Re: [Bug #15124] PCI host bridge windows ignored (works with pci=use_crs)

From: Jesse Barnes <jbarnes@virtuousgeek.org>
To: Yinghai Lu <yinghai@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
	Bjorn Helgaas <bjorn.helgaas@hp.com>,
	"Rafael J. Wysocki" <rjw@sisk.pl>,
	Jeff Garrett <jeff@jgarrett.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Kernel Testers List <kernel-testers@vger.kernel.org>,
	Linux PCI <linux-pci@vger.kernel.org>,
	Myron Stowe <myron.stowe@hp.com>,
	Matthew Garrett <mjg59@srcf.ucam.org>,
	Ingo Molnar <mingo@elte.hu>
Subject: Re: [Bug #15124] PCI host bridge windows ignored (works with pci=use_crs)
Date: Tue, 26 Jan 2010 10:35:11 -0800	[thread overview]
Message-ID: <20100126103511.16781111@jbarnes-piketon> (raw)
In-Reply-To: <4B5F32A9.7090105@kernel.org>

On Tue, 26 Jan 2010 10:21:29 -0800
Yinghai Lu <yinghai@kernel.org> wrote:

> On 01/26/2010 10:16 AM, Linus Torvalds wrote:
> > 
> > 
> > On Tue, 26 Jan 2010, Bjorn Helgaas wrote:
> >>
> >> which IS big enough, and we know the bridge is in fact forwarding the
> >> [mem 0xd0000000-0xdfffffff 64bit pref] region, because the Radeon works
> >> when Jeff boots with "pci=use_crs".
> > 
> > I bet it's a subtractive decode thing. Sure, it could be just another 
> > undocumented range register (does anybody have the datasheet for that 
> > thing?) but Intel tends to often have subtractive decode.
> > 
> > That system in question has three PCI express root ports, but two of them 
> > have IO and memory disabled according to the lspci info. So maybe it's as 
> > simple as that "I/O Hub PCI Express Root Port 7" just catching anything 
> > that nobody else does, and the single IOH host chip doing the same?
> > 
> >> I think we should remove intel_bus.c before .33.  It's breaking boxes
> >> and we don't know how to fix it.  Even if we do find out how to fix it,
> >> I think we should move toward using _CRS instead, because that's what
> >> Windows uses and it's an easy way for the firmware to tell us about
> >> platform quirks.
> > 
> > I suspect that for 33 it is indeed best to just revert. But somebody is 
> > bound to have information on how the actual hardware works. Yinghai?
> 
> I have asked intel if there is any bit that could be enabled the routing.
> there is no info about for their documentations.

I could probably dig something up in our confidential database, but this
is the main problem with intel_bus.c.  It'll always be behind with _CRS
provides.  Sure _CRS may be wrong sometimes, but it'll always work well
enough to bring Windows up, so we ought not to ignore it.

The underlying problems with our _CRS support still aren't fixed
though, so switching that on for 2.6.33 isn't an option.

-- 
Jesse Barnes, Intel Open Source Technology Center