amd_iommu_attach_device() checks if device id is within the limits
of amd_iommu_last_bdf and instead checking if devid > amd_iommu_last_bdf,
it checks devid >= amd_iommu_last_bdf. As a result the last device attach
fails because amd_iommu_attach_device() returns an -EINVAL.
This bug is in linux-2.6.32 and an equivalent fix in linux-2.6.33 and has been
carried forward to later kernels and is in the upstream kernel. This equivalent
fix includes restructuring and consolidating device checks into a routine
check_device(). Instead of back-porting all of that work, spot-fixed the bug in
amd_iommu_attach_device() for linux-2.6.32.
Signed-off-by: Shuah Khan <[email protected]>
CC: <[email protected]> v2.6.32
---
arch/x86/kernel/amd_iommu.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 3a44b75..67de7d7 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -2288,7 +2288,7 @@ static int amd_iommu_attach_device(struct iommu_domain *dom,
devid = calc_devid(pdev->bus->number, pdev->devfn);
- if (devid >= amd_iommu_last_bdf ||
+ if (devid > amd_iommu_last_bdf ||
devid != amd_iommu_alias_table[devid])
return -EINVAL;
--
1.7.9.5
On Fri, 2012-10-12 at 10:29 -0600, Shuah Khan wrote:
> amd_iommu_attach_device() checks if device id is within the limits
> of amd_iommu_last_bdf and instead checking if devid > amd_iommu_last_bdf,
> it checks devid >= amd_iommu_last_bdf. As a result the last device attach
> fails because amd_iommu_attach_device() returns an -EINVAL.
>
> This bug is in linux-2.6.32 and an equivalent fix in linux-2.6.33 and has been
> carried forward to later kernels and is in the upstream kernel. This equivalent
> fix includes restructuring and consolidating device checks into a routine
> check_device(). Instead of back-porting all of that work, spot-fixed the bug in
> amd_iommu_attach_device() for linux-2.6.32.
>
> Signed-off-by: Shuah Khan <[email protected]>
> CC: <[email protected]> v2.6.32
> ---
> arch/x86/kernel/amd_iommu.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
> index 3a44b75..67de7d7 100644
> --- a/arch/x86/kernel/amd_iommu.c
> +++ b/arch/x86/kernel/amd_iommu.c
> @@ -2288,7 +2288,7 @@ static int amd_iommu_attach_device(struct iommu_domain *dom,
>
> devid = calc_devid(pdev->bus->number, pdev->devfn);
>
> - if (devid >= amd_iommu_last_bdf ||
> + if (devid > amd_iommu_last_bdf ||
> devid != amd_iommu_alias_table[devid])
> return -EINVAL;
>
Looks correct to me, this case was clearly inconsistent with other users
of amd_iommu_last_bdf.
Reviewed-by: Alex Williamson <[email protected]>
Shuah Khan wrote:
> This bug is in linux-2.6.32 and an equivalent fix in linux-2.6.33 and has been
> carried forward to later kernels and is in the upstream kernel. This equivalent
> fix includes restructuring and consolidating device checks into a routine
> check_device(). Instead of back-porting all of that work, spot-fixed the bug in
> amd_iommu_attach_device() for linux-2.6.32.
To save Willy time: am I correct in guessing the upstream commit you
are referring to is 98fc5a693bbdda498a556654c70d1e31a186c988
(x86/amd-iommu: Use get_device_id and check_device where appropriate,
2009-11-24)?
Thanks,
Jonathan
On Fri, 2012-10-12 at 11:38 -0700, Jonathan Nieder wrote:
> Shuah Khan wrote:
>
> > This bug is in linux-2.6.32 and an equivalent fix in linux-2.6.33 and has been
> > carried forward to later kernels and is in the upstream kernel. This equivalent
> > fix includes restructuring and consolidating device checks into a routine
> > check_device(). Instead of back-porting all of that work, spot-fixed the bug in
> > amd_iommu_attach_device() for linux-2.6.32.
>
> To save Willy time: am I correct in guessing the upstream commit you
> are referring to is 98fc5a693bbdda498a556654c70d1e31a186c988
> (x86/amd-iommu: Use get_device_id and check_device where appropriate,
> 2009-11-24)?
Yes that is one. I looked that up again and it has the change to
amd_iommu_attach_device() to use check_device() and removes the check
- if (devid >= amd_iommu_last_bdf ||
- devid != amd_iommu_alias_table[devid])
- return -EINVAL;
+ devid = get_device_id(dev);
Ref: http://www.spinics.net/lists/linux-tip-commits/msg06129.html
Thanks,
-- Shuah
Shuah Khan wrote:
> On Fri, 2012-10-12 at 11:38 -0700, Jonathan Nieder wrote:
>> To save Willy time: am I correct in guessing the upstream commit you
>> are referring to is 98fc5a693bbdda498a556654c70d1e31a186c988
>> (x86/amd-iommu: Use get_device_id and check_device where appropriate,
>> 2009-11-24)?
>
> Yes that is one.
Great, thanks for the quick confirmation.
Ciao,
Jonathan
On Fri, 2012-10-12 at 12:35 -0700, Jonathan Nieder wrote:
> Shuah Khan wrote:
> > On Fri, 2012-10-12 at 11:38 -0700, Jonathan Nieder wrote:
>
> >> To save Willy time: am I correct in guessing the upstream commit you
> >> are referring to is 98fc5a693bbdda498a556654c70d1e31a186c988
> >> (x86/amd-iommu: Use get_device_id and check_device where appropriate,
> >> 2009-11-24)?
> >
> > Yes that is one.
>
> Great, thanks for the quick confirmation.
Is this patch in the queue to be pulled into 2.6.32 tree?
Thanks,
-- Shuah