[PATCH x86-64] Calgary - Fix mis-handled PCI topology

[PATCH x86-64] Calgary - Fix mis-handled PCI topology

[Muli Ben-Yehuda]

Andrew, can you please push this Calgary bug-fix to 2.6.23 in your
next merge? it fixes a showstopper bug in the recently merged CalIOC2
support that hits machines with multiple levels of PCI-to-PCI bridges.

Thanks,
Muli
---
From: Murillo Fernandes Bernardes <bernarde@br.ibm.com>
Subject: x86-64: Calgary - Fix mis-handled PCI topology

Current code assumed that devices were directly connected to a Calgary
bridge, as it tried to get the iommu table directly from the parent bus 
controller.

When we have another bridge between the Calgary/CalIOC2 bridge and the 
device we should look upwards until we get to the top (Calgary/CalIOC2 
bridge), where the iommu table resides.

Signed-off-by: Murillo Fernandes Bernardes <mfb@br.ibm.com>
Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com>

--
diff -ruN linux-2.6.23-rc1.orig/arch/x86_64/kernel/pci-calgary.c linux-2.6.23-rc1/arch/x86_64/kernel/pci-calgary.c
--- linux-2.6.23-rc1.orig/arch/x86_64/kernel/pci-calgary.c	2007-08-01 15:37:04.000000000 -0300
+++ linux-2.6.23-rc1/arch/x86_64/kernel/pci-calgary.c	2007-08-01 15:48:55.000000000 -0300
@@ -367,16 +367,15 @@
 
 	pdev = to_pci_dev(dev);
 
-	/* is the device behind a bridge? */
-	if (unlikely(pdev->bus->parent))
-		pbus = pdev->bus->parent;
-	else
-		pbus = pdev->bus;
+	pbus = pdev->bus;
+
+	/* is the device behind a bridge? Look for the root bus */
+	while (pbus->parent)
+		pbus = pbus->parent;
 
 	tbl = pci_iommu(pbus);
 
-	BUG_ON(pdev->bus->parent &&
-	       (tbl->it_busno != pdev->bus->parent->number));
+	BUG_ON(tbl && (tbl->it_busno != pbus->number));
 
 	return tbl;
 }

Re: [PATCH x86-64] Calgary - Fix mis-handled PCI topology

[Andrew Morton]

On Thu, 2 Aug 2007 16:39:26 +0300 Muli Ben-Yehuda <muli@il.ibm.com> wrote:

> Andrew, can you please push this Calgary bug-fix to 2.6.23 in your
> next merge? it fixes a showstopper bug in the recently merged CalIOC2
> support that hits machines with multiple levels of PCI-to-PCI bridges.
> 
> Thanks,
> Muli
> ---
> From: Murillo Fernandes Bernardes <bernarde@br.ibm.com>
> Subject: x86-64: Calgary - Fix mis-handled PCI topology
> 
> Current code assumed that devices were directly connected to a Calgary
> bridge, as it tried to get the iommu table directly from the parent bus 
> controller.
> 
> When we have another bridge between the Calgary/CalIOC2 bridge and the 
> device we should look upwards until we get to the top (Calgary/CalIOC2 
> bridge), where the iommu table resides.
> 
> Signed-off-by: Murillo Fernandes Bernardes <mfb@br.ibm.com>
> Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com>
> 
> --
> diff -ruN linux-2.6.23-rc1.orig/arch/x86_64/kernel/pci-calgary.c linux-2.6.23-rc1/arch/x86_64/kernel/pci-calgary.c
> --- linux-2.6.23-rc1.orig/arch/x86_64/kernel/pci-calgary.c	2007-08-01 15:37:04.000000000 -0300
> +++ linux-2.6.23-rc1/arch/x86_64/kernel/pci-calgary.c	2007-08-01 15:48:55.000000000 -0300
> @@ -367,16 +367,15 @@
>  
>  	pdev = to_pci_dev(dev);
>  
> -	/* is the device behind a bridge? */
> -	if (unlikely(pdev->bus->parent))
> -		pbus = pdev->bus->parent;
> -	else
> -		pbus = pdev->bus;
> +	pbus = pdev->bus;
> +
> +	/* is the device behind a bridge? Look for the root bus */
> +	while (pbus->parent)
> +		pbus = pbus->parent;
>  
>  	tbl = pci_iommu(pbus);
>  
> -	BUG_ON(pdev->bus->parent &&
> -	       (tbl->it_busno != pdev->bus->parent->number));
> +	BUG_ON(tbl && (tbl->it_busno != pbus->number));
>  
>  	return tbl;
>  }
> 

hmm, sorry, I missed the this-is-urgent factor.

Andi, should I include this in my next batch to Linus?

btw, my list of "patches which should be in 2.6.23 but go via subsystem
maintainers" is presently

dma_free_coherent-needs-irqs-enabled-sigh.patch
usb-serial-fix-oti6858c-segfault-in-termios-handling.patch
sparsemem-ensure-we-initialise-the-node-mapping-for-sparsemem_static.patch
sparsemem-ensure-we-initialise-the-node-mapping-for-sparsemem_static-fix.patch
tpmdd-maintainers.patch
nfs-fix-put_nfs_open_context.patch
make-oprofile-call-shutdown-only-once-per-session.patch
perfctr_watchdog-do-not-bug_on-when-msr-is-unknown.patch
acpi-bay-send-envp-with-uevent-fix.patch
acpi-dock-send-key=value-pair-instead-of-plain-value.patch
acpi_ec_remove-fix-use-after-free.patch
acpi_battery_add-use-after-free.patch
kernel-auditscc-fix-an-off-by-one.patch
document-linux-memory-policy-v3.patch
futex_unlock_pi-hurts-my-brain-and-may-cause.patch
block-hide-the-contents-of-linux-bioh-if-config_block=n.patch
cdc-acm-fix-sysfs-attribute-registration-bug.patch
rtc-periodic-irq-fix.patch
at91-include-atmel_lcdch-in-at91sam92613_devicesc.patch
fix-local_irq_-macro-definition-thinkos-for-the-trace_irqflags_support-case.patch
calgary-fix-mis-handled-pci-topology.patch

several of which are x86-related.  Please consider...