Date: Thu, 8 Jan 2009 16:08:08 -0800
From: Gary Hade <garyhade@us.ibm.com>
To: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>, Connor Behan <connor.behan@gmail.com>,
       linux-kernel@vger.kernel.org, Linux PCI <linux-pci@vger.kernel.org>,
       Linus Torvalds <torvalds@linux-foundation.org>
Subject: Re: Can't allocate resources for PCI video card behind bridge
Message-ID: <20090109000808.GA7703@us.ibm.com>
References: <49665493.4070808@gmail.com> <200901082057.00170.rjw@sisk.pl> <200901081226.38909.jbarnes@virtuousgeek.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <200901081226.38909.jbarnes@virtuousgeek.org>
User-Agent: Mutt/1.5.17+20080114 (2008-01-14)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 4589
Lines: 98

On Thu, Jan 08, 2009 at 12:26:38PM -0800, Jesse Barnes wrote:
> On Thursday, January 8, 2009 11:56 am Rafael J. Wysocki wrote:
> > Cc-ing linux-pci and Jesse.
> >
> > On Thursday 08 January 2009, Connor Behan wrote:
> > > Hello, I am trying to use two video cards. One is the built in ATI Rage
> > > Mobility M3 AGP card which drives the screen of my laptop (Thinkpad a22m
> > > 2628-S1U if that matters), the other is the ATI Radeon X1550 PCI card in
> > > the Thinkpad Dock II which is supposed to drive an external monitor.
> > > X1550 is a PCI-E chipset but my card is a PCI card with an onboard PCI
> > > to PCI-E bridge. I first tried this with kernel 2.6.27 as packaged by
> > > Archlinux and got the following startup errors:
> > >
> > > pci 0000:00:04.0: BAR 7: can't allocate resource
> > > pci 0000:00:04.0: BAR 8: can't allocate resource
> > > pci 0000:00:04.0: BAR 9: can't allocate resource
> > >
> > > I've seen this problem discussed in the archives and couldn't find what
> > > looked like a fix. I WAS able to fix this by loading dock, pci_slot,
> > > pci_hotplug and acpiphp and appending pci=assign-busses. However since
> > > then I have run a package update to kernel 2.6.27.10 and I am once again
> > > getting the same problem even with the same modules and parameters. I
> > > have since tried appending irqpoll, pci=bios, pci=routeirq and acpi=off
> > > with no success. I have even rolled back my kernel to 2.6.27 (and the
> > > headers and a few other packages) setup my system the way it was setup
> > > before to the best of my knowledge but I still can't get the PCI card
> > > working again. Another subtle change must be responsible.
> > >
> > > Certain PCI devices in the dock are working such as the Cardbus slots...
> > > but I could care less about them... is there a way I could tell the
> > > kernel to prioritize the video card and sacrifice Cardbus slots first if
> > > it can't allocate enough resources? Also my lspci output shows a large
> > > gap 01:00.0 to 06:00.0 so some part of my system must have expected
> > > other devices in that range. When I had the video card working it was
> > > 02:00.0. I have tried changing the four IRQ numbers in the BIOS but this
> > > is pointless at best and sometimes causes my soundcard to be disabled.
> > > My dmesg, lspci and (trimmed) kernel config are pasted:
> > >
> > > dmesg: http://pastebin.com/f3ae15a58
> > > lspci: http://pastebin.com/f5d7dee6e
> > > .config: http://pastebin.com/fb925760
> > >
> > > I'm hoping for a solution that doesn't require recompiling the kernel
> > > but I definitely would if that's necessary. In any event I'll be much
> > > more vigilant about system updates if I get this working again.
> > > Thank-you very much for your time.
> 
> Can you post the output of lspci -vvv somewhere as well after trying 2.6.28?  
> It may be that we want the "don't allocate resources for transparent bridges" 
> patch after all:
> 
> diff --git a/drivers/pci/setup-bus.c b/drivers/pci/setup-bus.c
> index ea979f2..586451c 100644
> --- a/drivers/pci/setup-bus.c
> +++ b/drivers/pci/setup-bus.c
> @@ -467,8 +467,12 @@ void __ref pci_bus_size_bridges(struct pci_bus *bus)
>                 }
>         }
> 
> -       /* The root bus? */
> -       if (!bus->self)
> +       /*
> +        * We don't need to allocate PCI bridging windows
> +        * for a root bus (everything bridged) or for a
> +        * transparent one.
> +        */
> +       if (!bus->self || bus->self->transparent)
>                 return;
> 
>         switch (bus->self->class >> 8) {

Jesse, I think this change is doing the same thing as
  http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=8fa5913d54f3b1e09948e6a0db34da887e05ff1f
which was reverted by
  http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=12c22d6ef299ccf0955e5756eb57d90d7577ac68
because of
  http://bugzilla.kernel.org/show_bug.cgi?id=10080
  http://marc.info/?l=linux-kernel&m=120620382316329&w=2
  http://marc.info/?l=linux-kernel&m=120653981716729&w=2

It definitely seems like the right thing to do but it apparently
breaks stuff on systems with oddball transparent bridges. 

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

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