On Thu, Jul 14, 2022 at 12:22:36PM +0200, Jan Beulich wrote:
> On 06.07.2022 17:32, Marek Marczykowski-Górecki wrote:
> > --- a/xen/drivers/passthrough/amd/iommu_acpi.c
> > +++ b/xen/drivers/passthrough/amd/iommu_acpi.c
> > @@ -1078,6 +1078,20 @@ static inline bool_t is_ivmd_block(u8 type)
> >              type == ACPI_IVRS_TYPE_MEMORY_IOMMU);
> >  }
> >  
> > +static int __init cf_check add_one_extra_ivmd(xen_pfn_t start, xen_ulong_t nr, u32 id, void *ctxt)
> > +{
> > +    struct acpi_ivrs_memory ivmd;
> > +
> > +    ivmd.start_address = start << PAGE_SHIFT;
> > +    ivmd.memory_length = nr << PAGE_SHIFT;
> 
> Aren't these at risk of truncation on 32-bit architectures? We have
> suitable wrappers for such conversions, avoiding such issues.

Well, this (and the vtd equivalent) is x86-only, so it's always 64-bit.
Anyway, what are those wrappers?

> > +    ivmd.header.flags = ACPI_IVMD_UNITY |
> > +                        ACPI_IVMD_READ | ACPI_IVMD_WRITE;
> > +    ivmd.header.length = sizeof(ivmd);
> > +    ivmd.header.device_id = id;
> > +    ivmd.header.type = ACPI_IVRS_TYPE_MEMORY_ONE;
> 
> Please make these the variable's initializer.
> 
> Jan

-- 
Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab