Subject: [PATCH] x86: amd_iommu: fix nasty bug that caused ILLEGAL_DEVICE_TABLE_ENTRY errors

We are on 64-bit so better use u64 instead of u32 to deal with
addresses:

static void __init iommu_set_device_table(struct amd_iommu *iommu)
{
u64 entry;
...
entry = virt_to_phys(amd_iommu_dev_table);
...

(I am wondering why gcc 4.2.x did not warn about the assignment
between u32 and unsigned long.)

Cc: [email protected]
Cc: Joerg Roedel <[email protected]>
Signed-off-by: Andreas Herrmann <[email protected]>

---
arch/x86/kernel/amd_iommu_init.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)

Patch is against today's tip/x86/amd-iommu
Joerg is on vacation. That is why I post this patch
directly to you.

Regards,

Andreas

diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index a69cc0f..bccd0ef 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -210,7 +210,7 @@ static void __init iommu_set_exclusion_range(struct amd_iommu *iommu)
/* Programs the physical address of the device table into the IOMMU hardware */
static void __init iommu_set_device_table(struct amd_iommu *iommu)
{
- u32 entry;
+ u64 entry;

BUG_ON(iommu->mmio_base == NULL);

--
1.6.0.2



2008-10-16 15:00:11

by Joerg Roedel

[permalink] [raw]
Subject: Re: [PATCH] x86: amd_iommu: fix nasty bug that caused ILLEGAL_DEVICE_TABLE_ENTRY errors

Argh. Very stupid bug. This explains why it worked on my system with 1G
RAM and failed on some others with more. Thanks for the fix. I also CC
[email protected].

On Thu, Oct 16, 2008 at 04:27:36PM +0200, Andreas Herrmann wrote:
> We are on 64-bit so better use u64 instead of u32 to deal with
> addresses:
>
> static void __init iommu_set_device_table(struct amd_iommu *iommu)
> {
> u64 entry;
> ...
> entry = virt_to_phys(amd_iommu_dev_table);
> ...
>
> (I am wondering why gcc 4.2.x did not warn about the assignment
> between u32 and unsigned long.)
>
> Cc: [email protected]
> Cc: Joerg Roedel <[email protected]>
> Signed-off-by: Andreas Herrmann <[email protected]>

Signed-off-by: Joerg Roedel <[email protected]>

>
> ---
> arch/x86/kernel/amd_iommu_init.c | 2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
>
> Patch is against today's tip/x86/amd-iommu
> Joerg is on vacation. That is why I post this patch
> directly to you.
>
> Regards,
>
> Andreas
>
> diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
> index a69cc0f..bccd0ef 100644
> --- a/arch/x86/kernel/amd_iommu_init.c
> +++ b/arch/x86/kernel/amd_iommu_init.c
> @@ -210,7 +210,7 @@ static void __init iommu_set_exclusion_range(struct amd_iommu *iommu)
> /* Programs the physical address of the device table into the IOMMU hardware */
> static void __init iommu_set_device_table(struct amd_iommu *iommu)
> {
> - u32 entry;
> + u64 entry;
>
> BUG_ON(iommu->mmio_base == NULL);
>
> --
> 1.6.0.2
>
>
>
> _______________________________________________
> iommu mailing list
> [email protected]
> https://lists.linux-foundation.org/mailman/listinfo/iommu

2008-10-20 16:06:25

by Ingo Molnar

[permalink] [raw]
Subject: Re: [PATCH] x86: amd_iommu: fix nasty bug that caused ILLEGAL_DEVICE_TABLE_ENTRY errors


* Joerg Roedel <[email protected]> wrote:

> Argh. Very stupid bug. This explains why it worked on my system with 1G
> RAM and failed on some others with more. Thanks for the fix. I also CC
> [email protected].
>
> On Thu, Oct 16, 2008 at 04:27:36PM +0200, Andreas Herrmann wrote:
> > We are on 64-bit so better use u64 instead of u32 to deal with
> > addresses:
> >
> > static void __init iommu_set_device_table(struct amd_iommu *iommu)
> > {
> > u64 entry;
> > ...
> > entry = virt_to_phys(amd_iommu_dev_table);
> > ...
> >
> > (I am wondering why gcc 4.2.x did not warn about the assignment
> > between u32 and unsigned long.)
> >
> > Cc: [email protected]
> > Cc: Joerg Roedel <[email protected]>
> > Signed-off-by: Andreas Herrmann <[email protected]>
>
> Signed-off-by: Joerg Roedel <[email protected]>

applied to tip/x86/urgent, thanks guys!

Ingo