Re: Make b43 driver fall back gracefully to PIO mode after fatal DMA errors

[Michael Buesch <mb@bu3sch.de>]

On Friday 26 February 2010 22:50:35 Linus Torvalds wrote:
> 
> On Fri, 26 Feb 2010, Linus Torvalds wrote:
> > 
> > so even if mmiotrace fixes it for others, it's not working for me.
> 
> Side note: In case this is timing-related: the machine in question is a 
> dual-core 1.3Ghz CPU (Core 2 Duo U1400). I think SMP has always been 
> enabled, I don't think I ever tried UP.
> 
> Also, in addition to the PIO fallback code, that particular mmiotrace test 
> was done with the "udelay(10)" added to the b43_write*() functions. So it 
> was even slower.
> 
> Maybe the bug only happens on slow machines (and then the udelay obviously 
> wouldn't help!),

I'm pretty sure it's triggered by some kind of BIOS background service.
Power saving or something like that. There must be some interference
that fucks up the memory link between machine->ssb->wireless.

However, did you add the udelay to b43_write()? Please try to add it
further down in the callchain into the lowlevel SSB read and write functions.
drivers/ssb/pci.c

I don't think it'll help at all, but it's worth a try.

-- 
Greetings, Michael.