Bug 42312 - Memory corruption revealed by repeated slabinfo -v
Summary: Memory corruption revealed by repeated slabinfo -v
Status: ASSIGNED
Alias: None
Product: Memory Management
Classification: Unclassified
Component: Slab Allocator (show other bugs)
Hardware: All Linux
: P1 normal
Assignee: Andrew Morton
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2011-09-03 16:10 UTC by Alex Villacis Lasso
Modified: 2014-09-02 06:39 UTC (History)
3 users (show)

See Also:
Kernel Version: 3.12
Subsystem:
Regression: Yes
Bisected commit-id:


Attachments
dmesg output showing memory corruption reports (88.39 KB, text/plain)
2011-09-03 16:10 UTC, Alex Villacis Lasso
Details
Kernel configuration file for 3.1-rc4 (116.93 KB, text/plain)
2011-09-03 16:12 UTC, Alex Villacis Lasso
Details
dmesg output for 3.0 (58.37 KB, text/plain)
2011-09-03 16:13 UTC, Alex Villacis Lasso
Details

Description Alex Villacis Lasso 2011-09-03 16:10:24 UTC
Created attachment 71532 [details]
dmesg output showing memory corruption reports

Machine is a Fedora 14 x86_64 running kernel 3.1-rc4 with 2 GB of memory, Intel KMS.

In 3.1-rc4, seemingly simple operations are now triggering memory corruption reports, as shown by slabinfo -v. What I did to reproduce:

Compiled 3.1-rc4 with zcache support and EXT4 mounting of EXT2/3 filesystems
Booted into 3.1-rc4 with zcache and slub_debug=FZPU (although I have seen memory corruptions even without zcache), into init 3 (textmode, KMS, no graphical desktop)
In one console, I ran (as root): while true ; do ./slabinfo -v ; sleep 3 ; done
In another console, as an ordinary user, I did some operations: git pull on kernel tree, mplayer with MP3 file, mplayer with video file. Since my user does not have access to /dev/fb0 for framebuffer output, I logged out from the ordinary user, and logged back in as root in the second console. When I ran usermod -a -G video <myuser> , I got the first memory corruption report. After that, corruption reports appeared regularly even with idle system.

I also got a circular dependency warning when mplayer managed to use the framebuffer for the video, and then was shut down normally.


The same sequence fails to trigger memory corruption errors in 3.0 even with zcache active. However, I could trigger memory corruptions in 3.1-rc3 too.

The circular locking warning is reproducible in 3.0 too, so I believe it is a separate problem.
Comment 1 Alex Villacis Lasso 2011-09-03 16:12:06 UTC
Created attachment 71542 [details]
Kernel configuration file for 3.1-rc4
Comment 2 Alex Villacis Lasso 2011-09-03 16:13:14 UTC
Created attachment 71552 [details]
dmesg output for 3.0

Notice there are no corruption reports, but the circular dependency is still reported.
Comment 3 Alan 2012-08-30 13:35:19 UTC
Can you dplicate this without virtualbox loaded ?
Comment 4 Paulo Zanoni 2013-10-03 21:07:29 UTC
(In reply to Alan from comment #3)
> Can you dplicate this without virtualbox loaded ?

I can reproduce it without virtualbox, with a relatively small 3.12-rc3 Kernel, on different machines. The error messages are always very similar to the one attached by Alex.

Note You need to log in before you can comment on or make changes to this bug.