Re: if_sk patch to get more info from people with problems

On Mon, 1 Nov 2004, Arjan de Vet wrote:

> In article <1099326109.23508.133.camel_at_leguin> you write:
> 
> >On Mon, 2004-11-01 at 05:54, User Valera wrote:
> 
> [...]
> 
> >debug.mpsafenet=0 has been my cure here as well for basically the same
> >symptoms.  I'll apply this patch and get info on what hardware I have
> >soon.
> 
> I filed a bug report for if_sk.c for possibly related symptoms: 
> 
>     http://www.freebsd.org/cgi/query-pr.cgi?pr=kern%2F73038 The
> IFF_NEEDSGIANT flag seems to solve the problem but I need to do much
> more testing with it this week (had to go back to 5.2.1 because 5.3
> hangs for me after 1-1.5 days; still need to debug that one as well :(). 

Note that it could well be a locking bug in if_sk, but that
debug.mpsafenet and IFF_NEEDSGIANT both substantially change the timing in
the driver, so these changes could just mask an existing bug in your
configuration (yet still leave it present in other's).  The if_de bug
recently, btw, turned out to be an existing (real) bug that was unmasked
due to timing changes from running with IFF_NEEDSGIANT, but was in fact
(we believe) present even with Giant over the stack...

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert_at_fledge.watson.org      Principal Research Scientist, McAfee Research