Using drm-radeon-testing, mesa, and ddx from git, libdrm in Fedora 12 is new enough. REbooting back into linux seems to make the corruption go away. Really bad in apps like xsane under compiz
seb_: logs report: fb0: radeondrmfb frame buffer device
seb_: registered panic notifier
Zajec: agd5f: just tried to find it with hexdumping R600_rlc.bin but this file does not seem to contain registers
Zajec: :)
agd5f: Zajec: nothing to do with firmware
agd5f: Zajec: I can release the code once I get approval
Zajec: agd5f: em, i thought firmware reports that... but ok :)
Zajec: agd5f: didn't mean to really dig into this :) was just curios if I can find register value :)
Zajec: just hacking train :)
eosie: must r300_vap_vtx_out_fmt be specified for texture-stuffed coordinates?
eosie: i see, it shouldn't
agd5f: eosie: I htink you just have to set the rasterizer properly
eosie: agd5f: that's what I am doing
agd5f: eosie: and of course handle the inputs in the pixel shader
seb_: noobish question how to be sure to get rid of everything involving frame buffer please ?
eosie: I am assuming the pixel shader is ok, since the same pixel shader works on swrast
eosie: this is what I have: http://paste.pocoo.org/show/157234/ It doesn't work so I must be missing something
edt: looks like 779720a3209849be202ac36a811e934865c50971 causes font corruption here. Confirming now by rebuilding drm-radeon-testing without this commit.
agd5f: eosie: check your pixel shader. you are adding additional inputs to the pixel shader that aren't outputs from the vertex shader
airlied: agd5f: any ideas on that ^^ a few ppl have given out
Zajec: me for example
Zajec: :)
Zajec: locking up since that commit
Zajec: i can workaround this... but don't know if you acccept work around
Zajec: (just use old order of initialization)
airlied: Zajec: it wuold be nice to know the root cause
Zajec: yup, believe to
airlied: I was going to look after I fixed all the regressions in my tree, but that is taking a lot long that I thought it would
Zajec: airlied: on the other hand it's not usable for me for about week now
edt: bbiab - rebooting
airlied: Zajec: I haven't had a desktop since last week
airlied: kinda makes fixing anything other than my desktop less important ;-)
Zajec: :P
airlied: hopefully TH TTM fixes help a bit
agd5f: airlied: I don't see how it could cause a problem it just moves the loading of the ucode from file slightly later, but nothing uses the ucode until later in the process
agd5f: unless it's some race or the reading of the ucode adds some delay that masks another issue
airlied: agd5f: yeah it sounds like that smoehow
edt: airlie with 779720a3209849be202ac36a811e934865c50971 reverted drm-radeon-testing does not have desktop font corruption (btw manual bisects are a pita!)
edgecase: ok so it turns out i'm the dumbass who upgraded a bunch of xorg stuff to try KMS, and wiggled the cable loose at the back of the monitor at the same time
edgecase: whoops can I can "ass" in this chan :)
amarks: perhaps the description is deserving in this case :)
edt: agd5f a race could explain what I see. Sometimes the char is ok. What is drawn, corrupt or not, is the correct fg and bg colors.
edt: airlie want me to post on lkml incase anyone else is having drm fun?
agd5f: edt: you might try adding a udelay at the top of r600_startup() or rv770_startup() depending on what chip you are on
edt: rs780
agd5f: r600_startup then
mikkoc: edt: this kind of corruption? http://bugs.freedesktop.org/attachment.cgi?id=32032
edt: mikkoc sort of - mine was worst than what yours...
mikkoc: edt: can you please report it here: http://bugs.freedesktop.org/show_bug.cgi?id=25559
mikkoc: seems the same
edt: agd5f how big a udelay do you suggest?
agd5f: edt: 100 maybe
agd5f: edt: actually add the udelay before r600_pcie_gart_init in r600_init()
edt: ok testing
edgecase: ok compiz can't handle many apps open on my puny 16M VRAM, but it does work
eosie: agd5f: well, in r300g, the pixel shader doesn't know about the vertex shader and vice versa (unlike r300c), both are built separately (no input and output is optimized out) and linked together in the RS unit. I re-checked and it really gets written to the pixel shader (rs_inst_tex_cn_write). I suspect there might be something else busted.
edgecase: vlc fullscreen with texturedvideo is black, non-fullscreen the colors are like CMY instead of RGB
edt: mikkoc added a comment
agd5f: eosie: right, but does the pixel shader has to have instructions to deal with the stuffed coords?
agd5f: eosie: e.g. if you are trying to implement aa lines, you need to blend the fragment fetched from the textures specified with the stuffed coordinates
edgecase: ok n, i only have textured video, adaptor #1 in vlc goes all freaky
eosie: agd5f: I am implementing point sprites
edgecase: textured video looks way better
agd5f: eosie: then you need to blend the aa mask texture (whose coords are stuffed) with the point prim
agd5f: eosie: and set up the additional tex for the mask
eosie: thanks, that seems to be the missing part
evil_core: 0anybody can recommend some GlideWrapper working with radeon driver?
evil_core: I was using zeneck in wine in past even on r200, but now it hangs on r500 :/
edt: bbiab
agd5f: airlied: I think I found the problem, although I'm not sure how this ever worked. r600_blit_init() mmaps vram to write the state and shaders before the mc has been programmed
agd5f: luck I guess
airlied: agd5f: excellent, magic is always good
edt: agd5f the udelay did not help.
agd5f: edt: I'm working on the real fix
edt: grins
dileX_: # cat /sys/kernel/debug/dri/0/radeon_pm_info
dileX_: default engine clock: 392000 kHz
dileX_: current engine clock: 391500 kHz
dileX_: default memory clock: 300000 kHz
dileX_: current memory clock: 297000 kHz
dileX_: Zajec: ^^
spreeuw: airlied: theres a post on lwn about nouveau again
spreeuw: how unfree is that nvidia blob compared to the blobs used in ati
spreeuw: like now with irqs
spreeuw: and the other 2 things in firmware
kdekorte: spreeuw, how I read it was that the nvidia blob was part of the code and not separated out like the ati blob, and so the nvidia code rather than building it in, needed to use the firmware loader like ati
agd5f: spreeuw: AMD allows redistribution of our ucode
spreeuw: ucode being the sourcecode of that binary?
dileX_: lol, "Sir, we are surrounded!" - "Great - we can attack in any direction."
soreau: heh
agd5f: spreeuw: ucode is an abreviation for microcode
Zajec: dileX_: thanks for test :)
Zajec: dileX_: i think i have reclocking patch ready... compiling right now
Zajec: dileX_: can I pass your Tested-by to airlied? :)
dileX_: yupp
Zajec: he should get it faster ;)
Zajec: thx
Zajec: rebooting to test patch
Zajec: dileX_: have V4 :)
evil_core: cs16 works slower than with stable drivers
evil_core: maybe because kernel is stable?
evil_core: generally everything works a bit slower, somebody told me about superious c16 performance with mastwr ;)
soreau: eosie: Did you ever figure out what else is needed for alpha blur to work? Something missing in the frag prog perhaps?
evil_core: and why mine radeon is detected as geforce 3 in wine games?
BioTube: wine only checks for proprietary drivers to properly check the card
BioTube: it just assigns an nvidia model based on extensions for FOSS cards
Ghworg: Unknown opengl vendor string makes wine pretend the card is the geforce
Ghworg: I hacked a version that picks an ATI card instead, didn't make any practical difference in games that I could see
dileX_: Zajec: testing v4
Zajec: dileX_: :)
evil_core: Ghworg: what card? Games runs in wine good for you?
evil_core: All I can say for now, is that radeons sucks ;)
eosie: soreau: I am about to send another round of patches to MAD, though I haven't tested compiz yet, some other things are fixed now (e.g. mesa/gearbox, glean/texCombine, subtests of glean/fbo)
Zajec: eosie: do you work on r600?
MostAwesomeDude: Mm, delicious patches.
evil_core: is there anything interesting to test on r500?
MostAwesomeDude: evil_core: Stop using the Gallium driver, use Mesa.
soreau: MostAwesomeDude: No doubt :)
Ghworg: evil_core: hd2600, old games play fine (half-life, diablo 2, warcraft 3). Nothing new runs at a playable rate last time I tried
soreau: eosie: Ok thanks
eosie: Zajec: r300g only
evil_core: MostAwesomeDude: I am not using it for now
Zajec: ah, ok :)
evil_core: And I dont play new games
eosie: Zajec: I don't have r600 hw
evil_core: but glide wrapper is must have IMHO, allwos to play old games at UXGA, and were 100x quicker than d3d mode
evil_core: Dethkarz rulez, but lags on mine V5200 now
eosie: soreau: neverball works again but I don't think I did anything to fix it, there appears to be some kind of magic
osiris_: airlied: here's the patch for radeon drm to allow for tiled textures http://pastebin.ca/1716322
soreau: eosie: Magic is awesome! :D
osiris_: airlied: I wonder whether we should bump the minor version
Ghworg: I've got more games working now than I did back when I used fglrx. Though that may be partly wine getting better
evil_core: zeneck glide wrapper nearly hangs computer like nfshp2 after upgrade, in fullscreen even magic sysrq doesnt work
Zajec: hm, wine fails totaly under KMS :|
Zajec: works somehow on non-KMS
evil_core: but wine worked for everything I anted on nvidia
Zajec: err:d3d:CreateContext SetPixelFormat failed on HDC=0x3580 for iPixelFormat=110
Zajec: err:d3d:IWineD3DDeviceImpl_CreateSwapChain Failed to create a new context
Zajec: err:ddraw:IDirectDrawImpl_CreateSurface IDirectDrawImpl_AttachD3DDevice failed, hr = 8876086a
Ghworg: Zajec: Works fine for me under KMS. I don't ever run UMS anymore
evil_core: Zajec: equal errors
evil_core: under nonkms
Zajec: oh, weird
evil_core: but works
Zajec: ok, i'll really compare non-KMS vs. KMS
Zajec: now I just citied KMS, maybe get same on non-KMS
evil_core: this iPixelFormat happens when set to pbuffer
evil_core: but works while set to fbo or backbuffer
evil_core: OffscreenRendering in wine registry
Ghworg: Zajec: Set OffscreenRenderingMode to backbuffer. That works the best
Zajec: oh, can I?
Zajec: will try :0
Ghworg: fbo works too now (didn't used to), but is far slower
evil_core: Zajec: yes, and it works also when set to fbo, but theres not fbo
evil_core: Ghworg: for r500?
Ghworg: evil_core: r600
Zajec: what is FBO?
Ghworg: Framebuffer Objects
evil_core: FrameBufferObject
Zajec: frame buffer object
Zajec: ok :)
Zajec: so what should be final&ideal solution? use FBO and make it faster?
evil_core: no, theres no idesal solution
evil_core: different games works with different settings
evil_core: especialy source based needed pbuffer probably or something
evil_core: not everything wil lbe working with backbuffer
dileX_: Zajec: must build a complete kernel
evil_core: Should I also abandon 2.6.32?
Ghworg: fbo is the default I believe, so making that faster would be great
eosie: for those interested: http://storm.unas.cz/r300g-status/
Ghworg: But making everything work correctly is still the priority :-)
evil_core: inte got working currently nearly everything liek nvidia?
evil_core: intel* like*
evil_core: I winder if the worst choice for linux is radeon <=r500
agd5f: edt: http://www.botchco.com/alex/xorg/0001-drm-radeon-kms-make-sure-mc-is-initialized-before-m.patch
evil_core: wonder*
Zajec: edt: what does it fix/
Ghworg: For me the worst choice is nvidia, but that's 'cos I want a completely free kernel (not counting firmware, which I'm fine with being non-free)
Zajec: 7797 commit maybe?
cxo: evil_core, if you want to play games the only way to go is nvidia, both intel and amd suck really badly in that respect
agd5f: Zajec: http://www.botchco.com/alex/xorg/0001-drm-radeon-kms-make-sure-mc-is-initialized-before-m.patch
Zajec: agd5f: what does it fix? 7797?
MostAwesomeDude: eosie: Very nice. I'll help you tackle those GLSL bugs; they'll take some work.
Zajec: agd5f: "drm/radeon/kms/r600/r700: fallback gracefully on ucode failure"?
evil_core: but intel got GLSL, FBO and pbufffer, right?
agd5f: Zajec: related to that
Zajec: agd5f: let me check
evil_core: and r500 will got that in gallium which will be usable in 'unknown feature'?
agd5f: Zajec: I don't really see how 779720a3209849be202ac36a811e934865c50971 can cause corruption or hangs as it just moves the reading of the ucode from file slightly later in the process, however the ucode isn't used until much later.
agd5f: as I said before, I suspect it's just masking another issue
evil_core: Can I got two xorg sessions, one with Mesa and one with Gallium?
soreau: evil_core: You can specify gallium or classic mesa driver on a per app basis in a single X session
MostAwesomeDude: evil_core: FBO and pbuffer work with Mesa.
MostAwesomeDude: evil_core: The worst choice is the S3 Trio, or maybe the Trident.
MostAwesomeDude: The worst choice, out of chipsets actually being made today, is probably the Unichrome.
MostAwesomeDude: Or the Poulsbo.
lordheavy: i vote for poulsbo :-)
dileX_: hmm, need both patches
dileX_: dri-devel/0001-drm-radeon-kms-init-pm-on-all-chipsets.patch
dileX_: dri-devel/0001-drm-radeon-kms-add-dynamic-engine-reclocking-v4.patch
evil_core: I got many s3trios :)
osiris_: MrCooper: here's the patch I'm going to push http://pastebin.ca/1716355
evil_core: one machine I was using over 8 years
evil_core: MostAwesomeDude: it works for r500?
lordheavy: got trio and trident
osiris_: MrCooper: basically I was sending wrong texcoords when destination size != source size
MrCooper: right
osiris_: MrCooper: unfortunately I still not 100% sure that it is correct now
osiris_: MrCooper: but it works for gearbox and prey now
Zajec: agd5f: great work
agd5f: Zajec: works?
Zajec: yup
Zajec: of course
Zajec: agd5f: also I today noticed some single points corruption after switching to VT1
agd5f: cool
Zajec: agd5f: thought it's due to my PM
Zajec: agd5f: so turned it off...but didn't help
Zajec: agd5f: your patch fixes that single points corruptions
agd5f: Zajec: excellent
evil_core: how to check how much memory gfx card got?
evil_core: or maybe ius usable for games
MrCooper: osiris_: looks like the texture contains the topmost line of the colour buffer repeated
Zajec: eosie: nice status, will you update it?
osiris_: MrCooper: hmm, just like the pitch was 0?
MrCooper: osiris_: yeah, or one of the texture coordinates clamped to 1.0...
MrCooper: osiris_: that's on Mobility RV350, did you test on something similar?
osiris_: MrCooper: IIRC I've tested gearbox on rv530 only
hagabaka: I upgraded to kde4.4 beta packages from kubuntu ppa, and desktop effects refuse to work again. anyone else have the problem?
MrCooper: osiris_: could there be a problem with the texture coordinate setup on pre-R5xx?
eosie: Zajec: probably not, since anyone can generate such a page using piglit, I am testing on RV530, the test results for non-R5xx hw may differ
osiris_: MrCooper: I don't know. remind me tomorrow I'll play with it on rv380
alazyworkaholic: I have a 4670 & just downloaded Fedora 12, for experimentation. What's the easiest way to try the newest goodies in the open source driver?
XenoPL: @hagabaka It's KDE known bug
XenoPL: see KDE 4.4 betta anouncement
hagabaka: hmm "compositing support in the window manager being disabled on some hardware"? do you know if they have a list of actual hardware?
XenoPL: haven't looked up for one but it doesn't wokr for me (Radeon 9550) too
cxo: that must suck
XenoPL: also this question was asked few times in last days, so I guess it applies o oss radeon drivem cards
XenoPL: and prolly some other
hagabaka: oh
adamk: Heh... Some guy on linuxquestions.org is looking for help getting 2D and 3D acceleration going on RHEL3, update 9.
airlied: adamk: nice
airlied: somene should send him some beer instead
adamk: I told him that trying to get it working would be a waste of time and to use a more up-to-date distribution.
soreau: But.. its update *9*! ;)
MostAwesomeDude: Just wait. At update 10, ajax will switch to a harder drink.
cxo: yeah, beer++, someone who's trying to do that has likely got bigger problems in life
mentor: Is there a 3D test suite of sorts?
soreau: piglit or mesa demos?
soreau: unless you want the over rated PTS
MostAwesomeDude: piglit is the test suite we use.
mentor: tries to find piglit
MostAwesomeDude: mentor: Why, though? It's not exactly useful.
MostAwesomeDude: It's only a conformance suite; it doesn't address speed.
mentor: Oh well, I just thought since I'd gone to all the trouble to compile the 3d stuff for this (r620) I might as well see if it works.
MostAwesomeDude: Try something more fun. Compiz, OA.
mentor: mmmk
mentor: Tends to be less repeatable though
BioTube: mesa comes with all sorts of demos
mentor: will have a play with those
cxo: overrated :)
roysjosh_home: mjg59, random question: I'm trying to get an old compaq presario (2598) to resume from suspend - it has very similar hardware to a nc4000 (bugzilla.kernel.org 6058, your comment in #4). do you suppose all similar models suffer from that suspend bug?
Ronis_BR: roysjosh_home: I can't suspend here also
Ronis_BR: btw, I'm using RS7800
Ronis_BR: RS780
Ronis_BR: have already talked to airlied, he will give a look at suspend bugs
roysjosh_home: Ronis_BR, this is a rs200 :)
roysjosh_home: it has never been able to suspend. my rs780 is suspended right now, and has successfully for a while, which I am very happy about.
Ronis_BR: roysjosh_home: really?! are you using KMS?
roysjosh_home: Ronis_BR, yep.
Ronis_BR: roysjosh_home: what magic have you done? Can you help me?
roysjosh_home: Ronis_BR, sure :)
Ronis_BR: roysjosh_home: first, what distro are you using?
roysjosh_home: Ronis_BR, I recently helped debug an issue in legacy hw... fedora 12. see https://bugzilla.redhat.com/show_bug.cgi?id=537140
Ronis_BR: can you pastebin your kernel configurations?
roysjosh_home: Ronis_BR, the fix got pushed into airlieds git a bit ago. it's stock f12
Ronis_BR: and what are you using to suspend? tuxonice or uswsusp?
roysjosh_home: other than the patch I'm carrying right now
roysjosh_home: Ronis_BR, oh you're talking about the rs780?
roysjosh_home: Ronis_BR, still stock f12
Ronis_BR: yes I'm :)
Ronis_BR: rs780 i mean
Ronis_BR: can you suspend on that?
roysjosh_home: yep
roysjosh_home: Ronis_BR, it's probably something hw specific or even worse timing related
Ronis_BR: roysjosh_home: so, are you using uswsusp or tuxonice?
roysjosh_home: Ronis_BR, whatever the default is. I just click "Suspend"
Ronis_BR: roysjosh_home: after resume I got ring test fail and a totally corrupted X with KMS enable
roysjosh_home: I think pm-tools or something
Ronis_BR: roysjosh_home: can you pastebin your .config?
Ronis_BR: roysjosh_home: and verify if you have any scripts on /etc/pm/sleep.d/ ?
roysjosh_home: Ronis_BR, it's the default f12 x86_64. see koji.fedoraproject.org? I have to go to sleep too soon to go turn it on - the wife is asking me to turn off the laptop :)
roysjosh_home: mjg59, if you think of anything please reply to the nick 'roysjosh' - thanks
roysjosh_home: Ronis_BR, I'll be on tomorrow and maybe I'll have more time? sorry :)
Ronis_BR: ok thanks
Ronis_BR: no problems fellow :D
mjg59: roysjosh: Quite possible. I never managed to get to the bottom of that, despite talking to HP
Ronis_BR: airlied: that's wierd
MostAwesomeDude: eosie: I've acked 8 of 9 patches. I had a report that some things got weird with patch #1 (buffer_is_referenced), did you see anything like that?
MostAwesomeDude: eosie: Whatever. I'll push all 9, and I think I figured out the thing with my kernel, so I should be able to get back in the game.
soreau: :)
eosie: MostAwesomeDude: cool
edt: agd5f your fix to make sure mc initializes correctly seems to be doing the trick - no font corruption at startup here now. Thanks
edt: agd5f this is with the udelay removed
eosie: MostAwesomeDude: concerning buffer_is_referenced, I think that all of our bo->space_accounted issues might be caused by this function not working properly
eosie: MostAwesomeDude: may I see the report you had?
MostAwesomeDude: eosie: soreau was the reporter IIRC.
MostAwesomeDude: agd5f, airlied : Currently running with this patch: http://paste.pocoo.org/show/157311/
MostAwesomeDude: My rv410 comes up fine, gonna go play Flash games in Fx for a few minutes and see if it still locks up. :3
soreau: You mean this one? neverball: radeon_cs_gem.c:121: cs_gem_write_reloc: Assertion `bo->space_accounted' failed.\nAborted
MostAwesomeDude: Yep.
MostAwesomeDude: Does that still happen with mesa master?
eosie: not here
soreau: It only happens with gallium
MostAwesomeDude: Well, I just pushed 9 patches. Does it still happen with all of them applied?
soreau: And I just compiled latest mesa and it still happens, but i am also in the middle of compiling system updates. Will recompile mesa after that again and retest to be sure
soreau: I am running drm-radeon-testing 2.6.32-38952-gd63ca9d if that matters
eosie: when I first used buffer_is_referenced, it fixed overlapping occlusion queries (I was actually trying to fix something else, but it was nice that these had started to work too), a week later they suddenly stopped working as well as neverball, and now everything works again
MostAwesomeDude: agd5f, airlied : ...And lockup. Any ideas?
eosie: I can't explain it this time
soreau: eosie: Well iirc, you said it didn't happen initially for you, then it did and now it doesn't, though you hadn't changed anything. So I'm thinking possibly it happens intermittently for you
eosie: yeah, it seems to depend on wheather or something
eosie: *weather
soreau: heh
soreau: I am updating everything now, will test again after the smoke clears here
basilgohar: Hey all. I am getting graphical glitches when trying to us a Radeon 1600 or HD3650 in Fedora 12, 64-bit. Can anyone offer some tips on what I can do to get it working or how I can help test it?
DanaG: weird... every time I try nexuiz on my r350, the system freezes.
DanaG: Is that likely a radeon issue, or just an issue with only having 512 megs of system memory?
airlied: radeon
cxo: its frozen many times on my box too
MostAwesomeDude: airlied: Any idea on these lockups? The above patch did nothing.
airlied: MostAwesomeDude: whcich one?
airlied: oh I don't nick higlight for some reason in sceond place
airlied: MostAwesomeDude: not surpriusing
airlied: if you only do it at startup
airlied: you need to do it every bo
airlied: every CS even
MostAwesomeDude: Ugh. Is there a good spot for that? Should I do it every time we dispatch an IB?
MostAwesomeDude: What about for the in-kernel blits?
airlied: MostAwesomeDude: when we emit fences most likely
MostAwesomeDude: airlied: Hm. Doesn't the resync have to be on the pipeline before that, though?
airlied: well you can do it before the fence is actually emitted
MostAwesomeDude: Oh, the fence scratch reg writeback is what's actually causing this?
MostAwesomeDude: In that case, I could RESYNC, do the fence emit, then DC_FINISH right after.
airlied: MostAwesomeDude: I think so, not sure