Subject : 2.6.32-rc6-git4 -- BUG kmalloc-8192: Poison overwritten - INFO: Allocated in ath_rxbuf_alloc+0x1c/0x70 [ath] age=526999 cpu=1 pid=1275
Submitter : Miles Lane <email@example.com>
Date : 2009-11-12 4:58
References : http://marc.info/?l=linux-kernel&m=125800196520396&w=4
Notify-Also : "Luis R. Rodriguez" <firstname.lastname@example.org>
This entry is being used for tracking a regression from 2.6.31. Please don't
close it until the problem is fixed in the mainline.
Asked user to see if he could test wireless-testing + some patches.
Patches supplied, waiting on user.
Created attachment 24393 [details]
2.6.33-rc2-git3 -- Recurrence of this problem
Created attachment 24394 [details]
The .config file for this build
User reports that this problem persists as of 2.6.33-rc2-git3. I'm guessing that would equate to the testing Luis requested in comment 1?
The message thread linked in the description ended with the following:
>> How can you trigger it?
>I will do my best to determine that and get back to you.
Did we get that determination? Can we record it here?
Created attachment 26116 [details]
2.6.34-rc5 problem log
I also hit this bug in current latest linus's git(2.6.34-rc5+).
seems it's not easy to reproduce.
Anyway, is there a patch (either fixing this bug or put debug log) I can try?
I just realized that one of my patches that's a part of the AR9300 support patch series might have an effect on this bug, maybe it could even fix it.
Please test a recent version of wireless-testing, because the change is not in 2.6.34-rc5 yet.
Felix what patch did you have in mind?
As part of the EDMA changes, I changed an existing dma_sync_single_for_cpu() line into dma_sync_single_for_device().
In other tests, I sometimes saw weird memory corruption issues when the cache wasn't properly synced, so maybe this change would help on x86 under some circumstances as well.
Did this get fixed?
No, but we had no intstructions to reproduce. We finally got someone who might have instructions to reproduce and I am looking at it now:
seems like fun :)
Patch series that fixes this issue:
fixed by mainline commits (marked for stable):