| alkisg | KitsuWhooa: hi, so, any idea where should we report this? To nouveau upstream? To xorg? To ubuntu? | 13:17 |
|---|---|---|
| KitsuWhooa | I'm still working on it | 13:17 |
| alkisg | Ah great | 13:18 |
| KitsuWhooa | bisecting goes from one bug to another | 13:18 |
| alkisg | Btw I haven't heard of any instability issues with pageflip off, from a lot of schools | 13:18 |
| alkisg | Haha | 13:18 |
| KitsuWhooa | https://tasossah.com/s/126f55d9cdb3.jpg | 13:18 |
| alkisg | Are you trying on tnt2 with the segfault, or on the other ones with the corruption? | 13:19 |
| KitsuWhooa | tnt2 | 13:19 |
| KitsuWhooa | Looking at the code through gdb, the issue is probably somewhere in DIX. It's a bit of a maze | 13:19 |
| alkisg | Hmm, well, even if you fix the real issue that is the segfault, you wouldn't fix the issue that exposed the segfault, i.e. the lack of initialization somewhere, right? | 13:20 |
| KitsuWhooa | That is exactly what I am trying not to do | 13:20 |
| KitsuWhooa | if I wanted to do that, I could add a check for a valid pointer, and if not return | 13:20 |
| KitsuWhooa | but I doubt that would work | 13:21 |
| KitsuWhooa | nouveau calls this https://cgit.freedesktop.org/xorg/xserver/tree/exa/exa.c#n66 and this returns an invalid pointer | 13:23 |
| KitsuWhooa | and ExaPixmapPriv is a macro | 13:24 |
| KitsuWhooa | which is here | 13:24 |
| KitsuWhooa | https://cgit.freedesktop.org/xorg/xserver/tree/exa/exa_priv.h#n283 | 13:24 |
| KitsuWhooa | and this in turn calls DIX functions/macros | 13:25 |
| alkisg | KitsuWhooa: would it be possible to do this? (1) dump vars (2) run xrandr 640x480; xrandr 1366x768; (3) dump vars | 13:56 |
| alkisg | (1) would be different from (3) in the vars that are not initialized | 13:56 |
| KitsuWhooa | not really | 13:56 |
| KitsuWhooa | the function gets called multiple times | 13:56 |
| alkisg | I'm not talking about the function | 13:57 |
| alkisg | Suppose xorg somewhere has a variabled named, I dont know, other_page_resolution | 13:57 |
| alkisg | And this in turn messes things up, and breaks where you see it, but you can't trace it back to the wrong initialization, | 13:57 |
| alkisg | so, since xrandr does initialize that internal xorg variable, wouldn't comparing before/after help locate the uninitialized var? | 13:58 |
| KitsuWhooa | If it's an uninitialized variable, I can probably use valgrind to detect it | 13:58 |
| KitsuWhooa | dumping every single variable would be very impractical, if not impossible | 13:58 |
| alkisg | Well since 2 xrandr calls fix it, wouldn't that be the case? | 13:58 |
| alkisg | I haven't programmed in C/assembly in recent days, but in the old days, I could dump the whole data/stack etc segments and compare | 13:59 |
| alkisg | And then check the address in the linker to find the variable there... | 14:00 |
| * alkisg googles about valgrind... | 14:00 | |
| KitsuWhooa | if you go low level like that, possibly | 14:00 |
| KitsuWhooa | but due to the "refresh rate", you'd probably get a lot of false positives | 14:01 |
| alkisg | Wow, just as easy as "valigrind myprogram"? ! | 14:03 |
| KitsuWhooa | yup | 14:04 |
| alkisg | That sounds too good! | 14:04 |
| KitsuWhooa | I just switched to one of the cards that have just corruption | 14:04 |
| KitsuWhooa | so let's see what valgrind says | 14:04 |
| alkisg | Great | 14:04 |
| KitsuWhooa | alkisg: does epoptes-client restart whenever a user logs in? | 14:04 |
| alkisg | There are 2 epoptes-client processes. The system one and the session one. The session one starts on login, so yeah. | 14:05 |
| KitsuWhooa | Because I found that if I kill the one running already and run it manually from a tty, it doesn't kill the remote sessions | 14:05 |
| alkisg | You can't kill epoptes (to avoid students doing so), | 14:05 |
| KitsuWhooa | from a root vt? | 14:05 |
| alkisg | pkill -QUIT -U "$UID" -f '^epoptes-client$' | 14:05 |
| alkisg | This sends it a QUIT signal, which it respects | 14:05 |
| alkisg | Otherwise it restarts | 14:05 |
| KitsuWhooa | I kill -9 its PID | 14:05 |
| KitsuWhooa | oops | 14:05 |
| KitsuWhooa | it doesn't restart for me | 14:06 |
| alkisg | There's a timeout in the restarting | 14:06 |
| KitsuWhooa | Ah, I see | 14:06 |
| alkisg | So you may start it 20 seconds later, and it might be restarted 60 seconds later, killing your "20 seconds later" instance | 14:06 |
| * alkisg wants to simplify this now with systemd which supports restarting | 14:07 | |
| KitsuWhooa | it doesn't seem to kill it | 14:07 |
| KitsuWhooa | all I know is that I do that and I can use the shell across X restarts/crashes | 14:07 |
| alkisg | Make sure UID is defined | 14:08 |
| alkisg | Eh, you should be able to use the root epoptes-client without restarting it | 14:08 |
| alkisg | It survives xorg crashes | 14:08 |
| KitsuWhooa | Once I am done with this I can probably tell you how to reproduce it | 14:08 |
| KitsuWhooa | and I checked stdout/err and it says nothing when it dies | 14:09 |
| KitsuWhooa | something complains about socat dying | 14:09 |
| KitsuWhooa | but I don't think that's the cause | 14:09 |
| alkisg | Btw if it helps, you can also enable sshd on ltsp clients | 14:10 |
| alkisg | I never had that issue though, with root epoptes-client dying | 14:10 |
| KitsuWhooa | it would help, but this ugly workaround seems to be okay so far | 14:10 |
| KitsuWhooa | so no need yet | 14:10 |
| KitsuWhooa | would also be nice if there was an easy file transfer option | 14:12 |
| KitsuWhooa | I just pipe tar to netcat at the moment | 14:12 |
| KitsuWhooa | to send compiled binaries to clients | 14:12 |
| KitsuWhooa | that is a lot of uninitalised byte warnings | 14:14 |
| alkisg | I use python -m SimpleHTTPServer on the server, and wget server:8000/file on the clients, or scp | 14:19 |
| alkisg | or sshfs | 14:20 |
| alkisg | There's also LOCAL_APPS_EXTRAMOUNTS=/some/folder in lts.conf, which sshfs's it for you on login | 14:20 |
| alkisg | Or you can login and /home/username is automounted, and put binaries there on the server, and run them on the client | 14:21 |
| alkisg | ...the downside being that you need X to login :D | 14:21 |
| KitsuWhooa | yeeeeah | 14:21 |
| KitsuWhooa | was about to say that there is nothing in /home/ | 14:21 |
| KitsuWhooa | I don't see anything immediately related in the valgrind output | 14:27 |
| === mamarley_ is now known as mamarley | ||
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!
