[v1] intel-iommu: Accept 64-bit writes to FEADDR

[Qemu-devel] [PATCH] intel-iommu: Accept 64-bit writes to FEADDR

Posted by Jan Kiszka 7 years, 8 months ago

From: Jan Kiszka <jan.kiszka@siemens.com>

Xen is doing this [1] and currently triggers an abort.

[1] http://xenbits.xenproject.org/gitweb/?p=xen.git;a=blob;f=xen/drivers/passthrough/vtd/iommu.c;h=daaed0abbdd06b6ba3d948ea103aadf02651e83c;hb=refs/heads/master#l1108

Reported-by: Luis Lloret <luis_lloret@mentor.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
---
 hw/i386/intel_iommu.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c
index 2e841cde27..b61d0da270 100644
--- a/hw/i386/intel_iommu.c
+++ b/hw/i386/intel_iommu.c
@@ -2129,7 +2129,12 @@ static void vtd_mem_write(void *opaque, hwaddr addr,
 
     /* Fault Event Address Register, 32-bit */
     case DMAR_FEADDR_REG:
-        assert(size == 4);
+        /*
+         * While the register is 32-bit only, some guests (Xen...) write to it
+         * with 64-bit. Ignore the upper part, that's likely what the hardware
+         * does as well (plus the upper part is not used by our model anyway).
+         */
+        assert(size >= 4);
         vtd_set_long(s, addr, val);
         break;
 
-- 
2.13.6

Re: [Qemu-devel] [PATCH] intel-iommu: Accept 64-bit writes to FEADDR

Posted by Peter Xu 7 years, 8 months ago

On Sat, Feb 17, 2018 at 12:26:19PM +0100, Jan Kiszka wrote:
> From: Jan Kiszka <jan.kiszka@siemens.com>
> 
> Xen is doing this [1] and currently triggers an abort.
> 
> [1] http://xenbits.xenproject.org/gitweb/?p=xen.git;a=blob;f=xen/drivers/passthrough/vtd/iommu.c;h=daaed0abbdd06b6ba3d948ea103aadf02651e83c;hb=refs/heads/master#l1108
> 
> Reported-by: Luis Lloret <luis_lloret@mentor.com>
> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
> ---
>  hw/i386/intel_iommu.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c
> index 2e841cde27..b61d0da270 100644
> --- a/hw/i386/intel_iommu.c
> +++ b/hw/i386/intel_iommu.c
> @@ -2129,7 +2129,12 @@ static void vtd_mem_write(void *opaque, hwaddr addr,
>  
>      /* Fault Event Address Register, 32-bit */
>      case DMAR_FEADDR_REG:
> -        assert(size == 4);
> +        /*
> +         * While the register is 32-bit only, some guests (Xen...) write to it
> +         * with 64-bit. Ignore the upper part, that's likely what the hardware
> +         * does as well (plus the upper part is not used by our model anyway).
> +         */
> +        assert(size >= 4);
>          vtd_set_long(s, addr, val);
>          break;

(Sorry for the late response due to Chinese Spring Festival)

I agree with the problem there, but do we still better provide a
conditional vtd_set_quad()?  Since from the spec 10.4.13 the upper 32
bits may still be used when x2apic (Extended Interrupt Mode) is
enabled?

Thanks,

-- 
Peter Xu

[Qemu-devel] [PATCH v2] intel-iommu: Accept 64-bit writes to FEADDR

Posted by Jan Kiszka 7 years, 8 months ago

From: Jan Kiszka <jan.kiszka@siemens.com>

Xen is doing this [1] and currently triggers an abort.

[1] http://xenbits.xenproject.org/gitweb/?p=xen.git;a=blob;f=xen/drivers/passthrough/vtd/iommu.c;h=daaed0abbdd06b6ba3d948ea103aadf02651e83c;hb=refs/heads/master#l1108

Reported-by: Luis Lloret <luis_lloret@mentor.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
---
 hw/i386/intel_iommu.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c
index 2e841cde27..fb31de9416 100644
--- a/hw/i386/intel_iommu.c
+++ b/hw/i386/intel_iommu.c
@@ -2129,8 +2129,15 @@ static void vtd_mem_write(void *opaque, hwaddr addr,
 
     /* Fault Event Address Register, 32-bit */
     case DMAR_FEADDR_REG:
-        assert(size == 4);
-        vtd_set_long(s, addr, val);
+        if (size == 4) {
+            vtd_set_long(s, addr, val);
+        } else {
+            /*
+             * While the register is 32-bit only, some guests (Xen...) write to
+             * it with 64-bit.
+             */
+            vtd_set_quad(s, addr, val);
+        }
         break;
 
     /* Fault Event Upper Address Register, 32-bit */

Re: [Qemu-devel] [PATCH v2] intel-iommu: Accept 64-bit writes to FEADDR

Posted by Peter Xu 7 years, 7 months ago

On Sat, Feb 24, 2018 at 09:30:12AM +0100, Jan Kiszka wrote:
> From: Jan Kiszka <jan.kiszka@siemens.com>
> 
> Xen is doing this [1] and currently triggers an abort.
> 
> [1] http://xenbits.xenproject.org/gitweb/?p=xen.git;a=blob;f=xen/drivers/passthrough/vtd/iommu.c;h=daaed0abbdd06b6ba3d948ea103aadf02651e83c;hb=refs/heads/master#l1108
> 
> Reported-by: Luis Lloret <luis_lloret@mentor.com>
> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Thanks, Jan.

Reviewed-by: Peter Xu <peterx@redhat.com>

> ---
>  hw/i386/intel_iommu.c | 11 +++++++++--
>  1 file changed, 9 insertions(+), 2 deletions(-)
> 
> diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c
> index 2e841cde27..fb31de9416 100644
> --- a/hw/i386/intel_iommu.c
> +++ b/hw/i386/intel_iommu.c
> @@ -2129,8 +2129,15 @@ static void vtd_mem_write(void *opaque, hwaddr addr,
>  
>      /* Fault Event Address Register, 32-bit */
>      case DMAR_FEADDR_REG:
> -        assert(size == 4);
> -        vtd_set_long(s, addr, val);
> +        if (size == 4) {
> +            vtd_set_long(s, addr, val);
> +        } else {
> +            /*
> +             * While the register is 32-bit only, some guests (Xen...) write to
> +             * it with 64-bit.
> +             */
> +            vtd_set_quad(s, addr, val);
> +        }
>          break;
>  
>      /* Fault Event Upper Address Register, 32-bit */

-- 
Peter Xu