Reilithion: Any reason why the radeon driver would make it impossible to switch to a virtual terminal while using 2 graphics cards?
spstarr_work: hmm
spstarr_work: --- Comment #3 from Lubos Lunak
spstarr_work: Sorry, but if X locks up, it's X problem. We can't even know how to work around
spstarr_work: that without X.Org developers analysing the problem.
spstarr_work: [Bug 178117] X locks up with VT switching when kwin composite is enabled
spstarr_work: yay!
spstarr_work: may DRI2 fix this issue
glisse: DRI2 is not the answer to everythings
glisse: problem is how we deal with the GPU
felipec: nope, it's 42
glisse: what we need is a solid way to reset GPU then we need to avoid sending stupid cmd to it
otaylor: still thinks that "hang reports" with a trace of the last stuff that got sent to the GPU would go a long way to getting these cases fixed much sooner
glisse: otaylor: things is what ring is reading might be way ahead of what gpu is doing
glisse: and sometimes even dumping the fifo doesn't show the source of the hang
glisse: also ddx should be less stupid and stop wait for gpu in infinite loop and fallback to software rendering
otaylor: glisse: I would guess that a command buffer dump would be informative a good portion of the time, even if you didn't have exact "IP" information but then again I haven't succesfully diagnosed many hangs, so I could be wrong
otaylor: glisse: you'd have to figure out how to get the "please reboot now" to user space
glisse: i think userspace should understand that after waiting for a second for GPU idle things have just gone wrong
glisse: well of course i can come up with a patch for that :)
otaylor: glisse: sorry, by "user space" I meant "desktop land"
otaylor: glisse: not X server
otaylor: glisse: it's good if your machine doens't hang, but not as good as it could be if it mysteriousy becomes slow with no message to the user
glisse: rebooting is not a necessity, we can reset the card we just haven't figured out how to do it yet
jcarlos: spstarr_work: One question: You have the same graphics chip that mine (RV350) ... Does kwin automatically enable Composite mode on your KDE 4.2 ?
spstarr_work: it would yes, but i turn it off
spstarr_work: since i can't use it right now with KMS
jcarlos: spstarr_work: But in my system it doesn't enable Composite. And the autoself test fail.
jcarlos: spstarr_work: I use the radeon free drivers ...
jcarlos: spstarr_work: 6.9.0 with Xserver 1.4 (debian unstable)
spstarr_work: likely too old KDE
jcarlos: KDE from trunk
jcarlos: Compile by myself just now
spstarr_work: hmm
spstarr_work: check your xorg.conf
spstarr_work: no ricer options
spstarr_work: it will abort
jcarlos: No options in xorg.conf
jcarlos: What does you mean by "ricer options" ?
spstarr_work: pastebin your xorg.conf
jcarlos: OK
jcarlos: spstarr_work: http://www.pastebin.ca/1309216
jcarlos: spstarr_work: It's an autogenerated xorg.conf
spstarr_work: erm
spstarr_work: whats your Xorg.0.log
jcarlos: spstarr_work: http://www.pastebin.ca/1309220
jcarlos: spstarr_work: The xserver is using XAA mode ... but if I remember correctly under EXA occurs the same ...
spstarr_work: (II) RADEON(0): XAA Render acceleration unsupported on Radeon 9500/9700 and newer. Please use EXA instead.
spstarr_work: use eXA
spstarr_work: #
spstarr_work: (II) AIGLX: Loaded and initialized /usr/lib/dri/r300_dri.so
spstarr_work: #
spstarr_work: (II) GLX: Initialized DRI GL provider for screen 0
spstarr_work: looks fine to me
spstarr_work: kwin sometimes needs to be kicked
jcarlos: I will try with EXA when I can, but I have low expectations ... :-(
glisse: airlied: is it normal to get warning about kobject drm missing a release function ?
airlied: glisse: not that I know off.
airlied: unless I noob something.
glisse: according to kobject debug printing it's missing
glisse: parent: module & set:module
airlied: oops I should turn that on :)
glisse: well i turned it on because i broke ttm somehow
glisse: :)
glisse: but i don't think i introduced this kobject stuff
airlied: yeah doubt its causing the issue
glisse: well kobject don't allow to reload module properly it seems
glisse: so it might be local to my tree haven't tested master
glisse: airlied: only sysfs stuff are kobject ? nothings in drm_device is ?
airlied: glisse: the new master stuff should be.
airlied: wonders did I push the correct version :)
airlied: oh it isn't, I'm confused I thought it was.
airlied: ah it uses krefs no kobjs
airlied: gem also uses krefs
airlied: so only sysfs is kobject
glisse: that shrink investigations
airlied: and connectors stuff in sysfs.
glisse: yup i think it's connectors stuff
glisse: basicly my radeon module fail to successfully load so maybe connector are created but drm_sysfs_connector_add is never call
glisse: so release callback is never set
glisse: and tada error
glisse: maybe tada is only expressive in frenc :)
glisse: french
glisse: airlied: the way i see it, it's not possible for kdev to be registered without release being set
glisse: but of course log show me the opposite
glisse: Einstein was right about theory and practise
glisse: airlied: could it be because struct are freed without the device being unregistered ?
airlied: glisse: anythings possible :)
airlied: glisse: might ask jbarnes he wrote that sysfs crack.