bugzilla-daemon at freedesktop.org
2013-Nov-01 08:40 UTC
[Nouveau] [Bug 71116] New: [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 Priority: medium Bug ID: 71116 Assignee: nouveau at lists.freedesktop.org Summary: [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume. QA Contact: xorg-team at lists.x.org Severity: normal Classification: Unclassified OS: Linux (All) Reporter: ronald645 at gmail.com Hardware: x86 (IA32) Status: NEW Version: 7.7 (2011) Component: Driver/nouveau Product: xorg Created attachment 88462 --> https://bugs.freedesktop.org/attachment.cgi?id=88462&action=edit Attempt to resume [ 0.197865] nouveau [ DEVICE][0000:01:00.0] Chipset: NV34 (NV34) [ 0.197869] nouveau [ DEVICE][0000:01:00.0] Family : NV30 After *only upgrading userspace* from Ubuntu 13.04 to Ubuntu 13.10 the card fails to properly resume from s2disk. Important upgraded packages: xserver-xorg-core:i386 2:1.13.3-0ubuntu6.2 -> 2:1.14.3-3ubuntu2 xserver-xorg-video-nouveau:i386 1:1.0.7-0ubuntu1 -> 1:1.0.9-2ubuntu1 libdrm2:i386 2.4.43-0ubuntu1.1 -> 2.4.46-1 libpciaccess0:i386 0.13.1-2 -> 0.13.2-1 -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131101/4092e665/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-01 08:41 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #1 from Ronald <ronald645 at gmail.com> --- Created attachment 88463 --> https://bugs.freedesktop.org/attachment.cgi?id=88463&action=edit Attempt to resume (nouveau.agpmode=4) Suggested by imirkin. No dice. -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131101/56230795/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-01 08:41 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #2 from Ronald <ronald645 at gmail.com> --- Created attachment 88464 --> https://bugs.freedesktop.org/attachment.cgi?id=88464&action=edit Full list of all upgraded packages -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131101/9b455da9/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-02 14:59 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #3 from Ronald <ronald645 at gmail.com> --- Woa, tried nvidia-173 on - 3.11 - 3.10 - 3.4 All failed to resume with a black screen. I guess nouveau used to have an edge on this one! So, no mmiotraces =( ... -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131102/fcfa6432/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-02 16:34 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #4 from Ronald <ronald645 at gmail.com> --- Using agp=off for the Linux 3.11 kernel and NvAGP=1 in xorg.conf it works with the blob again. Let me know if you would like to have a trace and all. https://bugs.launchpad.net/suse/+bug/255601 -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131102/6afd80aa/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-02 19:12 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #5 from Emil Velikov <emil.l.velikov at gmail.com> --- Hi Ronald AFAICS there are two routes you/we can take this * Find out which package update caused the breakage * Disable AGP with nouveau (similar to what you did with the blob) If you're going for the first one, here is a list of packages that you'll need to check (other may not me related): * xserver-xorg-core * xserver-xorg-video-nouveau * All *mesa* The above can be rather messy and hard, as you'll need to track that the dependencies are met and/or rebuild packages. The latter route is much easier * Append nouveau.agpmode=0 to your kernel command line (i.e. grub/lilo) -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131102/6d8a98d5/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-03 07:38 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #6 from Ronald <ronald645 at gmail.com> --- Created attachment 88546 --> https://bugs.freedesktop.org/attachment.cgi?id=88546&action=edit 2_Attempt to resume (nouveau.agpmode=0) Survives one cycle since I did this from SSH with `pm-hibernate`. Apparently Xorg only locksup on resume because of xscreensaver if I press the button in XFCE. -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131103/72129650/attachment-0001.html>
bugzilla-daemon at freedesktop.org
2013-Nov-03 07:39 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #7 from Ronald <ronald645 at gmail.com> --- Created attachment 88547 --> https://bugs.freedesktop.org/attachment.cgi?id=88547&action=edit 2_Attempt to resume (agp=off) Also tried agp=off -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131103/2f9079c2/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-03 07:40 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #8 from Ronald <ronald645 at gmail.com> --- Created attachment 88548 --> https://bugs.freedesktop.org/attachment.cgi?id=88548&action=edit 2_Attempt to resume To confirm my hypotheses that disabling AGP has no effect I tried it without kernel parameters that alter the driver's behaviour. Same effect. -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131103/fef4af9d/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-03 07:40 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #9 from Ronald <ronald645 at gmail.com> --- (see above three logfiles) I tried your second suggestion, did not work(!). - nouveau.agpmode=0 -> crash - agp=off -> crash - no kernel parameters -> typing pm-hibernate from SSH -> success(!) . Clicking somewhere on desktop -> crash (Only moving cursor did not make it crash). I'm not totally against bisecting these source packages, but that will then be a longterm thing. Furthermore, is all of mesa required? (i.e. I'm not using a compositing desktop for that matter (just XFCE)). Conclusion, it appears that disabling agp has no effect. Machine only crashes after resume if the screen is repainted with the exception of mouse movements. -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131103/bd23a364/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-03 11:49 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #10 from Emil Velikov <emil.l.velikov at gmail.com> --- (In reply to comment #6)> Created attachment 88546 [details] > 2_Attempt to resume (nouveau.agpmode=0) > > Survives one cycle since I did this from SSH with `pm-hibernate`. Apparently > Xorg only locksup on resume because of xscreensaver if I press the button in > XFCE.I had a strange feeling that mesa is involved :\ If you're up for it you can {re,}move nouveau_dri.so (the 3g/gl "driver") and confirm it that fixes your s2ram problems. Note your CPU usage may sky-rocket (the system will fall back to software GL rendering). (In reply to comment #9) [...]> I'm not totally against bisecting these source packages, but that will then > be a longterm thing. Furthermore, is all of mesa required? (i.e. I'm not > using a compositing desktop for that matter (just XFCE)). >Before bisecting anything I would suggest that you absolutely make sure that you know which package broke things - mesa, xserver-xorg-core, xserver-xorg-video-nouveau and/or other. When going through mesa, make sure that you keep all the packages at the same version. Here is a crash course of what the different mesa packages provide Essential parts * libgl1-mesa-glx -> libGL provider * libgl1-mesa-dri -> provides the "hardware accelerated drivers" * libglapi-mesa -> dispatch library Extra (your system may require one or both depending on what software you have) * libgbm1 * libegl1-mesa -> libEGL provider Other * libopenvg1-mesa -> libOpenVG, afaik noone uses this * mesa-common-dev -> headers Not part of mesa (but have mesa in the package name :P) * libglu1-mesa -> minor suspect * mesa-utils -> highly unlikely to be related Have fun :) -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131103/0255ffaa/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-03 13:28 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #11 from Ronald <ronald645 at gmail.com> --- I relocated /usr/lib/i386-linux-gnu/dri/nouveau_dri.so to nouveau_dri_off.so and restarted the computer with AGP. I confirmed the AIGLX error in that it switches to swrast. Did not fix the s2disk issue (dmesg attached). So, it's not mesa nor the kernel. All that remains is (just...) the X server, xf86-video-nouveau and libdrm. Right? -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131103/bea91d64/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-03 13:29 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #12 from Ronald <ronald645 at gmail.com> --- Created attachment 88558 --> https://bugs.freedesktop.org/attachment.cgi?id=88558&action=edit 3_Attempt to resume (nouveau_dri.so disabled) -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131103/d436e2f4/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-03 13:59 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #13 from Emil Velikov <emil.l.velikov at gmail.com> --- (In reply to comment #11)> I relocated /usr/lib/i386-linux-gnu/dri/nouveau_dri.so to nouveau_dri_off.so > and restarted the computer with AGP. I confirmed the AIGLX error in that it > switches to swrast. > > Did not fix the s2disk issue (dmesg attached). > > So, it's not mesa nor the kernel. All that remains is (just...) the X > server, xf86-video-nouveau and libdrm. Right?Sounds like it. Do you know which gcc version is/was used to compile the new libdrm ? Might it be related to this http://gcc.gnu.org/ml/gcc-help/2013-07/msg00103.html? -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131103/9c0cfc7a/attachment-0001.html>
bugzilla-daemon at freedesktop.org
2013-Nov-03 15:10 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #14 from Ronald <ronald645 at gmail.com> --- 13.10 (saucy) is using 4.8. 4.7 seems available, rebuilding libdrm with the source package should be easy. Just need to modify CC var I think... -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131103/4d8a6c4b/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-03 15:49 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #15 from Ronald <ronald645 at gmail.com> --- Recompiled libdrm with gcc-4.7 and BOOM the issue did p00f... =) I'll be reporting this to Ubuntu upstream. Thanks a ton! -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131103/ff7876bb/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-03 15:57 UTC
[Nouveau] [Bug 71116] [Ubuntu userspace upgrade 13.04 -> 13.10][regression] NV34 card fails to properly resume.
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #16 from Ronald <ronald645 at gmail.com> --- Filed a bugreport here: https://bugs.launchpad.net/ubuntu/+source/libdrm/+bug/1247607 -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131103/4a3898c5/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-03 16:16 UTC
[Nouveau] [Bug 71116] [NV34] libdrm built with gcc4.8 causes crashes/lockups
https://bugs.freedesktop.org/show_bug.cgi?id=71116 Emil Velikov <emil.l.velikov at gmail.com> changed: What |Removed |Added ---------------------------------------------------------------------------- Summary|[Ubuntu userspace upgrade |[NV34] libdrm built with |13.04 -> 13.10][regression] |gcc4.8 causes |NV34 card fails to properly |crashes/lockups |resume. | --- Comment #17 from Emil Velikov <emil.l.velikov at gmail.com> --- Ronald all the thanks goes to the Andreas Radke (Arch developer) for finding and bisecting the issue. FWIW Arch has been using clang for ~3 months to build libdrm and it seems to work great. Ideally one of the Ubuntu/GCC devs can take a look and provide more bit more clarity than "This hints at something depending on undefined evaluation order between sequence points" [1] Would that be an issue with nouveau code, gcc and/or both will hopefully be determined soon. Cheers Emil [1] http://gcc.gnu.org/ml/gcc-help/2013-07/msg00151.html -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131103/8d71dbca/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-07 09:58 UTC
[Nouveau] [Bug 71116] [NV34] libdrm built with gcc4.8 causes crashes/lockups
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #18 from Maarten Lankhorst <bugs at mblankhorst.nl> --- Wow this is definitely a weird interesting bug. It affects suspend-to-mem on my pci-e nv43 too. <after resume from suspend to memory> [ 40.107810] nouveau E[ PGRAPH][0000:01:00.0] ERROR nsource: DATA_ERROR nstatus: BAD_ARGUMENT [ 40.108059] nouveau E[ PGRAPH][0000:01:00.0] ch 3 [0x000f7000 compiz[1713]] subc 2 class 0x0039 mthd 0x0314 data 0x0051f000 (error repeats) I can confirm that rebuilding only libdrm-nouveau2 with the raring toolchain fixes it. -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131107/8d7b04f6/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-07 10:02 UTC
[Nouveau] [Bug 71116] [any/all(?) cards below nv50(?)] libdrm built with gcc4.8 causes crashes/lockups
https://bugs.freedesktop.org/show_bug.cgi?id=71116 Ronald <ronald645 at gmail.com> changed: What |Removed |Added ---------------------------------------------------------------------------- Summary|[NV34] libdrm built with |[any/all(?) cards below |gcc4.8 causes |nv50(?)] libdrm built with |crashes/lockups |gcc4.8 causes | |crashes/lockups -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131107/5c1e00c5/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-07 10:04 UTC
[Nouveau] [Bug 71116] [any/all(?) cards below nv50(?)] libdrm built with gcc4.8 causes crashes/lockups
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #19 from Ronald <ronald645 at gmail.com> --- So, we have nv34, nv43 and nv44 affected by this. I updated the bugreport title so hopefully more people will stumble on it earlier (hope that is okay). - nv44: GCC thread: During normal operation. - nv43: Resuming from suspend-to-mem - nv34: Resuming from suspend-to-disk -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131107/d4ba66f9/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-07 10:05 UTC
[Nouveau] [Bug 71116] [any/all(?) cards below nv50(?)] libdrm built with gcc4.8 causes crashes/lockups normal operation, and resume from disk/memory
https://bugs.freedesktop.org/show_bug.cgi?id=71116 Ronald <ronald645 at gmail.com> changed: What |Removed |Added ---------------------------------------------------------------------------- Summary|[any/all(?) cards below |[any/all(?) cards below |nv50(?)] libdrm built with |nv50(?)] libdrm built with |gcc4.8 causes |gcc4.8 causes |crashes/lockups |crashes/lockups normal | |operation, and resume from | |disk/memory -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131107/d4991ebd/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-07 10:05 UTC
[Nouveau] [Bug 71116] [any/all(?) cards below nv50(?)] libdrm built with gcc4.8 causes crashes/lockups during normal operation, and resume from disk/memory
https://bugs.freedesktop.org/show_bug.cgi?id=71116 Ronald <ronald645 at gmail.com> changed: What |Removed |Added ---------------------------------------------------------------------------- Summary|[any/all(?) cards below |[any/all(?) cards below |nv50(?)] libdrm built with |nv50(?)] libdrm built with |gcc4.8 causes |gcc4.8 causes |crashes/lockups normal |crashes/lockups during |operation, and resume from |normal operation, and |disk/memory |resume from disk/memory -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131107/04bc9469/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-07 10:31 UTC
[Nouveau] [Bug 71116] [any/all(?) cards below nv50(?)] libdrm built with gcc4.8 causes crashes/lockups during normal operation, and resume from disk/memory
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #20 from Maarten Lankhorst <bugs at mblankhorst.nl> --- Created attachment 88817 --> https://bugs.freedesktop.org/attachment.cgi?id=88817&action=edit hack to force defined behavior As I feared, recompiling only pushbuf.c with gcc-4.7 fixes this bug. So something in pushbuf.c is miscompiled. Unfortunately it's also the biggest abuser of post increment ops. A simple workaround seems to be this patch. -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131107/2c7ff594/attachment-0001.html>
bugzilla-daemon at freedesktop.org
2013-Nov-07 10:51 UTC
[Nouveau] [Bug 71116] [any/all(?) cards below nv50(?)] libdrm built with gcc4.8 causes crashes/lockups during normal operation, and resume from disk/memory
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #21 from Maarten Lankhorst <bugs at mblankhorst.nl> --- Compiled code with this patch: 00000000000015b0 <nouveau_pushbuf_reloc>: 15b0: 55 push %rbp 15b1: 48 8b 6f 30 mov 0x30(%rdi),%rbp 15b5: 53 push %rbx 15b6: 48 89 fb mov %rdi,%rbx 15b9: e8 42 ea ff ff callq 0 <pushbuf_krel> 15be: 89 45 00 mov %eax,0x0(%rbp) 15c1: 48 83 43 30 04 addq $0x4,0x30(%rbx) 15c6: 5b pop %rbx 15c7: 5d pop %rbp 15c8: c3 retq Without: 0000000000001650 <nouveau_pushbuf_reloc>: 1650: 48 89 5c 24 f0 mov %rbx,-0x10(%rsp) 1655: 48 89 6c 24 f8 mov %rbp,-0x8(%rsp) 165a: 48 83 ec 10 sub $0x10,%rsp 165e: 48 8b 6f 30 mov 0x30(%rdi),%rbp 1662: 48 89 fb mov %rdi,%rbx 1665: e8 96 e9 ff ff callq 0 <pushbuf_krel> 166a: 89 45 00 mov %eax,0x0(%rbp) 166d: 48 83 c5 04 add $0x4,%rbp 1671: 48 89 6b 30 mov %rbp,0x30(%rbx) 1675: 48 8b 1c 24 mov (%rsp),%rbx 1679: 48 8b 6c 24 08 mov 0x8(%rsp),%rbp 167e: 48 83 c4 10 add $0x10,%rsp 1682: c3 retq It seems that the nouveau_pushbuf_reloc difference is harmless. Code is equivalent, except that without the patch applied it uses another temp variable causing less optimal code generation. Remaining candidates: pushbuf_validate, nouveau_pushbuf_data, nouveau_pushbuf_del. -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131107/bff92539/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-07 11:57 UTC
[Nouveau] [Bug 71116] [any/all(?) cards below nv50(?)] libdrm built with gcc4.8 causes crashes/lockups during normal operation, and resume from disk/memory
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #22 from Maarten Lankhorst <bugs at mblankhorst.nl> --- Oops, so gcc is broken here after all, look at this.. 0000000000003860 <nouveau_pushbuf_reloc>: 3860: 53 push %rbx 3861: 48 8b 5f 30 mov 0x30(%rdi),%rbx 3865: 48 8d 43 04 lea 0x4(%rbx),%rax 3869: 48 89 47 30 mov %rax,0x30(%rdi) 386d: e8 3e ea ff ff callq 22b0 <pushbuf_krel> 3872: 89 03 mov %eax,(%rbx) 3874: 5b pop %rbx 3875: c3 retq Source: void nouveau_pushbuf_reloc(struct nouveau_pushbuf *push, struct nouveau_bo *bo, uint32_t data, uint32_t flags, uint32_t vor, uint32_t tor) { *push->cur++ = pushbuf_krel(push, bo, data, flags, vor, tor); } gcc-4.8 evaluates it as: void nouveau_pushbuf_reloc(struct nouveau_pushbuf *push, struct nouveau_bo *bo, uint32_t data, uint32_t flags, uint32_t vor, uint32_t tor) { push->cur++; *push->cur[-1] = pushbuf_krel(push, bo, data, flags, vor, tor); } -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131107/6374f133/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-07 11:59 UTC
[Nouveau] [Bug 71116] [any/all(?) cards below nv50(?)] libdrm built with gcc4.8 causes crashes/lockups during normal operation, and resume from disk/memory
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #23 from Maarten Lankhorst <bugs at mblankhorst.nl> --- while gcc-4.7 evaluates it as: void nouveau_pushbuf_reloc(struct nouveau_pushbuf *push, struct nouveau_bo *bo, uint32_t data, uint32_t flags, uint32_t vor, uint32_t tor) { *push->cur = pushbuf_krel(push, bo, data, flags, vor, tor); push->cur++; } -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131107/d4f8dd15/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-07 12:07 UTC
[Nouveau] [Bug 71116] [any/all(?) cards below nv50(?)] libdrm built with gcc4.8 causes crashes/lockups during normal operation, and resume from disk/memory
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #24 from Maarten Lankhorst <bugs at mblankhorst.nl> --- Because the function also increments the push->cur pointer, it's a subtle but real difference: gcc-4.7: u32 ret = func(); // May change push->cur ptr *push->cur = ret; push->cur++; gcc-4.8: u32 *ptr = push->cur; push->cur++; *ptr = func(); // Already sees the push->cur ptr I'm not a language expert, so no idea if I'm right, but it seems the updated gcc-4.8 behavior is wrong here. -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131107/800faffa/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-07 12:24 UTC
[Nouveau] [Bug 71116] [any/all(?) cards below nv50(?)] libdrm built with gcc4.8 causes crashes/lockups during normal operation, and resume from disk/memory
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #25 from Maarten Lankhorst <bugs at mblankhorst.nl> --- Well, pedantically gcc-4.7 behavior is this: u32 *ptr = push->cur; *ptr = func(); push->cur++; But since func only reads push->cur it's the same. -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131107/c2a96cf7/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-07 12:57 UTC
[Nouveau] [Bug 71116] [any/all(?) cards below nv50(?)] libdrm built with gcc4.8 causes crashes/lockups during normal operation, and resume from disk/memory
https://bugs.freedesktop.org/show_bug.cgi?id=71116 --- Comment #26 from Kelly Doran <kel.p.doran at gmail.com> --- I am pretty sure the order of execution is considered undefined behavior here in C, thus the nouveau code is wrong (it is a side effect). Check out chapter 2.12 in "The C Programming Language" if you have it. -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131107/11524e83/attachment.html>
bugzilla-daemon at freedesktop.org
2013-Nov-08 08:04 UTC
[Nouveau] [Bug 71116] [any/all(?) cards below nv50(?)] libdrm built with gcc4.8 causes crashes/lockups during normal operation, and resume from disk/memory
https://bugs.freedesktop.org/show_bug.cgi?id=71116 Maarten Lankhorst <bugs at mblankhorst.nl> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEW |RESOLVED Resolution|--- |FIXED --- Comment #27 from Maarten Lankhorst <bugs at mblankhorst.nl> --- http://cgit.freedesktop.org/mesa/drm/commit/?id=482abbfafb56cbceaf5355c026434e638cddd0f1 -- You are receiving this mail because: You are the assignee for the bug. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20131108/d2e6be55/attachment-0001.html>
Seemingly Similar Threads
- [Bug 74613] New: [v3.14-rc1] [nv34] nouveau: get 0x10000000 put 0x0000ed30 state 0xc0000000 (err: MEM_FAULT) push 0x00000000
- [Bug 99354] New: [G71] "Assertion `bkref' failed" reproducible with glmark2
- [PATCH] nouveau: add coherent BO attribute
- [Bug 73358] New: [nv34] adobe flash + firefox -> DATA_ERROR
- [Bug 50121] New: [Regression] NV34 Fx5200 fails to resume with commit 5d720f2450