Skip to contentSkip to conversation

RISC-V: KVM: Fix guest page fault within HLV* instructions

PATCH In review [email protected] 2 participants 5 messages Sep 12, 2025 – Nov 6, 2025

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

PATCH 1/1 RISC-V: KVM: Fix guest page fault within HLV* instructions

From: Fangyu Yu <[email protected]>

When executing HLV* instructions at the HS mode, a guest page fault may occur when a g-stage page table migration between triggering the virtual instruction exception and executing the HLV* instruction.

This may be a corner case, and one simpler way to handle this is to re-execute the instruction where the virtual instruction exception occurred, and the guest page fault will be automatically handled.

Fixes 9f7013265112 ("RISC-V: KVM: Handle MMIO exits for VCPU")
---
 arch/riscv/kvm/vcpu_insn.c | 21 ++++++++++++++++++---
 1 file changed, 18 insertions(+), 3 deletions(-)
arch/riscv/kvm/vcpu_insn.c +18 -3
index 97dec18e6989..a8b93aa4d8ec 100644
@@ -448,7 +448,12 @@ int kvm_riscv_vcpu_virtual_insn(struct kvm_vcpu *vcpu, struct kvm_run *run,
 			insn = kvm_riscv_vcpu_unpriv_read(vcpu, true,
 							  ct->sepc,
 							  &utrap);
-			if (utrap.scause) {
+			switch (utrap.scause) {
+			case 0:
+				break;
+			case EXC_LOAD_GUEST_PAGE_FAULT:
+				return KVM_INSN_CONTINUE_SAME_SEPC;
+			default:
 				utrap.sepc = ct->sepc;
 				kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
 				return 1;
@@ -503,7 +508,12 @@ int kvm_riscv_vcpu_mmio_load(struct kvm_vcpu *vcpu, struct kvm_run *run,
 		 */
 		insn = kvm_riscv_vcpu_unpriv_read(vcpu, true, ct->sepc,
 						  &utrap);
-		if (utrap.scause) {
+		switch (utrap.scause) {
+		case 0:
+			break;
+		case EXC_LOAD_GUEST_PAGE_FAULT:
+			return KVM_INSN_CONTINUE_SAME_SEPC;
+		default:
 			/* Redirect trap if we failed to read instruction */
 			utrap.sepc = ct->sepc;
 			kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
@@ -629,7 +639,12 @@ int kvm_riscv_vcpu_mmio_store(struct kvm_vcpu *vcpu, struct kvm_run *run,
 		 */
 		insn = kvm_riscv_vcpu_unpriv_read(vcpu, true, ct->sepc,
 						  &utrap);
-		if (utrap.scause) {
+		switch (utrap.scause) {
+		case 0:
+			break;
+		case EXC_LOAD_GUEST_PAGE_FAULT:
+			return KVM_INSN_CONTINUE_SAME_SEPC;
+		default:
 			/* Redirect trap if we failed to read instruction */
 			utrap.sepc = ct->sepc;
 			kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
Signature · 1 line
2.49.0
#1 lore raw

PATCH 1/1 RISC-V: KVM: Fix guest page fault within HLV* instructions

Quoted 49 lines

From: Fangyu Yu <[email protected]>

When executing HLV* instructions at the HS mode, a guest page fault may occur when a g-stage page table migration between triggering the virtual instruction exception and executing the HLV* instruction.

This may be a corner case, and one simpler way to handle this is to re-execute the instruction where the virtual instruction exception occurred, and the guest page fault will be automatically handled.

Fixes 9f7013265112 ("RISC-V: KVM: Handle MMIO exits for VCPU")

---
arch/riscv/kvm/vcpu_insn.c | 21 ++++++++++++++++++---

1 file changed, 18 insertions(+), 3 deletions(-)
arch/riscv/kvm/vcpu_insn.c +12 -2
index 97dec18e6989..a8b93aa4d8ec 100644
@@ -448,7 +448,12 @@ int kvm_riscv_vcpu_virtual_insn(struct kvm_vcpu *vcpu, struct kvm_run *run,
 			insn = kvm_riscv_vcpu_unpriv_read(vcpu, true,
 							  ct->sepc,
 							  &utrap);
-			if (utrap.scause) {
+			switch (utrap.scause) {
+			case 0:
+				break;
+			case EXC_LOAD_GUEST_PAGE_FAULT:
+				return KVM_INSN_CONTINUE_SAME_SEPC;
+			default:
 				utrap.sepc = ct->sepc;
 				kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
 				return 1;
@@ -503,7 +508,12 @@ int kvm_riscv_vcpu_mmio_load(struct kvm_vcpu *vcpu, struct kvm_run *run,
 		 */
 		insn = kvm_riscv_vcpu_unpriv_read(vcpu, true, ct->sepc,
 						  &utrap);
-		if (utrap.scause) {
+		switch (utrap.scause) {
+		case 0:
+			break;
+		case EXC_LOAD_GUEST_PAGE_FAULT:
+			return KVM_INSN_CONTINUE_SAME_SEPC;
+		default:
 			/* Redirect trap if we failed to read instruction */
 			utrap.sepc = ct->sepc;
 			kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_mmio_store(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 645

Quoted 8 lines
arch/riscv/kvm/vcpu_insn.c +4 -1
@@ -629,7 +639,12 @@ int kvm_riscv_vcpu_mmio_store(struct kvm_vcpu *vcpu, struct kvm_run *run,
 		 */
 		insn = kvm_riscv_vcpu_unpriv_read(vcpu, true, ct->sepc,
 						  &utrap);
-		if (utrap.scause) {
+		switch (utrap.scause) {
+		case 0:
+			break;
+		case EXC_LOAD_GUEST_PAGE_FAULT:

Here should be EXC_STORE_GUEST_PAGE_FAULT, I will fix it next version.

Quoted 7 lines
arch/riscv/kvm/vcpu_insn.c +2 -0
+			return KVM_INSN_CONTINUE_SAME_SEPC;
+		default:
 			/* Redirect trap if we failed to read instruction */
 			utrap.sepc = ct->sepc;
 			kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
Signature · 1 line
2.49.0
Signature · 3 lines
kvm-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kvm-riscv
#2 lore raw
Quoted 51 lines
Quoted 51 lines

From: Fangyu Yu <[email protected]>

When executing HLV* instructions at the HS mode, a guest page fault may occur when a g-stage page table migration between triggering the virtual instruction exception and executing the HLV* instruction.

This may be a corner case, and one simpler way to handle this is to re-execute the instruction where the virtual instruction exception occurred, and the guest page fault will be automatically handled.

Fixes 9f7013265112 ("RISC-V: KVM: Handle MMIO exits for VCPU")

---
arch/riscv/kvm/vcpu_insn.c | 21 ++++++++++++++++++---

1 file changed, 18 insertions(+), 3 deletions(-)
arch/riscv/kvm/vcpu_insn.c +12 -2
index 97dec18e6989..a8b93aa4d8ec 100644
@@ -448,7 +448,12 @@ int kvm_riscv_vcpu_virtual_insn(struct kvm_vcpu *vcpu, struct kvm_run *run,
 			insn = kvm_riscv_vcpu_unpriv_read(vcpu, true,
 							  ct->sepc,
 							  &utrap);
-			if (utrap.scause) {
+			switch (utrap.scause) {
+			case 0:
+				break;
+			case EXC_LOAD_GUEST_PAGE_FAULT:
+				return KVM_INSN_CONTINUE_SAME_SEPC;
+			default:
 				utrap.sepc = ct->sepc;
 				kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
 				return 1;
@@ -503,7 +508,12 @@ int kvm_riscv_vcpu_mmio_load(struct kvm_vcpu *vcpu, struct kvm_run *run,
 		 */
 		insn = kvm_riscv_vcpu_unpriv_read(vcpu, true, ct->sepc,
 						  &utrap);
-		if (utrap.scause) {
+		switch (utrap.scause) {
+		case 0:
+			break;
+		case EXC_LOAD_GUEST_PAGE_FAULT:
+			return KVM_INSN_CONTINUE_SAME_SEPC;
+		default:
 			/* Redirect trap if we failed to read instruction */
 			utrap.sepc = ct->sepc;
 			kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
@@ -629,7 +639,12 @@ int kvm_riscv_vcpu_mmio_store(struct kvm_vcpu *vcpu, struct kvm_run *run,
 		 */
 		insn = kvm_riscv_vcpu_unpriv_read(vcpu, true, ct->sepc,
Quoted 8 lines

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_virtual_insn(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 454

Quoted 6 lines
arch/riscv/kvm/vcpu_insn.c +4 -1
 						  &utrap);
-		if (utrap.scause) {
+		switch (utrap.scause) {
+		case 0:
+			break;
+		case EXC_LOAD_GUEST_PAGE_FAULT:

Here should be EXC_STORE_GUEST_PAGE_FAULT, I will fix it next version.

Please ignore this comment, EXC_LOAD_GUEST_PAGE_FAULT is correct.

Quoted 9 lines
Quoted 8 lines
arch/riscv/kvm/vcpu_insn.c +2 -1
+			return KVM_INSN_CONTINUE_SAME_SEPC;
+		default:
 			/* Redirect trap if we failed to read instruction */
 			utrap.sepc = ct->sepc;
 			kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
--

2.49.0

maintainer
#3 lore raw
In reply to [email protected]
Quoted 22 lines · On Fri, Sep 12, 2025 at 7:13 PM <[email protected]> wrote:

From: Fangyu Yu <[email protected]>

When executing HLV* instructions at the HS mode, a guest page fault may occur when a g-stage page table migration between triggering the virtual instruction exception and executing the HLV* instruction.

This may be a corner case, and one simpler way to handle this is to re-execute the instruction where the virtual instruction exception occurred, and the guest page fault will be automatically handled.

Fixes 9f7013265112 ("RISC-V: KVM: Handle MMIO exits for VCPU")
---
 arch/riscv/kvm/vcpu_insn.c | 21 ++++++++++++++++++---
 1 file changed, 18 insertions(+), 3 deletions(-)
arch/riscv/kvm/vcpu_insn.c +0 -0
index 97dec18e6989..a8b93aa4d8ec 100644

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_virtual_insn(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 453

Quoted 7 lines
arch/riscv/kvm/vcpu_insn.c +3 -1
@@ -448,7 +448,12 @@ int kvm_riscv_vcpu_virtual_insn(struct kvm_vcpu *vcpu, struct kvm_run *run,
                        insn = kvm_riscv_vcpu_unpriv_read(vcpu, true,
                                                          ct->sepc,
                                                          &utrap);
-                       if (utrap.scause) {
+                       switch (utrap.scause) {
+                       case 0:
+                               break;

This looks like an unrelated change so drop it or send a separate patch with proper explanation.

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_virtual_insn(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 455

Quoted 2 lines
arch/riscv/kvm/vcpu_insn.c +2 -0
+                       case EXC_LOAD_GUEST_PAGE_FAULT:
+                               return KVM_INSN_CONTINUE_SAME_SEPC;

The KVM_INSN_xyz enum values are only for functions called via system_opcode_insn() so return 1 over here just like the below default case.

Also, add some comments over here about why we are simply continuing the guest.

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_mmio_load(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 513

Quoted 12 lines
arch/riscv/kvm/vcpu_insn.c +4 -1
+                       default:
                                utrap.sepc = ct->sepc;
                                kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
                                return 1;
@@ -503,7 +508,12 @@ int kvm_riscv_vcpu_mmio_load(struct kvm_vcpu *vcpu, struct kvm_run *run,
                 */
                insn = kvm_riscv_vcpu_unpriv_read(vcpu, true, ct->sepc,
                                                  &utrap);
-               if (utrap.scause) {
+               switch (utrap.scause) {
+               case 0:
+                       break;

This looks like an unrelated change so drop it or send a separate patch with proper explanation.

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_mmio_load(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 515

Quoted 2 lines
arch/riscv/kvm/vcpu_insn.c +2 -0
+               case EXC_LOAD_GUEST_PAGE_FAULT:
+                       return KVM_INSN_CONTINUE_SAME_SEPC;

Same comment as kvm_riscv_vcpu_virtual_insn().

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_mmio_store(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 644

Quoted 12 lines
arch/riscv/kvm/vcpu_insn.c +4 -1
+               default:
                        /* Redirect trap if we failed to read instruction */
                        utrap.sepc = ct->sepc;
                        kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
@@ -629,7 +639,12 @@ int kvm_riscv_vcpu_mmio_store(struct kvm_vcpu *vcpu, struct kvm_run *run,
                 */
                insn = kvm_riscv_vcpu_unpriv_read(vcpu, true, ct->sepc,
                                                  &utrap);
-               if (utrap.scause) {
+               switch (utrap.scause) {
+               case 0:
+                       break;

This looks like an unrelated change so drop it or send a separate patch with proper explanation.

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_mmio_store(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 646

Quoted 2 lines
arch/riscv/kvm/vcpu_insn.c +2 -0
+               case EXC_LOAD_GUEST_PAGE_FAULT:
+                       return KVM_INSN_CONTINUE_SAME_SEPC;

Same comment as kvm_riscv_vcpu_virtual_insn().

Quoted 8 lines
arch/riscv/kvm/vcpu_insn.c +1 -1
+               default:
                        /* Redirect trap if we failed to read instruction */
                        utrap.sepc = ct->sepc;
                        kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
--

2.49.0

Regards,
Anup

#4 lore raw
Quoted 93 lines
Quoted 21 lines

From: Fangyu Yu <[email protected]>

When executing HLV* instructions at the HS mode, a guest page fault may occur when a g-stage page table migration between triggering the virtual instruction exception and executing the HLV* instruction.

This may be a corner case, and one simpler way to handle this is to re-execute the instruction where the virtual instruction exception occurred, and the guest page fault will be automatically handled.

Fixes 9f7013265112 ("RISC-V: KVM: Handle MMIO exits for VCPU")
---
 arch/riscv/kvm/vcpu_insn.c | 21 ++++++++++++++++++---
 1 file changed, 18 insertions(+), 3 deletions(-)
arch/riscv/kvm/vcpu_insn.c +0 -0
index 97dec18e6989..a8b93aa4d8ec 100644

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_virtual_insn(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 453

Quoted 7 lines
arch/riscv/kvm/vcpu_insn.c +3 -1
@@ -448,7 +448,12 @@ int kvm_riscv_vcpu_virtual_insn(struct kvm_vcpu *vcpu, struct kvm_run *run,
                        insn = kvm_riscv_vcpu_unpriv_read(vcpu, true,
                                                          ct->sepc,
                                                          &utrap);
-                       if (utrap.scause) {
+                       switch (utrap.scause) {
+                       case 0:
+                               break;

This looks like an unrelated change so drop it or send a separate patch with proper explanation.

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_virtual_insn(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 455

Quoted 2 lines
arch/riscv/kvm/vcpu_insn.c +2 -0
+                       case EXC_LOAD_GUEST_PAGE_FAULT:
+                               return KVM_INSN_CONTINUE_SAME_SEPC;

The KVM_INSN_xyz enum values are only for functions called via system_opcode_insn() so return 1 over here just like the below default case.

Also, add some comments over here about why we are simply continuing the guest.

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_mmio_load(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 513

Quoted 12 lines
arch/riscv/kvm/vcpu_insn.c +4 -1
+                       default:
                                utrap.sepc = ct->sepc;
                                kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
                                return 1;
@@ -503,7 +508,12 @@ int kvm_riscv_vcpu_mmio_load(struct kvm_vcpu *vcpu, struct kvm_run *run,
                 */
                insn = kvm_riscv_vcpu_unpriv_read(vcpu, true, ct->sepc,
                                                  &utrap);
-               if (utrap.scause) {
+               switch (utrap.scause) {
+               case 0:
+                       break;

This looks like an unrelated change so drop it or send a separate patch with proper explanation.

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_mmio_load(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 515

Quoted 2 lines
arch/riscv/kvm/vcpu_insn.c +2 -0
+               case EXC_LOAD_GUEST_PAGE_FAULT:
+                       return KVM_INSN_CONTINUE_SAME_SEPC;

Same comment as kvm_riscv_vcpu_virtual_insn().

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_mmio_store(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 644

Quoted 12 lines
arch/riscv/kvm/vcpu_insn.c +4 -1
+               default:
                        /* Redirect trap if we failed to read instruction */
                        utrap.sepc = ct->sepc;
                        kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
@@ -629,7 +639,12 @@ int kvm_riscv_vcpu_mmio_store(struct kvm_vcpu *vcpu, struct kvm_run *run,
                 */
                insn = kvm_riscv_vcpu_unpriv_read(vcpu, true, ct->sepc,
                                                  &utrap);
-               if (utrap.scause) {
+               switch (utrap.scause) {
+               case 0:
+                       break;

This looks like an unrelated change so drop it or send a separate patch with proper explanation.

arch/riscv/kvm/vcpu_insn.c int kvm_riscv_vcpu_mmio_store(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 646

Quoted 2 lines
arch/riscv/kvm/vcpu_insn.c +2 -0
+               case EXC_LOAD_GUEST_PAGE_FAULT:
+                       return KVM_INSN_CONTINUE_SAME_SEPC;

Same comment as kvm_riscv_vcpu_virtual_insn().

Quoted 8 lines
arch/riscv/kvm/vcpu_insn.c +1 -1
+               default:
                        /* Redirect trap if we failed to read instruction */
                        utrap.sepc = ct->sepc;
                        kvm_riscv_vcpu_trap_redirect(vcpu, &utrap);
--

2.49.0

Quoted 2 lines

Regards,
Anup

Hi Anup:

Thanks for the review.
I will follow your suggestions in the next version.

Thanks,
Fangyu