Skip to contentSkip to conversation

kvm/x86: intercept guest changes to X86_CR4_LA57

RFC PATCH In review Lai Jiangshan 2 participants 6 messages Sep 28, 2020 – Sep 29, 2020

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/2 kvm/x86: intercept guest changes to X86_CR4_LA57

From: Lai Jiangshan <[email protected]>

When shadowpaping is enabled, guest should not be allowed to toggle X86_CR4_LA57. And X86_CR4_LA57 is a rarely changed bit, so we can just intercept all the attempts to toggle it no matter shadowpaping is in used or not.

Fixes fd8cb433734ee ("KVM: MMU: Expose the LA57 feature to VM.")
Cc Sean Christopherson <[email protected]>
Cc Yu Zhang <[email protected]>
Cc Paolo Bonzini <[email protected]>

---
No test to toggle X86_CR4_LA57 in guest since I can't access to any CPU supports it. Maybe it is not a real problem.

 arch/x86/kvm/kvm_cache_regs.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kvm/kvm_cache_regs.h +1 -1
index cfe83d4ae625..ca0781b41df9 100644
@@ -7,7 +7,7 @@
 #define KVM_POSSIBLE_CR0_GUEST_BITS X86_CR0_TS
 #define KVM_POSSIBLE_CR4_GUEST_BITS				  \
 	(X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR  \
-	 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_PGE | X86_CR4_TSD)
+	 | X86_CR4_OSXMMEXCPT | X86_CR4_PGE | X86_CR4_TSD)
 
 #define BUILD_KVM_GPR_ACCESSORS(lname, uname)				      \
 static __always_inline unsigned long kvm_##lname##_read(struct kvm_vcpu *vcpu)\
Signature · 1 line
2.19.1.6.gb485710b
#1 lore raw

RFC PATCH 2/2 kvm/x86: allow guest to toggle X86_CR4_FSGSBASE

From: Lai Jiangshan <[email protected]>

There is no reason to force VM-Exit on toggling
X86_CR4_FSGSBASE.

---
 arch/x86/kvm/kvm_cache_regs.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kvm/kvm_cache_regs.h +1 -1
index ca0781b41df9..a889563ad02d 100644
@@ -7,7 +7,7 @@
 #define KVM_POSSIBLE_CR0_GUEST_BITS X86_CR0_TS
 #define KVM_POSSIBLE_CR4_GUEST_BITS				  \
 	(X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR  \
-	 | X86_CR4_OSXMMEXCPT | X86_CR4_PGE | X86_CR4_TSD)
+	 | X86_CR4_OSXMMEXCPT | X86_CR4_PGE | X86_CR4_TSD | X86_CR4_FSGSBASE)
 
 #define BUILD_KVM_GPR_ACCESSORS(lname, uname)				      \
 static __always_inline unsigned long kvm_##lname##_read(struct kvm_vcpu *vcpu)\
Signature · 1 line
2.19.1.6.gb485710b
#2 lore raw
In reply to Lai Jiangshan
Quoted 7 lines · Lai Jiangshan

From: Lai Jiangshan <[email protected]>

When shadowpaping is enabled, guest should not be allowed to toggle X86_CR4_LA57. And X86_CR4_LA57 is a rarely changed bit, so we can just intercept all the attempts to toggle it no matter shadowpaping is in used or not.

Quoted 8 lines
Fixes fd8cb433734ee ("KVM: MMU: Expose the LA57 feature to VM.")
Cc Sean Christopherson <[email protected]>
Cc Yu Zhang <[email protected]>
Cc Paolo Bonzini <[email protected]>

---
No test to toggle X86_CR4_LA57 in guest since I can't access to any CPU supports it. Maybe it is not a real problem.

LA57 doesn't need to be intercepted. It can't be toggled in 64-bit mode (causes a #GP), and it's ignored in 32-bit mode. That means LA57 can only take effect when 64-bit mode is enabled, at which time KVM will update its MMU context accordingly.

Quoted 19 lines
 arch/x86/kvm/kvm_cache_regs.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kvm/kvm_cache_regs.h +1 -1
index cfe83d4ae625..ca0781b41df9 100644
@@ -7,7 +7,7 @@
 #define KVM_POSSIBLE_CR0_GUEST_BITS X86_CR0_TS
 #define KVM_POSSIBLE_CR4_GUEST_BITS				  \
 	(X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR  \
-	 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_PGE | X86_CR4_TSD)
+	 | X86_CR4_OSXMMEXCPT | X86_CR4_PGE | X86_CR4_TSD)
 
 #define BUILD_KVM_GPR_ACCESSORS(lname, uname)				      \
 static __always_inline unsigned long kvm_##lname##_read(struct kvm_vcpu *vcpu)\
Signature · 1 line
2.19.1.6.gb485710b
#3 lore raw
Quoted 9 lines · Sean Christopherson
Quoted 7 lines · Lai Jiangshan

From: Lai Jiangshan <[email protected]>

When shadowpaping is enabled, guest should not be allowed to toggle X86_CR4_LA57. And X86_CR4_LA57 is a rarely changed bit, so we can just intercept all the attempts to toggle it no matter shadowpaping is in used or not.

Quoted 9 lines
Quoted 8 lines
Fixes fd8cb433734ee ("KVM: MMU: Expose the LA57 feature to VM.")
Cc Sean Christopherson <[email protected]>
Cc Yu Zhang <[email protected]>
Cc Paolo Bonzini <[email protected]>

---
No test to toggle X86_CR4_LA57 in guest since I can't access to any CPU supports it. Maybe it is not a real problem.

Hello

Thanks for reviewing.

Quoted 5 lines

LA57 doesn't need to be intercepted. It can't be toggled in 64-bit mode (causes a #GP), and it's ignored in 32-bit mode. That means LA57 can only take effect when 64-bit mode is enabled, at which time KVM will update its MMU context accordingly.

Oh, I missed that part which is so obvious that the patch seems impertinent.

But X86_CR4_LA57 is so fundamental that it makes me afraid to give it over to guests. And it is rarely changed too. At least, there is no better reason to give it to the guest than intercepting it.

There might be another reason that this patch is still needed with an updated changelog.

When a user (via VMM such as qemu) launches a VM with LA57 disabled in its cpuid on a LA57 enabled host. The hypervisor, IMO, needs to intercept guest's changes to X86_CR4_LA57 even when the guest is still in the non-paging mode. Otherwise the hypervisor failed to detective such combination when the guest changes paging mode later.

Anyway, maybe it is still not a real problem.

Thanks
Lai

#4 lore raw
Quoted 43 lines · Lai Jiangshan
Quoted 9 lines · Sean Christopherson
Quoted 7 lines · Lai Jiangshan

From: Lai Jiangshan <[email protected]>

When shadowpaping is enabled, guest should not be allowed to toggle X86_CR4_LA57. And X86_CR4_LA57 is a rarely changed bit, so we can just intercept all the attempts to toggle it no matter shadowpaping is in used or not.

Quoted 9 lines
Quoted 8 lines
Fixes fd8cb433734ee ("KVM: MMU: Expose the LA57 feature to VM.")
Cc Sean Christopherson <[email protected]>
Cc Yu Zhang <[email protected]>
Cc Paolo Bonzini <[email protected]>

---
No test to toggle X86_CR4_LA57 in guest since I can't access to any CPU supports it. Maybe it is not a real problem.

Hello

Thanks for reviewing.

Quoted 5 lines

LA57 doesn't need to be intercepted. It can't be toggled in 64-bit mode (causes a #GP), and it's ignored in 32-bit mode. That means LA57 can only take effect when 64-bit mode is enabled, at which time KVM will update its MMU context accordingly.

Oh, I missed that part which is so obvious that the patch seems impertinent.

But X86_CR4_LA57 is so fundamental that it makes me afraid to give it over to guests. And it is rarely changed too. At least, there is no better reason to give it to the guest than intercepting it.

There might be another reason that this patch is still needed with an updated changelog.

Quoted 7 lines

When a user (via VMM such as qemu) launches a VM with LA57 disabled in its cpuid on a LA57 enabled host. The hypervisor, IMO, needs to intercept guest's changes to X86_CR4_LA57 even when the guest is still in the non-paging mode. Otherwise the hypervisor failed to detective such combination when the guest changes paging mode later.

Anyway, maybe it is still not a real problem.

Oof, the above is a KVM bug, though in a more generic manner. All reserved bits should be intercepted, not just LA57. LA57 is the only affected bit at the moment, but proper support is needed as the follow-on patch to let the guest toggle FSGSBASE would introduce the same bug.

Sadly, fixing this is a bit of a mess. Well, fixing LA57 is easy, e.g. this patch will do the trick. But actually refreshing the CR4 guest/host mask when the guest's CPUID is updated is a pain, and that's what's needed for proper FSGSBASE support.

I'll send a series, bookended by these two RFC patches, with patches to intercept CR4 reserved bits smushed in between. I agree there's no point in letting the guest write LA57 directly, it's almost literally a once-per-boot thing. I wouldn't be surprised if intercepting it is a net win (but still inconsequential), e.g. due to the MMU having to grab CR4 out of the VMCS.

#5 lore raw
Quoted 58 lines · Sean Christopherson
Quoted 43 lines · Lai Jiangshan
Quoted 9 lines · Sean Christopherson
Quoted 7 lines · Lai Jiangshan

From: Lai Jiangshan <[email protected]>

When shadowpaping is enabled, guest should not be allowed to toggle X86_CR4_LA57. And X86_CR4_LA57 is a rarely changed bit, so we can just intercept all the attempts to toggle it no matter shadowpaping is in used or not.

Quoted 9 lines
Quoted 8 lines
Fixes fd8cb433734ee ("KVM: MMU: Expose the LA57 feature to VM.")
Cc Sean Christopherson <[email protected]>
Cc Yu Zhang <[email protected]>
Cc Paolo Bonzini <[email protected]>

---
No test to toggle X86_CR4_LA57 in guest since I can't access to any CPU supports it. Maybe it is not a real problem.

Hello

Thanks for reviewing.

Quoted 5 lines

LA57 doesn't need to be intercepted. It can't be toggled in 64-bit mode (causes a #GP), and it's ignored in 32-bit mode. That means LA57 can only take effect when 64-bit mode is enabled, at which time KVM will update its MMU context accordingly.

Oh, I missed that part which is so obvious that the patch seems impertinent.

But X86_CR4_LA57 is so fundamental that it makes me afraid to give it over to guests. And it is rarely changed too. At least, there is no better reason to give it to the guest than intercepting it.

There might be another reason that this patch is still needed with an updated changelog.

Quoted 7 lines

When a user (via VMM such as qemu) launches a VM with LA57 disabled in its cpuid on a LA57 enabled host. The hypervisor, IMO, needs to intercept guest's changes to X86_CR4_LA57 even when the guest is still in the non-paging mode. Otherwise the hypervisor failed to detective such combination when the guest changes paging mode later.

Anyway, maybe it is still not a real problem.

Oof, the above is a KVM bug, though in a more generic manner. All reserved bits should be intercepted, not just LA57. LA57 is the only affected bit at the moment, but proper support is needed as the follow-on patch to let the guest toggle FSGSBASE would introduce the same bug.

Quoted 6 lines

Sadly, fixing this is a bit of a mess. Well, fixing LA57 is easy, e.g. this patch will do the trick. But actually refreshing the CR4 guest/host mask when the guest's CPUID is updated is a pain, and that's what's needed for proper FSGSBASE support.

I'll send a series, bookended by these two RFC patches, with patches to

Thanks for illustrating deep inside.
I'm looking forward to the series.

Quoted 4 lines

intercept CR4 reserved bits smushed in between. I agree there's no point in letting the guest write LA57 directly, it's almost literally a once-per-boot thing. I wouldn't be surprised if intercepting it is a net win (but still inconsequential), e.g. due to the MMU having to grab CR4 out of the VMCS.