|Summary:||[NV40] GPU lockup and segmentation fault of Nouveau-1.0.10 on PowerMac G5 with acceleration|
|Component:||Driver/nouveau||Assignee:||Nouveau Project <nouveau>|
|Status:||NEW ---||QA Contact:||Xorg Project Team <xorg-team>|
|i915 platform:||i915 features:|
Description dehua.yang 2013-11-22 02:16:10 UTC
Created attachment 89611 [details] Dmesg I tried to use nouvear driver with acceleration on gentoo PPC64 running on Apple PowerMac G5, and failed with GPU lockup and segmentation fault. The linux kernel booted successfuly into a console, then the screen was filled with rubbish(colorful blocks or the like) and stuck there when I started "startx" in the console. After 5mins I pressed "Ctrl+Alt+F2" to switch to another TTY and succeeded, I got a message "nouveau E[ DRM] GPU lockup - switch to software fbcon". When I came again to the console X was launched, the X terminated and system fell back to the TTY with many errors(see dmesg and Xorg.0.log attached). This only happens when nouveau has acceleration, when I disable it by putting "Option 'NoAccel' 'true'" in xorg.conf, X and Gnome run normally without problems. My computer is Apple PowerMac G5 of late 2003(PowerMac7,2) with GeForce 6800Ultra DDL graphics card(No any problem on Mac OXS 10.4.11). The followings are installed: The linux kernel 3.12.0, xorg-server 1.14.4, mesa 9.1.6, xf86-video-nouveau 1.0.10
Comment 2 dehua.yang 2013-11-22 02:23:20 UTC
I tried also Fedora 20 Beta PPC64 and Lubuntu 13.10 Powerpc Desktop, they all have the same problem.
Comment 3 Ilia Mirkin 2013-11-22 02:25:51 UTC
I'm not sufficiently familiar with PPC dmesg -- are you using 4K pages? If not, can you try it with 4K pages (CONFIG_PPC_4K_PAGES).
Comment 4 dehua.yang 2013-11-22 10:05:25 UTC
Yes. In my kernel .config file, there is "CONFIG_PPC_4K_PAGES=y". I am glad to provide further info as required.
Comment 5 dehua.yang 2013-11-22 12:18:06 UTC
I updated my system, now with: kernel 3.12.1 mesa 9.2.3 nouveau 1.9.10 xorg-server 1.14.4 After I put option '"ShadowBF" "true"' in xorg.conf, the nouveau driver works with acceleration. In glxinfo it has "direct rendering: Yes", and no nouveau error in dmesg and Xorg.0.log.
Comment 6 dehua.yang 2013-11-22 12:36:47 UTC
Sorry for the error above. nouveau 1.0.10, not 1.9.10.
Comment 7 dehua.yang 2013-11-23 01:09:48 UTC
Actually this is not HW acceleration by using ShadowFB. In Xorg.0.log there is "Using Shadow Framebuffer - acceleration disabled", so I still have a broken nouveau driver with acceleration. When using ShadowFB it seems the graphics operations are faster than that NoAccel option is enabled on my system. The 3D seems not supported, there is no difference in FPS running glxgears in both situations.
Comment 8 Emil Velikov 2013-12-13 20:12:08 UTC
There was a patch on the ML, which should handle some of the mayhem on PPC. Ben had a couple of comments so you might want to try the original patch and/or address the points raised. PPC is not the only arch affected by the "funny" page sizes, SPARC seems to be in the list as well see bug 58984.  http://lists.freedesktop.org/archives/nouveau/2013-November/015226.html  http://lists.freedesktop.org/archives/nouveau/2013-December/015358.html
Comment 9 Brock Wittrock 2015-07-06 01:47:33 UTC
Is there any update on this bug/issue with pagesize issues on 64-bit PowerPC and the nouveau driver? I assume now as this is still marked as open, but thought I would ask anyways. I've been able to partly get the driver working by using a kernel with 4kb page sizes and MSI interrupts disabled on Debian Jessie. I'll try to download, install, and build the latest nouveau driver to see if there are any changes. Also, I do see this is marked as a to-do on under DRM on the nouveau website. I can provide any sort of output/logs necessary. Just wanted to try and revive this somehow.
Comment 10 Brock Wittrock 2015-07-31 03:40:31 UTC
From what I was reading, Ben's patches weren't necessarily ideal for final inclusion into the driver. Still trying to wrap my head around the comments from that thread in the mailing list archives to see what changes need to be made to get 64k page sizes working for G5 owners with Nvidia cards. Anybody able to provide any assistance with this somehow?
Comment 11 Ben Skeggs 2015-07-31 05:17:49 UTC
(In reply to Brock Wittrock from comment #10) > From what I was reading, Ben's patches weren't necessarily ideal for final > inclusion into the driver. Still trying to wrap my head around the comments > from that thread in the mailing list archives to see what changes need to be > made to get 64k page sizes working for G5 owners with Nvidia cards. > > Anybody able to provide any assistance with this somehow? I plan on attempting to untangle the vm mess in the next few weeks, and this will be one of the things I'll attempt to address while doing so.
Comment 12 Alex Perez 2017-08-09 16:02:18 UTC
Ping, this bug is zombified. It's been open for nearly five years now, and hasn't had a comment in over two years. Ben, two years ago, you wrote "I plan on attempting to untangle the vm mess in the next few weeks, and this will be one of the things I'll attempt to address while doing so." Did anything ever come of that?
Comment 13 Ilia Mirkin 2017-08-09 16:21:20 UTC
Alex, which issue are you talking about? Like almost every issue where someone says "me too", they actually have a totally different issue. Here's a list of problems: - MSI is broken - 64k pages are broken - Undiagnosed issue with the NV40 AGP AFAIK all these issues are wholly unrelated to each other.