KVM: SVM: Final C-bit fixes?
Sean Christopherson
maintainercover KVM: SVM: Final C-bit fixes?
Patch 01 reverts the C-bit truncation patch as the reserved #PF was confirmed to be due to a magic HyperTransport region (how many magic addresses are there!?!). Hopefully the original patch simply be dropped, but just in case...
Patch 02 reverts the C-bit clearing in the #NPF handler. If that somehow turns out to be incorrect, i.e. there are flows where the CPU doesn't mask off the C-bit, then it can be conditional on a SEV guest.
I'll be offline for the next two weeks, fingers crossed I've undone all the damage. :-)
Thanks!
Sean Christopherson (2):
Revert "KVM: x86: Truncate reported guest MAXPHYADDR to C-bit if SEV
is supported"
KVM: SVM: Revert clearing of C-bit on GPA in #NPF handler
arch/x86/kvm/cpuid.c | 11 ----------- arch/x86/kvm/svm/svm.c | 39 +++++++++------------------------------ arch/x86/kvm/x86.c | 3 --- arch/x86/kvm/x86.h | 1 - 4 files changed, 9 insertions(+), 45 deletions(-)
Signature · 1 line
2.32.0.93.g670b81a890-goog
Sean Christopherson
maintainerPATCH 1/2 Revert "KVM: x86: Truncate reported guest MAXPHYADDR to C-bit if SEV is supported"
It turns out that non-SEV guest can indeed use bit 47, and the unexpected reserved #PF observed when setting bit 47 is due to a magic 12gb HyperTransport region being off limits, even for GPAs. Per Tom:
I think you may be hitting a special HT region that is at the top 12GB of the 48-bit memory range and is reserved, even for GPAs. Can you somehow get the test to use an address below 0xfffd_0000_0000? That would show that bit 47 is valid for the legacy guest while staying out of the HT region.
And indeed, accessing 0xfffd00000000 generates a reserved #PF, while dropping down a single page to 0xfffcfffff000 does not.
This reverts commit 3675f005c87c4026713c9f863924de511fdd36c4.
--- arch/x86/kvm/cpuid.c | 11 ----------- arch/x86/kvm/svm/svm.c | 37 ++++++++----------------------------- arch/x86/kvm/x86.c | 3 --- arch/x86/kvm/x86.h | 1 - 4 files changed, 8 insertions(+), 44 deletions(-)
arch/x86/kvm/cpuid.c +0 -11
@@ -955,17 +955,6 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)else if (!g_phys_as) g_phys_as = phys_as; -/* -* The exception to the exception is if hardware supports SEV, -* in which case the C-bit is reserved for non-SEV guests and -* isn't a GPA bit for SEV guests. -* -* Note, KVM always reports '0' for the number of reduced PA -* bits (see 0x8000001F). -*/ -if (tdp_enabled && sev_c_bit) - g_phys_as =min (g_phys_as, (unsigned int )sev_c_bit); - entry->eax = g_phys_as | (virt_as <<8 ); entry->edx =0 ;cpuid_entry_override (entry,CPUID_8000_0008_EBX );
arch/x86/kvm/svm/svm.c +8 -29
@@ -860,26 +860,6 @@ static void shrink_ple_window(struct kvm_vcpu *vcpu) } } -static __init u8 svm_get_c_bit (bool sev_only) -{ -unsigned int eax, ebx, ecx, edx; -u64 msr; - -if (cpuid_eax (0x80000000 ) <0x8000001f ) -return 0 ; - -if (rdmsrl_safe (MSR_AMD64_SYSCFG , &msr) || - !(msr &MSR_AMD64_SYSCFG_MEM_ENCRYPT )) -return 0 ; - -cpuid (0x8000001f , &eax, &ebx, &ecx, &edx); - -if (sev_only && !(eax &feature_bit (SEV ))) -return 0 ; - -return ebx &0x3f ; -} -/* * The default MMIO mask is a single bit (excluding the present bit), * which could conflict with the memory encryption bit. Check for
@@ -889,13 +869,18 @@ static __init u8 svm_get_c_bit(bool sev_only)static __init void svm_adjust_mmio_mask (void ) {unsigned int enc_bit, mask_bit; -u64 mask; +u64 msr, mask; + +/* If there is no memory encryption support, use existing mask */ +if (cpuid_eax (0x80000000 ) <0x8000001f ) +return ;/* If memory encryption is not enabled, use existing mask */ - enc_bit =svm_get_c_bit (false); -if (!enc_bit) +rdmsrl (MSR_AMD64_SYSCFG , msr); +if (!(msr &MSR_AMD64_SYSCFG_MEM_ENCRYPT ))return ; + enc_bit =cpuid_ebx (0x8000001f ) &0x3f ; mask_bit = boot_cpu_data.x86_phys_bits;/* Increment the mask bit if it is the same as the encryption bit */
@@ -1028,12 +1013,6 @@ static __init int svm_hardware_setup(void)kvm_configure_mmu (npt_enabled,get_max_npt_level (),PG_LEVEL_1G );pr_info ("kvm: Nested Paging %sabled\n" , npt_enabled ?"en" :"dis" ); -/* -* The SEV C-bit location is needed to correctly enumeration guest -* MAXPHYADDR even if SEV is not fully supported. -*/ - sev_c_bit =svm_get_c_bit (true); -/* Note, SEV setup consumes npt_enabled. */ sev_hardware_setup ();
arch/x86/kvm/x86.c +0 -3
@@ -209,9 +209,6 @@ static struct kvm_user_return_msrs __percpu *user_return_msrs; | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \ | XFEATURE_MASK_PKRU) -u8 __read_mostly sev_c_bit; -EXPORT_SYMBOL_GPL (sev_c_bit); -u64 __read_mostly host_efer;EXPORT_SYMBOL_GPL (host_efer);
arch/x86/kvm/x86.h +0 -1
@@ -328,7 +328,6 @@ extern u64 host_xcr0;extern u64 supported_xcr0;extern u64 host_xss;extern u64 supported_xss; -extern u8 sev_c_bit;static inline bool kvm_mpx_supported (void ) {
Signature · 1 line
2.32.0.93.g670b81a890-goog
Re: [PATCH 1/2] Revert "KVM: x86: Truncate reported guest MAXPHYADDR to C-bit if SEV is supported"
Quoted 13 lines · Sean Christopherson
It turns out that non-SEV guest can indeed use bit 47, and the unexpected reserved #PF observed when setting bit 47 is due to a magic 12gb HyperTransport region being off limits, even for GPAs. Per Tom:
I think you may be hitting a special HT region that is at the top 12GB of the 48-bit memory range and is reserved, even for GPAs. Can you somehow get the test to use an address below 0xfffd_0000_0000? That would show that bit 47 is valid for the legacy guest while staying out of the HT region.
And indeed, accessing
0xfffd00000000generates a reserved #PF, while dropping down a single page to0xfffcfffff000does not.
Quoted 6 lines
This reverts commit
3675f005c87c4026713c9f863924de511fdd36c4.Cc Peter Gonda <[email protected]>Cc Brijesh Singh <[email protected]>Cc Tom Lendacky <[email protected]>Signed-off-by Sean Christopherson <[email protected]>
Quoted 122 lines
--- arch/x86/kvm/cpuid.c | 11 ----------- arch/x86/kvm/svm/svm.c | 37 ++++++++----------------------------- arch/x86/kvm/x86.c | 3 --- arch/x86/kvm/x86.h | 1 - 4 files changed, 8 insertions(+), 44 deletions(-)arch/x86/kvm/cpuid.c +0
-11@@ -955,17 +955,6 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)else if (!g_phys_as) g_phys_as = phys_as; -/* -* The exception to the exception is if hardware supports SEV, -* in which case the C-bit is reserved for non-SEV guests and -* isn't a GPA bit for SEV guests. -* -* Note, KVM always reports '0' for the number of reduced PA -* bits (see 0x8000001F). -*/ -if (tdp_enabled && sev_c_bit) - g_phys_as =min (g_phys_as, (unsigned int )sev_c_bit); - entry->eax = g_phys_as | (virt_as <<8 ); entry->edx =0 ;cpuid_entry_override (entry,CPUID_8000_0008_EBX );arch/x86/kvm/svm/svm.c +8
-29@@ -860,26 +860,6 @@ static void shrink_ple_window(struct kvm_vcpu *vcpu) } } -static __init u8 svm_get_c_bit (bool sev_only) -{ -unsigned int eax, ebx, ecx, edx; -u64 msr; - -if (cpuid_eax (0x80000000 ) <0x8000001f ) -return 0 ; - -if (rdmsrl_safe (MSR_AMD64_SYSCFG , &msr) || - !(msr &MSR_AMD64_SYSCFG_MEM_ENCRYPT )) -return 0 ; - -cpuid (0x8000001f , &eax, &ebx, &ecx, &edx); - -if (sev_only && !(eax &feature_bit (SEV ))) -return 0 ; - -return ebx &0x3f ; -} -/* * The default MMIO mask is a single bit (excluding the present bit), * which could conflict with the memory encryption bit. Check for @@ -889,13 +869,18 @@ static __init u8 svm_get_c_bit(bool sev_only)static __init void svm_adjust_mmio_mask (void ) {unsigned int enc_bit, mask_bit; -u64 mask; +u64 msr, mask; + +/* If there is no memory encryption support, use existing mask */ +if (cpuid_eax (0x80000000 ) <0x8000001f ) +return ;/* If memory encryption is not enabled, use existing mask */ - enc_bit =svm_get_c_bit (false); -if (!enc_bit) +rdmsrl (MSR_AMD64_SYSCFG , msr); +if (!(msr &MSR_AMD64_SYSCFG_MEM_ENCRYPT ))return ; + enc_bit =cpuid_ebx (0x8000001f ) &0x3f ; mask_bit = boot_cpu_data.x86_phys_bits;/* Increment the mask bit if it is the same as the encryption bit */ @@ -1028,12 +1013,6 @@ static __init int svm_hardware_setup(void)kvm_configure_mmu (npt_enabled,get_max_npt_level (),PG_LEVEL_1G );pr_info ("kvm: Nested Paging %sabled\n" , npt_enabled ?"en" :"dis" ); -/* -* The SEV C-bit location is needed to correctly enumeration guest -* MAXPHYADDR even if SEV is not fully supported. -*/ - sev_c_bit =svm_get_c_bit (true); -/* Note, SEV setup consumes npt_enabled. */ sev_hardware_setup ();arch/x86/kvm/x86.c +0
-3@@ -209,9 +209,6 @@ static struct kvm_user_return_msrs __percpu *user_return_msrs; | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \ | XFEATURE_MASK_PKRU) -u8 __read_mostly sev_c_bit; -EXPORT_SYMBOL_GPL (sev_c_bit); -u64 __read_mostly host_efer;EXPORT_SYMBOL_GPL (host_efer);arch/x86/kvm/x86.h +0
-1@@ -328,7 +328,6 @@ extern u64 host_xcr0;extern u64 supported_xcr0;extern u64 host_xss;extern u64 supported_xss; -extern u8 sev_c_bit;static inline bool kvm_mpx_supported (void ) {
Sean Christopherson
maintainerPATCH 2/2 KVM: SVM: Revert clearing of C-bit on GPA in #NPF handler
In reply to Sean ChristophersonDon't clear the C-bit in the #NPF handler, as it is a legal GPA bit for non-SEV guests, and for SEV guests the C-bit is dropped before the GPA hits the NPT in hardware. Clearing the bit for non-SEV guests causes KVM to mishandle #NPFs with that collide with the host's C-bit.
Although the APM doesn't explicitly state that the C-bit is not reserved for non-SEV, Tom Lendacky confirmed that the following snippet about the effective reduction due to the C-bit does indeed apply only to SEV guests.
Note that because guest physical addresses are always translated through the nested page tables, the size of the guest physical address space is not impacted by any physical address space reduction indicated in CPUID 8000_001F[EBX]. If the C-bit is a physical address bit however, the guest physical address space is effectively reduced by 1 bit.
And for SEV guests, the APM clearly states that the bit is dropped before walking the nested page tables.
If the C-bit is an address bit, this bit is masked from the guest physical address when it is translated through the nested page tables. Consequently, the hypervisor does not need to be aware of which pages the guest has chosen to mark private.
Note, the bogus C-bit clearing was removed from legacy #PF handler in commit 6d1b867d0456 ("KVM: SVM: Don't strip the C-bit from CR2 on #PF interception").
--- arch/x86/kvm/svm/svm.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kvm/svm/svm.c +1 -1
@@ -1923,7 +1923,7 @@ static int npf_interception(struct kvm_vcpu *vcpu) {struct vcpu_svm *svm =to_svm (vcpu); -u64 fault_address =__sme_clr (svm->vmcb->control.exit_info_2); +u64 fault_address = svm->vmcb->control.exit_info_2;u64 error_code = svm->vmcb->control.exit_info_1;trace_kvm_page_fault (fault_address, error_code);
Signature · 1 line
2.32.0.93.g670b81a890-goog
Re: [PATCH 2/2] KVM: SVM: Revert clearing of C-bit on GPA in #NPF handler
Quoted 27 lines · Sean Christopherson
Don't clear the C-bit in the #NPF handler, as it is a legal GPA bit for non-SEV guests, and for SEV guests the C-bit is dropped before the GPA hits the NPT in hardware. Clearing the bit for non-SEV guests causes KVM to mishandle #NPFs with that collide with the host's C-bit.
Although the APM doesn't explicitly state that the C-bit is not reserved for non-SEV, Tom Lendacky confirmed that the following snippet about the effective reduction due to the C-bit does indeed apply only to SEV guests.
Note that because guest physical addresses are always translated through the nested page tables, the size of the guest physical address space is not impacted by any physical address space reduction indicated in CPUID 8000_001F[EBX]. If the C-bit is a physical address bit however, the guest physical address space is effectively reduced by 1 bit.
And for SEV guests, the APM clearly states that the bit is dropped before walking the nested page tables.
If the C-bit is an address bit, this bit is masked from the guest physical address when it is translated through the nested page tables. Consequently, the hypervisor does not need to be aware of which pages the guest has chosen to mark private.
Note, the bogus C-bit clearing was removed from legacy #PF handler in commit
6d1b867d0456("KVM: SVM: Don't strip the C-bit from CR2 on #PF interception").
Quoted 6 lines
Fixes 0ede79e13224 ("KVM: SVM: Clear C-bit from the page fault address")Cc Peter Gonda <[email protected]>Cc Brijesh Singh <[email protected]>Cc Tom Lendacky <[email protected]>Signed-off-by Sean Christopherson <[email protected]>
Yep, definitely wasn't correct to be using an SME based macro on a guest address. And as the APM states, the encryption bit is stripped for SEV guests, so looks correct to me.
Quoted 18 lines
--- arch/x86/kvm/svm/svm.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)arch/x86/kvm/svm/svm.c +1
-1@@ -1923,7 +1923,7 @@ static int npf_interception(struct kvm_vcpu *vcpu) {struct vcpu_svm *svm =to_svm (vcpu); -u64 fault_address =__sme_clr (svm->vmcb->control.exit_info_2); +u64 fault_address = svm->vmcb->control.exit_info_2;u64 error_code = svm->vmcb->control.exit_info_1;trace_kvm_page_fault (fault_address, error_code);
Quoted 19 lines · Sean Christopherson
Patch 01 reverts the C-bit truncation patch as the reserved #PF was confirmed to be due to a magic HyperTransport region (how many magic addresses are there!?!). Hopefully the original patch simply be dropped, but just in case...
Patch 02 reverts the C-bit clearing in the #NPF handler. If that somehow turns out to be incorrect, i.e. there are flows where the CPU doesn't mask off the C-bit, then it can be conditional on a SEV guest.
I'll be offline for the next two weeks, fingers crossed I've undone all the damage. :-)
Thanks!
Sean Christopherson (2):
Revert "KVM: x86: Truncate reported guest MAXPHYADDR to C-bit if SEV
is supported"
KVM: SVM: Revert clearing of C-bit on GPA in #NPF handler
Quoted 6 lines
arch/x86/kvm/cpuid.c | 11 ----------- arch/x86/kvm/svm/svm.c | 39 +++++++++------------------------------ arch/x86/kvm/x86.c | 3 --- arch/x86/kvm/x86.h | 1 - 4 files changed, 9 insertions(+), 45 deletions(-)
Queued, thanks.
Paolo