Nightwulf|work: hi all
haeger: hello i have great performance regressions when using mutter with my ati firegl (r300) under xorg-server 1.8.0 with xf86-video-ati 6.13.0... can someone confirm this and/or has a solution for that?
rxt0: does tvout work on KMS?
glisse: MrCooper: do you often have lockup ?
glisse: or do you trigger them ? to test my patch
glisse: i have been trying to get hardlockup with my patch applied but here it seems to workaround the issue properly, i will retest with AGP latter today
Ivanovic: hi
Ivanovic: using a 2.6.34-rc* kernel and the radeon module with dynpm=1, what is required to be able to monitor the current state it is in?
Ivanovic: that is: is there some file in sysfs and if i don't find it, which kernel option am i missing?
mjt: prefers a watt-meter for that task...
mjt: ;)
Ivanovic: i don't want to see the power consumption
Ivanovic: i just want to check that it does scale down and up again when under load
mjt: wattmeter will show you
Ivanovic: and "when under load" power usage will automatically go up due to the cpu utilization
Ivanovic: so no, it will *not* show me if the power state do work
mjt: i mean, say, with cpufreq (for the cpu, not gpu) and without - idle power consumption drops for quite a bit
Ivanovic: oaky, lets assume that i don't have a freaking power meter at hand and just want to see what the driver does set
mjt: but actually that was sort of a joke, note the smile
mjt: it works, but it's not what you want
Ivanovic: okay, repeating the question:
Ivanovic: how to check the current state (clock speed for ram and gpu) using a .34-rc kernel?
Ivanovic: yes, at booting it does list several supported power states and i have dynpm=1 set with the radeon module
Ivanovic: i just don't find the corresponding entry in /sys where i can check this
Ivanovic: i think i once read at phoronix that there was such a thing, though it might require some specific params built into the kernel
Ivanovic: if that is the case: which ones
dileX: Ivanovic: there should be infos in dmesg
Ivanovic: (even better: is this powersaving stuff documented somewher?)
Ivanovic: dileX: the supported states are listed in dmesg when starting
Ivanovic: dileX: the question is the current mode when eg starting a 3d app
mjt: and also if it works with a built-in (780g) gpu...
mjt: Ivanovic: how it looks like in dmesg?
dileX: Ivanovic: me currently using d-r-t. but have a look into /sys/class/drm/card0/device/
Ivanovic: the dmesg stuff for the radeon/drm part: http://pastebin.com/PL6x1X63
dileX: [drm] radeon: dynamic power management enabled
dileX: [drm] radeon: power management initialized
mjt: aha
mjt: found it @here too
mjt: 2 power states for the 780g thing
DanaG: That PM code (as I've commented before): needs to be able to choose a STATE, and let clocks vary within that state.
dileX: Ivanovic: IIRC debugfs has some infos
Ivanovic: dileX: okay, then i need debugfs...
dileX: (on debian-kernel right now - which has no dynpm-support)
Ivanovic: gentoo system with a vanilla kernel
mjt: DanaG: so does it work? :)
DanaG: Haven't tried that new pm3 set.
Ivanovic: dileX: do you know if anything else is required beside debug filesystem?
dileX: Ivanovic: hehe. found an old email from me with instructions:
Ivanovic: dileX: cool
dileX: Ivanovic: was my first time with debugfs. the pm stuff changed status msg - feel free to adapt.
Ivanovic: dileX: so are both dynclks and dynpm required to have all the stuff?
Ivanovic: or does dynpm include dynclks?
mjt: there's also radeon X module options for that
dileX: dynclk is IIRC for UMS and could be set in xorg.conf
dileX: but the module-option name changed fron dyn* to dynpm now
Ivanovic: okay
dileX: you have CONFIG_DEBUG_FS=y set?
Ivanovic: have not done so before, set it now
Ivanovic: will recompile and reboot (directly switching from rc5 to rc6
dileX: Ivanovic: I have a single d-r-t patch for .34-rc6 if you want
Ivanovic: what does d-r-t do?
dileX: drm-radeon-testing GIT branch
dileX: test branch
dileX: http://git.kernel.org/?p=linux/kernel/git/airlied/drm-2.6.git;a=shortlog;h=refs/heads/drm-radeon-testing
dileX: see also radeon build-howto
dileX: (breakfast time)
dileX: afk for a while
MrCooper: glisse: usually still between around one day to one week; to test the patch I set agpmode=4 or 2 which still locks up immediately, and it doesn't fall back to PCI anymore but hardlocks
glisse: MrCooper: so it hardlock as module load ?
MrCooper: I have it built into the kernel, but yeah
glisse: or just very quickly under X
MrCooper: GPU lockup on IB test
glisse: and if you comment out the reset code in r300 it works ?
glisse: well at least doesn't freeze the computer
MrCooper: haven't tried that, but it worked before the GPU reset rework
glisse: if you could comment out the inside of r300_asic_reset and just print the RBBM_STATUS value that could be helpull
glisse: maybe i will see similar thing on one of my agp bridge
dileX: Ivanovic: http://files.iniza.org/drm-radeon-kms/ (2 patches from mjg59 are in d-r-t now, plus one additional patch). see also README_drm-radeon-kms.txt.
dileX: ...and series file.
Ivanovic: okay, looks like the reclocking is not really working for me with rc6
maligor: oh cool, I'm getting constant gpu lockups running blender, but it's not crashing! :P
maligor: wonderful recovery
Ivanovic: at least i get the same info all the time no matter if i just idle on the desktop or run sacred-gold or glxgears: http://pastebin.com/engci7Ec
dileX: Ivanovic: do a tail on /var/log/messages
Ivanovic: dileX: uhm, could i not just fetch the radeon-testing git and replace the respective part of the vanilla sources with this stuff?
Ivanovic: nothing about powermanagement in the logs
dileX: Ivanovic: I had merge conflicts while pulling d-r-t into linux-2.6 GIT
Ivanovic: outsch
dileX: (IIRC i915_drv.h)
Ivanovic: okay, a merge conflict with this stuff does not matter for me at all
Ivanovic: since i don't have the intel driver stuff active over here...
MrCooper: glisse: I'll try that; I had already added some debugging output there but unfortunately netconsole hasn't caught any of it yet
glisse: MrCooper: yeah i tend to add mdelay(50) or even 500
glisse: to get info
glisse: hard lockup are painfull
glisse: i need setup my firewire stuff again
glisse: MrCooper: btw different issue, can you use non native resolution on your powerbook ?
Ivanovic: hmm, should it be possible to just drop the whole gpu folder from drm-radeon-testing into the rc6 folder?
Ivanovic: okay, does not work
MrCooper: glisse: yes
yoshi314: hi there, does anybody know what exactly does this mean :
yoshi314: Mesa warning: external dxt library not available: texstore_rgba_dxt3
yoshi314: same for dxt1
yoshi314: it's preceeded by "Mesa: 247 similar GL_INVALID_VALUE errors" at beginning
Ivanovic: looks like it does not scale up the gpu even when running drm-radeon-testing
glisse: MrCooper: on weird, i send a patch that i need on M7 to be able to do so
Zaba: hm
Zaba: I'm trying to run spring, from http://springrts.org, on a radeon hd 4200, using the radeon driver and mesa from git. I get a segfault in r600_dri.so, and the top of the backtrace (below is stuff from spring) looks like http://sprunge.us/JSbI
Ivanovic: interesting
Zaba: the call the game makes is glFramebufferTexture2DEXT
Ivanovic: using drm-radeon-testing it does not make a difference in fps if my graphics card is clocked at 300MHz engine clock or the full 670MHz
Ivanovic: that is at least the case with sacred-gold and (yes, it is no benchmark) glxgears
glisse: Ivanovic: it means driver is cpu limited
glisse: at least for this application
Ivanovic: okaz
Zaba: hm
Ivanovic: glisse: btw great work on the open source driver
Ivanovic: glisse: sacred gold seems to render without a single glitch (was rather problematic several month ago)
Ivanovic: and even speedwise the game is playable now (though not really fullspeed, but the 10 to 15fps that i get at the moment when zoomed out are okay to play it a little
twnqx: helloes...
twnqx: airlied: after 53 kernel build sicne yesterday evening i conclude that 7547a917fa5f3b2406f52c7dcf7ec9ad3c8532eb Merge branch 'drm-ttm-unmappable' into drm-core-next causes display corruption on resume
papillon81: hi there. the following commit leads to problems in OSG: http://cgit.freedesktop.org/mesa/mesa/commit/?id=ad7d36e91933a49ce3f6ed2c14ae585a8a84fdf5
papillon81: i just bisected it and reverting this solves an "Not able to create requested visual." error
mentor: papillon81: I suspect the answer is to implement the tag there, though
papillon81: mentor: so you say this is a shortcoming in OSG?
mentor: er
glisse: agd5f: for scaled mode it seems we definitly need the code i ported from ddx
agd5f: glisse: did you see the new patch I posted?
glisse: haven't got doc to check but it seems all we need to keep from native mode is over, syncwidth & blankwidth
agd5f: avivo needs some other adjustments as well
glisse: agd5f: newer one than previous ?
agd5f: yeah
agd5f: just sent
glisse: you just send it
glisse: ?
agd5f: replied to the previous email
agd5f: glisse: http://people.freedesktop.org/~agd5f/0001-drm-radeon-kms-fix-panel-scaling-adjusted-mode-setu.patch
glisse: apply on top of previous one ?
agd5f: instead of
agd5f: glisse: unforunately, the panel on my legacy laptop is pretty forgiving and takes just about any mode you throw at it
glisse: agd5f: ok this patch works here
agd5f: cool
agd5f: I'll test avivo this afternoon
agd5f: gotta run. bbiab
mjg59: Ok, that seems to basically work. Now I just need to deal with the fact that the gui interrupts don't
twnqx: agd5f: you were right about the bug you linked.
mjg59: agd5f: http://www.codon.org.uk/~mjg59/tmp/radeon_pm/0001-radeon-Reclock-r500-memory-by-hand.patch seems to basically work on rv515 now
mjg59: Need to sort out the gui interrupts issue to test it properly
twnqx: ugh
twnqx: how could commit 7547a917fa5f3b2406f52c7dcf7ec9ad3c8532eb possible affect radeon
twnqx: or is "git show
agd5f: twnqx: it's the merge of the unmappable vram code
agd5f: It was designed for radeon
twnqx: but the git show command only shows 50 lines of nouveau stuff
agd5f: twnqx: it's a merge commit
twnqx: ugh
twnqx: so... how would i see the actual changes?
agd5f: looks up the actual commits
twnqx: because after that patch my resume breaks
twnqx: or merge
twnqx: how... do i make git tell me the commits?
agd5f: twnqx: http://git.kernel.org/?p=linux/kernel/git/airlied/drm-2.6.git;a=commit;h=0a2d50e3a8faaf36cde36920431586090411ea15 and http://git.kernel.org/?p=linux/kernel/git/airlied/drm-2.6.git;a=commitdiff;h=6b8b1786a8c29ce6e32298b93ac8d4a18a2b11c4
twnqx: thanks
agd5f: twnqx: but the whole series involves other patches as well
twnqx: mh
twnqx: so better to use git diff between the two ids?
agd5f: twnqx: starts with drm/ttm: ttm_fault callback to allow driver to handle... and goes to drm/radeon/kms: enable use of unmappable VRAM V2: http://git.kernel.org/?p=linux/kernel/git/airlied/drm-2.6.git;a=shortlog;h=refs/heads/drm-radeon-testing
twnqx: agd5f: i see those patches as sinles, yes
twnqx: singles*
agd5f: mjg59: I'll take a look and test it today
Zaba_: uh oh. What does this mean.. http://sprunge.us/CHFB
twnqx: agd5f: https://bugs.freedesktop.org/show_bug.cgi?id=27822 see my comment therE?
dileX: twnqx: use tools like tig (ncurses-based) tools to browse through commits
Zaba_: it caused my monitor to go 'out of range' for a bit and then the box rebooted..
twnqx: the TTM patches themselves work
agd5f: twnqx: all the patches between those two are that patch series
twnqx: but it's the one with the merge comment that makes my resume fail
agd5f: the first two I mentioned are the ones that touch radeon
twnqx: they don't cause problems, see my comment to the bug
evil_core: dynpm works for any r500 user?
twnqx: i built and tested almost every single patch
agd5f: twnqx: those patches don't exist if the merge commit has been removed
twnqx: huh? but i got different kernel versions with "git checkout
twnqx: gah git is just too confusing
agd5f: twnqx: because different versions have been merged at various points
twnqx: meh
agd5f: if you go back a few commits it may be prior to when the latest upstream version was merged
twnqx: yeah, i noticed i head some rc2 kernels AFTER rc5
twnqx: it's like... totally never what you expect
twnqx: and i have no clue how to tell it "just get me those patches in, one by one, in my rc5 tree"
twnqx: all i can say is: the merge causes erratic PCI BAR mapping
twnqx: (as notice in dmesg)
twnqx: agd5f: just for my understanding.. when i checked out those commits it means i got the original kernel where these patches were applied (somewhere in 2.6.34rc2), and where they worked
twnqx: but after the merge into 2.6.34rc5 soemthing broke - either the merge was incorrect, or something in the kernel changed in between and broke the patches?
edwin: twnqx: if you have a good and a bad commit you can use git bisect, it can cope with merges
edwin: well usually it can...
twnqx: well the problem is
twnqx: it requires a patch after the merge
twnqx: so the kernel doesn't crash during the boot >_>
edwin: you can patch it
edwin: git bisect just gives you a checkout
edwin: its up to you to compile/test it
edwin: so may as well patch it
evil_core: is there tig for svn/cvs too?
edwin: there is git-svn
edwin: that can import a svn tree into git
edwin: useful if you want to bisect a svn repo
edwin: cvs I don't know its been long since I used it, but there is probably some interface between cvs and git
evil_core: but I would prefer native ones, than playing with importing, etc
evil_core: mine distro is using still CVS for SPECS
edwin: evil_core: sorry I don't understand what you want to do then
evil_core: and we use svn at school, but IMHO svn sucks a bit
edwin: weren't you asking for 'git for svn/cvs'?
evil_core: tig
edwin: what is that?
evil_core: somebody there told about tig
evil_core: its curses frontend for git
evil_core: for viewing commits
edwin: ah I'm sure svn has some tools for that
evil_core: i saw it first time minute ago, so mine desc sucks ;)
edwin: svn is usually better in terms of user-friendly tools (see tortoisesvn)
edwin: depending on platform of course
evil_core: but it doesnt got real ranches, etc
maligor: user-friendliness is such a long word
maligor: users who need tortoisesvn for usin' shouldn't be using in the first place
edwin: don't tell me, I'm a git user :D
maligor: the brotherhood of git
evil_core: I dont like git too
evil_core: I like CVS
evil_core: its quick and easy to use, and takes not so much space like git
maligor: are you serious?
maligor: CVS is a nightmare
evil_core: why?
maligor: it just wants to break down
evil_core: I am using it for SPECS and patches only as distro developer ;)
maligor: svn is similar but much better
maligor: but git is the holy word
hnsr: well i use git, and the person using my code is on Windows and right now to get the latest he just fetches a zip through the cgit webinterface, it would be nice if there was something like tortoisesvn for git that would let him just right click on a gitdir and click 'pull' or something
hnsr: there actually is something liek that
hnsr: but it has a fullblown ui with the 'pull' thingy hidden somewhere that I didn't want to expose the users of my code to.. yet :p
evil_core: he can use .bat files :P
edwin: tehre is msysgit, and it has a git shell where you type 'git pull'
hnsr: hrm, yeah I guess that should work well enough
hnsr: thanks
seanius: hi everyeone -- is HDMI audio working yet?
seanius: just compiled the radeon driver from master and have a 2.6.34 r6 kernel fwiw
seanius: it seems like my system sees the device anyway, but i can't get any sound from it
glisse: it should work, some people are using it
glisse: make sure you select the gpu audio controller as ouput
agd5f: mjg59, airlied: http://people.freedesktop.org/~agd5f/pm-drt/ enable voltage drop, other misc pm stuff on r1xx-r4xx
seanius: glisse: that's be HDA-Intel - HDA ATI HDMI right?
glisse: yes select hda ati hdmi
glisse: also double check output volume & similar setup
glisse: and last your tv might need some tweaking
glisse: but i am not hdmi expert
seanius: hm... the only input i see on that card is s/pdif
mjg59: agd5f: I'm still occasionally managing to get corruption on memory reclock on 515, but at this point I'm wondering if it's because things aren't idle to start with
seanius: i've been using the radeonhd driver before and it worked
mjg59: Anyway, I'm testing fences for dynpm now
agd5f: mjg59: perhaps we should be checking the mc idle bits
agd5f: in addition to gui idle
seanius: it's really hard to google this, i get pages either about windows or about the radeonhd driver :/
mjg59: agd5f: Ok, memory reclocking on 515 seems to approximate working...
mjg59: Still some flashes
agd5f: mjg59: missing the vbl or implementation issues?
mjg59: agd5f: I think we need to reclock on adjacent frames
mjg59: Rather than trying to do both engine and memory in one
agd5f: yeah, agreed
mjg59: It almost looks like engine reclocking may also be too slow on 515
mjg59: But that's probably fixable
agd5f: not sure how much effort it's worth. i.e., to have it be stable everywhere
agd5f: might be better to just not adjust the mclk automatically
mjg59: My measurements suggest that mclk is where most of the power saving is
agd5f: yeah, but if it's going to be problematic in a lot of cases due to customized reclocking code...
agd5f: I'd rather have stable and using slightly more power than unstable and using less power
agd5f: I guess we'll have to see how stable it turns out to be
mjg59: Right
evil_core: theres significant difference between V5200 and V5250?
evil_core: in power usage i.e.
mjg59: Turns out that the memory reclocking works a lot better if I program mpll the way the BIOS does
agd5f: evil_core: I think they're pretty similar
evil_core: at power usage also?
mjg59: agd5f: Ok, looks like there's still a certain level of us missing vbl
agd5f: evil_core: probably
evil_core: http://www.notebookcheck.net/Mobile-Graphics-Cards-Benchmark-List.844.0.html
evil_core: its strange
evil_core: according to this table V5200 got higher CPU clock, but lower points in benchmarks :/
agd5f: evil_core: differences in the rest of the platform possibly
mjg59: agd5f: http://www.codon.org.uk/~mjg59/tmp/radeon_pm
evil_core: mjg59: will not hang mine r500? ;)
maligor: evil_core, http://en.wikipedia.org/wiki/Comparison_of_AMD_graphics_processing_units is a very convenient list
agd5f: mjg59: looks good
mjg59: evil_core: Might not hang your r500
mjg59: Give it a go, anyway
mjg59: It may be that we can't get memory reclocking stable on r500
evil_core: equal nm for both? I was reading many times i nthe past that 'nm' are the only difference, so V5250 uses less power and gets less hot
evil_core: I was very scared when waited for mine T60p after reading about firegl heating problems(and that you cannot play games at it), and mine gets hot, but awlays stable :)
maligor: if you mean the wikipedia list, well, it could be incorrect
evil_core: mjg59: but Zajec patches didnt hanged mine r500
evil_core: I got only frame loses when reclocking
agd5f: mjg59: http://people.freedesktop.org/~agd5f/pm-drt/0004-drm-radeon-kms-atom-disable-the-encoders-in-encoder.patch
evil_core: why V5200 according to this table having higher memory clock got lower bandwith, something like dual channel?
agd5f: should save some more power
agd5f: depends on the oem.
mjg59: agd5f: Nice!
evil_core: 15.2so what patxhes I need to apply, mjg59 1st and layer agd5f?
agd5f: evil_core: yeah, that should work. on top of d-r-t
evil_core: and why pm2 is only for KMS?
agd5f: evil_core: note that the lowest clock mode on my t60p causes issues with the panel still, so avoid that mode until we sort it out
evil_core: it depends so much on it?
agd5f: evil_core: http://people.freedesktop.org/~agd5f/pm-drt/
evil_core: agd5f: lspci!
evil_core: you got V5200 or V5250?
agd5f: evil_core: 5200
evil_core: I got V5200
evil_core: hmm..so it whould not hang mine if doesnt your, right?
agd5f: right
evil_core: when you got your T60p?
agd5f: although I haven't tested all of mjg59's latest patches yet
evil_core: omg, I was thinking I am talking now tih mjg59 ;)
evil_core: al lyou got strange nicks with numers between ;)
agd5f: evil_core: the pm stuff is for kms because it's not really feasible on ums; you need interrupts and control of the full card state
evil_core: but KMS sucks in wine :/
agd5f: so fix it :)
ajax: ITYM "wine sucks in KMS". HTH. HAND.
evil_core: I am too dumb
evil_core: agd5f: I should apply all your patches, only only one mentioned there?
agd5f: evil_core: they should all be fine
evil_core: ajax: if something works in UMs, then kits regression
evil_core: 0003 patch failed
agd5f: evil_core: are you sure you have the right tree checked out?
evil_core: mj0003-radeon-Use-fences-to-gate-entry-to-reclocking-on-r60.patch <
evil_core: patching file drivers/gpu/drm/radeon/radeon_pm.c
evil_core: Hunk #1 FAILED at 77.
evil_core: never sure of that, git is strange
agd5f: evil_core: oh, mjg59's patch 3. you need my 0002 patch first
User_007: Hello. I am using the new kernel with modesetting, with last stable mesa.
evil_core: I got mjg 0001 and 0002, and agd5f 0001 and your 0002 failed
User_007: it happens that i can get almost 600 fps with full power in processor, but when i try to play a simple game like Neverball it appears to have a lot of bugs
User_007: and glxgears get a lot of bugs too...
agd5f: evil_core: is your d-r-t branch up to date?
evil_core: yes
agd5f: git fetch; git pull origin drm-radeon-testing
evil_core: but can be 'broken' a put ;)
evil_core: Already up-to-date.
evil_core: Maybe I tried to patch in bad order?
agd5f: evil_core: try all of mine, then all of mjg59's
User_007: like: Modesetting works "well" (runs at a considerably speed, like before) but it started to cash my games.. Can anyone help me with this? maybe i am doind something really whrong here..
evil_core: I applied all your except 0002
evil_core: buit I got r500, so it doesnt matter probably
agd5f: evil_core: it does. you need mjg59's fence patch
evil_core: I cannot apply your 002 and mjg 003, everything else applied
agd5f: evil_core: does your tree have b94756abf96e624201e0c575ba4997de9aa942fa ?
agd5f: git show b94756abf96e624201e0c575ba4997de9aa942fa
evil_core: -#if 0
evil_core: I dont get suchs statements
evil_core: # git show b94756abf96e624201e0c575ba4997de9aa942fa
evil_core: fatal: bad object b94756abf96e624201e0c575ba4997de9aa942fa
agd5f: evil_core: your tree is not up to date
evil_core: I executed your commands
agd5f: git pull origin drm-radeon-testing
evil_core: I am resetting it now
evil_core: # git pull origin drm-radeon-testing
evil_core: From /mnt/download/soft/drivers/gfx_card/radeon/drm-2.6
evil_core: * branch drm-radeon-testing -> FETCH_HEAD
evil_core: Already up-to-date.
evil_core: but still bad object :/
agd5f: evil_core: what does git branch show?
evil_core: # git branch
evil_core: * drm-radeon-testing
evil_core: commit 65cab33d813e684accd4c1ac3b9132b08ffc1976
evil_core: its last commit I see
agd5f: evil_core: yeah, that's two days old
evil_core: I got old pm2 tree merged there, maybe it causes problems
evil_core: its fresh?
agd5f: ugh, yeah get rid of that
evil_core: 65cab33d813e684accd4c1ac3b9132b08ffc1976
evil_core: it causes problem probably
agd5f: evil_core: ?
evil_core: because it removes if conditions
evil_core: commit that braks 0002 patch
agd5f: evil_core: different files
agd5f: evil_core: http://git.kernel.org/?p=linux/kernel/git/airlied/drm-2.6.git;a=shortlog;h=refs/heads/drm-radeon-testing
agd5f: you are missing 3 commits
evil_core: it caused problems, I added manually #if 0 and #endif and applied it all
evil_core: agd5f: tell me how to reset it ;)
agd5f: evil_core: delete the branch and start over: git branch -D drm-radeon-testing
agd5f: git fetch
agd5f: git checkout -b drm-radeon-testing origin/drm-radeon-testing
glisse: or git reset --hard origin/drm-radeon-testing
MostAwesomeDude: 41ee49
evil_core: error: Cannot delete the branch 'drm-radeon-testing' which you are currently on.
MostAwesomeDude: Oops, ignore that.
evil_core: so I need to make tem pbranch??
agd5f: evil_core: git checkout master
evil_core: I cannot reset it
agd5f: first
evil_core: reseting doesnt work
evil_core: # git reset --hard origin/drm-radeon-testing
evil_core: HEAD is now at 65cab33 radeon: Enable memory reclocking on r100-500
evil_core: maybe some fu*** proxy?
MostAwesomeDude: $ git fetch origin
MostAwesomeDude: $ git reset --hard origin/drm-radeon-testing
evil_core: MostAwesomeDude: I am resettinjg it all the time ;)
evil_core: # git fetch origin
evil_core: [root@sinistar drm-2.6_pageflip]# git reset --hard origin/drm-radeon-testing
evil_core: Checking out files: 100% (17676/17676), done.
evil_core: HEAD is now at 65cab33 radeon: Enable memory reclocking on r100-500
evil_core: will try fetching master, deleting drt and fetching it again
MostAwesomeDude: Is origin actually the right remote?
evil_core: # git checkout master
evil_core: error: Untracked working tree file 'Documentation/ABI/testing/sysfs-class-usb_host' would be overwritten by merge.
evil_core: MostAwesomeDude: I dunno anything about git
MostAwesomeDude: Clearly.
MostAwesomeDude: What's origin point to? $ git remote -v
agd5f: evil_core: git checkout -f master
evil_core: # git remote -v
evil_core: origin /mnt/download/soft/drivers/gfx_card/radeon/drm-2.6 (fetch)
evil_core: origin /mnt/download/soft/drivers/gfx_card/radeon/drm-2.6 (push)
evil_core: its a local copy saving space ;)
agd5f: there's your problem :)
evil_core: so maybe I need to always checkout drm-2.6 only?
evil_core: ok, thanks
agd5f: origin needs to be git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6.git
evil_core: but I need to update both?
MostAwesomeDude: I have something like this in my tree: $ git remote add airlied git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6.git
MostAwesomeDude: And then the branches are copied from airlied instead of origin.
evil_core: but will they still sabve space on hdd?
evil_core: I did a local copy because after messing in drm-2.6 I got problem updating it
evil_core: so I got copy I can mess and always easily delete and recopy it
MostAwesomeDude: In an ideal world, git trees would never need to be re-cloned.
MostAwesomeDude: Ever.
MostAwesomeDude: Because you can always reset them.
MostAwesomeDude: But do whatever you like.
evil_core: aybe I shiuld learn git ;)
airlied: agd5f: do you have a hack to stop rv530 using the lowest power mode?
agd5f: airlied: not yet
agd5f: airlied: I was going to try bumpping the sclk up a bit to see if that fixed it
agd5f: I suspect the spll and ppll are interfering or something. probably needs some spread spectrum fix. maybe dpms cycle
evil_core: It does hanged mine machine ~10seconds after starting xorg
evil_core: but I will try to clean and build kernel fro mscratch
evil_core: agd5f: you told me to do something with not setting to lowe power mode or something
agd5f: evil_core: yeah, you'd need to add a hack, or just manually set the power states
evil_core: agd5f: but can you share that hacks?
agd5f: evil_core: I haven't written it yet
evil_core: so how are you running it on T60p w/o hangs?
agd5f: evil_core: manually setting power states
evil_core: how?
evil_core: echo 3.0?
evil_core: or 2.0
agd5f: yeah
evil_core: so I need to add it to script loading radeon, right?
agd5f: echo 1.0 > /sys/class/drm/card0/device/power_state
lwhacker: I have an ATI FirePro 2260. Whenever I try doing an "X -retro -config /root/xorg.conf.new", the screen goes blank and I can't access the puter physically until I reboot. Here's my info: http://pastebin.com/spam.php?i=pafpw7Ca
agd5f: no
agd5f: evil_core: you can do it at run time
evil_core: to send it imidietaly after loading module
evil_core: yes, but I cannot do it after GPU hang ;)
agd5f: evil_core: you can do it anytime
evil_core: when GPU hangs, I cannot do anything
agd5f: mjg59: hmm... we probably need to check if the cp is enabled or not and if not, assume we don't need fences
agd5f: or gui idle ints
agd5f: so we can still do pm even if accel is off
evil_core: agd5f: I understand GPU shouldnt hang w/o that, only messy screen?
agd5f: evil_core: messy screen in power_state 0.0
agd5f: evil_core: also, you don't need to pass dynpm=1 to the drm anymore at load time
agd5f: you can enable it on the fly with sysfs
evil_core: echo 1.0 > /sys/class/drm/card0/device/power_state
evil_core: causes total mess
evil_core: but no hang at least
evil_core: i needed to call radeon_restart anyway
mjg59: agd5f: Mm, true.
evil_core: agd5f: even sysrq+b doesnt work after enabling dynpm in xorg if not 1.p0 is not used
evil_core: agd5f: yiu dont have hard lockup?
agd5f: evil_core: works fine here
Obscene_CNN: any idea what would cause this?
Obscene_CNN: #
Obscene_CNN: (II) RADEON(0): I2C device "DisplayPort-1:ddc2" registered at address 0xA0.
Obscene_CNN: #
Obscene_CNN: (II) RADEON(0): Output: DisplayPort-1, Detected Monitor Type: 0
Obscene_CNN: #
Obscene_CNN: finished output detect: 0
Obscene_CNN: #
Obscene_CNN: (II) RADEON(0): I2C device "DisplayPort-0:ddc2" registered at address 0xA0.
Obscene_CNN: #
Obscene_CNN: (II) RADEON(0): Output: DisplayPort-0, Detected Monitor Type: 0
Obscene_CNN: #
Obscene_CNN: finished output detect: 1
Obscene_CNN: #
Obscene_CNN: (EE) RADEON(0): No connected devices found!
agd5f: Obscene_CNN: nothing connected?
Obscene_CNN: a monitor is connected
agd5f: displayport monitor?
mancha: you should pastebin in the future i think, the scroll's a bit of a pita
agd5f: displayport doesn't work with the 6.12.x series of the driver
Obscene_CNN: so do a git install?
agd5f: git or 6.13.x
Obscene_CNN: okay, thanks
Obscene_CNN: mancha, lwhacker did pastebin it but noby helped him
mancha: did he ask for his money back?
evil_core: mancha: whats going on?
agd5f: evil_core: you could try without http://www.codon.org.uk/~mjg59/tmp/radeon_pm/0001-radeon-Reclock-r500-memory-by-hand.patch
evil_core: agd5f: I dont understand one thing. GPU doesnt hang when you echo 1.0, or even without that xorg works with dynpm?
agd5f: evil_core: yes
evil_core: and what about dynclks?
agd5f: evil_core: I'm not using that
evil_core: I have it also set to 0
evil_core: biut all its weird
evil_core: but*
dileX: [22:40:08]
evil_core: dynpm is for reclocking, and dyclks for disabling parts of chip?
agd5f: dynclks is for clock gating which stops the clock to different blocks when they aren't in use
evil_core: dileX: no, I dont have problem with applying patches
dileX: nice, new pm patches
evil_core: dileX: ?
evil_core: something new I dont have yet?
dileX: evil_core: I was reading backlog
evil_core: agd5f: no hang w/o 0001
evil_core: agd5f: what to do now?
agd5f: evil_core: that's fine. it means that the manual memory reclocking code isn't stable on your system, but the atom version is
evil_core: agd5f: but 0001 works for you?
agd5f: evil_core: I haven't tried it yet
evil_core: why you are not using dynclks, wouldnt it help much in power saving?
agd5f: evil_core: not really my focus at the moment
evil_core: and by not using 001 I am not losing anything, only using atom version, right?
agd5f: right. the reclock just might take a bit longer which may result in it happens
agd5f: since it might take longer than the vbl period
agd5f: s/happens/flickering/
evil_core: 17W
dileX: mjg59 0001 causing problems?
dileX: I mean 0001-radeon-Reclock-r500-memory-by-hand.patch
evil_core: yes
evil_core: for r500 in T60p
evil_core: w/o it theres no hangs
dileX: same patch (with the version 24hrs ago) I could not start X - so I am excluding it
evil_core: dileX: what machine do you have?
dileX: hp compaq nc6400 with rv515 (radeon mobility x1300 pci-e)
evil_core: now sevs there got T60p, hp isnt as pr0 :P
dileX: pr0?
evil_core: thinkpads, terminal, wmaker, vim are pr0!
evil_core: ideapads, GUI, gnome, gedt arent
evil_core: understand? :P
chithead: some thinkpads (l/sl series) are actually closer to ideapads than to other thinkpads
dileX: then you dont need X and even radeon
evil_core: but I got IBM branded T60p, not SL!
evil_core: or edge
evil_core: with FlexView, thinklight. etc, so its last real tghinkpad ;)
evil_core: I would make frankenspad,but in Poland it was hard to buy this one
evil_core: peoples there hate matt screen, noblue leds, "square screen" and nonglossy finnish
mjt: um blue leds. hate hate hate hate... :)
evil_core: they love there Dells at most, later pavilions, and probably man want vaio
evil_core: and now theres ideapads
evil_core: lenovo is winning
evil_core: people thinks that idea is "beaty" and better thinkpads
evil_core: and for notebooks there are msi winds mostly and eepc a bit
evil_core: theres no blue led in T60
evil_core: dunno about T61
evil_core: but T61 has ugly wider left edge, dunnp why
agd5f: mjg59: 0001-radeon-Reclock-r500-memory-by-hand.patch hoses my laptop as well
evil_core: he only writes patches, not test ;)
agd5f: well, likely the mem reclock sequence is different on different oem boards
agd5f: also, rv515 vs. rv530, etc.
evil_core: this patch osnt eating power while reclocking, its changing all the time
evil_core: ~100 reclocks per seconds
evil_core: so the only problems now are flickering and fuzzy screen(too much clocked down)
evil_core: Top causes for wakeups:
evil_core: 41.7% (201.2)
evil_core: 10.4% ( 50.2)
evil_core: it sucks
evil_core: after killinf wmtictactoe, wmpong and wmMAtrix screen is fizuzzy all the time
evil_core: disabling pm doesnt work :/
evil_core: it doesnt save power on T60p :P
evil_core: maybe its helping on shits like GNOME/KDE that throws IRQs all the time, but there radeon driver is waking up CPU all the time
evil_core: didnt got this problem in november
AndrewR: evil_core, hello. Hello to all.
AndrewR: testing new drm-radeon-testing i found it unstable, and while trying agpmode=-1 i also found non-default gartsize (64, not 512) was more stable on low-memory machine (mem=192m, physically installed 768 mb). But limiting gartsize has funny effect after few hours of seamonkey usage: starting tvtime killed X , guess it was trying to allocate buffers (in VRAM then in GTT, but everything was full, so kernel -22 us ...)
AndrewR: -12, not -22. [drm:radeon_gem_object_create] *ERROR* Failed to allocate GEM object (847872, 2, 4096, -12)
evil_core: I went to 13.4W on T60p with screen dimmed and ethernet wo/ dynpm
evil_core: and 31.1W with dynpm
evil_core: its not impressive
Obscene_CNN: declares war on assert statements to see if removing them helps
evil_core: hi AndrewR
evil_core: AndrewR: agp sucks
AndrewR: http://pastebin.com/70Zx9Nr9 - more full dmesg error ...one can blame ramzswap .... but ... is there really no (easy) way to protect at least 2d driver from crashing like this? And may be related question - if i remmeber correctly, there was two places for Xv buffers with KMS, in VRAM and GTT. I _think_ i saw something i can describe as texturing from PCI gart space, i.e. unusually slow tvtime on one ocassion. can ddx detect what kind of gart kernel us
AndrewR: es, until we get dual-gart support?
evil_core: can I set gpu/mem manually?
AndrewR: even more questions .... PCI (not PCIe) gart tables are saved in VRAM, right?
evil_core: I want to get rid of fuzzy screen
AndrewR: evil_core, if you hack driver .... try to expose memclock/core clock vars via sysfs ? (or it might blow up ... setting wrong clocks can hang card)
evil_core: agd5f told we ce can set states currentlyy
Ronis_BR: hi guys
Ronis_BR: is there any improvements with kernel 2.6.34-rc6?
evil_core: use d-r-t
AndrewR: evil_core, my desctop card has just one defined state :)
AndrewR: *desktop
AndrewR: evil_core, http://pastebin.com/pAUAP0P5
AndrewR: this is rv280
evil_core: echo 1.0 > /sys/class/drm/card0/device/power_state
evil_core: what it does?
AndrewR: evil_core, guess set state, and then clock inside selected state ..... ? (as defined by video-BIOS)
evil_core: but clocks no longer vchanges
evil_core: can i leave memoy-only or gpu recklocking
evil_core: memory*
evil_core: to foind out whats causes fuzzy screen
AndrewR: evil_core, may be if you or someone else adds few more sysfs files, or module params ... you can just try to "if 0 ... endif" parts of code, or make somethink like this ...
AndrewR: *something.. like // comment gpu_reclocking_func()
mjt: umm. Just installed 2.6.34-rc6, and gave it a try with nouveau driver. That one is _fast_ on the fb console, unlike radeon... ;)
evil_core: 12900 causes that
AndrewR: evil_core, be careful, removing too much code can hang you. (i assume you talked about KMS, so wrong step .... and kernel go to heaven.)
evil_core: no, I am setting it using /sys
evil_core: agd5f: shouldnt 0 state be ommited for r500?
AndrewR: evil_core, guess agd5f was paged out by users pressure ( ;) )
AndrewR: stability testing can be time-consuming ... especially if you have just soft-hang without anything in logs (like it was in my case)
ketema: hello i have been directed her from #gentoo. i have a the following device : http://www.pastebin.org/195707 i simply want to startx and I get error setting MTRR (base = 0xe0000000, size = 0x01000000, type = 1) Inappropriate ioctl for device (25). I have tried many things i found on the net nothing works. hoping someone here can help. thanks
evil_core: mine system is stable
AndrewR: wait 4 or even 12 hours, use you computer lightly and heavy, and then _may be_ it will hang ... or may be not. change one parameter .. nad repet ... So, this is one reason why writing stable driver can be hard task ...... (evil_core, nice to hear)
AndrewR: keyboard ate few letters from above
evil_core: Iam using heavily mine machine all the day
AndrewR: ketema, i has this error too, on X exit ....for me it was harmless (until something killed old UMS mode, so i run KMS-only setup, and this message don't appear anymore. Sorry for so much words ...)
evil_core: maybe AGP too
AndrewR: ketema, ES1000 - something in server?
lwhacker: (II) RADEON(0): I2C device "DisplayPort-1-DP:ddc2" registered at address 0xA0.
lwhacker: failed to get auxch 0030 40 30 20
lwhacker: any ideas what that means?
AndrewR: evil_core, especially if you consider physically escaping (from slot) card in my machine .... yes, it can be not even software (at least not all times)... Anyway, it is really good to hear when someone has NO stability problems .... keep it this way
ketema: AndrewR: mine happens on X start and it dies, yes its a simple server, I don't need 3d or anything
ketema: AndrewR: in fact its the first time i ever started X
evil_core: radeon is generally stable
evil_core: especially in UMS
evil_core: te worst is suspends
AndrewR: ketema, usually people here want X log and dmesg .... sorry if you posted them and i missed ....
ketema: AndrewR: no problem : http://www.pastebin.org/195711 and how much of dmesg is appropiate ?
evil_core: anyone got cs16?
AndrewR: ketema, full ? If you can ...
ketema: AndrewR: done. http://www.pastebin.org/195714
lwhacker: ping agd5f
lwhacker: Here is my Xorg log showing my problem with: failed to get auxch http://pastebin.com/Acd3bzQ3
evil_core: http://pld.pastebin.com/cq6Nszsc
lwhacker: I using the git radeon driver
evil_core: its cs16 output
evil_core: it doesnt run with latest driver
evil_core: in UMs and KMS mode
evil_core: before worked even under Gallium(perfect FPS in UXGA res)
AndrewR: ketema, MTRR setup seems more complicated, compared to mine ... may be you can open new bug on bugs.freedesktop.org ? It may take some time until developers understand your problem ...
ketema: AndrewR: ok. i'll give it a shot
AndrewR: ketema, or if bug already there .. one moment, searching
AndrewR: ketema, just X from console gives nothing, too? (or X -retro)
ketema: AndrewR: trying....
ketema: AndrewR: just X gives me expected black screen, retro gives the expected X cursor....on backspace exit both give the error...
ketema: startx still give error and no desktop is started....btw my registered wm is openbox and its not configured just basic install
Droste: ketema: video=vesafb:ywrap,mtrr,vga=0x31X,vram:128 <-- do you need this?
Droste: (kernel boot line)
ketema: Droste: its just the basic kernel line from gentoo set up...gives nice resolution on console
Droste: try booting without it
ketema: Droste: need? i guess not really its nice to have a little screen realestate....ok
Droste: ketema: if your kernel is >2.6.33 you can get a nice resolution on console with the "radeon" kernel module
Droste: >=
ketema: Droste: ok reboot yielded same results just crappy resolution
Droste: ok
AndrewR: sorry, /me hangs .... so, it was not gartsize ..... in my case.
ketema: X and X -retro do what they are supposed to but give the error on exit
ketema: startx does not start X and exits with the same error error setting MTRR (base = 0xe0000000, size = 0x01000000, type = 1) Inappropriate ioctl for device (25)
ketema: am i trying to use some feature my card does not support ?
ketema: if so i'll disable it....i don't need 3d or anything
ketema: i have also tried the basic vesa driver with same result
AndrewR: ketema, my be startx just doesn't have any programs it want to run in path?
AndrewR: *may be
ketema: AndrewR: openbox &
ketema: aterm -geometry 80x20+494-0 & are in the /etc/X11/xinit/xinitrc file
ketema: should i move that to my home dir ?
AndrewR: ketema, from your X log - server correctly disables 3d (dri), but keep XAA (2d accel, old but hopefully more useful on such "card" (EXA still really likes VRAM).)
AndrewR: ketema, can you start aterm like this: export DISPLAY=":0" (from local console) then X -retro& sleep 2&& aterm ?
AndrewR: ketema, may be they just segfaults on start, and so server exit automatically
ketema: AndrewR: ha..aterm segfaulted
ketema: AndrewR: going to remove it replace with xterm
ketema: AndrewR: ok i did the following...startx with just openbox & and xterm& i got same error...so thinking open box was broke i changed to twm still no go
ketema: AndrewR: how is it the retor works but not startx ? this has to be a config thing right ?
AndrewR: ketema, i never looked at X startup too close .... sorry.
AndrewR: ketema, but yes, i have .xinitrc in my home dir ....
AndrewR: exec /usr/bin/starte16 - not really long
ketema: AndrewR: well i would have no idea what to look for configs are auto generated...http://www.pastebin.org/195766 and http://www.pastebin.org/195767
ketema: AndrewR: i do not have this command : /usr/bin/starte16 and starte16 is not in my path anywhere
AndrewR: ketema, try just openbox, without "&" at the end ?
AndrewR: like /usr/bin/openbox or where it located on your system .....
ketema: AndrewR: woot that did it
ketema: AndrewR: jeez i never would have thought of that....al the examples throw the WM into the background
evil_core: export LIBGL_AWLAYSSOFTWARE=1
evil_core: how should I set it?
evil_core: its not LIBGL_ALWAYS_SOFTWARE
evil_core: wtf it doesnt work?