Re: Regression with commit f9cde5f in 2.6.30-gitX

[Gary Hade <garyhade@us.ibm.com>]

On Wed, Jun 24, 2009 at 02:24:24PM -0700, Gary Hade wrote:
> On Wed, Jun 24, 2009 at 03:05:12PM -0500, Larry Finger wrote:
> > Gary Hade wrote:
> > > On Wed, Jun 24, 2009 at 08:45:31PM +0200, Thomas Gleixner wrote:
> > >> Can we just bring the limit check back and increase the number for now
> > >> until folks come up with a better solution ?
> > > 
> > > Another possible option is leaving in the limit check (still valid
> > > IMO for correct behavior of previous 'pci=use_crs') and reverting
> > >  http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=9e9f46c44e487af0a82eb61b624553e2f7118f5b
> > > until the better solution for the fixed size array issue is
> > > available.
> > 
> > Either increasing PCI_BUS_NUM_RESOURCES from 16 to 20 or 24
> 
> Yes, it looks to me like 20 might be the absolute minimum
> that your system could tolerate.  The big mystery is how much
> to increase it so we don't see the same problem on other
> systems where _CRS may return some unknown amount more that
                                                         ^^^^
                                                         than

> the 17 resources that are being returned on your system.
> 
> > or reverting f9cde5f will work for my system.
> 
> This would not be good without also reverting
> 9e9f46c44e487af0a82eb61b624553e2f7118f5b since devices
> below the transparent bridge on your's and other's systems
> may have trouble getting the resources they need.
> 
> Removing the check would only hide possibly more insidious
> and difficult to debug problems that could show up later.
> This was the main reason I provided the patch.  This kind
> of check should have actually existed prior to
> 9e9f46c44e487af0a82eb61b624553e2f7118f5b when 'pci=use_crs'
> was needed to enable use of the _CRS data.

One additional thing that is interesting about _CRS
returning 17 resources on your system is that even in
the absence of the transparent bridge constraint where
the last 3 slots of the resource array would be available,
there still isn't enough room in that 16 element array for
all 17 resources.  Without the check that 17th resource
(apparently not needed for your current configuration)
would be silently ignored.

Gary

-- 
Gary Hade
System x Enablement
IBM Linux Technology Center
503-578-4503  IBM T/L: 775-4503
garyhade@us.ibm.com
http://www.ibm.com/linux/ltc