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

[Rafael J. Wysocki]

On Tuesday 26 January 2010, Bjorn Helgaas wrote:
> On Tuesday 26 January 2010 05:48:59 am Rafael J. Wysocki wrote:
> > On Tuesday 26 January 2010, Jeff Garrett wrote:
> > > On Sun, Jan 24, 2010 at 11:04:38PM +0100, Rafael J. Wysocki wrote:
> > > > The following bug entry is on the current list of known regressions
> > > > from 2.6.32.  Please verify if it still should be listed and let me know
> > > > (either way).
> > > > 
> > > > Bug-Entry	: http://bugzilla.kernel.org/show_bug.cgi?id=15124
> > > > Subject		: PCI host bridge windows ignored (works with pci=use_crs)
> > > > Submitter	: Jeff Garrett <jeff@xxxxxxxxxxxx>
> > > > Date		: 2010-01-13 5:37 (12 days old)
> > > > References	: http://marc.info/?l=linux-kernel&m=126336296600307&w=4
> > > > Handled-By	: Yinghai Lu <yinghai@xxxxxxxxxx>
> > > > 		  Bjorn Helgaas <bjorn.helgaas@xxxxxx>
> > > 
> > > This regression should still be listed.  No patch to test yet.
> > ...
> > IIRC, we already know how to fix this ...
> 
> As far as I know, we do NOT know how to fix this.
> 
> This regression occurred when we added intel_bus.c because it's not
> yet smart enough to determine the correct host bridge apertures.
> Here's what it thinks the bridge aperture is and the Radeon BAR:
> 
>   IOH bus: 00 index 1 mmio: [e0000000, fdffffff]
>   pci 0000:04:00.0: reg 10: [mem 0xd0000000-0xdfffffff 64bit pref]
> 
> The IOH aperture is obviously not big enough to cover the Radeon BAR.
> But the host bridge _CRS tells us this:
> 
>   pci_root PNP0A08:00: host bridge window [mem 0xc0000000-0xdfffffff]
>   pci_root PNP0A08:00: host bridge window [mem 0xf0000000-0xfed8ffff]
> 
> 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'm quite concerned about this for .33 because I don't think Jeff's
> configuration (Dell desktop with Intel x58 and large graphics device)
> is unusual.
> 
> The benefit of intel_bus.c is on machines with multiple IOHs, where we
> need to figure out which address ranges go to which IOHs so we can
> program downstream devices correctly.  But even there, _CRS should give
> us the information we need, so "pci=use_crs" should make these machines
> work.
> 
> 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.

Perhaps it would be sufficient to make pci=use_crs the default and leave the
option to use intel_bus.c for whoever needs that?

Rafael
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/