Message ID | 20221220200923.1532710-3-maz@kernel.org (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | KVM: arm64: Fix handling of S1PTW S2 fault on RO memslots | expand |
Hello, On Tue, Dec 20, 2022 at 08:09:22PM +0000, Marc Zyngier wrote: > As a minor optimisation, we can retrofit the "S1PTW is a write > even on translation fault" concept *if* the vcpu is using the > HW-managed Access Flag, as setting TCR_EL1.HA is guaranteed > to result in an update of the PTE. > > However, we cannot do the same thing for DB, as it would require > us to parse the PTs to find out if the DBM bit is set there. > This is not going to happen. > > Signed-off-by: Marc Zyngier <maz@kernel.org> > --- > arch/arm64/include/asm/kvm_emulate.h | 20 +++++++++++++++++++- > 1 file changed, 19 insertions(+), 1 deletion(-) > > diff --git a/arch/arm64/include/asm/kvm_emulate.h b/arch/arm64/include/asm/kvm_emulate.h > index fd6ad8b21f85..4ee467065042 100644 > --- a/arch/arm64/include/asm/kvm_emulate.h > +++ b/arch/arm64/include/asm/kvm_emulate.h > @@ -374,6 +374,9 @@ static __always_inline int kvm_vcpu_sys_get_rt(struct kvm_vcpu *vcpu) > static inline bool kvm_is_write_fault(struct kvm_vcpu *vcpu) > { > if (kvm_vcpu_abt_iss1tw(vcpu)) { > + unsigned int afdb; > + u64 mmfr1; > + > /* > * Only a permission fault on a S1PTW should be > * considered as a write. Otherwise, page tables baked > @@ -385,12 +388,27 @@ static inline bool kvm_is_write_fault(struct kvm_vcpu *vcpu) > * to map the page containing the PT (read only at > * first), then a permission fault to allow the flags > * to be set. > + * > + * We can improve things if the guest uses AF, as this > + * is guaranteed to result in a write to the PTE. For > + * DB, however, we'd need to parse the guest's PTs, > + * and that's not on. DB is crap anyway. > */ > switch (kvm_vcpu_trap_get_fault_type(vcpu)) { Nit: fault_status is calculated once when taking the fault, and passed around to all users (like user_mem_abort()). Not sure if this is because of the extra cycles needed to get it, or just style. Anyway, maybe it applies here. > case ESR_ELx_FSC_PERM: > return true; > default: > - return false; > + /* Can't introspect TCR_EL1 with pKVM */ > + if (kvm_vm_is_protected(vcpu->kvm)) > + return false; > + > + mmfr1 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1); > + afdb = cpuid_feature_extract_unsigned_field(mmfr1, ID_AA64MMFR1_EL1_HAFDBS_SHIFT); > + > + if (afdb == ID_AA64MMFR1_EL1_HAFDBS_NI) > + return false; > + > + return (vcpu_read_sys_reg(vcpu, TCR_EL1) & TCR_HA); Also tested this specific case using page_fault_test when the PT page is marked for dirty logging with and without AF. In both cases there's a single _FSC_FAULT (no PERM_FAUT) as expected, and the PT page is marked dirty in the AF case. The RO and UFFD cases also work as expected. Need to send some changes for page_fault_test as many tests assume that any S1PTW is always a PT write, and are failing. Also need to add some new tests for PTs in RO memslots (as it didn't make much sense before this change). > } > } > > -- > 2.34.1 > > _______________________________________________ > kvmarm mailing list > kvmarm@lists.cs.columbia.edu > https://lists.cs.columbia.edu/mailman/listinfo/kvmarm Reviewed-by: Ricardo Koller <ricarkol@google.com> Thanks, Ricardo
Hi Ricardo, On Wed, 21 Dec 2022 16:46:06 +0000, Ricardo Koller <ricarkol@google.com> wrote: > > Hello, > > On Tue, Dec 20, 2022 at 08:09:22PM +0000, Marc Zyngier wrote: > > As a minor optimisation, we can retrofit the "S1PTW is a write > > even on translation fault" concept *if* the vcpu is using the > > HW-managed Access Flag, as setting TCR_EL1.HA is guaranteed > > to result in an update of the PTE. > > > > However, we cannot do the same thing for DB, as it would require > > us to parse the PTs to find out if the DBM bit is set there. > > This is not going to happen. > > > > Signed-off-by: Marc Zyngier <maz@kernel.org> > > --- > > arch/arm64/include/asm/kvm_emulate.h | 20 +++++++++++++++++++- > > 1 file changed, 19 insertions(+), 1 deletion(-) > > > > diff --git a/arch/arm64/include/asm/kvm_emulate.h b/arch/arm64/include/asm/kvm_emulate.h > > index fd6ad8b21f85..4ee467065042 100644 > > --- a/arch/arm64/include/asm/kvm_emulate.h > > +++ b/arch/arm64/include/asm/kvm_emulate.h > > @@ -374,6 +374,9 @@ static __always_inline int kvm_vcpu_sys_get_rt(struct kvm_vcpu *vcpu) > > static inline bool kvm_is_write_fault(struct kvm_vcpu *vcpu) > > { > > if (kvm_vcpu_abt_iss1tw(vcpu)) { > > + unsigned int afdb; > > + u64 mmfr1; > > + > > /* > > * Only a permission fault on a S1PTW should be > > * considered as a write. Otherwise, page tables baked > > @@ -385,12 +388,27 @@ static inline bool kvm_is_write_fault(struct kvm_vcpu *vcpu) > > * to map the page containing the PT (read only at > > * first), then a permission fault to allow the flags > > * to be set. > > + * > > + * We can improve things if the guest uses AF, as this > > + * is guaranteed to result in a write to the PTE. For > > + * DB, however, we'd need to parse the guest's PTs, > > + * and that's not on. DB is crap anyway. > > */ > > switch (kvm_vcpu_trap_get_fault_type(vcpu)) { > > Nit: fault_status is calculated once when taking the fault, and passed > around to all users (like user_mem_abort()). Not sure if this is because > of the extra cycles needed to get it, or just style. Anyway, maybe it > applies here. All these things are just fields in ESR_EL2, which we keep looking at all the time. The compiler actually does a pretty good job at keeping that around, specially considering that this function is inlined (at least here, kvm_handle_guest_abort and kvm_user_mem_abort are merged into a single monster). So passing the parameter wouldn't change a thing, and I find the above more readable (I know that all the information in this function are derived from the same data structure). > > > case ESR_ELx_FSC_PERM: > > return true; > > default: > > - return false; > > + /* Can't introspect TCR_EL1 with pKVM */ > > + if (kvm_vm_is_protected(vcpu->kvm)) > > + return false; > > + > > + mmfr1 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1); > > + afdb = cpuid_feature_extract_unsigned_field(mmfr1, ID_AA64MMFR1_EL1_HAFDBS_SHIFT); > > + > > + if (afdb == ID_AA64MMFR1_EL1_HAFDBS_NI) > > + return false; > > + > > + return (vcpu_read_sys_reg(vcpu, TCR_EL1) & TCR_HA); > > Also tested this specific case using page_fault_test when the PT page is > marked for dirty logging with and without AF. In both cases there's a > single _FSC_FAULT (no PERM_FAUT) as expected, and the PT page is marked dirty > in the AF case. The RO and UFFD cases also work as expected. Ah, thanks for checking this. > > Need to send some changes for page_fault_test as many tests assume that > any S1PTW is always a PT write, and are failing. Also need to add some new > tests for PTs in RO memslots (as it didn't make much sense before this > change). I think this is what I really quite didn't grok in these tests. They seem to verify the KVM behaviour, which is not what we should check for. Instead, we should check for the architectural behaviour, which is that if HAFDBS is enabled, we can observe updates to the PTs even when we do not write to them directly. > > > } > > } > > > > -- > > 2.34.1 > > > > _______________________________________________ > > kvmarm mailing list > > kvmarm@lists.cs.columbia.edu > > https://lists.cs.columbia.edu/mailman/listinfo/kvmarm > > Reviewed-by: Ricardo Koller <ricarkol@google.com> Thanks! M.
On Wed, Dec 21, 2022 at 08:46:06AM -0800, Ricardo Koller wrote: [...] > > - return false; > > + /* Can't introspect TCR_EL1 with pKVM */ > > + if (kvm_vm_is_protected(vcpu->kvm)) > > + return false; > > + > > + mmfr1 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1); > > + afdb = cpuid_feature_extract_unsigned_field(mmfr1, ID_AA64MMFR1_EL1_HAFDBS_SHIFT); > > + > > + if (afdb == ID_AA64MMFR1_EL1_HAFDBS_NI) > > + return false; > > + > > + return (vcpu_read_sys_reg(vcpu, TCR_EL1) & TCR_HA); > > Also tested this specific case using page_fault_test when the PT page is > marked for dirty logging with and without AF. In both cases there's a > single _FSC_FAULT (no PERM_FAUT) as expected, and the PT page is marked dirty > in the AF case. The RO and UFFD cases also work as expected. > > Need to send some changes for page_fault_test as many tests assume that > any S1PTW is always a PT write, and are failing. Also need to add some new > tests for PTs in RO memslots (as it didn't make much sense before this > change). So I actually wanted to bring up the issue of user visibility, glad your test picked up something. This has two implications, which are rather odd. - When UFFD is in use, translation faults are reported to userspace as writes when from a RW memslot and reads when from an RO memslot. - S1 page table memory is spuriously marked as dirty, as we presume a write immediately follows the translation fault. That isn't entirely senseless, as it would mean both the target page and the S1 PT that maps it are both old. This is nothing new I suppose, just weird. Marc, do you have any concerns about leaving this as-is for the time being? At least before we were doing the same thing (write fault) every time. -- Thanks, Oliver
On Wed, 21 Dec 2022 17:46:24 +0000, Oliver Upton <oliver.upton@linux.dev> wrote: > > On Wed, Dec 21, 2022 at 08:46:06AM -0800, Ricardo Koller wrote: > > [...] > > > > - return false; > > > + /* Can't introspect TCR_EL1 with pKVM */ > > > + if (kvm_vm_is_protected(vcpu->kvm)) > > > + return false; > > > + > > > + mmfr1 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1); > > > + afdb = cpuid_feature_extract_unsigned_field(mmfr1, ID_AA64MMFR1_EL1_HAFDBS_SHIFT); > > > + > > > + if (afdb == ID_AA64MMFR1_EL1_HAFDBS_NI) > > > + return false; > > > + > > > + return (vcpu_read_sys_reg(vcpu, TCR_EL1) & TCR_HA); > > > > Also tested this specific case using page_fault_test when the PT page is > > marked for dirty logging with and without AF. In both cases there's a > > single _FSC_FAULT (no PERM_FAUT) as expected, and the PT page is marked dirty > > in the AF case. The RO and UFFD cases also work as expected. > > > > Need to send some changes for page_fault_test as many tests assume that > > any S1PTW is always a PT write, and are failing. Also need to add some new > > tests for PTs in RO memslots (as it didn't make much sense before this > > change). > > So I actually wanted to bring up the issue of user visibility, glad your > test picked up something. > > This has two implications, which are rather odd. > > - When UFFD is in use, translation faults are reported to userspace as > writes when from a RW memslot and reads when from an RO memslot. Not quite: translation faults are reported as reads if TCR_EL1.HA isn't set, and as writes if it is. Ignoring TCR_EL1.HD for a moment, this matches exactly the behaviour of the page-table walker, which will update the S1 PTs only if this bit is set. Or is it what userfaultfd does on its own? That'd be confusing... > > - S1 page table memory is spuriously marked as dirty, as we presume a > write immediately follows the translation fault. That isn't entirely > senseless, as it would mean both the target page and the S1 PT that > maps it are both old. This is nothing new I suppose, just weird. s/old/young/ ? I think you're confusing the PT access with the access that caused the PT access (I'll have that printed on a t-shirt, thank you very much). Here, we're not considering the cause of the PT access anymore. If TCR_EL1.HA is set, the S1 PT page will be marked as accessed even on a read, and only that page. TCR_EL1.HD is what muddies the waters a bit. If it is set without HA being set, we still handle the translation fault as a read, followed by a write permission fault. But again, that's solely for the purpose of the S1 PT. What happens for the mapped page is completely independent. > Marc, do you have any concerns about leaving this as-is for the time > being? At least before we were doing the same thing (write fault) every > time. I have the ugly feeling we're talking at cross purpose here, mostly because I don't get how userfaultfd fits in that picture. Can you shed some light here? Thanks, M.
On Thu, Dec 22, 2022 at 09:01:15AM +0000, Marc Zyngier wrote: > On Wed, 21 Dec 2022 17:46:24 +0000, Oliver Upton <oliver.upton@linux.dev> wrote: > > - When UFFD is in use, translation faults are reported to userspace as > > writes when from a RW memslot and reads when from an RO memslot. > > Not quite: translation faults are reported as reads if TCR_EL1.HA > isn't set, and as writes if it is. Ignoring TCR_EL1.HD for a moment, > this matches exactly the behaviour of the page-table walker, which > will update the S1 PTs only if this bit is set. My bad, yes you're right. I conflated the use case here with the architectural state. I'm probably being way too pedantic, but I just wanted to make sure we agree about the ensuing subtlety. More below: > Or is it what userfaultfd does on its own? That'd be confusing... > > > > > - S1 page table memory is spuriously marked as dirty, as we presume a > > write immediately follows the translation fault. That isn't entirely > > senseless, as it would mean both the target page and the S1 PT that > > maps it are both old. This is nothing new I suppose, just weird. > > s/old/young/ ? > > I think you're confusing the PT access with the access that caused the > PT access (I'll have that printed on a t-shirt, thank you very much). I'd buy it! > Here, we're not considering the cause of the PT access anymore. If > TCR_EL1.HA is set, the S1 PT page will be marked as accessed even on a > read, and only that page. I think this is where the disconnect might be. TCR_EL1.HA == 1 suggests a write could possibly follow, but I don't think it requires it. The page table walker must first load the S1 PTE before writing to it. From AArch64.S1Translate() (DDI0487H.a): (fault, descaddress, walkstate, descriptor) = AArch64.S1Walk(fault, walkparams, va, regime, ss, acctype, iswrite, ispriv); [...] new_desc = descriptor; if walkparams.ha == '1' && AArch64.FaultAllowsSetAccessFlag(fault) then // Set descriptor AF bit new_desc<10> = '1'; [...] // Either the access flag was clear or AP<2> is set if new_desc != descriptor then if regime == Regime_EL10 && EL2Enabled() then s1aarch64 = TRUE; s2fs1walk = TRUE; aligned = TRUE; iswrite = TRUE; (s2fault, descupdateaddress) = AArch64.S2Translate(fault, descaddress, s1aarch64, ss, s2fs1walk, AccType_ATOMICRW, aligned, iswrite, ispriv); if s2fault.statuscode != Fault_None then return (s2fault, AddressDescriptor UNKNOWN); else descupdateaddress = descaddress; (fault, mem_desc) = AArch64.MemSwapTableDesc(fault, descriptor, new_desc, walkparams.ee, descupdateaddress) Buried in AArch64.S1Walk() is a stage-2 walk for a read to fetch the descriptor. The second stage-2 walk for write is conditioned on having already fetched the stage-1 descriptor and determining the AF needs to be set. Relating back to UFFD: if we expect KVM to do exactly what hardware does, UFFD should see an attempted read when the first walk fails because of an S2 translation fault. Based on this patch, though, we'd promote it to a write if TCR_EL1.HA == 1. This has the additional nuance of marking the S1 PT's IPA as dirty, even though it might not actually have been written to. Having said that, the false positive rate should be negligible given that S1 PTs ought to account for a small amount of guest memory. Like I said before, I'm probably being unnecessarily pedantic :) It just seems to me that the view we're giving userspace of S1PTW aborts isn't exactly architectural and I want to make sure that is explicitly intentional. -- Thanks, Oliver
Hi Marc, On Wed, Dec 21, 2022 at 05:43:03PM +0000, Marc Zyngier wrote: > Hi Ricardo, > > On Wed, 21 Dec 2022 16:46:06 +0000, > Ricardo Koller <ricarkol@google.com> wrote: > > > > Hello, > > > > On Tue, Dec 20, 2022 at 08:09:22PM +0000, Marc Zyngier wrote: > > > As a minor optimisation, we can retrofit the "S1PTW is a write > > > even on translation fault" concept *if* the vcpu is using the > > > HW-managed Access Flag, as setting TCR_EL1.HA is guaranteed > > > to result in an update of the PTE. > > > > > > However, we cannot do the same thing for DB, as it would require > > > us to parse the PTs to find out if the DBM bit is set there. > > > This is not going to happen. > > > > > > Signed-off-by: Marc Zyngier <maz@kernel.org> > > > --- > > > arch/arm64/include/asm/kvm_emulate.h | 20 +++++++++++++++++++- > > > 1 file changed, 19 insertions(+), 1 deletion(-) > > > > > > diff --git a/arch/arm64/include/asm/kvm_emulate.h b/arch/arm64/include/asm/kvm_emulate.h > > > index fd6ad8b21f85..4ee467065042 100644 > > > --- a/arch/arm64/include/asm/kvm_emulate.h > > > +++ b/arch/arm64/include/asm/kvm_emulate.h > > > @@ -374,6 +374,9 @@ static __always_inline int kvm_vcpu_sys_get_rt(struct kvm_vcpu *vcpu) > > > static inline bool kvm_is_write_fault(struct kvm_vcpu *vcpu) > > > { > > > if (kvm_vcpu_abt_iss1tw(vcpu)) { > > > + unsigned int afdb; > > > + u64 mmfr1; > > > + > > > /* > > > * Only a permission fault on a S1PTW should be > > > * considered as a write. Otherwise, page tables baked > > > @@ -385,12 +388,27 @@ static inline bool kvm_is_write_fault(struct kvm_vcpu *vcpu) > > > * to map the page containing the PT (read only at > > > * first), then a permission fault to allow the flags > > > * to be set. > > > + * > > > + * We can improve things if the guest uses AF, as this > > > + * is guaranteed to result in a write to the PTE. For > > > + * DB, however, we'd need to parse the guest's PTs, > > > + * and that's not on. DB is crap anyway. > > > */ > > > switch (kvm_vcpu_trap_get_fault_type(vcpu)) { > > > > Nit: fault_status is calculated once when taking the fault, and passed > > around to all users (like user_mem_abort()). Not sure if this is because > > of the extra cycles needed to get it, or just style. Anyway, maybe it > > applies here. > > All these things are just fields in ESR_EL2, which we keep looking at > all the time. The compiler actually does a pretty good job at keeping > that around, specially considering that this function is inlined (at > least here, kvm_handle_guest_abort and kvm_user_mem_abort are merged > into a single monster). > > So passing the parameter wouldn't change a thing, and I find the above > more readable (I know that all the information in this function are > derived from the same data structure). > Got it, thanks for the info. > > > > > case ESR_ELx_FSC_PERM: > > > return true; > > > default: > > > - return false; > > > + /* Can't introspect TCR_EL1 with pKVM */ > > > + if (kvm_vm_is_protected(vcpu->kvm)) > > > + return false; > > > + > > > + mmfr1 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1); > > > + afdb = cpuid_feature_extract_unsigned_field(mmfr1, ID_AA64MMFR1_EL1_HAFDBS_SHIFT); > > > + > > > + if (afdb == ID_AA64MMFR1_EL1_HAFDBS_NI) > > > + return false; > > > + > > > + return (vcpu_read_sys_reg(vcpu, TCR_EL1) & TCR_HA); > > > > Also tested this specific case using page_fault_test when the PT page is > > marked for dirty logging with and without AF. In both cases there's a > > single _FSC_FAULT (no PERM_FAUT) as expected, and the PT page is marked dirty > > in the AF case. The RO and UFFD cases also work as expected. > > Ah, thanks for checking this. > > > > > Need to send some changes for page_fault_test as many tests assume that > > any S1PTW is always a PT write, and are failing. Also need to add some new > > tests for PTs in RO memslots (as it didn't make much sense before this > > change). > > I think this is what I really quite didn't grok in these tests. They > seem to verify the KVM behaviour, which is not what we should check > for. > > Instead, we should check for the architectural behaviour, which is > that if HAFDBS is enabled, we can observe updates to the PTs even when > we do not write to them directly. There are some tests checking that case (e.g., AF set by HW), but they also do it while interacting with dirty-logging, userfaultfd, and/or RO memslots. Some checks are clearly dealing with architectural behavior, while others are not that clear. Let me use this sample test to get more specific. This test deals with HW setting the AF bit on a punched hole backed by userfaultfd: TEST_UFFD(guest_exec, with_af, CMD_HOLE_PT, ...): 1. set TCR_EL1.HA and clear the AF in the test-data PTE, 2. punch a hole on the test-data PTE page, and register it for userfaultfd, 3. execute code in the test-data page; this triggers a S1PTW, 4. assert that there is a userfaultfd _write_ fault on the PT page, 5. assert that the test-data instruction was executed, 6. assert that the AF is set on the test-data PTE, IIUC, only checking for architectural behavior implies skipping step 4. And I agree, it might be a good idea to skip 4, as it actually depends on whether the kernel has only the previous commit, or both the previous and this one. The fault is a _write_ at this commit, and a _read_ at the previous. The issue is that there are some cases where checking KVM behavior could be useful. For example, this dirty_logging test can also help (besides testing dirty-logging) for checking regressions of commit c4ad98e4b72c ("KVM: arm64: Assume write fault on S1PTW permission fault on instruction fetch"): TEST_DIRTY_LOG(guest_exec, with_af, ...) 1. set TCR_EL1.HA and clear the AF in the test-data PTE, 2. enable dirty logging on the PT memslot, 3. execute code in the test-data page; this triggers a S1PTW, 4. assert that the test-data PTE page is dirty on the log, 5. assert that the test-data instruction was executed, 6. assert that the AF is set on the test-data PTE, Step 4 above is not exactly checking architectural behavior, but I think it's still useful. So, regarding what to do with page_fault_test. As a start, I need to go through all tests and make sure they pass at both this and the previous commit. Next, I have to identify other tests that also need to be relaxed a bit (removing some test asserts). Thanks, Ricardo > > > > > > } > > > } > > > > > > -- > > > 2.34.1 > > > > > > _______________________________________________ > > > kvmarm mailing list > > > kvmarm@lists.cs.columbia.edu > > > https://lists.cs.columbia.edu/mailman/listinfo/kvmarm > > > > Reviewed-by: Ricardo Koller <ricarkol@google.com> > > Thanks! > > M. > > -- > Without deviation from the norm, progress is not possible.
On Thu, Dec 22, 2022 at 08:58:40PM +0000, Oliver Upton wrote: > On Thu, Dec 22, 2022 at 09:01:15AM +0000, Marc Zyngier wrote: > > On Wed, 21 Dec 2022 17:46:24 +0000, Oliver Upton <oliver.upton@linux.dev> wrote: > > > - When UFFD is in use, translation faults are reported to userspace as > > > writes when from a RW memslot and reads when from an RO memslot. > > > > Not quite: translation faults are reported as reads if TCR_EL1.HA > > isn't set, and as writes if it is. Ignoring TCR_EL1.HD for a moment, > > this matches exactly the behaviour of the page-table walker, which > > will update the S1 PTs only if this bit is set. > > My bad, yes you're right. I conflated the use case here with the > architectural state. > > I'm probably being way too pedantic, but I just wanted to make sure we > agree about the ensuing subtlety. More below: > > > Or is it what userfaultfd does on its own? That'd be confusing... > > > > > > > > - S1 page table memory is spuriously marked as dirty, as we presume a > > > write immediately follows the translation fault. That isn't entirely > > > senseless, as it would mean both the target page and the S1 PT that > > > maps it are both old. This is nothing new I suppose, just weird. > > > > s/old/young/ ? > > > > I think you're confusing the PT access with the access that caused the > > PT access (I'll have that printed on a t-shirt, thank you very much). > > I'd buy it! > > > Here, we're not considering the cause of the PT access anymore. If > > TCR_EL1.HA is set, the S1 PT page will be marked as accessed even on a > > read, and only that page. > > I think this is where the disconnect might be. TCR_EL1.HA == 1 suggests > a write could possibly follow, but I don't think it requires it. The > page table walker must first load the S1 PTE before writing to it. > > From AArch64.S1Translate() (DDI0487H.a): > > (fault, descaddress, walkstate, descriptor) = AArch64.S1Walk(fault, walkparams, va, regime, > ss, acctype, iswrite, ispriv); > > [...] > > new_desc = descriptor; > if walkparams.ha == '1' && AArch64.FaultAllowsSetAccessFlag(fault) then > // Set descriptor AF bit > new_desc<10> = '1'; > > [...] > > // Either the access flag was clear or AP<2> is set > if new_desc != descriptor then > if regime == Regime_EL10 && EL2Enabled() then > s1aarch64 = TRUE; > s2fs1walk = TRUE; > aligned = TRUE; > iswrite = TRUE; > (s2fault, descupdateaddress) = AArch64.S2Translate(fault, descaddress, s1aarch64, > ss, s2fs1walk, AccType_ATOMICRW, > aligned, iswrite, ispriv); > > if s2fault.statuscode != Fault_None then > return (s2fault, AddressDescriptor UNKNOWN); > else > descupdateaddress = descaddress; > > (fault, mem_desc) = AArch64.MemSwapTableDesc(fault, descriptor, new_desc, > walkparams.ee, descupdateaddress) > > Buried in AArch64.S1Walk() is a stage-2 walk for a read to fetch the > descriptor. The second stage-2 walk for write is conditioned on having > already fetched the stage-1 descriptor and determining the AF needs > to be set. > > Relating back to UFFD: if we expect KVM to do exactly what hardware > does, UFFD should see an attempted read when the first walk fails > because of an S2 translation fault. Based on this patch, though, we'd > promote it to a write if TCR_EL1.HA == 1. > > This has the additional nuance of marking the S1 PT's IPA as dirty, even > though it might not actually have been written to. Having said that, > the false positive rate should be negligible given that S1 PTs ought to > account for a small amount of guest memory. Another false positive is TCR_EL1.HA == 1 and having the AF bit set in the PTE. This results on a write, when I don't think it should. > > Like I said before, I'm probably being unnecessarily pedantic :) It just > seems to me that the view we're giving userspace of S1PTW aborts isn't > exactly architectural and I want to make sure that is explicitly > intentional. > > -- > Thanks, > Oliver
On Thu, 22 Dec 2022 20:58:40 +0000, Oliver Upton <oliver.upton@linux.dev> wrote: > > On Thu, Dec 22, 2022 at 09:01:15AM +0000, Marc Zyngier wrote: > > On Wed, 21 Dec 2022 17:46:24 +0000, Oliver Upton <oliver.upton@linux.dev> wrote: > > > - When UFFD is in use, translation faults are reported to userspace as > > > writes when from a RW memslot and reads when from an RO memslot. > > > > Not quite: translation faults are reported as reads if TCR_EL1.HA > > isn't set, and as writes if it is. Ignoring TCR_EL1.HD for a moment, > > this matches exactly the behaviour of the page-table walker, which > > will update the S1 PTs only if this bit is set. > > My bad, yes you're right. I conflated the use case here with the > architectural state. > > I'm probably being way too pedantic, but I just wanted to make sure we > agree about the ensuing subtlety. More below: > > > Or is it what userfaultfd does on its own? That'd be confusing... > > > > > > > > - S1 page table memory is spuriously marked as dirty, as we presume a > > > write immediately follows the translation fault. That isn't entirely > > > senseless, as it would mean both the target page and the S1 PT that > > > maps it are both old. This is nothing new I suppose, just weird. > > > > s/old/young/ ? > > > > I think you're confusing the PT access with the access that caused the > > PT access (I'll have that printed on a t-shirt, thank you very much). > > I'd buy it! > > > Here, we're not considering the cause of the PT access anymore. If > > TCR_EL1.HA is set, the S1 PT page will be marked as accessed even on a > > read, and only that page. > > I think this is where the disconnect might be. TCR_EL1.HA == 1 suggests > a write could possibly follow, but I don't think it requires it. The > page table walker must first load the S1 PTE before writing to it. Ah, you're talking of the write to the PTE. Too many writes! My reasoning is based on Rule LFTXR in DDI0487I.a, which says: "When the PE performs a hardware update of the AF, it sets the AF to 1 in the corresponding descriptor in memory, in a coherent manner, using an atomic read-modify-write of that descriptor." An atomic-or operation fits this description, and I cannot see anything in the architecture that would prevent the write of a PTE even if AF is already set, such as mandating something like a test-and-set or compare-and-swap. I'm not saying this is the only possible implementation, or even a good one. But I don't think this is incompatible with what the architecture mandates. > > From AArch64.S1Translate() (DDI0487H.a): > > (fault, descaddress, walkstate, descriptor) = AArch64.S1Walk(fault, walkparams, va, regime, > ss, acctype, iswrite, ispriv); > > [...] > > new_desc = descriptor; > if walkparams.ha == '1' && AArch64.FaultAllowsSetAccessFlag(fault) then > // Set descriptor AF bit > new_desc<10> = '1'; > > [...] > > // Either the access flag was clear or AP<2> is set > if new_desc != descriptor then > if regime == Regime_EL10 && EL2Enabled() then > s1aarch64 = TRUE; > s2fs1walk = TRUE; > aligned = TRUE; > iswrite = TRUE; > (s2fault, descupdateaddress) = AArch64.S2Translate(fault, descaddress, s1aarch64, > ss, s2fs1walk, AccType_ATOMICRW, > aligned, iswrite, ispriv); > > if s2fault.statuscode != Fault_None then > return (s2fault, AddressDescriptor UNKNOWN); > else > descupdateaddress = descaddress; > > (fault, mem_desc) = AArch64.MemSwapTableDesc(fault, descriptor, new_desc, > walkparams.ee, descupdateaddress) > > Buried in AArch64.S1Walk() is a stage-2 walk for a read to fetch the > descriptor. The second stage-2 walk for write is conditioned on having > already fetched the stage-1 descriptor and determining the AF needs > to be set. The question is whether this is one possible implementation, or the only possible implementation. My bet is on the former. > Relating back to UFFD: if we expect KVM to do exactly what hardware > does, UFFD should see an attempted read when the first walk fails > because of an S2 translation fault. Based on this patch, though, we'd > promote it to a write if TCR_EL1.HA == 1. > > This has the additional nuance of marking the S1 PT's IPA as dirty, even > though it might not actually have been written to. Having said that, > the false positive rate should be negligible given that S1 PTs ought to > account for a small amount of guest memory. > > Like I said before, I'm probably being unnecessarily pedantic :) It just > seems to me that the view we're giving userspace of S1PTW aborts isn't > exactly architectural and I want to make sure that is explicitly > intentional. I think it is perfectly fine to be pedantic about these things, because they really matter. I still think that this change doesn't violate the architecture. But at the same time, I see some value in strictly following what the HW does, specially given that this only optimises the case where: - S1 PTs do not have a pre-existing S2 mapping (either placed there by the VMM, or swapped out) - TCR_EL1.HA==1 which is such a corner case that nobody will loose any sleep over it (and I'll buy beer to anyone who can come up with a real workload where this optimisation actually matters). So my suggestion is to drop the change altogether, and stick with the original fix. Thanks, M.
diff --git a/arch/arm64/include/asm/kvm_emulate.h b/arch/arm64/include/asm/kvm_emulate.h index fd6ad8b21f85..4ee467065042 100644 --- a/arch/arm64/include/asm/kvm_emulate.h +++ b/arch/arm64/include/asm/kvm_emulate.h @@ -374,6 +374,9 @@ static __always_inline int kvm_vcpu_sys_get_rt(struct kvm_vcpu *vcpu) static inline bool kvm_is_write_fault(struct kvm_vcpu *vcpu) { if (kvm_vcpu_abt_iss1tw(vcpu)) { + unsigned int afdb; + u64 mmfr1; + /* * Only a permission fault on a S1PTW should be * considered as a write. Otherwise, page tables baked @@ -385,12 +388,27 @@ static inline bool kvm_is_write_fault(struct kvm_vcpu *vcpu) * to map the page containing the PT (read only at * first), then a permission fault to allow the flags * to be set. + * + * We can improve things if the guest uses AF, as this + * is guaranteed to result in a write to the PTE. For + * DB, however, we'd need to parse the guest's PTs, + * and that's not on. DB is crap anyway. */ switch (kvm_vcpu_trap_get_fault_type(vcpu)) { case ESR_ELx_FSC_PERM: return true; default: - return false; + /* Can't introspect TCR_EL1 with pKVM */ + if (kvm_vm_is_protected(vcpu->kvm)) + return false; + + mmfr1 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1); + afdb = cpuid_feature_extract_unsigned_field(mmfr1, ID_AA64MMFR1_EL1_HAFDBS_SHIFT); + + if (afdb == ID_AA64MMFR1_EL1_HAFDBS_NI) + return false; + + return (vcpu_read_sys_reg(vcpu, TCR_EL1) & TCR_HA); } }
As a minor optimisation, we can retrofit the "S1PTW is a write even on translation fault" concept *if* the vcpu is using the HW-managed Access Flag, as setting TCR_EL1.HA is guaranteed to result in an update of the PTE. However, we cannot do the same thing for DB, as it would require us to parse the PTs to find out if the DBM bit is set there. This is not going to happen. Signed-off-by: Marc Zyngier <maz@kernel.org> --- arch/arm64/include/asm/kvm_emulate.h | 20 +++++++++++++++++++- 1 file changed, 19 insertions(+), 1 deletion(-)