On Sun, 2012-08-12 at 09:03 -0600, Jesse Rhodes wrote: > Yes, it does! Confirmed on both bcf3655... and 3.2.0-3 from sid. Why > would a cpufreq module break a PCI wireless card like that? According to a comment in the code that logs the error message: * If we are in a noisy environment, AGC calibration may time * out and/or noise floor calibration might timeout. Maybe the CPU frequency changes create additional noise in the wifi band. If so, I don't know how this can reasonably be avoided. CPU frequency scaling should normally be enabled, and Debian has for long a time enabled it by default on any computer detected as being a laptop or similar. (The kernel change just automated this at a slightly lower level.) ath5k developers, have you seen this failure mode before? The comment suggests there is a known hardware bug that can cause it; is it possible to work around that by retrying? Or is the timeout actually too low? Ben. > On Sat, Aug 11, 2012 at 9:43 PM, Ben Hutchings <ben@decadent.org.uk> > wrote: > On Sat, 2012-08-11 at 21:17 -0600, Jesse Rhodes wrote: > > sney@bivouac:~/data/debian/linux$ git bisect good > > bcf3655971d24d7f08f373ed5830e8e11010088a is the first bad > commit > > commit bcf3655971d24d7f08f373ed5830e8e11010088a > > Author: Andi Kleen <ak@linux.intel.com> > > Date: Thu Jan 26 00:09:12 2012 +0100 > > > > cpufreq: Add support for x86 cpuinfo auto loading v4 > > [...] > > So does the problem go away if you unload the cpufreq driver > (powernow_k8)? > > Ben. > > -- > Ben Hutchings > Sturgeon's Law: Ninety percent of everything is crap. > -- Ben Hutchings Sturgeon's Law: Ninety percent of everything is crap.
Attachment:
signature.asc
Description: This is a digitally signed message part