RE: [PATCH v4 09/12] iommu/vt-d: Add iotlb flush for nested domain

From: Liu, Yi L
Date: Wed Aug 09 2023 - 23:28:23 EST


> From: Nicolin Chen <nicolinc@xxxxxxxxxx>
> Sent: Thursday, August 10, 2023 4:17 AM
>
> On Wed, Aug 09, 2023 at 04:19:01PM -0300, Jason Gunthorpe wrote:
> > On Wed, Aug 09, 2023 at 12:12:25PM -0700, Nicolin Chen wrote:
> > > On Wed, Aug 09, 2023 at 01:24:56PM -0300, Jason Gunthorpe wrote:
> > > > On Wed, Aug 09, 2023 at 09:30:12AM +0000, Liu, Yi L wrote:
> > > >
> > > > > Yeah, adding new structures to ucmd_buffer may increase the size as
> > > > > well if the new one is larger. While for an array, if there is new entry,
> > > > > it is for sure to increase the size. I remember there is one tricky thing
> > > > > when handling the selftest type. E.g. it is defined as 0xbadbeef, if using
> > > > > it to index array, it would expire. So we have some special handling on
> > > > > it. If defining the things in iommu_ops, it is simpler. Selftest may be
> > > > > not so critical to determining the direction though.
> > > >
> > > > Maybe we are trying too hard to make it "easy" on the driver.
> > > >
> > > > Can't we just have the driver invoke some:
> > > >
> > > > driver_iommufd_invalidate_op(??? *opaque)
> > > > {
> > > > struct driver_base_struct args;
> > > >
> > > > rc = iommufd_get_args(opaque, &args, sizeof(args),
> > > > offsetof(args, last));
> > >
> > > OK. So, IIUIC, the opaque should be:
> > >
> > > struct iommu_user_data {
> > > void __user *data_uptr;
> > > size_t data_len;
> > > }user_data;
> > >
> > > And core does basic sanity of data_uptr != NULL and data_len !=0
> > > before passing this to driver, and then do a full sanity during
> > > the iommufd_get_args (or iommufd_get_user_data?) call.
> >
> > Don't even need to check datA_uptr and data_len, the helper should
> > check the size and null is caught by copy from user
>
> I see. I was worried about the alloc path since its data input is
> optional upon IOMMU_DOMAIN_UNMANAGED. But this helper should work
> for that also.
>
> In that case, we might not even need to define the union with all
> structures, in iommu.h.
>
> > > > Similarly for managing the array of invalidation commands.
> > >
> > > You mean an embedded uptr inside a driver user data struct right?
> > > Sure, that should go through the new helper too.
> >
> > If we are committed that all drivers have to process an array then put
> > the array in the top level struct and pass it in the same user_data
> > struct and use another helper to allow the driver to iterate through
> > it.
>
> I see. Both VTD and SMMU pass uptr to the arrays of invalidation
> commands/requests. The only difference is that SMMU's array is a
> ring buffer other than a plain one indexing from the beginning.
> But the helper could take two index inputs, which should work for
> VTD case too. If another IOMMU driver only supports one request,
> rather than a array of requests, we can treat that as a single-
> entry array.
>
> Then, the driver-specific data structure will be the array entry
> level only.
>
> @Yi,
> This seems to be a bigger rework than the top level struct. Along
> with Jason's request for fail_nth below, we'd need to bisect the
> workload between us, or can just continue each other's daily work.
> Let me know which one you prefer.
> https://lore.kernel.org/linux-iommu/ZNPCtPTcHvITt6fk@xxxxxxxxxx/

Let me address the fail_nth request first. You may rework the
iommufd_get_user_data(). If I can finish the fail_nth soon,
then may help to lift the array to the top level. If not, you
may make it as well. 😊 I guess I need some study on nth
as well.

Regards,
Yi Liu