lkml.org 
[lkml]   [2021]   [Oct]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
SubjectRe: [PATCH Part2 v5 39/45] KVM: SVM: Introduce ops for the post gfn map and unmap
From
Date

On 10/12/21 5:23 PM, Sean Christopherson wrote:
> On Fri, Aug 20, 2021, Brijesh Singh wrote:
>> When SEV-SNP is enabled in the guest VM, the guest memory pages can
>> either be a private or shared. A write from the hypervisor goes through
>> the RMP checks. If hardware sees that hypervisor is attempting to write
>> to a guest private page, then it triggers an RMP violation #PF.
>>
>> To avoid the RMP violation, add post_{map,unmap}_gfn() ops that can be
>> used to verify that its safe to map a given guest page. Use the SRCU to
>> protect against the page state change for existing mapped pages.
> SRCU isn't protecting anything. The synchronize_srcu_expedited() in the PSC code
> forces it to wait for existing maps to go away, but it doesn't prevent new maps
> from being created while the actual RMP updates are in-flight. Most telling is
> that the RMP updates happen _after_ the synchronize_srcu_expedited() call.
>
> This also doesn't handle kvm_{read,write}_guest_cached().

Hmm, I thought the kvm_{read_write}_guest_cached() uses the
copy_{to,from}_user(). Writing to the private will cause a #PF and will
fail the copy_to_user(). Am I missing something?


>
> I can't help but note that the private memslots idea[*] would handle this gracefully,
> e.g. the memslot lookup would fail, and any change in private memslots would
> invalidate the cache due to a generation mismatch.
>
> KSM is another mess that would Just Work.
>
> I'm not saying that SNP should be blocked on support for unmapping guest private
> memory, but I do think we should strongly consider focusing on that effort rather
> than trying to fix things piecemeal throughout KVM. I don't think it's too absurd
> to say that it might actually be faster overall. And I 100% think that having a
> cohesive design and uABI for SNP and TDX would be hugely beneficial to KVM.
>
> [*] https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Flkml.kernel.org%2Fr%2F20210824005248.200037-1-seanjc%40google.com&data=04%7C01%7Cbrijesh.singh%40amd.com%7Cd1717d511a1f473cedc408d98ddfb027%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637696814148744591%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=3LF77%2BcqmpUdiP6YAk7LpImisBzjRGUzdI3raqjJxl0%3D&reserved=0
>
>> +int sev_post_map_gfn(struct kvm *kvm, gfn_t gfn, kvm_pfn_t pfn, int *token)
>> +{
>> + struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
>> + int level;
>> +
>> + if (!sev_snp_guest(kvm))
>> + return 0;
>> +
>> + *token = srcu_read_lock(&sev->psc_srcu);
>> +
>> + /* If pfn is not added as private then fail */
> This comment and the pr_err() are backwards, and confused the heck out of me.
> snp_lookup_rmpentry() returns '1' if the pfn is assigned, a.k.a. private. That
> means this code throws an error if the page is private, i.e. requires the page
> to be shared. Which makes sense given the use cases, it's just incredibly
> confusing.
Actually I followed your recommendation from the previous feedback that
snp_lookup_rmpentry() should return 1 for the assigned page, 0 for the
shared and -negative for invalid. I can clarify it here  again.
>
>> + if (snp_lookup_rmpentry(pfn, &level) == 1) {
> Any reason not to provide e.g. rmp_is_shared() and rmp_is_private() so that
> callers don't have to care as much about the return values? The -errno/0/1
> semantics are all but guarantee to bite us in the rear at some point.

If we look at the previous series, I had a macro rmp_is_assigned() for
exactly the same purpose but the feedback was to drop those macros and
return the state indirectly through the snp_lookup_rmpentry(). I can
certainly add a helper rmp_is_{shared,private}() if it makes code more
readable.


> Actually, peeking at other patches, I think it already has. This usage in
> __unregister_enc_region_locked() is wrong:
>
> /*
> * The guest memory pages are assigned in the RMP table. Unassign it
> * before releasing the memory.
> */
> if (sev_snp_guest(kvm)) {
> for (i = 0; i < region->npages; i++) {
> pfn = page_to_pfn(region->pages[i]);
>
> if (!snp_lookup_rmpentry(pfn, &level)) <-- attempts make_shared() on non-existent entry
> continue;
>
> cond_resched();
>
> if (level > PG_LEVEL_4K)
> pfn &= ~(KVM_PAGES_PER_HPAGE(PG_LEVEL_2M) - 1);
>
> host_rmp_make_shared(pfn, level, true);
> }
> }
>
>
>> + srcu_read_unlock(&sev->psc_srcu, *token);
>> + pr_err_ratelimited("failed to map private gfn 0x%llx pfn 0x%llx\n", gfn, pfn);
>> + return -EBUSY;
>> + }
>> +
>> + return 0;
>> +}
>> static struct kvm_x86_init_ops svm_init_ops __initdata = {
>> diff --git a/arch/x86/kvm/svm/svm.h b/arch/x86/kvm/svm/svm.h
>> index d10f7166b39d..ff91184f9b4a 100644
>> --- a/arch/x86/kvm/svm/svm.h
>> +++ b/arch/x86/kvm/svm/svm.h
>> @@ -76,16 +76,22 @@ struct kvm_sev_info {
>> bool active; /* SEV enabled guest */
>> bool es_active; /* SEV-ES enabled guest */
>> bool snp_active; /* SEV-SNP enabled guest */
>> +
>> unsigned int asid; /* ASID used for this guest */
>> unsigned int handle; /* SEV firmware handle */
>> int fd; /* SEV device fd */
>> +
>> unsigned long pages_locked; /* Number of pages locked */
>> struct list_head regions_list; /* List of registered regions */
>> +
>> u64 ap_jump_table; /* SEV-ES AP Jump Table address */
>> +
>> struct kvm *enc_context_owner; /* Owner of copied encryption context */
>> struct misc_cg *misc_cg; /* For misc cgroup accounting */
>> +
> Unrelated whitespace changes.
>
>> u64 snp_init_flags;
>> void *snp_context; /* SNP guest context page */
>> + struct srcu_struct psc_srcu;
>> };

\
 
 \ /
  Last update: 2021-10-13 20:11    [W:0.260 / U:0.092 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site