Bug 97367 - k520 card produce high volume of IRQ
Summary: k520 card produce high volume of IRQ
Status: RESOLVED MOVED
Alias: None
Product: xorg
Classification: Unclassified
Component: Driver/nouveau (show other bugs)
Version: unspecified
Hardware: Other All
: medium normal
Assignee: Nouveau Project
QA Contact: Xorg Project Team
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2016-08-16 14:53 UTC by Mathieu Maret
Modified: 2019-12-04 09:15 UTC (History)
0 users

See Also:
i915 platform:
i915 features:


Attachments
dmesg (38.41 KB, text/plain)
2016-08-16 14:53 UTC, Mathieu Maret
no flags Details
vbios ROM (192.00 KB, application/octet-stream)
2016-08-16 14:53 UTC, Mathieu Maret
no flags Details
dmesg with debug (nouveau.debug=debug drm.debug=0xe) (81.30 KB, text/plain)
2016-08-16 16:28 UTC, Mathieu Maret
no flags Details
perf report from perf record -ag (1.07 MB, text/plain)
2016-08-18 08:16 UTC, Mathieu Maret
no flags Details
perf report from perf record -g -C 7 (1.08 MB, text/plain)
2016-08-18 08:17 UTC, Mathieu Maret
no flags Details
result from 'cat /proc/interrupts > irq ; sleep 1; cat /proc/interrupts >>irq' (19.56 KB, text/plain)
2016-08-18 08:17 UTC, Mathieu Maret
no flags Details
Tracing using ftrace during 1sec (809.71 KB, application/x-gtar)
2017-02-08 15:44 UTC, Mathieu Maret
no flags Details

Description Mathieu Maret 2016-08-16 14:53:01 UTC
Created attachment 125818 [details]
dmesg

Amazon propose computer with k520(GK104) GPU for virtualisation (Xen).
You can use Ubuntu LTS on it for example.
When inserting the nouveau driver, /proc/interrupts report that nvkm is doing ~100000 interrupts/sec.

This have been tested on kernel 4.4.0 and 4.4.13.

On amazon, you do not control the real hardware where your system will be run. It appear that sometimes, this issue do not happen. May be this could be due to vbios version ?

Forcing nouveau in headless mode (modeset=2) produce a normal amount of IRQ.
Comment 1 Mathieu Maret 2016-08-16 14:53:33 UTC
Created attachment 125819 [details]
vbios ROM
Comment 2 Mathieu Maret 2016-08-16 16:28:30 UTC
Created attachment 125821 [details]
dmesg with debug (nouveau.debug=debug drm.debug=0xe)
Comment 3 Mathieu Maret 2016-08-16 16:31:26 UTC
With another try having modeset=2 does not produce a normal amount of IRQ.
Sorry for the error, it should be due to the randomness of the issue :/
Comment 4 Mathieu Maret 2016-08-18 08:16:18 UTC
Created attachment 125864 [details]
perf report from perf record -ag
Comment 5 Mathieu Maret 2016-08-18 08:17:14 UTC
Created attachment 125865 [details]
perf report from perf record -g -C 7
Comment 6 Mathieu Maret 2016-08-18 08:17:43 UTC
Created attachment 125866 [details]
result from 'cat /proc/interrupts > irq ; sleep 1; cat /proc/interrupts >>irq'
Comment 7 Mathieu Maret 2016-08-18 08:22:08 UTC
The irq file show that the IRQ on nvkm happens 140954 times in 1 sec.
The perf reports does not show things related to nvkm, but I may not have use the right options for perf
Comment 8 Mathieu Maret 2017-02-06 15:35:11 UTC
Disabling MSI interrupts (with config="NvMSI=0") reduce the number of IRQ to about 1k/s
Comment 9 Mathieu Maret 2017-02-08 15:44:20 UTC
Created attachment 129412 [details]
Tracing using ftrace during 1sec
Comment 10 Mathieu Maret 2017-02-08 16:54:05 UTC
This issue can be reproduce on 4.10-rc10
Comment 11 Martin Peres 2019-12-04 09:15:56 UTC
-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/xorg/driver/xf86-video-nouveau/issues/280.


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.