Skip to contentSkip to conversation

KVM: allow KVM_BUG/KVM_BUG_ON to handle 64-bit cond

PATCH Applied Wei Wang 4 participants 12 messages Mar 7, 2023 – Jun 5, 2023 KVM

lore
Share & tools
lore

Wei Wang

author
#0 lore raw

PATCH v2 1/1 KVM: allow KVM_BUG/KVM_BUG_ON to handle 64-bit cond

Current KVM_BUG and KVM_BUG_ON assume that 'cond' passed from callers is 32-bit as it casts 'cond' to the type of int. This will be wrong if 'cond' provided by a caller is 64-bit, e.g. an error code of 0xc0000d0300000000 will be converted to 0, which is not expected.

Improves the implementation by using bool in KVM_BUG and KVM_BUG_ON. 'bool' is preferred to 'int' as __ret is essentially used as a boolean and coding-stytle.rst documents that use of bool is encouraged to improve readability and is often a better option than 'int' for storing boolean values.

Fixes 0b8f11737cff ("KVM: Add infrastructure and macro to mark VM as bugged")
---
 include/linux/kvm_host.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
include/linux/kvm_host.h +2 -2
index f06635b24bd0..0221a48b3e3d 100644
@@ -881,7 +881,7 @@ static inline void kvm_vm_bugged(struct kvm *kvm)
 
 #define KVM_BUG(cond, kvm, fmt...)				\
 ({								\
-	int __ret = (cond);					\
+	bool __ret = !!(cond);					\
 								\
 	if (WARN_ONCE(__ret && !(kvm)->vm_bugged, fmt))		\
 		kvm_vm_bugged(kvm);				\
@@ -890,7 +890,7 @@ static inline void kvm_vm_bugged(struct kvm *kvm)
 
 #define KVM_BUG_ON(cond, kvm)					\
 ({								\
-	int __ret = (cond);					\
+	bool __ret = !!(cond);					\
 								\
 	if (WARN_ON_ONCE(__ret && !(kvm)->vm_bugged))		\
 		kvm_vm_bugged(kvm);				\
Signature · 1 line
2.27.0
Reviewed-by on v2 1/1
#1 lore raw
Quoted 12 lines · Wei Wang

Current KVM_BUG and KVM_BUG_ON assume that 'cond' passed from callers is 32-bit as it casts 'cond' to the type of int. This will be wrong if 'cond' provided by a caller is 64-bit, e.g. an error code of 0xc0000d0300000000 will be converted to 0, which is not expected.

Improves the implementation by using bool in KVM_BUG and KVM_BUG_ON. 'bool' is preferred to 'int' as __ret is essentially used as a boolean and coding-stytle.rst documents that use of bool is encouraged to improve readability and is often a better option than 'int' for storing boolean values.

Quoted 3 lines
Fixes 0b8f11737cff ("KVM: Add infrastructure and macro to mark VM as bugged")

---

reviewer Reviewed-by on v2 1/1
#2 lore raw
In reply to Wei Wang
Quoted 2 lines · Wei Wang

Current KVM_BUG and KVM_BUG_ON assume that 'cond' passed from callers is 32-bit as it casts 'cond' to the type of int.

You're very generous, I would have led with "Fix a badly done copy+paste ..." ;-)

Quoted 3 lines
Fixes 0b8f11737cff ("KVM: Add infrastructure and macro to mark VM as bugged")

---

#3 lore raw
Quoted 8 lines · Sean Christopherson
Quoted 1 line · Wei Wang

Current KVM_BUG and KVM_BUG_ON assume that 'cond' passed from

callers

Quoted 1 line

is 32-bit as it casts 'cond' to the type of int.

You're very generous, I would have led with "Fix a badly done copy+paste ..." ;-)

Quoted 6 lines
Quoted 4 lines
Fixes 0b8f11737cff ("KVM: Add infrastructure and macro to mark VM as

bugged")

---

Kind ping on this patch.
Seems it wasn't noticed for months. Just check if it would be good to be merged or not proper for any reason?

Thanks,
Wei

#4 lore raw
Quoted 14 lines · Wei W Wang
Quoted 13 lines · Sean Christopherson
Quoted 1 line · Wei Wang

Current KVM_BUG and KVM_BUG_ON assume that 'cond' passed from

callers

Quoted 1 line

is 32-bit as it casts 'cond' to the type of int.

You're very generous, I would have led with "Fix a badly done copy+paste ..." ;-)

Quoted 4 lines
Fixes 0b8f11737cff ("KVM: Add infrastructure and macro to mark VM as

bugged")

---

Quoted 5 lines
Quoted 1 line

Kind ping on this patch.
Seems it wasn't noticed for months. Just check if it would be good to be merged or not proper for any reason?

I'll grab it for 6.5.

#5 lore raw
Quoted 16 lines · Sean Christopherson
Quoted 15 lines · Wei W Wang
Quoted 14 lines · Sean Christopherson
Quoted 1 line · Wei Wang

Current KVM_BUG and KVM_BUG_ON assume that 'cond' passed from

callers

Quoted 1 line

is 32-bit as it casts 'cond' to the type of int.

You're very generous, I would have led with "Fix a badly done copy+paste ..." ;-)

Quoted 5 lines
Fixes 0b8f11737cff ("KVM: Add infrastructure and macro to mark VM

as
bugged")

---

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

Kind ping on this patch.
Seems it wasn't noticed for months. Just check if it would be good to be merged or not proper for any reason?

I'll grab it for 6.5.

OK, thanks. Please check if these two are ready to go into 6.5 if possible:
https://lore.kernel.org/kvm/[email protected]/
https://lore.kernel.org/kvm/[email protected]/

reviewer Applied
#6 lore raw
In reply to Wei Wang
Quoted 5 lines · Wei Wang

Current KVM_BUG and KVM_BUG_ON assume that 'cond' passed from callers is 32-bit as it casts 'cond' to the type of int. This will be wrong if 'cond' provided by a caller is 64-bit, e.g. an error code of 0xc0000d0300000000 will be converted to 0, which is not expected.

Quoted 7 lines

Improves the implementation by using bool in KVM_BUG and KVM_BUG_ON. 'bool' is preferred to 'int' as __ret is essentially used as a boolean and coding-stytle.rst documents that use of bool is encouraged to improve readability and is often a better option than 'int' for storing boolean values.

[...]

Applied to kvm-x86 generic, thanks!

[1/1] KVM: allow KVM_BUG/KVM_BUG_ON to handle 64-bit cond

      https://github.com/kvm-x86/linux/commit/c9d601548603

--
https://github.com/kvm-x86/linux/tree/next
https://github.com/kvm-x86/linux/tree/fixes

#7 lore raw
Quoted 12 lines · Sean Christopherson
Quoted 11 lines · Wei Wang

Current KVM_BUG and KVM_BUG_ON assume that 'cond' passed from callers is 32-bit as it casts 'cond' to the type of int. This will be wrong if 'cond' provided by a caller is 64-bit, e.g. an error code of 0xc0000d0300000000 will be converted to 0, which is not expected.

Improves the implementation by using bool in KVM_BUG and KVM_BUG_ON. 'bool' is preferred to 'int' as __ret is essentially used as a boolean and coding-stytle.rst documents that use of bool is encouraged to improve readability and is often a better option than 'int' for storing boolean values.

Quoted 6 lines
Quoted 1 line

[...]

Applied to kvm-x86 generic, thanks!

[1/1] KVM: allow KVM_BUG/KVM_BUG_ON to handle 64-bit cond

      https://github.com/kvm-x86/linux/commit/c9d601548603

I guess this makes the !! in kvm_vm_ioctl_create_vcpu() unnecessary:

KVM_BUG_ON(!!xa_store(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, 0)...

Is it worth a patch (perhaps along with chopping off !! in kvm_msr_allowed() and few other places)?

#8 lore raw
Quoted 15 lines · Michal Luczaj
Quoted 14 lines · Sean Christopherson
Quoted 12 lines · Wei Wang

Current KVM_BUG and KVM_BUG_ON assume that 'cond' passed from callers is 32-bit as it casts 'cond' to the type of int. This will be wrong if 'cond' provided by a caller is 64-bit, e.g. an error code of 0xc0000d0300000000 will be converted to 0, which is not expected.

Improves the implementation by using bool in KVM_BUG and KVM_BUG_ON. 'bool' is preferred to 'int' as __ret is essentially used as a boolean and coding-stytle.rst documents that use of bool is encouraged to improve readability and is often a better option than 'int' for storing boolean values.

[...]

Quoted 8 lines
Quoted 4 lines

Applied to kvm-x86 generic, thanks!

[1/1] KVM: allow KVM_BUG/KVM_BUG_ON to handle 64-bit cond

      https://github.com/kvm-x86/linux/commit/c9d601548603

I guess this makes the !! in kvm_vm_ioctl_create_vcpu() unnecessary:

KVM_BUG_ON(!!xa_store(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, 0)...

Ya, I saw that, which in addition to Wei's ping, is what reminded me that the KVM_BUG_ON() fix hadn't been merged.

Quoted 2 lines

Is it worth a patch (perhaps along with chopping off !! in kvm_msr_allowed() and few other places)?

Yes, I think so.

#9 lore raw
Quoted 5 lines · Sean Christopherson
Quoted 3 lines · Michal Luczaj

I guess this makes the !! in kvm_vm_ioctl_create_vcpu() unnecessary:

KVM_BUG_ON(!!xa_store(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, 0)...

Quoted 7 lines

Ya, I saw that, which in addition to Wei's ping, is what reminded me that the KVM_BUG_ON() fix hadn't been merged.

Quoted 2 lines

Is it worth a patch (perhaps along with chopping off !! in kvm_msr_allowed() and few other places)?

Yes, I think so.

OK, so xa_store() aside[*], I see some bool-to-bools:

arch/x86/kvm/x86.c:

	kvm_msr_allowed():allowed = !!test_bit(index - start, bitmap);
arch/x86/kvm/hyperv.c:
	kvm_hv_hypercall():hc.rep = !!(hc.rep_cnt || hc.rep_idx);
arch/x86/kvm/mmu/mmu.c:
	update_pkru_bitmask():
		pkey_bits = !!check_pkey;
		pkey_bits |= (!!check_write) << 1;
arch/x86/kvm/svm/svm.c:
	msr_write_intercepted():return !!test_bit(bit_write,  &tmp);
	svm_vcpu_after_set_cpuid():
		2x set_msr_interception...
tools/testing/selftests/kvm/x86_64/vmx_exception_with_invalid_guest_state.c:
	set_or_clear_invalid_guest_state():sregs.tr.unusable = !!set;

But perhaps this is a matter of style and those were meant to be this kind-of explicit?

[*] https://lore.kernel.org/kvm/[email protected]/

#10 lore raw
Quoted 31 lines · Michal Luczaj
Quoted 5 lines · Sean Christopherson
Quoted 3 lines · Michal Luczaj

I guess this makes the !! in kvm_vm_ioctl_create_vcpu() unnecessary:

KVM_BUG_ON(!!xa_store(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, 0)...

Quoted 7 lines

Ya, I saw that, which in addition to Wei's ping, is what reminded me that the KVM_BUG_ON() fix hadn't been merged.

Quoted 2 lines

Is it worth a patch (perhaps along with chopping off !! in kvm_msr_allowed() and few other places)?

Yes, I think so.

OK, so xa_store() aside[*], I see some bool-to-bools:

arch/x86/kvm/x86.c:

	kvm_msr_allowed():allowed = !!test_bit(index - start, bitmap);
arch/x86/kvm/hyperv.c:
	kvm_hv_hypercall():hc.rep = !!(hc.rep_cnt || hc.rep_idx);
arch/x86/kvm/mmu/mmu.c:
	update_pkru_bitmask():
		pkey_bits = !!check_pkey;
		pkey_bits |= (!!check_write) << 1;
arch/x86/kvm/svm/svm.c:
	msr_write_intercepted():return !!test_bit(bit_write,  &tmp);
	svm_vcpu_after_set_cpuid():
		2x set_msr_interception...
tools/testing/selftests/kvm/x86_64/vmx_exception_with_invalid_guest_state.c:
	set_or_clear_invalid_guest_state():sregs.tr.unusable = !!set;
Quoted 2 lines

But perhaps this is a matter of style and those were meant to be this kind-of explicit?

I doubt it, I'm guessing most cases are due to the author being overzealous for one reason or another, e.g. I suspect the test_bit() ones are due to the original author incorrectly assuming test_bit() returned an unsigned long, i.e. the bit, as opposed to the bool.

If you want to clean these up, I'd say "fix" the test_bit() cases, but leave the others alone. The test_bit() ones are clearly redundant, and IMO can be actively due to implying test_bit() returns something other than a bool.

Quoted 1 line

[*] https://lore.kernel.org/kvm/[email protected]/

#11 lore raw
Quoted 21 lines · Sean Christopherson
Quoted 17 lines · Michal Luczaj

OK, so xa_store() aside[*], I see some bool-to-bools:

arch/x86/kvm/x86.c:

	kvm_msr_allowed():allowed = !!test_bit(index - start, bitmap);
arch/x86/kvm/hyperv.c:
	kvm_hv_hypercall():hc.rep = !!(hc.rep_cnt || hc.rep_idx);
arch/x86/kvm/mmu/mmu.c:
	update_pkru_bitmask():
		pkey_bits = !!check_pkey;
		pkey_bits |= (!!check_write) << 1;
arch/x86/kvm/svm/svm.c:
	msr_write_intercepted():return !!test_bit(bit_write,  &tmp);
	svm_vcpu_after_set_cpuid():
		2x set_msr_interception...
tools/testing/selftests/kvm/x86_64/vmx_exception_with_invalid_guest_state.c:
	set_or_clear_invalid_guest_state():sregs.tr.unusable = !!set;
Quoted 2 lines

But perhaps this is a matter of style and those were meant to be this kind-of explicit?

Quoted 8 lines

I doubt it, I'm guessing most cases are due to the author being overzealous for one reason or another, e.g. I suspect the test_bit() ones are due to the original author incorrectly assuming test_bit() returned an unsigned long, i.e. the bit, as opposed to the bool.

If you want to clean these up, I'd say "fix" the test_bit() cases, but leave the others alone. The test_bit() ones are clearly redundant, and IMO can be actively due to implying test_bit() returns something other than a bool.

Done: https://lore.kernel.org/kvm/[email protected]/