Skip to contentSkip to conversation

KVM: MMU: Use 'INVALID_GPA' and 'INVALID_GFN' properly

PATCH In review Yu Zhang 5 participants 14 messages Dec 16, 2022 – Dec 22, 2022

Subsystems (7)
x86 KVM Kselftest arm64 +3 more subsystems
lore
Share & tools
lore

Yu Zhang

author
#0 lore raw

v4 cover KVM: MMU: Use 'INVALID_GPA' and 'INVALID_GFN' properly

Pacth 1 adds a new definition, 'INVALID_GFN', so that GFN values can use it, instead of the 'GPA_INVALID'.

Pacth 2 makes the definition of 'INVALID_GPA' shared by different archs.

Both are tested by rebuilding KVM for x86 and for ARM64.

---
V4:
- Put the addition of 'INVALID_GFN' into a seperate patch. V3:
- Added 'INVALID_GFN' and use it.
v2:
- Renamed 'GPA_INVALID' to 'INVALID_GPA' and modify _those_ users. v1:
https://lore.kernel.org/lkml/[email protected]/

Yu Zhang (2):
KVM: MMU: Introduce 'INVALID_GFN' and use it for GFN values
KVM: MMU: Make the definition of 'INVALID_GPA' common

 arch/arm64/include/asm/kvm_host.h                  |  4 ++--
 arch/arm64/kvm/hypercalls.c                        |  2 +-
 arch/arm64/kvm/pvtime.c                            |  8 ++++----
 arch/x86/include/asm/kvm_host.h                    |  2 --
 arch/x86/kvm/xen.c                                 | 14 +++++++-------
 include/linux/kvm_types.h                          |  3 ++-
 .../testing/selftests/kvm/x86_64/xen_shinfo_test.c |  4 ++--
 7 files changed, 18 insertions(+), 19 deletions(-)
Signature · 1 line
2.25.1
author
#1 lore raw

PATCH v4 1/2 KVM: MMU: Introduce 'INVALID_GFN' and use it for GFN values

Currently, KVM xen and its shared info selftest code uses 'GPA_INVALID' for GFN values, but actually it is more accurate to use the name 'INVALID_GFN'. So just add a new definition and use it.

No functional changes intended.

---
 arch/x86/kvm/xen.c                                   | 4 ++--
 include/linux/kvm_types.h                            | 1 +
 tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c | 4 ++--
 3 files changed, 5 insertions(+), 4 deletions(-)
arch/x86/kvm/xen.c +2 -2
index d7af40240248..6908a74ab303 100644
@@ -41,7 +41,7 @@ static int kvm_xen_shared_info_init(struct kvm *kvm, gfn_t gfn)
 	int ret = 0;
 	int idx = srcu_read_lock(&kvm->srcu);
 
-	if (gfn == GPA_INVALID) {
+	if (gfn == INVALID_GFN) {
 		kvm_gpc_deactivate(gpc);
 		goto out;
 	}
@@ -659,7 +659,7 @@ int kvm_xen_hvm_get_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data)
 		if (kvm->arch.xen.shinfo_cache.active)
 			data->u.shared_info.gfn = gpa_to_gfn(kvm->arch.xen.shinfo_cache.gpa);
 		else
-			data->u.shared_info.gfn = GPA_INVALID;
+			data->u.shared_info.gfn = INVALID_GFN;
 		r = 0;
 		break;
 
include/linux/kvm_types.h +1 -0
index 76de36e56cdf..d21c0d7fee31 100644
@@ -41,6 +41,7 @@ typedef u64            gpa_t;
 typedef u64            gfn_t;
 
 #define GPA_INVALID	(~(gpa_t)0)
+#define INVALID_GFN	(~(gfn_t)0)
 
 typedef unsigned long  hva_t;
 typedef u64            hpa_t;
tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c +2 -2
index 721f6a693799..d65a23be88b1 100644
@@ -20,7 +20,7 @@
 #include <sys/eventfd.h>
 
 /* Defined in include/linux/kvm_types.h */
-#define GPA_INVALID		(~(ulong)0)
+#define INVALID_GFN		(~(ulong)0)
 
 #define SHINFO_REGION_GVA	0xc0000000ULL
 #define SHINFO_REGION_GPA	0xc0000000ULL
@@ -419,7 +419,7 @@ static void *juggle_shinfo_state(void *arg)
 
 	struct kvm_xen_hvm_attr cache_destroy = {
 		.type = KVM_XEN_ATTR_TYPE_SHARED_INFO,
-		.u.shared_info.gfn = GPA_INVALID
+		.u.shared_info.gfn = INVALID_GFN
 	};
 
 	for (;;) {
Signature · 1 line
2.25.1
#2 lore raw

Re: [PATCH v4 1/2] KVM: MMU: Introduce 'INVALID_GFN' and use it for GFN values

tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c line 23

Quoted 7 lines · Yu Zhang

+++ b/tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c

tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c +1 -1
@@ -20,7 +20,7 @@
 #include <sys/eventfd.h>
 
 /* Defined in include/linux/kvm_types.h */
-#define GPA_INVALID		(~(ulong)0)
+#define INVALID_GFN		(~(ulong)0)

Thank you for fixing the selftest!

Regarding xen_shinfo_test.c, a question to maintainers, would it be ok if I submit a simple patch fixing the misnamed cache_init/cache_destroy => cache_activate/cache_deactivate or it's just not worth the churn now?

Michal

#3 lore raw

Re: [PATCH v4 1/2] KVM: MMU: Introduce 'INVALID_GFN' and use it for GFN values

Quoted 12 lines · Michal Luczaj

tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c line 23

Quoted 7 lines · Yu Zhang

+++ b/tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c

tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c +1 -1
@@ -20,7 +20,7 @@
 #include <sys/eventfd.h>
 
 /* Defined in include/linux/kvm_types.h */
-#define GPA_INVALID		(~(ulong)0)
+#define INVALID_GFN		(~(ulong)0)

Thank you for fixing the selftest!

Quoted 3 lines

Regarding xen_shinfo_test.c, a question to maintainers, would it be ok if I submit a simple patch fixing the misnamed cache_init/cache_destroy => cache_activate/cache_deactivate or it's just not worth the churn now?

It's not too much churn. My only hesitation would be that chasing KVM names is usually a fruitless endeavor, but in this case I agree (de)activate is better terminology even if KVM changes again in the future.

#4 lore raw

Re: [PATCH v4 1/2] KVM: MMU: Introduce 'INVALID_GFN' and use it for GFN values

Quoted 13 lines · Sean Christopherson
Quoted 12 lines · Michal Luczaj

tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c line 23

Quoted 7 lines · Yu Zhang

+++ b/tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c

tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c +1 -1
@@ -20,7 +20,7 @@
 #include <sys/eventfd.h>
 
 /* Defined in include/linux/kvm_types.h */
-#define GPA_INVALID		(~(ulong)0)
+#define INVALID_GFN		(~(ulong)0)

Thank you for fixing the selftest!

Quoted 7 lines
Quoted 3 lines

Regarding xen_shinfo_test.c, a question to maintainers, would it be ok if I submit a simple patch fixing the misnamed cache_init/cache_destroy => cache_activate/cache_deactivate or it's just not worth the churn now?

It's not too much churn. My only hesitation would be that chasing KVM names is usually a fruitless endeavor, but in this case I agree (de)activate is better terminology even if KVM changes again in the future.

All right, I'll send the fix after Yu's patches land in queue.

Thanks,
Michal

#5 lore raw

Re: [PATCH v4 1/2] KVM: MMU: Introduce 'INVALID_GFN' and use it for GFN values

In reply to Yu Zhang
Quoted 20 lines · Yu Zhang

Currently, KVM xen and its shared info selftest code uses 'GPA_INVALID' for GFN values, but actually it is more accurate to use the name 'INVALID_GFN'. So just add a new definition and use it.

No functional changes intended.

---
 arch/x86/kvm/xen.c                                   | 4 ++--
 include/linux/kvm_types.h                            | 1 +
 tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c | 4 ++--
 3 files changed, 5 insertions(+), 4 deletions(-)
arch/x86/kvm/xen.c +0 -0
index d7af40240248..6908a74ab303 100644

arch/x86/kvm/xen.c static int kvm_xen_shared_info_init(struct kvm *kvm, gfn_t gfn) · line 44

Quoted 5 lines
arch/x86/kvm/xen.c +1 -1
@@ -41,7 +41,7 @@ static int kvm_xen_shared_info_init(struct kvm *kvm, gfn_t gfn)
 	int ret = 0;
 	int idx = srcu_read_lock(&kvm->srcu);
 
-	if (gfn == GPA_INVALID) {
+	if (gfn == INVALID_GFN) {

Grrr! This magic value is ABI, as "gfn == -1" yields different behavior than a random, garbage gfn.

So, sadly, we can't simply introduce INVALID_GFN here, and instead need to do something like:

include/uapi/linux/kvm.h +1 -0
index 20522d4ba1e0..2d31caaf812c 100644
@@ -1766,6 +1766,7 @@ struct kvm_xen_hvm_attr {
                __u8 vector;
                __u8 runstate_update_flag;
                struct {
+#define KVM_XEN_INVALID_GFN    (~0ull)
                        __u64 gfn;
                } shared_info;
                struct {
Quoted 3 lines
arch/x86/kvm/xen.c +0 -0
 		kvm_gpc_deactivate(gpc);
 		goto out;
 	}
author
#6 lore raw

Re: [PATCH v4 1/2] KVM: MMU: Introduce 'INVALID_GFN' and use it for GFN values

Quoted 24 lines · Sean Christopherson
Quoted 23 lines · Yu Zhang

Currently, KVM xen and its shared info selftest code uses 'GPA_INVALID' for GFN values, but actually it is more accurate to use the name 'INVALID_GFN'. So just add a new definition and use it.

No functional changes intended.

---
 arch/x86/kvm/xen.c                                   | 4 ++--
 include/linux/kvm_types.h                            | 1 +
 tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c | 4 ++--
 3 files changed, 5 insertions(+), 4 deletions(-)
arch/x86/kvm/xen.c +0 -0
index d7af40240248..6908a74ab303 100644
@@ -41,7 +41,7 @@ static int kvm_xen_shared_info_init(struct kvm *kvm, gfn_t gfn)
 	int ret = 0;
 	int idx = srcu_read_lock(&kvm->srcu);
 
Quoted 5 lines

arch/x86/kvm/xen.c static int kvm_xen_shared_info_init(struct kvm *kvm, gfn_t gfn) · line 44

Quoted 2 lines
arch/x86/kvm/xen.c +1 -1
-	if (gfn == GPA_INVALID) {
+	if (gfn == INVALID_GFN) {

Grrr! This magic value is ABI, as "gfn == -1" yields different behavior than a random, garbage gfn.

Thanks Sean. But I do not get it.
May I ask why ABI usages are different? Or is there any documentation describing the requirement? Thanks!

Quoted 9 lines

So, sadly, we can't simply introduce INVALID_GFN here, and instead need to do something like:

include/uapi/linux/kvm.h +0 -0
index 20522d4ba1e0..2d31caaf812c 100644

include/uapi/linux/kvm.h struct kvm_xen_hvm_attr { · line 1771

Quoted 6 lines
include/uapi/linux/kvm.h +1 -0
@@ -1766,6 +1766,7 @@ struct kvm_xen_hvm_attr {
                __u8 vector;
                __u8 runstate_update_flag;
                struct {
+#define KVM_XEN_INVALID_GFN    (~0ull)
                        __u64 gfn;
                } shared_info;

I guess above policy shall also be applied for the gpa inside struct kvm_xen_vcpu_attr. Instead of using INVALID_GPA (in patch 2), should be like:

include/uapi/linux/kvm.h +1 -0
index 61c052d51a64..c06ef8ed9680 100644
@@ -1823,6 +1823,7 @@ struct kvm_xen_vcpu_attr {
        __u16 type;
        __u16 pad[3];
        union {
+#define KVM_XEN_INVALID_GPA            (~0ull)
                __u64 gpa;
                __u64 pad[8];
                struct {

Also, xen.c should use KVM_XEN_INVALID_GPA for GPA values...

B.R.
Yu

maintainer
#7 lore raw

Re: [PATCH v4 1/2] KVM: MMU: Introduce 'INVALID_GFN' and use it for GFN values

In reply to Sean Christopherson
Quoted 16 lines · Sean Christopherson
Quoted 15 lines · Yu Zhang

Currently, KVM xen and its shared info selftest code uses 'GPA_INVALID' for GFN values, but actually it is more accurate to use the name 'INVALID_GFN'. So just add a new definition and use it.

No functional changes intended.

---
 arch/x86/kvm/xen.c                                   | 4 ++-- include/linux/kvm_types.h                            | 1 + tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c | 4 ++--

 3 files changed, 5 insertions(+), 4 deletions(-)
Quoted 17 lines
Quoted 10 lines
arch/x86/kvm/xen.c +0 -0
index d7af40240248..6908a74ab303 100644
@@ -41,7 +41,7 @@ static int kvm_xen_shared_info_init(struct kvm *kvm, gfn_t gfn)
         int ret = 0;
         int idx = srcu_read_lock(&kvm->srcu);


-       if (gfn == GPA_INVALID) {
+       if (gfn == INVALID_GFN) {

Grrr!  This magic value is ABI, as "gfn == -1" yields different behavior than a random, garbage gfn.

So, sadly, we can't simply introduce INVALID_GFN here, and instead need to do something like:

Well... you can still use INVALID_GFN as long as its value remains the same ((uint64_t)-1).

But yes, the KVM API differs here from Xen because Xen only allows a guest to *set* these (and later they invented SHUTDOWN_soft_reset). While KVM lets the userspace VMM handle soft reset, and needs to allow them to be *unset*. And since zero is a valid GPA/GFN, -1 is a reasonable value for 'INVALID'. But I do think that detail escaped the documentation and the uapi headers, so your suggestion below is a good one, although strictly we need a GPA one too.

Quoted 16 lines
include/uapi/linux/kvm.h +1 -0
index 20522d4ba1e0..2d31caaf812c 100644
@@ -1766,6 +1766,7 @@ struct kvm_xen_hvm_attr {
                 __u8 vector;
                 __u8 runstate_update_flag;
                 struct {
+#define KVM_XEN_INVALID_GFN    (~0ull)
                         __u64 gfn;
                 } shared_info;
                 struct {
Quoted 4 lines

                kvm_gpc_deactivate(gpc);
                goto out;
        }

#8 lore raw

Re: [PATCH v4 1/2] KVM: MMU: Introduce 'INVALID_GFN' and use it for GFN values

Quoted 35 lines · David Woodhouse
Quoted 16 lines · Sean Christopherson
Quoted 15 lines · Yu Zhang

Currently, KVM xen and its shared info selftest code uses 'GPA_INVALID' for GFN values, but actually it is more accurate to use the name 'INVALID_GFN'. So just add a new definition and use it.

No functional changes intended.

---
 arch/x86/kvm/xen.c                                   | 4 ++-- include/linux/kvm_types.h                            | 1 + tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c | 4 ++--

 3 files changed, 5 insertions(+), 4 deletions(-)
Quoted 17 lines
Quoted 10 lines
arch/x86/kvm/xen.c +0 -0
index d7af40240248..6908a74ab303 100644
@@ -41,7 +41,7 @@ static int kvm_xen_shared_info_init(struct kvm *kvm, gfn_t gfn)
         int ret = 0;
         int idx = srcu_read_lock(&kvm->srcu);


-       if (gfn == GPA_INVALID) {
+       if (gfn == INVALID_GFN) {

Grrr!  This magic value is ABI, as "gfn == -1" yields different behavior than a random, garbage gfn.

So, sadly, we can't simply introduce INVALID_GFN here, and instead need to do something like:

Quoted 8 lines

Well... you can still use INVALID_GFN as long as its value remains the same ((uint64_t)-1).

But yes, the KVM API differs here from Xen because Xen only allows a guest to *set* these (and later they invented SHUTDOWN_soft_reset). While KVM lets the userspace VMM handle soft reset, and needs to allow them to be *unset*. And since zero is a valid GPA/GFN, -1 is a reasonable value for 'INVALID'.

Oh, yeah, I'm not arguing against using '-1', just calling out that there is special meaning given to '-1' and so it needs to be formalized so that KVM doesn't accidentally break userspace.

Quoted 2 lines

But I do think that detail escaped the documentation and the uapi headers, so your suggestion below is a good one, although strictly we need a GPA one too.

Ah, right, for struct kvm_xen_vcpu_attr.

maintainer
#9 lore raw

Re: [PATCH v4 1/2] KVM: MMU: Introduce 'INVALID_GFN' and use it for GFN values

Quoted 45 lines · Sean Christopherson
Quoted 35 lines · David Woodhouse
Quoted 16 lines · Sean Christopherson
Quoted 15 lines · Yu Zhang

Currently, KVM xen and its shared info selftest code uses 'GPA_INVALID' for GFN values, but actually it is more accurate to use the name 'INVALID_GFN'. So just add a new definition and use it.

No functional changes intended.

---
 arch/x86/kvm/xen.c                                   | 4 ++-- include/linux/kvm_types.h                            | 1 + tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c | 4 ++--

 3 files changed, 5 insertions(+), 4 deletions(-)
Quoted 17 lines
Quoted 10 lines
arch/x86/kvm/xen.c +0 -0
index d7af40240248..6908a74ab303 100644
@@ -41,7 +41,7 @@ static int kvm_xen_shared_info_init(struct kvm *kvm, gfn_t gfn)
         int ret = 0;
         int idx = srcu_read_lock(&kvm->srcu);


-       if (gfn == GPA_INVALID) {
+       if (gfn == INVALID_GFN) {

Grrr!  This magic value is ABI, as "gfn == -1" yields different behavior than a random, garbage gfn.

So, sadly, we can't simply introduce INVALID_GFN here, and instead need to do something like:

Quoted 8 lines

Well... you can still use INVALID_GFN as long as its value remains the same ((uint64_t)-1).

But yes, the KVM API differs here from Xen because Xen only allows a guest to *set* these (and later they invented SHUTDOWN_soft_reset). While KVM lets the userspace VMM handle soft reset, and needs to allow them to be *unset*. And since zero is a valid GPA/GFN, -1 is a reasonable value for 'INVALID'.

Quoted 3 lines

Oh, yeah, I'm not arguing against using '-1', just calling out that there is special meaning given to '-1' and so it needs to be formalized so that KVM doesn't accidentally break userspace.

Indeed. I should make sure the xen_shinfo_test covers it too. We had been fairly diligent about selftests for all this, as I *hadn't* yet got round to posting the updated qemu support to go with it

Will update the docs and test accordingly. I have a couple of other minor doc fixes which I spotted as I was doing the qemu support. If nobody beats me to the uapi header part, I'll do that too. But I'm not *scheduled* to be in front of a real computer until next year now, and and time I do steal is likely to be spent on qemu itself.

#10 lore raw

Re: [PATCH v4 1/2] KVM: MMU: Introduce 'INVALID_GFN' and use it for GFN values

Quoted 5 lines · David Woodhouse

Will update the docs and test accordingly. I have a couple of other minor doc fixes which I spotted as I was doing the qemu support. If nobody beats me to the uapi header part, I'll do that too. But I'm not *scheduled* to be in front of a real computer until next year now, and and time I do steal is likely to be spent on qemu itself.

No rush, I'm about to disappear until next year too. I'm guessing Paolo will be fairly inactive next week too.

author
#11 lore raw

PATCH v4 2/2 KVM: MMU: Make the definition of 'INVALID_GPA' common

In reply to Yu Zhang

KVM already has a 'GPA_INVALID' defined as (~(gpa_t)0) in kvm_types.h, and it is used by ARM and X86 xen code. We do not need a specific definition of 'INVALID_GPA' for X86.

Instead of using the common 'GPA_INVALID' for X86, replace the definition of 'GPA_INVALID' with 'INVALID_GPA', and change the users of 'GPA_INVALID', so that the diff can be smaller. Also because the name 'INVALID_GPA' tells the user we are using an invalid GPA, while the name 'GPA_INVALID' is emphasizing the GPA is an invalid one.

Also, add definition of 'INVALID_GFN' because it is more proper than 'INVALID_GPA' for GFN variables.

No functional change intended.

---
 arch/arm64/include/asm/kvm_host.h |  4 ++--
 arch/arm64/kvm/hypercalls.c       |  2 +-
 arch/arm64/kvm/pvtime.c           |  8 ++++----
 arch/x86/include/asm/kvm_host.h   |  2 --
 arch/x86/kvm/xen.c                | 10 +++++-----
 include/linux/kvm_types.h         |  2 +-
 6 files changed, 13 insertions(+), 15 deletions(-)
arch/arm64/include/asm/kvm_host.h +2 -2
index 001c8abe87fc..fcf96e9cc8cd 100644
@@ -906,12 +906,12 @@ void kvm_arm_vmid_clear_active(void);
 
 static inline void kvm_arm_pvtime_vcpu_init(struct kvm_vcpu_arch *vcpu_arch)
 {
-	vcpu_arch->steal.base = GPA_INVALID;
+	vcpu_arch->steal.base = INVALID_GPA;
 }
 
 static inline bool kvm_arm_is_pvtime_enabled(struct kvm_vcpu_arch *vcpu_arch)
 {
-	return (vcpu_arch->steal.base != GPA_INVALID);
+	return (vcpu_arch->steal.base != INVALID_GPA);
 }
 
 void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome);
arch/arm64/kvm/hypercalls.c +1 -1
index c9f401fa01a9..64c086c02c60 100644
@@ -198,7 +198,7 @@ int kvm_hvc_call_handler(struct kvm_vcpu *vcpu)
 		break;
 	case ARM_SMCCC_HV_PV_TIME_ST:
 		gpa = kvm_init_stolen_time(vcpu);
-		if (gpa != GPA_INVALID)
+		if (gpa != INVALID_GPA)
 			val[0] = gpa;
 		break;
 	case ARM_SMCCC_VENDOR_HYP_CALL_UID_FUNC_ID:
arch/arm64/kvm/pvtime.c +4 -4
index 78a09f7a6637..4ceabaa4c30b 100644
@@ -19,7 +19,7 @@ void kvm_update_stolen_time(struct kvm_vcpu *vcpu)
 	u64 steal = 0;
 	int idx;
 
-	if (base == GPA_INVALID)
+	if (base == INVALID_GPA)
 		return;
 
 	idx = srcu_read_lock(&kvm->srcu);
@@ -40,7 +40,7 @@ long kvm_hypercall_pv_features(struct kvm_vcpu *vcpu)
 	switch (feature) {
 	case ARM_SMCCC_HV_PV_TIME_FEATURES:
 	case ARM_SMCCC_HV_PV_TIME_ST:
-		if (vcpu->arch.steal.base != GPA_INVALID)
+		if (vcpu->arch.steal.base != INVALID_GPA)
 			val = SMCCC_RET_SUCCESS;
 		break;
 	}
@@ -54,7 +54,7 @@ gpa_t kvm_init_stolen_time(struct kvm_vcpu *vcpu)
 	struct kvm *kvm = vcpu->kvm;
 	u64 base = vcpu->arch.steal.base;
 
-	if (base == GPA_INVALID)
+	if (base == INVALID_GPA)
 		return base;
 
 	/*
@@ -89,7 +89,7 @@ int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu,
 		return -EFAULT;
 	if (!IS_ALIGNED(ipa, 64))
 		return -EINVAL;
-	if (vcpu->arch.steal.base != GPA_INVALID)
+	if (vcpu->arch.steal.base != INVALID_GPA)
 		return -EEXIST;
 
 	/* Check the address is in a valid memslot */
arch/x86/include/asm/kvm_host.h +0 -2
index f35f1ff4427b..46e50cb6c9ca 100644
@@ -134,8 +134,6 @@
 #define INVALID_PAGE (~(hpa_t)0)
 #define VALID_PAGE(x) ((x) != INVALID_PAGE)
 
-#define INVALID_GPA (~(gpa_t)0)
-
 /* KVM Hugepage definitions for x86 */
 #define KVM_MAX_HUGEPAGE_LEVEL	PG_LEVEL_1G
 #define KVM_NR_PAGE_SIZES	(KVM_MAX_HUGEPAGE_LEVEL - PG_LEVEL_4K + 1)
arch/x86/kvm/xen.c +5 -5
index 6908a74ab303..ec893276681c 100644
@@ -705,7 +705,7 @@ int kvm_xen_vcpu_set_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
 		BUILD_BUG_ON(offsetof(struct vcpu_info, time) !=
 			     offsetof(struct compat_vcpu_info, time));
 
-		if (data->u.gpa == GPA_INVALID) {
+		if (data->u.gpa == INVALID_GPA) {
 			kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_info_cache);
 			r = 0;
 			break;
@@ -719,7 +719,7 @@ int kvm_xen_vcpu_set_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
 		break;
 
 	case KVM_XEN_VCPU_ATTR_TYPE_VCPU_TIME_INFO:
-		if (data->u.gpa == GPA_INVALID) {
+		if (data->u.gpa == INVALID_GPA) {
 			kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_time_info_cache);
 			r = 0;
 			break;
@@ -739,7 +739,7 @@ int kvm_xen_vcpu_set_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
 			r = -EOPNOTSUPP;
 			break;
 		}
-		if (data->u.gpa == GPA_INVALID) {
+		if (data->u.gpa == INVALID_GPA) {
 			r = 0;
 		deactivate_out:
 			kvm_gpc_deactivate(&vcpu->arch.xen.runstate_cache);
@@ -937,7 +937,7 @@ int kvm_xen_vcpu_get_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
 		if (vcpu->arch.xen.vcpu_info_cache.active)
 			data->u.gpa = vcpu->arch.xen.vcpu_info_cache.gpa;
 		else
-			data->u.gpa = GPA_INVALID;
+			data->u.gpa = INVALID_GPA;
 		r = 0;
 		break;
 
@@ -945,7 +945,7 @@ int kvm_xen_vcpu_get_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
 		if (vcpu->arch.xen.vcpu_time_info_cache.active)
 			data->u.gpa = vcpu->arch.xen.vcpu_time_info_cache.gpa;
 		else
-			data->u.gpa = GPA_INVALID;
+			data->u.gpa = INVALID_GPA;
 		r = 0;
 		break;
 
include/linux/kvm_types.h +1 -1
index d21c0d7fee31..b961043e664a 100644
@@ -40,7 +40,7 @@ typedef unsigned long  gva_t;
 typedef u64            gpa_t;
 typedef u64            gfn_t;
 
-#define GPA_INVALID	(~(gpa_t)0)
+#define INVALID_GPA	(~(gpa_t)0)
 #define INVALID_GFN	(~(gfn_t)0)
 
 typedef unsigned long  hva_t;
Signature · 1 line
2.25.1
maintainer Reviewed-by on v4 2/2
#12 lore raw

Re: [PATCH v4 2/2] KVM: MMU: Make the definition of 'INVALID_GPA' common

Quoted 11 lines · Yu Zhang

KVM already has a 'GPA_INVALID' defined as (~(gpa_t)0) in kvm_types.h, and it is used by ARM and X86 xen code. We do not need a specific definition of 'INVALID_GPA' for X86.

Instead of using the common 'GPA_INVALID' for X86, replace the definition of 'GPA_INVALID' with 'INVALID_GPA', and change the users of 'GPA_INVALID', so that the diff can be smaller. Also because the name 'INVALID_GPA' tells the user we are using an invalid GPA, while the name 'GPA_INVALID' is emphasizing the GPA is an invalid one.

Quoted 2 lines

Also, add definition of 'INVALID_GFN' because it is more proper than 'INVALID_GPA' for GFN variables.

Please wrap closer to ~75 chars, ~60 is too aggressive.

Quoted 5 lines

No functional change intended.

---

#13 lore raw

Re: [PATCH v4 2/2] KVM: MMU: Make the definition of 'INVALID_GPA' common

In reply to Yu Zhang
Quoted 2 lines · Yu Zhang

Also, add definition of 'INVALID_GFN' because it is more proper than 'INVALID_GPA' for GFN variables.

This perhaps is a leftover, since it belongs to patch 1 :)