Re: [PATCH v12 4/8] arm64: kvm: Introduce MTE VM feature

[Marc Zyngier]

On Wed, 19 May 2021 11:48:21 +0100,
Steven Price <steven.price@arm.com> wrote:
> 
> On 17/05/2021 17:45, Marc Zyngier wrote:
> > On Mon, 17 May 2021 13:32:35 +0100,
> > Steven Price <steven.price@arm.com> wrote:
> >>
> >> Add a new VM feature 'KVM_ARM_CAP_MTE' which enables memory tagging
> >> for a VM. This will expose the feature to the guest and automatically
> >> tag memory pages touched by the VM as PG_mte_tagged (and clear the tag
> >> storage) to ensure that the guest cannot see stale tags, and so that
> >> the tags are correctly saved/restored across swap.
> >>
> >> Actually exposing the new capability to user space happens in a later
> >> patch.
> > 
> > uber nit in $SUBJECT: "KVM: arm64:" is the preferred prefix (just like
> > patches 7 and 8).
> 
> Good spot - I obviously got carried away with the "arm64:" prefix ;)
> 
> >>
> >> Signed-off-by: Steven Price <steven.price@arm.com>
> >> ---
> >>  arch/arm64/include/asm/kvm_emulate.h |  3 +++
> >>  arch/arm64/include/asm/kvm_host.h    |  3 +++
> >>  arch/arm64/kvm/hyp/exception.c       |  3 ++-
> >>  arch/arm64/kvm/mmu.c                 | 37 +++++++++++++++++++++++++++-
> >>  arch/arm64/kvm/sys_regs.c            |  3 +++
> >>  include/uapi/linux/kvm.h             |  1 +
> >>  6 files changed, 48 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/arch/arm64/include/asm/kvm_emulate.h 
> >> b/arch/arm64/include/asm/kvm_emulate.h
> >> index f612c090f2e4..6bf776c2399c 100644
> >> --- a/arch/arm64/include/asm/kvm_emulate.h
> >> +++ b/arch/arm64/include/asm/kvm_emulate.h
> >> @@ -84,6 +84,9 @@ static inline void vcpu_reset_hcr(struct kvm_vcpu *vcpu)
> >>    if (cpus_have_const_cap(ARM64_MISMATCHED_CACHE_TYPE) ||
> >>        vcpu_el1_is_32bit(vcpu))
> >>            vcpu->arch.hcr_el2 |= HCR_TID2;
> >> +
> >> +  if (kvm_has_mte(vcpu->kvm))
> >> +          vcpu->arch.hcr_el2 |= HCR_ATA;
> >>  }
> >>  
> >>  static inline unsigned long *vcpu_hcr(struct kvm_vcpu *vcpu)
> >> diff --git a/arch/arm64/include/asm/kvm_host.h 
> >> b/arch/arm64/include/asm/kvm_host.h
> >> index 7cd7d5c8c4bc..afaa5333f0e4 100644
> >> --- a/arch/arm64/include/asm/kvm_host.h
> >> +++ b/arch/arm64/include/asm/kvm_host.h
> >> @@ -132,6 +132,8 @@ struct kvm_arch {
> >>  
> >>    u8 pfr0_csv2;
> >>    u8 pfr0_csv3;
> >> +  /* Memory Tagging Extension enabled for the guest */
> >> +  bool mte_enabled;
> >>  };
> >>  
> >>  struct kvm_vcpu_fault_info {
> >> @@ -769,6 +771,7 @@ bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu);
> >>  #define kvm_arm_vcpu_sve_finalized(vcpu) \
> >>    ((vcpu)->arch.flags & KVM_ARM64_VCPU_SVE_FINALIZED)
> >>  
> >> +#define kvm_has_mte(kvm) (system_supports_mte() && 
> >> (kvm)->arch.mte_enabled)
> >>  #define kvm_vcpu_has_pmu(vcpu)                                    \
> >>    (test_bit(KVM_ARM_VCPU_PMU_V3, (vcpu)->arch.features))
> >>  
> >> diff --git a/arch/arm64/kvm/hyp/exception.c 
> >> b/arch/arm64/kvm/hyp/exception.c
> >> index 73629094f903..56426565600c 100644
> >> --- a/arch/arm64/kvm/hyp/exception.c
> >> +++ b/arch/arm64/kvm/hyp/exception.c
> >> @@ -112,7 +112,8 @@ static void enter_exception64(struct kvm_vcpu *vcpu, 
> >> unsigned long target_mode,
> >>    new |= (old & PSR_C_BIT);
> >>    new |= (old & PSR_V_BIT);
> >>  
> >> -  // TODO: TCO (if/when ARMv8.5-MemTag is exposed to guests)
> >> +  if (kvm_has_mte(vcpu->kvm))
> >> +          new |= PSR_TCO_BIT;
> >>  
> >>    new |= (old & PSR_DIT_BIT);
> >>  
> >> diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
> >> index c5d1f3c87dbd..8660f6a03f51 100644
> >> --- a/arch/arm64/kvm/mmu.c
> >> +++ b/arch/arm64/kvm/mmu.c
> >> @@ -822,6 +822,31 @@ transparent_hugepage_adjust(struct kvm_memory_slot 
> >> *memslot,
> >>    return PAGE_SIZE;
> >>  }
> >>  
> >> +static int sanitise_mte_tags(struct kvm *kvm, unsigned long size,
> >> +                       kvm_pfn_t pfn)
> > 
> > Nit: please order the parameters as address, then size.
> 
> Sure
> 
> >> +{
> >> +  if (kvm_has_mte(kvm)) {
> >> +          /*
> >> +           * The page will be mapped in stage 2 as Normal Cacheable, so
> >> +           * the VM will be able to see the page's tags and therefore
> >> +           * they must be initialised first. If PG_mte_tagged is set,
> >> +           * tags have already been initialised.
> >> +           */
> >> +          unsigned long i, nr_pages = size >> PAGE_SHIFT;
> >> +          struct page *page = pfn_to_online_page(pfn);
> >> +
> >> +          if (!page)
> >> +                  return -EFAULT;
> > 
> > Under which circumstances can this happen? We already have done a GUP
> > on the page, so I really can't see how the page can vanish from under
> > our feet.
> 
> It's less about the page vanishing and more that pfn_to_online_page()
> will reject some pages. Specifically in this case we want to reject any
> sort of device memory (e.g. graphics card memory or other memory on the
> end of a bus like PCIe) as it is unlikely to support MTE.

OK. We really never should see this error as we check for device
mappings right before calling this, but I guess it doesn't hurt.

> 
> >> +
> >> +          for (i = 0; i < nr_pages; i++, page++) {
> >> +                  if (!test_and_set_bit(PG_mte_tagged, &page->flags))
> >> +                          mte_clear_page_tags(page_address(page));
> > 
> > You seem to be doing this irrespective of the VMA being created with
> > PROT_MTE. This is fine form a guest perspective (all its memory should
> > be MTE capable). However, I can't see any guarantee that the VMM will
> > actually allocate memslots with PROT_MTE.
> > 
> > Aren't we missing some sanity checks at memslot registration time?
> 
> I've taken the decision not to require that the VMM allocates with
> PROT_MTE, there are two main reasons for this:
> 
>  1. The VMM generally doesn't actually want a PROT_MTE mapping as the
> tags from the guest are almost certainly wrong for most usages (e.g.
> device emulation). So a PROT_MTE mapping actively gets in the way of the
> VMM using MTE for it's own purposes. However this then leads to the
> requirement for the new ioctl in patch 7.
> 
>  2. Because the VMM can change the pages in a memslot at any time and
> KVM relies on mmu notifiers to spot the change it's hard and ugly to
> enforce that the memslot VMAs keep having the PROT_MTE flag. When I
> tried this it meant we've discover that a page doesn't have the MTE flag
> at fault time and have no other option that to kill the VM at that time.
> 
> So the model is that non-PROT_MTE memory can be supplied to the memslots
> and KVM will automatically upgrade it to PG_mte_tagged if you supply it
> to a VM with MTE enabled. This makes the VMM implementation easier for
> most cases, and the new ioctl helps for migration. I think the kernel
> code is tidier too.

OK, I see your point. I guess we rely on the implicit requirement that
all the available memory is MTE-capable, although I'm willing to bet
that someone will eventually break this requirement.

> Of course even better would be a stage 2 flag to control MTE
> availability on a page-by-page basis, but that doesn't exist in the
> architecture at the moment.

Nah, that'd be way too good. Let's not do that.

> 
> >> +          }
> >> +  }
> >> +
> >> +  return 0;
> >> +}
> >> +
> >>  static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
> >>                      struct kvm_memory_slot *memslot, unsigned long hva,
> >>                      unsigned long fault_status)
> >> @@ -971,8 +996,13 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, 
> >> phys_addr_t fault_ipa,
> >>    if (writable)
> >>            prot |= KVM_PGTABLE_PROT_W;
> >>  
> >> -  if (fault_status != FSC_PERM && !device)
> >> +  if (fault_status != FSC_PERM && !device) {
> >> +          ret = sanitise_mte_tags(kvm, vma_pagesize, pfn);
> >> +          if (ret)
> >> +                  goto out_unlock;
> >> +
> >>            clean_dcache_guest_page(pfn, vma_pagesize);
> >> +  }
> >>  
> >>    if (exec_fault) {
> >>            prot |= KVM_PGTABLE_PROT_X;
> >> @@ -1168,12 +1198,17 @@ bool kvm_unmap_gfn_range(struct kvm *kvm, struct 
> >> kvm_gfn_range *range)
> >>  bool kvm_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
> >>  {
> >>    kvm_pfn_t pfn = pte_pfn(range->pte);
> >> +  int ret;
> >>  
> >>    if (!kvm->arch.mmu.pgt)
> >>            return 0;
> >>  
> >>    WARN_ON(range->end - range->start != 1);
> >>  
> >> +  ret = sanitise_mte_tags(kvm, PAGE_SIZE, pfn);
> >> +  if (ret)
> >> +          return ret;
> > 
> > Notice the change in return type?
> 
> I do now - I was tricked by the use of '0' as false. Looks like false
> ('0') is actually the correct return here to avoid an unnecessary
> kvm_flush_remote_tlbs().

Yup. BTW, the return values have been fixed to proper boolean types in
the latest set of fixes.

> 
> >> +
> >>    /*
> >>     * We've moved a page around, probably through CoW, so let's treat it
> >>     * just like a translation fault and clean the cache to the PoC.
> >> diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c
> >> index 76ea2800c33e..24a844cb79ca 100644
> >> --- a/arch/arm64/kvm/sys_regs.c
> >> +++ b/arch/arm64/kvm/sys_regs.c
> >> @@ -1047,6 +1047,9 @@ static u64 read_id_reg(const struct kvm_vcpu *vcpu,
> >>            break;
> >>    case SYS_ID_AA64PFR1_EL1:
> >>            val &= ~FEATURE(ID_AA64PFR1_MTE);
> >> +          if (kvm_has_mte(vcpu->kvm))
> >> +                  val |= FIELD_PREP(FEATURE(ID_AA64PFR1_MTE),
> >> +                                    ID_AA64PFR1_MTE);
> > 
> > Shouldn't this be consistent with what the HW is capable of
> > (i.e. FEAT_MTE3 if available), and extracted from the sanitised view
> > of the feature set?
> 
> Yes - however at the moment our sanitised view is either FEAT_MTE2 or
> nothing:
> 
>       {
>               .desc = "Memory Tagging Extension",
>               .capability = ARM64_MTE,
>               .type = ARM64_CPUCAP_STRICT_BOOT_CPU_FEATURE,
>               .matches = has_cpuid_feature,
>               .sys_reg = SYS_ID_AA64PFR1_EL1,
>               .field_pos = ID_AA64PFR1_MTE_SHIFT,
>               .min_field_value = ID_AA64PFR1_MTE,
>               .sign = FTR_UNSIGNED,
>               .cpu_enable = cpu_enable_mte,
>       },
> 
> When host support for FEAT_MTE3 is added then the KVM code will need
> revisiting to expose that down to the guest safely (AFAICS there's
> nothing extra to do here, but I haven't tested any of the MTE3
> features). I don't think we want to expose newer versions to the guest
> than the host is aware of. (Or indeed expose FEAT_MTE if the host has
> MTE disabled because Linux requires at least FEAT_MTE2).

What I was suggesting is to have something like this:

     pfr = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
     mte = cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR1_MTE_SHIFT);
     val |= FIELD_PREP(FEATURE(ID_AA64PFR1_MTE), mte);

which does the trick nicely, and doesn't expose more than the host
supports.

Thanks,

        M.

-- 
Without deviation from the norm, progress is not possible.