RE: [Bug 1685] 4.6.0 - sig 11 crash in or under __MESA_destroyBuffer
On Thu, 1 Nov 2007, John Lumby wrote: Thanks Marc. I finally got the chance to build a new World based on latest cvs 4.7 with your fixes and it does fix the problem - I've started X and run Lotus Notes and shut down several times with two different WM's and no problems while shutting down. Does the log show any messages about unknown buffers? I did notice one problem in one particular situation with this build but I can't say I've ever tried it with the old build and I won't get a chance to retry it for a while. I was in a location where there was an external monitor sitting there and I connected it to the external graphics socket on my IBM thinkpad T42p. I notice that, instead of what I've seen on other machines, where pressing the IBM Fn-F7 combination switches between the two monitors, with this particular radeon driver and XFree86, the image is simultaneously displayed on both monitors all the time and Fn-F7 has no effect. Also (and this surprises me but I don't really know much about multi-head displays) since the external monitor has a smaller resolution than the laptop's monitor which is 1600x1200, XFree86 displays just part of the T42 image on the external (at the displayable resolution) and the mouse can pan that image - although it's truly weird doing so since of course there's no panning on the laptop monitor. I don't know if this means anything at all to you. I would much prefer to be able to run the image at the lower resolution on the external and have nothing on the laptop under control of Fn-F7. I have tried fiddling with modes but it seems to ignore that. But that is by the This behaviour can be controlled with the driver's "MonitorLayout" option. See its man page. You could also generate a multihead configuration. way. I was running as described with this new build and then decided to VT-switch with Ctl-Alt-F1 back to the first console (which runs a framebuffer console with the fb radeon driver). That worked. But I then tried to Ctl-Alt-F7 back to X - and the whole machine froze. Nothing revived it - Atl-Alt-Del, Ctl-Alt-Bspc, etc etc. had to reboot. I have a XFree86.0.log of that but there's nothing significant in it - no crash, just the usual entries. You would need to trace this to figure out where in RADEONEnterVT() or its callees the hang occurs. You would have to login remotely and run a static debuggable server under `gdb`. I also noticed something strange concerning this XFree86 executable. I built it with the instructions you gave before for making a static exe: /* * Build a server that dynamically loads the modules by setting * this to YES. This defaults to YES on most platforms. A static server * can be built by setting this to NO. * */ #define DoLoadableServerNO but I'd say the exe that it built is not static - it's near the same size as the old dynamic 4.6 one and only 1/3 the size of the static one based on 4.6 that I built before without fixes. cd /usr/X11R6/bin;ls -lirt XFree86.dynload XFree86.statload XFree86; 22 -rws--x--x1 root root 1968583 May 8 2006 XFree86.dynload 202526 -rwsr-xr-x1 root root 7351275 Sep 1 17:41 XFree86.statload 199934 -rws--x--x1 root root 2087673 Oct 30 10:24 XFree86 I guess it doesn't matter since it fixed the problem but did you expect that? The log would say whether or not was generated by a loader server. Perhaps, DoLoadableServer wasn't picked up for some reason. I also notice the permissions are different. Is there a reason for that? I have one other question - unrelated. I see on the official XFree86 download site that pre-built binaries are available for up to (if I recall) glibc-2.4 or 2.5, I forget which. I was thinking of installing glibc-2.7 on one system. Now I know how to make World I guess I can do that but was wondering whether either one pre-built for an earlier glibc like 2.4 or 2.5 will run ok on a later glibc , or if not, I don't know. You'd have to try it. The reason we generate binaries for different glibc versions is that there have been incompatibilities in the past. BTW, the host.def files used to generate binary distributions are in 'xc/programs/Xserver/hw/xfree86/etc/bisdist/'. is XFree86 planning to build some at newer glibc levels? Not to my knowledge. Speaking for myself, I don't have access to glibc 2.7 system that would allow me to generate binaries for it. Do you want any more information or logs from either the freeze I mentioned or from running ok with this build? No. Given it's a hard hang, I doubt the log will show anything. Thanks. Marc. +--+--+ | Marc Aurele La France | work: 1-780-492-9310 | | Academic Information and| fax:1-780-492-1729 | |Communications Technologies | email: [EMAIL PROTECTED] | | 352 General Services Building +---
RE: [Bug 1685] 4.6.0 - sig 11 crash in or under __MESA_destroyBuffer
On Thu, 18 Oct 2007, Marc Aurele La France wrote: On Mon, 15 Oct 2007, John Lumby wrote: On Sun, 14 Oct 2007, John Lumby wrote: Thanks Marc - but it prompted me for pwd. And this id can happily ssh to another system of mine without pwd prompt. date;scp -qp /mnt/super9root/root/core.2856.X11crash.20071003122131 [EMAIL PROTECTED]:.;date;scp -qp /mnt/super9root/usr/X11R6/bin/XFree86.statload [EMAIL PROTECTED]:.;date; Sun Oct 14 22:10:48 EDT 2007 [EMAIL PROTECTED]'s password: Sun Oct 14 22:10:57 EDT 2007 [EMAIL PROTECTED]'s password: Sun Oct 14 22:11:00 EDT 2007 /home/lumby:0 ssh -l lumby date Sun Oct 14 22:19:51 EDT 2007 The problem turned out to be that jlumby wasn't listed in AllowUsers. Please try again. It worked this time; They should be there now (I hope) Thanks. These were (and still are) _most_ informative. I've uncovered a real bug here, one that affects not only GLX & Friends, but potentially other extensions also. The bug only occurs on server shutdown or reset. It is a definite candidate for causing this problem, but I can't be sure at this point, given the memory corruption this core file attests to. Fixing it will take some time (on my part). The bug has existed for quite some time, and from what I can tell, still exists not only in our repository, but X.Org's as well. Given that, I request that you update to the L&G sources, which you can get by following the instructions at http://xfree86.org/cvs. I hope to have a patch against that source ready for you to try in the next few days. Attached is a preliminary fix. As it turns out, this should also apply to 4.6.0, perhaps even 4.5.0. I say "preliminary" because this only deals with GLX/Mesa. I consider this instance of the problem to only be the tip of the iceberg of a more general design glitch. To fix that glitch, I'd have to change the order some things are done during server termination or reset. Doing so is likely to break several things which will take some time to go through. This fix does the following: - Fix initialisation of __GLXscreenInfo structures (not directly related to the problem at hand); - Fix Mesa to complain (on stderr), rather than segfault, when an attempt is made to free an unknown buffer; - Do not free all Mesa buffers upon GLX extension closedown. Instead these will be freed later, at FreeAllResources() time, when the drawable privates that reference these buffers are also freed. Please let me know if this fixes the segfault. Please `scp` to your id on my machine a capture of the server's stderr, the resulting /var/log/XFree86.0.log, and, should the server still segfault, another copy of the server binary and core file. Thanks. Marc. +--+--+ | Marc Aurele La France | work: 1-780-492-9310 | | Academic Information and| fax:1-780-492-1729 | |Communications Technologies | email: [EMAIL PROTECTED] | | 352 General Services Building +--+ | University of Alberta | | | Edmonton, Alberta |Standard disclaimers apply| | T6G 2H1 | | | CANADA | | +--+--+ XFree86 developer and VP. ATI driver and X server internals. cvs-devel.diff.gz Description: Binary data
RE: [Bug 1685] 4.6.0 - sig 11 crash in or under __MESA_destroyBuffer
On Mon, 15 Oct 2007, John Lumby wrote: On Sun, 14 Oct 2007, John Lumby wrote: Thanks Marc - but it prompted me for pwd. And this id can happily ssh to another system of mine without pwd prompt. date;scp -qp /mnt/super9root/root/core.2856.X11crash.20071003122131 [EMAIL PROTECTED]:.;date;scp -qp /mnt/super9root/usr/X11R6/bin/XFree86.statload [EMAIL PROTECTED]:.;date; Sun Oct 14 22:10:48 EDT 2007 [EMAIL PROTECTED]'s password: Sun Oct 14 22:10:57 EDT 2007 [EMAIL PROTECTED]'s password: Sun Oct 14 22:11:00 EDT 2007 /home/lumby:0 ssh -l lumby date Sun Oct 14 22:19:51 EDT 2007 The problem turned out to be that jlumby wasn't listed in AllowUsers. Please try again. It worked this time; They should be there now (I hope) Thanks. These were (and still are) _most_ informative. I've uncovered a real bug here, one that affects not only GLX & Friends, but potentially other extensions also. The bug only occurs on server shutdown or reset. It is a definite candidate for causing this problem, but I can't be sure at this point, given the memory corruption this core file attests to. Fixing it will take some time (on my part). The bug has existed for quite some time, and from what I can tell, still exists not only in our repository, but X.Org's as well. Given that, I request that you update to the L&G sources, which you can get by following the instructions at http://xfree86.org/cvs. I hope to have a patch against that source ready for you to try in the next few days. Thanks for your patience. Marc. +--+--+ | Marc Aurele La France | work: 1-780-492-9310 | | Academic Information and| fax:1-780-492-1729 | |Communications Technologies | email: [EMAIL PROTECTED] | | 352 General Services Building +--+ | University of Alberta | | | Edmonton, Alberta |Standard disclaimers apply| | T6G 2H1 | | | CANADA | | +--+--+ XFree86 developer and VP. ATI driver and X server internals. ___ Devel mailing list Devel@XFree86.Org http://XFree86.Org/mailman/listinfo/devel
