Bug 91705 - [NVE7] GPU crash (read fault) on boot w/ DMI+HDMI outputs
Summary: [NVE7] GPU crash (read fault) on boot w/ DMI+HDMI outputs
Status: NEW
Alias: None
Product: xorg
Classification: Unclassified
Component: Driver/nouveau (show other bugs)
Version: unspecified
Hardware: x86-64 (AMD64) Linux (All)
: medium normal
Assignee: Nouveau Project
QA Contact: Xorg Project Team
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2015-08-20 18:35 UTC by Zack Weinberg
Modified: 2015-10-27 08:26 UTC (History)
1 user (show)

See Also:
i915 platform:
i915 features:


Attachments
dmesg (crash) (93.56 KB, text/plain)
2015-08-20 18:35 UTC, Zack Weinberg
no flags Details
dmesg (no crash) (69.26 KB, text/plain)
2015-08-20 18:35 UTC, Zack Weinberg
no flags Details
xorg.log (crash) (38.12 KB, text/plain)
2015-08-20 18:36 UTC, Zack Weinberg
no flags Details
xorg.log (no crash) (38.89 KB, text/plain)
2015-08-20 18:36 UTC, Zack Weinberg
no flags Details
vbios.rom (161.55 KB, application/octet-stream)
2015-08-20 19:17 UTC, Zack Weinberg
no flags Details

Note You need to log in before you can comment on or make changes to this bug.
Description Zack Weinberg 2015-08-20 18:35:08 UTC
I have a brand new computer with an NVE7-based card (GT740); the card has three physical outputs, 1 each DVI, HDMI, and VGA.  I have two monitors.  If I connect one to the DVI and one to the VGA port, or if I just connect one monitor to the DVI port, it boots successfully.  But if I connect one monitor to the DVI port and the other to the HDMI port (using an HDMI-to-DVI adapter cable), the GPU and/or the kernel component of the nouveau driver crash immediately upon switching the console to nouveaufb.  (This happens before the X server starts.) Both monitors go into power save, the keyboard is unresponsive, and there is a flood of nouveau-related errors in the kernel log.  Later, X *thinks* it has started successfully -- there's nothing that looks like an error to me in xorg.0.log -- but the monitors do not wake up and the keyboard remains unresponsive.  I can, however, ssh into the computer.

(It now occurs to me to try connecting just one monitor to HDMI.  I will do that after submitting this bug report.)

I will attach dmesg and xorg logs from both successful and unsuccessful boots.  Please let me know if any additional information is required.

I'm using distribution packages of everything involved, from Debian unstable as of, er, today.  The Nvidia proprietary drivers have never been installed.  Version information gleaned from xorg.log:

Current Operating System: Linux kenaz 4.1.0-2-amd64 #1 SMP Debian 4.1.5-1 (2015-08-15) x86_64
xorg-server 2:1.17.2-1.1
Current version of pixman: 0.32.6
NOUVEAU driver Date:   Thu Aug 28 03:57:48 2014 +0200
[drm] nouveau interface version: 1.2.2
Comment 1 Zack Weinberg 2015-08-20 18:35:42 UTC
Created attachment 117821 [details]
dmesg (crash)
Comment 2 Zack Weinberg 2015-08-20 18:35:59 UTC
Created attachment 117822 [details]
dmesg (no crash)
Comment 3 Zack Weinberg 2015-08-20 18:36:11 UTC
Created attachment 117823 [details]
xorg.log (crash)
Comment 4 Zack Weinberg 2015-08-20 18:36:33 UTC
Created attachment 117824 [details]
xorg.log (no crash)
Comment 5 Zack Weinberg 2015-08-20 18:39:39 UTC
N.B. I have been unable to figure out whether or not any of the software involved is actually the newest available whatever.  I'm happy to test newer software, compiling it myself if necessary -- but I will need you to tell me exactly what is out of date and where the newer version can be found, because the various webpages that purport to explain this all contradict each other :-(
Comment 6 Zack Weinberg 2015-08-20 18:47:04 UTC
If there is one monitor connected to the HDMI port, and nothing connected to DVI or VGA, the computer boots successfully.
Comment 7 Ilia Mirkin 2015-08-20 18:53:14 UTC
errrr what?

[   12.163791] nouveau E[    PBUS][0000:01:00.0] MMIO read of 0x00000000 FAULT at 0x61d804 [ IBUS ]
[   26.237760] nouveau E[    PBUS][0000:01:00.0] MMIO read of 0x00000000 FAULT at 0x61a804 [ IBUS ]

That'd be PDISPLAY.SOR[0x3].PWR and PDISPLAY.DAC[0x1].DPMS_CTRL for those following at home.

No idea what we're (not) doing to make those mmio reads fail. As they're connected to "power", perhaps they're getting turned off for our convenience? Not sure.

Perhaps it just doesn't have a SOR[3]/DAC[1]? Can you include your VBIOS, perhaps there's something interesting in there? From a successful boot, grab /sys/kernel/debug/dri/0/vbios.rom and upload it here.
Comment 8 Zack Weinberg 2015-08-20 19:17:32 UTC
Created attachment 117825 [details]
vbios.rom

(In reply to Ilia Mirkin from comment #7)
> 
> No idea what we're (not) doing to make those mmio reads fail. As they're
> connected to "power", perhaps they're getting turned off for our
> convenience? Not sure.

Perhaps it *is* relevant, then, that when monitors are connected to VGA and DVI, both are activated (in mirrored mode) by the BIOS, but when monitors are connected to HDMI and DVI, *only* the HDMI-attached display is activated by the BIOS.

> Perhaps it just doesn't have a SOR[3]/DAC[1]? Can you include your VBIOS,
> perhaps there's something interesting in there? From a successful boot, grab
> /sys/kernel/debug/dri/0/vbios.rom and upload it here.

Attached.

This computer is in my office, and I need to go home shortly, but I will be able to do further experiments tomorrow morning.
Comment 9 Zack Weinberg 2015-08-20 19:21:21 UTC
Oh, one more thing: I do not have Windows on this computer at all, so I can't tell you what it does in the same configuration.  Sorry.  (I can probably dig up a Windows 7 install/recovery CD, though, if there's something useful I could do with it that doesn't involve repartitioning ;-)


Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.