Home Phoronix Phoronix Forums X.Org Videos From FOSDEM 2008

Radeon IRC Logs For 2009-6-19

Search This Log:

airlied: did someone with an rs690 have KMS issue with current code?
airlied: just wondering what it was
airlied: glisse: so my SiS AGP box works okay with current Linus finally.
airlied: glisse: maybe its the broken r100 VRAM check causing the issue
airlied: glisse: http://people.freedesktop.org/~airlied/radeon-kms-gart-hacks.patch
airlied: that has some hacks and fixes in it, its a bit WIP
airlied: still no rs690
ssieb: :-(
sylware: airlied: I tried the low power option... the fan is still spinning to the max :'( In the last resort I tried the evil/closed driver (but no kernel module because 2.6.30), nothing eitther. How can I gather info about fan control on my board (sapphire based on RV790)?
DanaG: hmm, did you do forcelowpowermode?
DanaG: or try sensors-detect, to see if the standard lm-sensors stuff sees the ... sensors... on the board?
sylware: DanaG: indeed
sylware: DanaG: I have a RV790, is the on die sensor of this chip is I2C?
sylware: (I'm not near that board... sorry)
DanaG: I don't think the on-die one would be, but some desktop cards have off-die sensors.
DanaG: Frankly, I'm just guessing.
glisse: airlied: the pci_set_dma is already call in radeon_device_init
glisse: airlied: Alex told me that not programming agp_base should be fine on rs690
yangman: sylware: the sensors have to be exposed. they're connected to the GPIO pins on the GPU
yangman: sylware: need driver support for it
sylware: yangman: yes, but airlied and bridgman told me that everything about sensors/fan control is on die for rv7xx. Maybe my board still usese off die controller.
yangman: sylware: needs driver support either way
DanaG: fglrx doesn't even give me GPU sensors for my mobility.
DanaG: mobility hd3650.
DanaG: The only thing that gives me GPU die temp is ATI Tray Tools.
sylware: yangman,DanaG: that's a problem... hope all boards manufacturers will go on die sensors/fan controller quickly on rv7xx board.
sylware: yangman, DanaG: I'll try to get more intel from my board... if I'm lucky my board uses the on die stuff and it's just a matter to have documented and published (bridgement told me they are working on it).
DanaG: Oh yeah, it would be good to get sensors info for the mobility ones, too.
DanaG: My previous ATI video card was a 9800 Pro, that didn't even HAVE a user-accessible sensor of any sort.
sylware: :( let's keep it on the bright side: things are improving...
glisse: airlied: and i am still seeing issue on sis
glisse: things is page that get corrupted are never allocated by ttm or radeon
sylware: DanaG: actually I would like to deal with fan control... it's sensitive since playing with it can fry the GPU...
airlied: glisse: but the set_dma mask is wrong for pcie and rs690
airlied: it should be 36-bits, and the table entries need to be fixed like the patch does
airlied: glisse: it mgiht be some other bug :)
airlied: or bad DMA.
glisse: airlied: well i thought that it didn't matter to only report 32bits and just left the upper bit unused
airlied: it causes issues with 4GB of RAM
airlied: since we don't ask for 32-bit pages
airlied: and 32-bit hw doesn't have remapping
airlied: or 32-bit kernel doesn't at least
glisse: you mean on 4g we likely have issue to get page in the 32bits area ?
airlied: I think we probably need to do 32-bit page allocs if we have a 32-bit mask
airlied: glisse: I get pages in the 36-bit area straight away on my rs690
airlied: I then get a fail doing pci_map_page
airlied: since I've only got 32-bits
glisse: this was my understanding of set_pci_dma_mask
airlied: it dosen't affect alloc_page
airlied: it only kicks in when you get to a dma operation
glisse: yeah it works only if you alloc page through dma_* api i guess
glisse: this sis stuff is really painfull
glisse: rawhide definitly work with agp on
glisse: 1hour with gears and firefox and no page corruption
airlied: I sent you my sis machine login internally
airlied: it sounds like an illegal DMA if anything
airlied: the rs690 is wierd, does yours do the same thing? ring test fails?
airlied: btw we don't do readback anymore? why not?
airlied: not readack, sorry, writebac
glisse: the code is their but i didn't pluged it in or use it
MrCooper: set_engine_clock is never called currently, right?
glisse: MrCooper: grep say no :)
airlied: MrCooper: btw I pushed most of your patches ot Linus (I hope you don't mind :)
airlied: the only one that worried me was I though the uninorth user memory changes might need some more bits.
airlied: when I compared it with the other agp user memory changes
MrCooper: airlied: sure thanks, guess I should try to remember adding my S-O-B from now on...
airlied: MrCooper: yeah I'll annoy you for it the next time :)
MrCooper: airlied: quite possibly, what did you compare to?
airlied: the nvidia agp changes TH did
MrCooper: are those in mainline?
airlied: a64d2b37c2259e169759c1701ac565f0a11dc0ea
glisse: airlied: for debugging rs690 looking at sysfs csq fifo and rbbminfo could be usefull
glisse: so you can see if card succeed to read system memory or not
airlied: glisse: yeah I suspect the ring isn't where it claism to be or something like that
airlied: I'll plug in my rs480 next week and see if it does any worse
glisse: need to plug that one too
glisse: but i think i will rather work on r6xx
glisse: i also need to do r1xx/r2xx cs checker
airlied: glisse: r6xx is nice, but it is good to finish the upstream code first :)
airlied: since we have no irqs for r600 yes
taiu: airlied: glisse btw kms fails to init on my RV570 with more than 4G of memory
glisse: well i would like to have fbcon and do some bo move benchmark
airlied: taiu: yeah I think its what I tried to fix with that hack patch
airlied: taiu: might be worth a test
taiu: where's that?
airlied: glisse: yeah fbcon and some bo movement would be useful
airlied: http://people.freedesktop.org/~airlied/radeon-kms-gart-hacks.patch
DanaG: cacafire!
DanaG: fun on KMS.
airlied: we have fbcon hacked up in rawhide.
taiu: airlied: nope [drm:radeon_ring_test] *ERROR* radeon: ring test failed (sracth(0x15E4)=0xCAFEDEAD)
phercek: Hi, I have artifacts in the image when a screen is rotated by 90 degrees. When it is not rotated it works well. You can see the picture of the bug here: http://www.hck.sk/users/peter/pub/xorg/ This is with git version of radeon from yesterday and drm from git://git.freedesktop.org/git/mesa/drm; branch r6xx-r7xx-support. The question is: is it worth reporting or should I expect this will get fixed with drm from 2.6.30 kernel?
phercek: oops, updated the image for a smaller vesion
phercek: it also seems that Option "Position" "1050 160" does not work well, two columns of pixels are show on both the first and the second monitor
airlied: taiu: same as I get on my rs690 with 4GB
taiu: radeon_gart.c does page_base = (uint32_t)rdev->gart.pages_addr[p];
taiu: radeon_ttm.c does gtt->offset = bo_mem->mm_node->start << PAGE_SHIFT;
taiu: where offset is: unsigned offset;
suokko: http://antin.net/wesnoth/xmoto.html :)
taiu: dunno if these are relevat
suokko: BUG: Bad page state in process gnome-keyring-d .....
suokko: That is in dmesg after data in vram is corrupted
airlied: taiu: oh that looks probable bad
airlied: taiu: the first one is bad, the second probably okay
suokko: ARGH! font corruption while I tryed to use gdb :(
glisse: suokko: we don't have any issue with lastest kernel
glisse: you should try those one
suokko: I have started to compile it already but it is takeing sometime
octoploid: Is there a public git tree where one could follow development of KMS for r600/r700, or does it happen behind closed doors?
glisse: octoploid: cgit.freedestkop.org drm-next r6xx wip branch
octoploid: glisse: Thanks.
scarabeus: i have tiny issue that when i run any opengl app/glxgears i see it lag every 2-3 seconds
scarabeus: which is kinda annoying when one try to play some game
scarabeus: card is r400
scarabeus: the mesa is 7.5_rc2
scarabeus: driver for 2d is from git
scarabeus: drm is from 2.6.30 kernel
scarabeus: erm mesa is 7.5_rc3 not rc2
scarabeus: what can i post or where i should look to find out what could cause this
scarabeus: ok fixed the above issue by tweaking the configs
scarabeus: now i have only one issue
scarabeus: err:x11settings:X11DRV_ChangeDisplaySettingsEx No matching mode found 640x480x8 @0! (XF86VidMode)
nanonyme: Doesn't sound too neat...
nanonyme: scarabeus: Which prog requires that mode, btw?
scarabeus: starcraft ;D
scarabeus: game older than 10 years :}
scarabeus: it used to work year back when i last tried ;P
nanonyme: Right... :)
scarabeus: well now i can play half life 2, but not this ;]
nanonyme: That mode isn't visible in xrandr either?
scarabeus: yep it shows 800x600 as minimal
scarabeus: http://dpaste.com/57166/
scarabeus: but it should not
scarabeus: based on the modes definition
nanonyme: I'm on a cell phone. :)
nanonyme: Well, I suppose you could try manually adding the mode with xrandr and hope it doesn't break anything. :)
nanonyme: If you can manage to add it and change to it, probably would work with Wine too.
nanonyme: (since Wine relies on xrandr)
scarabeus: nanonyme: mine wine use XVidMode
scarabeus: not xrandr
scarabeus: but i can add xrandr resolutions
nanonyme: You have a manual override for it?
nanonyme: That is, Wine defaults to xrandr.
nanonyme: It only uses that other with a registry override.
scarabeus: hm i will try the default and try to play with it more, will come back in the evening if i wont reach out something
nanonyme: You can google for "Wine registry keys" if you don't remember how to change it. :) There's a collection of them usually as the first hit.
stikonas: does blender segfaults for any of you with latest mesa and kernel modesetting?
woden: i.e. what's better, radeonhd or radeon?
[Enrico]: imho radeon
woden: Why? And why are the efforts between the two drivers split up? Why don't they join?
mjg59: Because there were certain differences in design philosophy
jcristau: because bonghits.
woden: design philosophy...such as?
MostAwesomeDude: bonghits++
hifi: radeonhd dev's took a few hits more than radeon? :)
woden: so what's the best card to get for 3d support?
woden: and I don't want an r500 series card they are too old
taiu: woden: most of us are on both channels
MostAwesomeDude: woden: "Old" might surprise you.
MostAwesomeDude: Additionally, it sounds like you already know what we're going to say, so why ask?
[Enrico]: woden: atm only <= r500 has 3d support, but r6xx and r7xx 3d support is a WIP
suokko: half hour testing with newer kernel (and airlied's vram size patch) I don't have anymore corruption
suokko: I gues that vram size detection patch fixed it. Now without it I still get corruption
arggg: Hmm.. I'm having trouble getting things up with my older RV250 card and the latest git ati driver. Here's my log: http://pastebin.ca/1466467.
arggg: xorg starts with the computer, the shortly the screen blanks
arggg: and from then on starting X results in a cycle of the screen blanking
arggg: I only get a display on the first start of a boot
arggg: I'm not allowing it to load radeon.ko due to an unrelated suspend issue on other chipsets as well
arggg: nothing in the log hints at what's going on to me, has anyone seen this?
DanaG: hmm, are any of you here on the "fprint" project's mailing list?
DanaG: http://lists.reactivated.net/pipermail/fprint/2009-June/001215.html
DanaG: I want to link to here, but don't want to bother subscribing to the list. Here's where they can get a board: http://partsurfer.hp.com/cgi-bin/spi/main?sel_flg=modinfo&HP_model=KW950AV -- under "PC Board"
DanaG: Unfortunately, it's $28.50 plus $11.25 shipping.
arggg: I just confirmed with an X800GTO that it's not an issue
arggg: does anyone else have RV250 to test?
suokko: arggg: I have rv280
suokko: But I'm using special branch with KMS just now
AndrewR: suokko, i have rv280 (64 mb vram, agp, but works in PCI mode currently) and after some time nearly all widwos start to show garbage .. or just pure black colour. may be this is e16 compositor's fault. I'm waiting for time whan Linus will pull another round of drm fixes
suokko: AndrewR fix for that is in DRI devel list
AndrewR: *windows **when
suokko: "[PATCH] drm/radeon: this VRAM vs aperture test is wrong, just remove it."
suokko: Now I have stable system but there is stil lsome 3D bugs
AndrewR: suokko, yes, but 2d more important for me currently (some Eterm with different make inside). Thanks for testing this, may be i should swap ati card with Nvidia one and test for nouveau instead .... (anyway, not before i will update all X bits)
Arvoreen: hi, I have a general question....I have a X1600 (rv530) card, that shows up in lspci with 2 lines....and I have 2 monitors attached. I've been able to easily get a single wide screen across the 2 monitors thanks to xrandr support, but I want to configure so that I get 2 seperate screens instead. But no matter what I try, I can't seem to have any luck...is this even possible?
adamk_: It is possible if you have two separate monitor, screen, and device sections.
adamk_: However, 3D acceleration will be disabled on both screens.
Arvoreen: hmm, I tried that (and I don't care about 3d), but I could never get it to work. The closest I could get would result in a seg fault
adamk_: What's your /etc/X11/xorg.conf file look like?
Arvoreen: I would have to recreate it,as I'm running now with xrandr to at least have a working system
Arvoreen: but I can talk through it
Arvoreen: I had the device section listed twice, and added screen 0/screen 1 to the options of each
Arvoreen: 2 monitor entries
Arvoreen: 2 screen entries
Arvoreen: and a layout that had the screens side by side
adamk_: I'd suggest recreating the xorg.conf file that you think should work, trying it, and then uploading it and the newly created /var/log/Xorg.0.log file to pastebin.
Arvoreen: ok, I will try that
DanaG: heh, we have an old VGA card around here -- yes, a VGA card.
DanaG: with 1 or 2 megs of video RAM.
mjr: I lack that, but I do have two ISA token ring cards
Arvoreen: ok, experiment done, here is link to pastebin: http://pastebin.com/m25ef8a65
Neo_The_User: ..where is agd5f?
rah: what does "DDX refer to in the phrase "latest DDX"?
rah: and for that matter, what does "latest" refer to?
ajax: 2d driver.
rah: s/DDX/DDX"/
rah: xf86-video-ati?
ajax: yes, that.
rah: ok, thanks
suokko: is happy with stable KMS/DRI2 system :)
stikonas: suokko: does framebuffer graphics (bootsplash, etc...) work for you with KMS?
suokko: usplash doesn't work and I haven't tested any graphics in console yet....
suokko: fbcon and vt switching works nicely (fbcon is a bit slow but usable)
suokko: Is it ossible to port sdl graphics to console now? :)
stikonas: on rs480 neither plymouth nor mplayer -vo fbdev nor fbview work for me
stikonas: though plymouth worked a couple of months ago
suokko: -vo fbdev doesn't work with my mplayer either
suokko: Time to test agp mode :)
arggg: hmm so RV250 fails to work -- the display keeps switching off for no good reason
arggg: and my R400 has serious hardware issues :/
arggg: http://img145.imageshack.us/i/bleh.png/ is the coruption I see -- unless only using the console (and not after starting X) -- I'm pretty sure it's a hardware issue
arggg: however sometimes I don't get it when I boot
suokko: Compiz&video works well in agpmode :)
suokko: flash doesn't like to play nice with compiz :( I hate it when choices are to use flash or wine
nanonyme: Eh?
nanonyme: suokko: That made little sense, I think. ^^
nanonyme: Neither Wine nor Flash work well with Compiz afaik.
suokko: I already thought that I could compiz on when videplay back and basic 3D seemed like working well
nanonyme: So the choice is more between using Compiz and everything else, not Flash and Wine. ;)
suokko: But then flash is doing something evil. And I want to use a program that ha flash version or windows version only
nanonyme: Hmm... Do you consider DRI2 to be basic 3D? ^^
suokko: There is still bugs in mesa so only basic 3D works
suokko: diffuse lightning is broken and at least torcs is crashing
nanonyme: Well, you need full DRI2 for Compiz with Flash or Wine.
nanonyme: (Full as in properly working)
nanonyme: Right.
nanonyme: (That is, if you want hardware rendering in Flash or Wine. Then again, I've heard Flash doesn't like detecting open drivers very much)
nanonyme: If not, could try Windows Flash over Wine. I've noted it in some cases runs programs that Linux Flash can't.
suokko: But at least it seems like DRI2 will be fully working soon. :)
suokko: Linux flash runs everything nicely but it causes rendering problems with compiz. There is sometimes flashing some bogus triangles all around the screen
[Enrico]: here flash work nice with kde4 composite
[Enrico]: works*
suokko: That might be also dri driver depend bug. I gues not too many others are using r200 cards anymore
suokko: http://antin.net/wesnoth/compiz.ogv I just played around a bit with compiz effects. Too bad recording pushes my system to the limits.
DanaG: hah, gpu caps viewer says I have -1 / -1 TUs / ROPs
DanaG: tHAT doesn't make sense.
ajavid: hello guys, how are you doing?
ajavid: I hope all is well
ajavid: I have build xorg 7.4 on debian stable lenny and I have glxinfo|grep ren telling me I have software rasterizer
ajavid: http://paste.debian.net/39729/ <- this is my xorg log file, Im looking through it
ajavid: could anyone kindly take a peek and help me to troubleshoot this?
ajavid: I am using X1900XT radeon driver, xorg 7.4, debian stable lenny, 2.6.29 custom
ajavid: when i had build xorg7.4 from the source in sid a while ago, from git, I had to also get this pacakge called firware-linux which had some radeonXXX.bin file
ajavid: a firmware that was required, I can't seem to find that package any loner, plus, nothing related to it in the log anyway so far
ajavid: I see something on line 770+ about that
ajavid: (WW) RADEON(0): Direct rendering disabled(II) RADEON(0): XAA Render acceleration unsupported on Radeon 9500/9700 and newer. Please use EXA instead.
ajavid: what does this mean?
dmb: wonders if glisse or airlied is around
ajavid: sorry got dc
ajavid: (II) RADEON(0): EXA Composite requires CP on R5xx/IGP
ajavid: http://paste.debian.net/39735/
ajavid: new log file
ajavid: what is cp?
ajavid: nm i got it guys
ajavid: dri section in xorg.conf mode66 fixed it
ajavid: thank you anyway, good evening
spstarr: Installing : kernel-2.6.31-0.11.rc0.git13.fc12.x86_64 163/373
spstarr: airlied: .31 hits rawhide :)
cooper_: ajavid: probably you should edit xorg.conf