| bryce | esr, in several examples of this class of bug, it's shown an edid-fail error. Not sure what's up with that but possibly a strong clue. | 00:00 |
|---|---|---|
| esr | OK. Running get-edid | parse-edid yields well-formed output... | 00:00 |
| bryce | esr, traditionally on -intel, these bugs have been caused by the need for monitor quirks. E.g. edid shows 0mmx0mm physical dimensions, or invert X and Y or mixes up mm with cm | 00:01 |
| bryce | we quirked a lot of that in X, so in theory some of those could have been lost going to KMS. However of the bugs I looked into only a few fit that classification. | 00:01 |
| esr | It got horiz and vert frequencies correct, but generated only one modeline - 1280x1024. | 00:01 |
| bryce | mm | 00:02 |
| esr | Sure looks like my guess was correct. | 00:02 |
| esr | EDID ain't delivering. | 00:02 |
| bryce | also compare with xrandr --verbose - that has different edid parsing code (same as is used in xserver) | 00:02 |
| esr | Trying... | 00:02 |
| esr | Now that's interesting. | 00:03 |
| bryce | get-edid pretty much bypasses X, so if you see an issue with that, it means something pretty far down the stack is not working right | 00:03 |
| bryce | ddcprobe is the third tool but that's hardly useful anymore | 00:04 |
| esr | OK, xrandr verbose generate a boatload of modelines but they top out at 1280x1024 too. | 00:05 |
| esr | I think out problem is right down at EDID level. | 00:06 |
| bryce | yeah | 00:06 |
| bryce | this is why I'm suspecting something at or just above the kernel level | 00:06 |
| bryce | booting an earlier kernel would rule out the kernel itself | 00:06 |
| esr | (Note: I maintained the XFree86 Video Timings HOWTO so this stuff isn't exactly alien to me.) | 00:07 |
| bryce | great | 00:07 |
| esr | (That's whyn I figured I could help you guys diagnose.) | 00:07 |
| esr | What do you recommend as a next diagnostic? | 00:08 |
| bryce | yeah I've so far been unable to reproduce the issue (and tied up with other priorities) so having someone that is clueful is quite helpful | 00:08 |
| bryce | well, we need to have a better understanding of the edid pipeline. What components are involved | 00:08 |
| bryce | and then vary those to identify which component has broken | 00:08 |
| esr | Agreed. | 00:08 |
| bryce | so booting a different kernel would probably be the next easiest step | 00:09 |
| bryce | e.g. if you have the jaunty kernel around | 00:09 |
| esr | I've got a stock 9.10 freshly installed here. I don't think I still have a 9.04 kernel but let me look at grub.conf | 00:10 |
| esr | Er, there is no grub.conf | 00:11 |
| bryce | you want /boot/grub/menu.lst | 00:11 |
| esr | Hm, I've got a vmliinuz.old. That *might* be the 9.04 kernel | 00:11 |
| kees | fresh 9.10 will use /boot/grub/grub.cfg | 00:12 |
| kees | you can add the jaunty repositories and install an old kernel by version, too. | 00:12 |
| esr | I shouldn't have said "fresh"; I upgraded. | 00:12 |
| bryce | get-edid uses a VESA VBE 2 interrupt service routine request to read a | 00:13 |
| bryce | 128 byte EDID version 1 structure from your graphics card, which retrieves | 00:13 |
| bryce | this information from the monitor via the Data Display Channel (DDC). | 00:13 |
| bryce | so what provides the VESA VBE 2 interrupt service ? | 00:13 |
| esr | That I never learned. | 00:13 |
| bryce | "real-mode x86 instructions on i386" | 00:14 |
| bryce | yeah me neither. I assume it's the kernel but maybe some other lib | 00:14 |
| bryce | kees, you'd know? | 00:14 |
| sconklin | bryce: I only have a few minutes. | 00:18 |
| bryce | heya sconklin | 00:18 |
| bryce | sconklin, do you have a pointer to that timing patch you mentioned? | 00:19 |
| sconklin | so where does the EDID coem from? The kernel driver reads it, at least in some cases, maybe all. | 00:19 |
| bryce | we're trying to figure out why edid seems to be coming through incomplete on karmic compared with jaunty | 00:19 |
| sconklin | huh, lemme look | 00:19 |
| sconklin | That patch was supposed to be only for a limited set of intel hardware, and after the last email you sent me about it I'm not sure it's related | 00:20 |
| bryce | ah ok, hmm. Still would be worth looking at | 00:21 |
| esr | My graphics chip is an Intel 965. | 00:21 |
| esr | Is that in the "limited set"? | 00:21 |
| bryce | weird, I also have 965 and haven't seen this problem | 00:22 |
| sconklin | it was supposed to be limited to the arrandale chipsets, but I think that's a lot of common code with the 945, not the 965 | 00:22 |
| esr | lspci sez: 00:02.0 VGA compatible controller: Intel Corporation 82Q963/Q965 Integrated Graphics Controller (rev 02) | 00:23 |
| bryce | esr, ahh Q965 is a different variant | 00:23 |
| sconklin | sorry, iron lake | 00:23 |
| esr | Lovely :-( | 00:23 |
| bryce | sconklin, oh | 00:23 |
| sconklin | http://kernel.ubuntu.com/git?p=ubuntu/ubuntu-karmic.git;a=commit;h=6a8ea128b38f4f0d3798bb09df4618c50d832c7c | 00:24 |
| sconklin | that's the one I was thinking of | 00:24 |
| sconklin | looks like only for new steppings, so I think that's not it | 00:24 |
| sconklin | sorry | 00:25 |
| esr | See why I didn't want to just drop aa bug report on the tracker? I had a feeling it was gonna be this kind of swamp... | 00:25 |
| sconklin | http://kernel.ubuntu.com/git?p=ubuntu%2Fubuntu-karmic.git&a=search&h=HEAD&st=commit&s=edid some edid-related patches were committed in august | 00:27 |
| bryce | drm/edid: fixup detailed timings like the X server. | 00:28 |
| bryce | that sounds interesting | 00:28 |
| sconklin | exactly what I was looking at - hold on a sec | 00:28 |
| sconklin | I have arlied's tree here | 00:28 |
| bryce | whoa | 00:29 |
| bryce | interesting this is in drivers/gpu/drm/ | 00:29 |
| bryce | so that suggests this patch applies to all hardware not just intel | 00:30 |
| bryce | however I'm not sure this would explain -nvidia... they kind of do their own thing | 00:30 |
| bryce | but maybe | 00:30 |
| bryce | wait no, I bet this is KMS-specific code | 00:30 |
| esr | I'm willing to test a custom kernel build if you guy can figure out a way to package it so I can install it. | 00:31 |
| esr | Advantage: For me this is 100% reproducible. get-edid fails every time. | 00:32 |
| sconklin | there's another patch that eliminates some entries in the edid based on hsync/vsync==0 or sync beyond blank | 00:33 |
| sconklin | when it does it emits a DRM_DEBUG_KMS message | 00:34 |
| sconklin | I can probably get a kernel built if we know what we need to test. I'm supposed to be off for the rest of the week so it may not be prompt | 00:35 |
| sconklin | I'm still browsing patches. | 00:35 |
| bryce | sconklin, it would be great to get a kernel ppa with some/all of the edid patches tossed in there | 00:36 |
| sconklin | bryce: you mean those patches removed? | 00:36 |
| esr | And then bisect the patch set... | 00:36 |
| sconklin | when was the first bug opened for this | 00:37 |
| sconklin | ? | 00:37 |
| esr | I haven't opened one yet. Wanted to screen for bogons first ;-) | 00:37 |
| bryce | sconklin, it's really hard to pinpoint I only noticed the surge of reports at release time | 00:37 |
| sconklin | ok, that could just be release | 00:38 |
| bryce | yeah | 00:38 |
| sconklin | bryce: by the way I'm working on getting a test kernel with the moblin patches to you | 00:38 |
| bryce | oh I see, we already have these patches in place, so it's a matter of identifying which to disable | 00:39 |
| sconklin | yeah, let me also pull the latest and see if there are any fixes or reverts | 00:39 |
| sconklin | Wait, that was drm-next, and may not have all hit linus's tree yet. | 00:40 |
| bryce | looks like several fixes are in linus' tree on top of this | 00:40 |
| sconklin | give me a few minutes to get this straight in my head | 00:40 |
| esr | I'm not going anywhere. | 00:41 |
| sconklin | haha, neither am I as it turns out | 00:41 |
| bryce | heh | 00:41 |
| bryce | sconklin, I'm looking through these upstream changes and they're really basic fixes... even *I* know these are needed to work around known edid problems | 00:42 |
| sconklin | I haven't seen anything that didn't look basically sane | 00:42 |
| bryce | wow, I am really surprised they didn't port this from the xserver, all this logic has been known for quite some time | 00:43 |
| bryce | sconklin, also I bet there are more in the fedora kernel | 00:44 |
| sconklin | the symptoms are that you get a wrong set or subset of what the monitor should support? | 00:44 |
| esr | Yup. | 00:44 |
| bryce | sconklin, that's correct | 00:44 |
| bryce | esr, can you pastebin your edid someplace? | 00:46 |
| bryce | I can probably figure out exactly which patch solves your issue | 00:46 |
| bryce | however this is proving my original suspicion correct - it's not just one bug but a whole bunch of separate ones because the kernel isn't carrying the quirks that the xserver had | 00:46 |
| esr | bryce Sure, will do | 00:47 |
| sconklin | that's pretty ugly | 00:47 |
| bryce | in theory, booting with i915.modeset = 0 (i.e. turning off KMS) would result in fixage | 00:47 |
| esr | http://pastebin.com/m274c46a1 | 00:48 |
| esr | That what you wanted? | 00:48 |
| bryce | sconklin, ok so we are going to need pretty much all of these edid patches. Plus more I think. There are a lot more quirks I know of, that aren't covered here. | 00:48 |
| sconklin | it looks like there are no further patches with "edid" in the comment queued for linus - they all landed in the last window | 00:48 |
| bryce | esr, let's see the xrandr --verbose too | 00:49 |
| esr | Hold on... | 00:49 |
| sconklin | bryce: ok, I'd be happy to work with you on making sure they're all in there and upstreamed. | 00:49 |
| bryce | sconklin, great | 00:49 |
| sconklin | can it wait until next week? Is the level of breakage high enough to make this urgent? | 00:50 |
| esr | http://pastebin.com/d1f70504e | 00:50 |
| esr | xrandr output | 00:50 |
| esr | I've got the problem solved for *me*...it just looks really bad that this got out in a release. | 00:51 |
| sconklin | esr: no question about that | 00:51 |
| bryce | sconklin, well, better to take time to do it right than roll out something that will cause worse problems | 00:51 |
| esr | Hard for me to tell people Linux is ready for end-users when we fuck up stuff this basic. | 00:52 |
| bryce | sconklin, I am also going to be gone thurs/fri anyway | 00:52 |
| esr | I repeat my offer to advance-test the fix. | 00:52 |
| bryce | esr, well these are highly monitor-specific things | 00:53 |
| sconklin | bryce: ok, let's make a review of what you know is broken, collect all the upstream patches and those in fedora, and then assemble them and upstream anything that needs to go. | 00:53 |
| esr | And you know as well as I do that end users don't buy those excuses. | 00:53 |
| bryce | esr, settle down | 00:55 |
| bryce | esr, hmm, neither of those showed the monitor model | 00:55 |
| bryce | by chance is it a 226BW or 225BW? | 00:55 |
| bryce | or 205BW | 00:55 |
| sconklin | hahaha my boys are in the next room talking about jailbreaking phones | 00:56 |
| bryce | esr, technically most users haven't bought anything ;-) | 00:56 |
| esr | bryce: Not sure how to tell, all that's on the bezel is SyncMaster 1100DF | 00:56 |
| bryce | esr, ok should say in the Xorg.0.log - mind posting that as well? | 00:57 |
| bryce | heh, we should have just had you file a bug originally, then all this would be there | 00:57 |
| sconklin | there are six patches upstream that are obviously edid related, and arrived after karmic | 00:57 |
| esr | Hold on... | 00:57 |
| esr | http://pastebin.com/d25e8244d | 00:59 |
| bryce | feh, I wish there were 3 of me | 00:59 |
| esr | Looking for monitor subtype. No match on "BW". | 01:00 |
| bryce | https://wiki.ubuntu.com/X/Quirks - monitor quirks section | 01:00 |
| bryce | Samsung SyncMaster's have required quirking before | 01:01 |
| bryce | one needed quirk_detailed_sync_pp, two needed quirk_prefer_large_60 | 01:01 |
| sconklin | great reference Bryce | 01:01 |
| bryce | thanks | 01:01 |
| bryce | sconklin, we need to update it for doing kernel quirks, but it'll give us a starting point | 01:01 |
| esr | Er. Product id 109? | 01:02 |
| bryce | btw, for reference all the X quirks are in hw/xfree86/modes/xf86EdidModes.c | 01:02 |
| bryce | we'll want to ensure all those get ported over to the kernel | 01:02 |
| bryce | esr, thanks | 01:02 |
| bryce | esr, ok yours is different. But I'd wager you need one of these quirks. | 01:03 |
| esr | I wouldn't be surprised if I did. | 01:03 |
| esr | quirk_prefer_large sounds suspiciously applicable. | 01:04 |
| bryce | getting called away to dinner. bbl | 01:05 |
| sconklin | later | 01:05 |
| esr | You guys need anything else from me? | 01:05 |
| albert23 | esr: x log says: Using hsync ranges from config file | 01:06 |
| sconklin | esr: I don't think so. Thanks | 01:06 |
| albert23 | esr: are you sure you are not blocking high res modes in xorg.conf? | 01:06 |
| esr | Yeah, that's because I had toi hand-hack the X config to get hight resolution. | 01:06 |
| esr | When I first saw the problem there was no custom stuff in it at all. | 01:07 |
| esr | I hand-patched in hsync/vsync and a mode line. Now it works. | 01:07 |
| ara | hey morning all! | 07:44 |
| === ripps_ is now known as ripps | ||
| ara | nvidia no longer discovers my external monitor (it shows as crt-0), any ideas? | 09:30 |
| MsMaco | i had a backup of /etc to avoid having to re-set-up everything and just restored it since i reinstalled. this /etc worked on amd64, but my new install is i386. does X get not-happy when this happens? i get kicked back out of X immediately on attempted login | 16:16 |
| MsMaco | i tried "dpkg-reconfigure xserver-xorg" in case it was something to do with the arch in /etc, but that doesnt seem to have helped any | 16:17 |
| MsMaco | (intel 965, karmic, dunno what else would be useful) | 16:20 |
| Ng | what does /var/log/Xorg.0.log say? | 16:21 |
| MsMaco | failed to load module i810 (module does not exist) | 16:22 |
| MsMaco | that and soething about virtual kbd are the only EE lines | 16:23 |
| Ng | the i810 error is fine | 16:23 |
| MsMaco | it should be -intel not -i810 though, i think | 16:23 |
| jcristau | it tries both | 16:23 |
| Ng | :) | 16:23 |
| jcristau | so the i810 error is expected | 16:24 |
| MsMaco | ok | 16:24 |
| Ng | a reasonable test is to log into a console, stop gdm and just run X, see if it's X or the gdm stuff on top of it | 16:24 |
| jcristau | sharing /etc across archs could make any number of things go wrong though. not particularly with X. | 16:25 |
| MsMaco | Ng: startx works. so something's bonkers with kdm | 16:26 |
| MsMaco | jcristau: yeah i never tried doing it across arch before, just across installs of the same arch because reconfiguring stuff is annoying | 16:26 |
| MsMaco | thanks for your time. sorry to bug you | 16:27 |
| kees | bryce: I'd like to copy xorg-server from the security-proposed PPA to -proposed, is that ok by you? https://edge.launchpad.net/~ubuntu-security-proposed/+archive/ppa/+sourcepub/868253/+listing-archive-extra | 16:41 |
| jcristau | kees: that reminds me i should apply your patch.. | 16:50 |
| kees | jcristau: heh, yes please! :) | 16:51 |
| jcristau | http://git.debian.org/?p=pkg-xorg/xserver/xorg-server.git;a=commit;h=81e2cbb3dea9ca293d22a3dfc56724513350cfbb | 16:55 |
| bryce | kees, fine by me | 17:08 |
| kees | bryce: okay | 17:08 |
| kees | jcristau: \o/ thanks :) | 17:09 |
| jcristau | kees: thanks for the patch :) | 17:09 |
| kees | jcristau: np :) | 17:09 |
| === Ng__ is now known as Ng | ||
| === wgrant_ is now known as wgrant | ||
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!