Re: ath0 no longer attaches, cardbus problems?

On Saturday, September 10, 2011 8:13:23 am Daniel Eischen wrote:
> On Fri, 9 Sep 2011, Warner Losh wrote:
> 
> >
> > On Sep 9, 2011, at 1:22 PM, Daniel Eischen wrote:
> >
> >> I found the commit that broke ath for me, r222753,
> >> specifically, the change to /dev/cardbus/cardbus_cis.c.
> >>
> >> To be sure, I updated to head using svn, and applied
> >> the patch included below.  ath attaches and works.  Without
> >> the patch, ath does not attach.
> >>
> >> On another note, I've no idea why updating from a local
> >> CVS repo lead me down a wrong path.  It seems wrong that
> >> a 'cvs update -P -d -A -D "31 Mar 2011"' works and
> >> a 'cvs update -P -d -A -D "1 Apr 2011"' does not work.
> >> r222753 did not occur until much later (June 6).
> >> Once John asked me to try r220195, I switched to using
> >> svn.  When that worked, it seemed strange to me because
> >> nothing else committed after that on Mar 31 should have
> >> broke ath.
> >>
> >> Anyway, culprit found.  Now what is the correct fix?
> >
> > Do you need both chunks?  The second one seems redundant given the 
definition of bus_alloc_reosurce_any does exactly that.
> 
> I tried it separately with the 2 chunks, and only the first
> chunk is needed.  To be pedantic, this was the change that
> made ath work again.
> 
> Index: sys/dev/cardbus/cardbus_cis.c
> ===================================================================
> --- sys/dev/cardbus/cardbus_cis.c	(revision 225463)
> +++ sys/dev/cardbus/cardbus_cis.c	(working copy)
> _at__at_ -441,6 +441,7 _at__at_
>   {
>   	if (res != CIS_CONFIG_SPACE) {
>   		bus_release_resource(child, SYS_RES_MEMORY, rid, res);
> +		bus_delete_resource(child, SYS_RES_MEMORY, rid);
>   	}
>   }
> 
> While debugging the problem a couple of weeks ago, I did
> seem to notice ath was trying to attach twice.  I seem to
> recall it was at different addresses.  Could this possibly
> cause the problem without the above patch?

No, but your patch is confusing to me.  Can a BAR change to a different
size after we read the CIS?  Or perhaps it should now be prefetchable when
it wasn't before?

Hmm, the bus_delete_resource() doesn't make us re-probe the BAR (perhaps
it should?).  I'm still not sure how exactly this fixes it.  However, I do
think this probably is more correct.

-- 
John Baldwin