Home Phoronix Phoronix Forums X.Org Videos From FOSDEM 2008

Radeon IRC Logs For 2008-11-19

Search This Log:


spstarr: airlied: still in EXAcomp + kms...no lockups 4 hours 37 mins
airlied: spstarr: I win :)
spstarr: uh huh
airlied: wonders if my blog post actually makes any sense.
spstarr: airlied: the agp corruption video doesn't explain to me the way that corruption i happening, so differently
spstarr: i see part of the plasma cashew on the menu spattered on it, tiny -- lines, and the broken square as a mouse cursor
spstarr: and various blitted data corruption
spstarr: reading your blog post
spstarr: 2. X does rendering without the vtSema, including hw calls. So if you invalidate the 3D state flags in EnterVT its too late, X has already sent command to the card without resending the state. So invalidate your state in LeaveVT as well. <-- interesting
airlied: yeah it fixes suspend/resume just got the fix in for GA.
spstarr: but since im a kms boy now..
spstarr: that however might fix the issue i saw which was reproduced on resume
spstarr: same result
spstarr: but with kms i didnt have vt switching problems - never tried suspend/resume with kms
spstarr: needless to say, your work puts things on a good track
agd5f: airlied: after working on r6xx, I'm thinking of rewriting all the exa code to use vertex buffers rather than vertexes embedded in the command stream
airlied: agd5f: might be a good plan.
spstarr: airlied: tomorrow I will be using COMPOSITE all day to see if i get any issues with kms
agd5f: not only is it cleaner, but I think newer hw prefers to be programmed that way and we are only making a single draw call for a whole set of primitives rather than a draw call per primitive
agd5f: so less flushing and such
agd5f: we should do the same for 2D. use packet3 blit/paint-multi and just accumulate rects, then submit at the end
airlied: we had some issue before with direction bits.
airlied: which is why we ended up doing it manually if memory serves
airlied: maybe MrCooper remembers better
airlied: agd5f: I expect the r600 is quite like the 965 in what it wants do do
airlied: the setup overhead is the painful bit.
agd5f: yeah there's actually not too much too it ~3600 dwords of common state setup and ~20 dwords of op specific stuff and a vertex buffer
agd5f: The tricky part is how to deal with directional blits
hifi: MostAwesomeDude: ah, didn't have time yesterday to do more extensive testing, I'll try today
MostAwesomeDude: hifi: 'k.
hifi: do you suspect the texcoord problem exists in R300 too?
spstarr: airlied: I am seeing minor corruption
airlied: yeah I haven't fixed all the minor ones
airlied: esp in firefox/evolution places
spstarr: I see it in some menus the image appear corrupt on small sizes
spstarr: except when i mouse over them they look right
spstarr: gets -54
spstarr: im going to try that VT fix w/o kms for fun
spstarr: airlied: its interesting to notice kms w/o X loaded will hang reboot, kms with X loaded,and a reboot initiated (within X) reboots ok
spstarr: i can use magic key to force a reboot though if its stuck
spstarr: testing VT fix in non-kms
spstarr: composite enabled... VT switch....................
spstarr_desk: wedged
spstarr_desk: although it wedged differently, the mouse cursor wont move
spstarr_desk: 12: /usr/lib/libdrm.so.2(drmCommandWriteRead+0x34) [0x65e934]
spstarr_desk: 13: /usr/lib/dri/r300_dri.so [0x80276e]
spstarr_desk: 14: /usr/lib/dri/r300_dri.so [0x802903]
spstarr_desk: 15: /usr/lib/dri/r300_dri.so(radeonCopyBuffer+0xd2) [0x802c6d]
spstarr_desk: 16: /usr/lib/dri/r300_dri.so(radeonSwapBuffers+0x54) [0x802441]
spstarr_desk: 17: /usr/lib/dri/r300_dri.so [0x7fd906]
spstarr_desk: Interrupt enable: 02000200
spstarr_desk: Interrupts received: 2904
spstarr_desk: Current sequence: 231 231
spstarr_desk: Counter sequence: 231
spstarr_desk: CS: 0
spstarr_desk: RADEON_CP_RB_WPTR 0002e620
spstarr_desk: RADEON_CP_RB_RPTR 0002e620
spstarr_desk: 0?
spstarr_desk: no command submissions?
spstarr_desk: back to kms
spstarr_home: im kms + exa comp + composite
spstarr_desk: crash
spstarr_desk: screen went blank
spstarr_desk: cannot ssh in, total wedgeout
spstarr_desk: sleep
boutz: agd5f: late response: It's a R350 and the resolution is 2400x1600, so i am probably hitting the 2k limit
adamk: airlied, Ping?
adamk: Hmmm.. So what does this mean: 'libGL error: drmMap of framebuffer failed (Invalid argument)' ?
soreau: adamk: Curious, which chip do you usually use?
adamk: I have an x1300 at home and an x1950 at work.
adamk: I have an x700 in a machine that is mostly headless, but that I do sometimes run X on.
adamk: Heh. If I disable kernel modesetting with the nomodeset option, I get that drmMap error and it reverts to the software rasterizer.
adamk: If I leave kernel modesetting enabled, and don't specify an AGP mode, X is completely unusuable.
adamk: If I leave kernel modesetting enabled and set radeon.agpmode to 2, I get lots of distortion, but X works, as does direct rendering.
soreau: adamk: Looking at the code in /mesa/src/glx/x11/dri_glx.c it would seem that drmMap(fd, hFB, framebuffer.size, (drmAddressPtr)&framebuffer.base); failed ;)
adamk: soreau, Yeah, thanks for that. That really helps me out.
soreau: Well, what resolution does it fail with?
adamk: Hmmm? This isn't resolution related.
soreau: ok
adamk: Alright, so I rebooted with kernel modesetting disabled again, and now DRI works without that drmMap error. Maybe the card was just in funky state before.,
adamk: It's still be nice to get KMS working here, though :-)
rnoland: adamk: are you cheating on me with another os?
adamk: lol.
adamk: Don't worry, I still have FreeBSD installed.
adamk: Just wanted to test the KMS goodness.
rnoland: yep, i have to look at that... along with the 1000 other things i have to do...
rnoland: i need like 2 month on a desert island with unlimited hardware...
rnoland: and lots of red bull....
adamk: Heh.
tuxcrafter: hello everygbody
tuxcrafter: witch xorg driver should i take for the
tuxcrafter: 01:05.0 VGA compatible controller: ATI Technologies Inc Radeon IGP 330M/340M/350M
tuxcrafter: xserver-xorg-video-ati or xserver-xorg-video-radeon
tuxcrafter: or xserver-xorg-video-radeonhd
agd5f: tuxcrafter: xserver-xorg-video-ati or xserver-xorg-video-radeon are the same driver. it's actually called xf86-video-ati
agd5f: tuxcrafter: radeonhd doesn't support your chip
tuxcrafter: agd5f: ok thanks
adamk: FYI, the game rune segfaults in r300_dri.so.... mesa-dri-drivers-7.2-0.13 from fc10. Not sure if you can tell anything from it, but here is the backtrace: http://pastebin.ca/1261530
spstarr_work: need symbols
spstarr_work: adamk: install mesa debuginfo RPMs then try again
adamk: Hmmm.. I'm not seeing debuginfo rpms in via 'yum search'... Let me see if I have all the repos enabled.
spstarr_work: you can use the command: debuginfo-install mesa
spstarr_work: or the RPM name itself explicitly
adamk: Oh, well I found the repo and enabled it :-)
spstarr_work: that works too :)
adamk: I see a mesa-debuginfo but not a mesa-dri-drivers-debuginfo. I take it that's all I need?
otaylor: adamk: Yeah,there's just one debuginfo pakcage for each source rpm, evenif it builds multiple binary rpms
adamk: Cool.
adamk: Almost done installing it.
adamk: http://pastebin.ca/1261538
jcarlos: I have found a bug in the radeon driver ... but I don't know if it is fixed ... I have found it running the application kstars using the EXA AccelMethod ...
jcarlos: I changed to XAA AccelMethod and kstars now flies. Before my X server put the CPU to 100% ...
jcarlos: Can I do anything to report the bug ?
adamk: I get a similar crash with ut2004, too: http://pastebin.ca/1261548
spstarr_work: adamk: you could check in freedesktop.org bugzilla or log one there
spstarr_work: assuming its upstream issue
spstarr_work: and not Fedora mesa
adamk: Yeah, I probably should. I usually like to run these kind of issues past the developers first, in case this is a known issue that they are working on, but it does seem like a pretty valid bug.
adamk: https://bugs.freedesktop.org/show_bug.cgi?id=17929
adamk: I'm guessing that's the bug.
adamk: I uploaded my backtrace to that report.
spstarr_work: aha
spstarr_work: you hit one of MAD's items
spstarr_work: FOG fog fog!
adamk: MAD?
zhasha: MostAwesomeDude
adamk: Ahhh
Rabenklaue: Hi, are there any known issues with the radeon driver with drm module and suspending to RAM?
Rabenklaue: Without the radeon-driver (rm /lib/modules/x11-drm and restart) it suspends quite fine
Rabenklaue: Also in raw terminal mode without X-server, only when having radeon
Rabenklaue: +drm loaded
spstarr: hmm