Messages in this thread | | | From | Shawn Nock <> | Subject | Re: rtlwifi: regression 39-rc5 (rtl8192ce) | Date | Mon, 02 May 2011 10:26:01 -0400 |
| |
Larry Finger <Larry.Finger@lwfinger.net> writes:
> On 05/01/2011 12:59 PM, Shawn Nock wrote: >> During heavy network traffic (esp. flash video) I see the attached >> NULL pointer dereference in 2.6.39-rc5 on an IBM Thinkpad >> x120e. Immediately afterward a "scheduling while atomic" bug is >> reported and the system becomes unresponsive. I am unable to produce >> this problem in 2.6.38.4. >> See attached backtrace and dmesg. Please let me know what I can >> collect to make this problem easier to troubleshoot. > > Could you also supply the instruction byte sequence for the oops? It > is the "Code:" line of the dump.
I had not been able to capture that, as the scheduler oops was triggered during the call stack dump (before the Code line).
> I think you are using a 32-bit system. Is that correct?
Correct.
> Is there one URL that exposes this problem? If so, please sent that as > well.
It had been speedtest.net (which *used* to crash this all the time). I've since re-configured the kernel to allow it to build faster (fedora .config trimmed to remove a lot of the unneeded modules). I can no longer trigger this BUG.
I originally thought that this may be because I enabled the 8192CU driver, but clean kernel without it also doesn't trigger the bug. My new hypothesis is that building the kernel for the AMD processor type was a mistake (the only other significant change was that I went back to Pentium Pro as the CPU target).
As far as I can tell this problem is resolved (20+ hours without issue, less than 20 min previously) and probably due to GCC not being aware of this (newish) AMD processor.
Sorry for the false alarm and thanks, Shawn
-- Shawn Nock (OpenPGP: 0x8132E623)
| |