svn commit: r261216 - head/sys/dev/pccbb

[John Baldwin]

On Wednesday, February 05, 2014 7:26:33 pm Gavin Atkinson wrote:
> On Tue, 4 Feb 2014, John Baldwin wrote:
> > On Sunday, February 02, 2014 5:34:58 pm Gavin Atkinson wrote:
> > > On Mon, 27 Jan 2014, John Baldwin wrote:
> > > > Author: jhb
> > > > Date: Mon Jan 27 19:49:52 2014
> > > > New Revision: 261216
> > > > URL: http://svnweb.freebsd.org/changeset/base/261216
> > > > 
> > > > Log:
> > > >   Explicitly enable I/O and memory decoding in the bridge's command register
> > > >   when activating an I/O or memory window on the CardBus bridge.
> > > 
> > > This fixes some, but not all of my machines.  One in particular, a Toshiba 
> > > M5 laptop, remains broken by r254263 even with this change.  Specificaly, 
> > > the laptop does not notice when a card is inserted.
> > > 
> > > The attached minimal patch gets things working again, though I don't know 
> > > if is the correct fix or if a more involved fix is required.
> > > 
> > > dmesg before and after that patch:
> > > 
> > > http://people.freebsd.org/~gavin/m5-dmesg-before.txt
> > > http://people.freebsd.org/~gavin/m5-dmesg-after.txt
> > > 
> > > The only difference is the cbb register dump, the one bit that I am 
> > > setting in the patch.
> > 
> > Your patch effectively reverts r254263.  It may be the correct thing to do,
> > but the question is why. :)  Can you provide 'pciconf -lbc' output for this
> > device?  (You can just do 'pciconf -lbc pccbb0' in HEAD now)
> 
> Full "pciconf -lbc" output at 
> http://people.freebsd.org/~gavin/m5-pciconf-lbc.txt
> 
> It's the same both with and without my hack-patch.

Humm, no I/O port BAR.  I found a copy of the cardbus spec online and there is
nothing special that says it requires I/O decoding to be enabled.  I guess add it
with an XXX comment of some sort to note that at least one chipset requires this.

-- 
John Baldwin