Re: [PATCH v13 10/26] KVM: SEV: Add KVM_SEV_SNP_LAUNCH_UPDATE command

[Michael Roth <michael.roth@amd.com>]

On Fri, Apr 19, 2024 at 06:12:11PM +0200, Paolo Bonzini wrote:
> On Fri, Apr 19, 2024 at 1:56 PM Paolo Bonzini <pbonzini@redhat.com> wrote:
> > > +       ret = kvm_gmem_populate(kvm, params.gfn_start, u64_to_user_ptr(params.uaddr),
> > > +                               npages, sev_gmem_post_populate, &sev_populate_args);
> > > +       if (ret < 0) {
> > > +               argp->error = sev_populate_args.fw_error;
> > > +               pr_debug("%s: kvm_gmem_populate failed, ret %d (fw_error %d)\n",
> > > +                        __func__, ret, argp->error);
> > > +       } else if (ret < npages) {
> > > +               params.len = ret * PAGE_SIZE;
> > > +               ret = -EINTR;
> >
> > This probably should 1) update also gfn_start and uaddr 2) return 0
> > for consistency with the planned KVM_PRE_FAULT_MEMORY ioctl (aka
> > KVM_MAP_MEMORY).
> 
> To be more precise, params.len should be set to the number of bytes *left*, i.e.
> 
>    params.len -= ret * PAGE_SIZE;
>    params.gfn_start += ret * PAGE_SIZE;
>    if (params.type != KVM_SEV_SNP_PAGE_TYPE_ZERO)
>        params.uaddr += ret * PAGE_SIZE;
> 
> Also this patch needs some other changes:
> 
> 1) snp_launch_update() should have something like this:
> 
>    src = params.type == KVM_SEV_SNP_PAGE_TYPE_ZERO ? NULL :
> u64_to_user_ptr(params.uaddr),;
> 
> so that then...
> 
> > +               vaddr = kmap_local_pfn(pfn + i);
> > +               ret = copy_from_user(vaddr, src + i * PAGE_SIZE, PAGE_SIZE);
> > +               if (ret) {
> > +                       pr_debug("Failed to copy source page into GFN 0x%llx\n", gfn);
> > +                       goto out_unmap;
> > +               }
> 
> ... the copy can be done only if src is non-NULL
> 
> 2) the struct should have some more fields
> 
> > +        struct kvm_sev_snp_launch_update {
> > +                __u64 gfn_start;        /* Guest page number to load/encrypt data into. */
> > +                __u64 uaddr;            /* Userspace address of data to be loaded/encrypted. */
> > +                __u32 len;              /* 4k-aligned length in bytes to copy into guest memory.*/
> > +                __u8 type;              /* The type of the guest pages being initialized. */
> 
> __u8 pad0;
> __u16 flags;   // must be zero
> __u64 pad1[5];
> 
> with accompanying flags check in snp_launch_update().

Have these all addressed in v14, but I ended up making 'len' a __u64, so the
final struct looks like this:

  struct kvm_sev_snp_launch_update {
          __u64 gfn_start;
          __u64 uaddr;
          __u64 len;
          __u8 type;
          __u8 pad0;
          __u16 flags;
          __u32 pad1;
          __u64 pad2[4];
  };

> 
> If you think IMI can be implemented already (with a bit in flags) go
> ahead and do it.

Migration will also need related flags in LAUNCH_START, and depending on how
we implement things, possibly in LAUNCH_FINISH. So for now I've left IMI
out, but added similar 'flags' and padding to those structs as well so we have
some flexibility with how we end up handling that.

-Mike

> 
> Paolo
>