Re: [Qemu-devel] [PATCH for 2.8 06/11] intel_iommu: support device iotlb descriptor

[Jason Wang]

On 2016年08月30日 21:16, Peter Xu wrote:
> On Tue, Aug 30, 2016 at 11:06:54AM +0800, Jason Wang wrote:
> > This patch enables device IOTLB support for intel iommu. The major
> > work is to implement QI device IOTLB descriptor processing and notify
> > the device through iommu notifier.
> IMHO it'll be nicer if we can mention the new bit in the commit
> message, like:
>
>      One new property "device-iotlb" is introduced to enable device
>      IOTLB capability for VT-d IOMMU device. By default, it is turned
>      off. To enable it:
>
>        -device intel-iommu,device-iotlb=on
>
> [...]

Yes, this looks better.

> > +static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s,
> > +                                          VTDInvDesc *inv_desc)
> > +{
> [...]
>
> > +    entry.target_as = &vtd_dev_as->as;
> > +    entry.addr_mask = sz - 1;
> > +    entry.iova = addr;
> Do we need to setup translated_addr and perm fields as well?
>
> [...]

Good question, tough vhost does not care about those two fileds now (we 
only care about invalidation). But I think it's better to add them, it 
may be used in the future (e.g CM which may also depends on notifier).

> > diff --git a/hw/i386/x86-iommu.c b/hw/i386/x86-iommu.c
> > index ce26b2a..a70a363 100644
> > --- a/hw/i386/x86-iommu.c
> > +++ b/hw/i386/x86-iommu.c
> > @@ -100,6 +100,18 @@ static void x86_iommu_intremap_prop_set(Object *o, bool 
> > value, Error **errp)
> >       s->intr_supported = value;
> >   }
> >   
> > +static bool x86_iommu_device_iotlb_prop_get(Object *o, Error **errp)
> > +{
> > +    X86IOMMUState *s = X86_IOMMU_DEVICE(o);
> > +    return s->dt_supported;
> > +}
> > +
> > +static void x86_iommu_device_iotlb_prop_set(Object *o, bool value, Error 
> > **errp)
> > +{
> > +    X86IOMMUState *s = X86_IOMMU_DEVICE(o);
> > +    s->dt_supported = value;
> > +}
> > +
> >   static void x86_iommu_instance_init(Object *o)
> >   {
> >       X86IOMMUState *s = X86_IOMMU_DEVICE(o);
> > @@ -108,6 +120,11 @@ static void x86_iommu_instance_init(Object *o)
> >       s->intr_supported = false;
> >       object_property_add_bool(o, "intremap", x86_iommu_intremap_prop_get,
> >                                x86_iommu_intremap_prop_set, NULL);
> > +    s->dt_supported = false;
> > +    object_property_add_bool(o, "device_iotlb",
> > +                             x86_iommu_device_iotlb_prop_get,
> > +                             x86_iommu_device_iotlb_prop_set,
> > +                             NULL);
> Nit 1: use "device-iotlb" instead of "device_iotlb"?

Yes.

> Nit 2: use Property bit (like vtd_properties)?

Not sure, I thought this may be reused by AMD IOMMU but maybe I was wrong.

> Thanks,
>
> -- peterx