bwn + BETA3/ppc = very unstable

[Roman Bogorodskiy]

  Adrian Chadd wrote:

> On 3 October 2011 12:43, Roman Bogorodskiy <novel at freebsd.org> wrote:
> >  Roman Bogorodskiy wrote:
> 
> > Hm, I have applied a fix similar to this one:
> >
> > http://www.openbsd.org/cgi-bin/cvsweb/src/sys/dev/ic/bwi.c.diff?r1=1.88;r2=1.89
> >
> > I'm not sure if it's a coincidence or not, but I had no panics since
> > then. Packet loss rate are still very high - first packets go through
> > without losses, but eventually loss rate goes up.
> 
> It sounds like a calibration issue.
> 
> > Now uptime is 1 hour and loss rate is about 80-85%.
> >
> > Any directions how can I debug this?
> 
> Well, firstly, can you please do up a patch against -HEAD? I'll commit
> it to FreeBSD-HEAD ASAP and then include it in the list of things I
> want to backmerge into -9.

Here's the  diff: http://people.freebsd.org/~novel/misc/bwi.diff

Uptime is 12 hours now and no panics so far with it.

> Debugging this - I'd first isolate it to whether it is:
> 
> * TX issues;
> * RX issues;
> * or both TX and RX issues.
> 
> I suggest liberal use of iperf in UDP mode to TX things. It's a bit
> difficult as TX'ing a one way UDP stream also involves RX'ing an ACK
> from the remote end, so it isn't a 100% guaranteed way of debugging
> it. But it hopefully will show it's either TX or RX related.
> 
> I'd also enable BWI_DEBUG and then flip on the relevant bits that let
> you debug the attach and configuration bits (BWI_DBG_RF,
> BWI_DBG_TXPOWER, BWI_DBG_ATTACH).
> In fact, enable both BWI_DEBUG and BWI_DEBUG_VERBOSE in your kernel
> config and rebuild.
> 
> The BWI_DBG bits are defined in if_bwivar.h.

Thanks for the info, will trying doing that!

Roman Bogorodskiy
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 488 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-wireless/attachments/20111003/7d329a56/attachment.pgp