Skip to contentSkip to conversation

RISC-V: KVM: fix the Zicbo[m|z|p] block sizes in GET_ONE_REG

PATCH In review Yong-Xuan Wang 2 participants 3 messages Sep 24

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

PATCH 1/1 RISC-V: KVM: fix the Zicbo[m|z|p] block sizes in GET_ONE_REG

The kvm_riscv_vcpu_get_reg_config() should return the block size when the host supports the corresponding extensions.

Fixes 298276da73ca ("RISC-V: KVM: Zicbo[m|z|p] block sizes should be always present in ONE_REG")
---
 arch/riscv/kvm/vcpu_onereg.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
arch/riscv/kvm/vcpu_onereg.c +3 -3
index 877ba823fc74..f2b125569449 100644
@@ -50,13 +50,13 @@ static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu,
 		reg_val = vcpu->arch.isa[0] & KVM_RISCV_BASE_ISA_MASK;
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(zicbom_block_size):
-		reg_val = (kvm_riscv_isa_check_host(ZICBOM)) ? 0 : riscv_cbom_block_size;
+		reg_val = (kvm_riscv_isa_check_host(ZICBOM)) ? riscv_cbom_block_size : 0;
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(zicboz_block_size):
-		reg_val = (kvm_riscv_isa_check_host(ZICBOZ)) ? 0 : riscv_cboz_block_size;
+		reg_val = (kvm_riscv_isa_check_host(ZICBOZ)) ? riscv_cboz_block_size : 0;
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(zicbop_block_size):
-		reg_val = (kvm_riscv_isa_check_host(ZICBOP)) ? 0 : riscv_cbop_block_size;
+		reg_val = (kvm_riscv_isa_check_host(ZICBOP)) ? riscv_cbop_block_size : 0;
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(mvendorid):
 		reg_val = vcpu->arch.mvendorid;

---
base-commit: b5060a4aa33d7d78a8c1837ab08ef0c81ea96650
change-id: 20260924-kvm_fix_cbosize-1bf2cdd282cb

#1 lore raw
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] kvm: riscv: reversed logic in GET_ONE_REG for Zicbo block sizes --

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] kvm: riscv: reversed logic in GET_ONE_REG for Zicbo block sizes
--

commit 66e4542b7a02ac3b1e7a491baa6d04283e1ccc3a
Author: Yong-Xuan Wang <[email protected]>

RISC-V: KVM: fix the Zicbo[m|z|p] block sizes in GET_ONE_REG

This commit modifies the ternary operator logic used to determine the Zicbo block sizes reported to KVM userspace via GET_ONE_REG. It changes the previous behavior to return the block size or 0 based on host extension support.

arch/riscv/kvm/vcpu_onereg.c static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu, · line 53

Quoted 10 lines
arch/riscv/kvm/vcpu_onereg.c +1 -1
index 877ba823fc747..f2b1255694491 100644
@@ -50,13 +50,13 @@ static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu,
 		reg_val = vcpu->arch.isa[0] & KVM_RISCV_BASE_ISA_MASK;
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(zicbom_block_size):
-		reg_val = (kvm_riscv_isa_check_host(ZICBOM)) ? 0 : riscv_cbom_block_size;
+		reg_val = (kvm_riscv_isa_check_host(ZICBOM)) ? riscv_cbom_block_size : 0;

[Severity: High]
Does this change reverse the intended logic and cause a regression in the KVM userspace ABI?

Looking at kvm_riscv_vcpu_get_reg_config() in arch/riscv/kvm/vcpu_onereg.c, kvm_riscv_isa_check_host() returns 0 on success (supported) and -ENOENT on failure.

Because 0 evaluates to false in C, the updated ternary operator returns 0 when the extension is supported, and the block size when unsupported (-ENOENT evaluates to true).

This causes KVM userspace to receive a 0 block size for supported extensions, which can prevent VMs from starting or cause guest crashes when performing cache operations.

arch/riscv/kvm/vcpu_onereg.c static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu, · line 56

Quoted 4 lines
arch/riscv/kvm/vcpu_onereg.c +1 -1
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(zicboz_block_size):
-		reg_val = (kvm_riscv_isa_check_host(ZICBOZ)) ? 0 : riscv_cboz_block_size;
+		reg_val = (kvm_riscv_isa_check_host(ZICBOZ)) ? riscv_cboz_block_size : 0;

[Severity: High]
Is this subject to the same logic reversal as above?

In kvm_riscv_vcpu_get_reg_config(), the ternary operator for the ZICBOZ extension similarly returns 0 when the host extension is supported.

arch/riscv/kvm/vcpu_onereg.c static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu, · line 59

Quoted 4 lines
arch/riscv/kvm/vcpu_onereg.c +1 -1
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(zicbop_block_size):
-		reg_val = (kvm_riscv_isa_check_host(ZICBOP)) ? 0 : riscv_cbop_block_size;
+		reg_val = (kvm_riscv_isa_check_host(ZICBOP)) ? riscv_cbop_block_size : 0;

[Severity: High]
Does this also share the same logic reversal?

For the ZICBOP extension in kvm_riscv_vcpu_get_reg_config(), the ternary operator evaluates to 0 when the host supports the extension.

Quoted 3 lines
arch/riscv/kvm/vcpu_onereg.c +0 -0
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(mvendorid):
 		reg_val = vcpu->arch.mvendorid;
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
#2 lore raw

oh I misunderstood the kvm_riscv_isa_check_host(...), please ignore this patch.

Quoted 70 lines · On Thu, Sep 24, 2026 at 4:16 PM <[email protected]> wrote:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] kvm: riscv: reversed logic in GET_ONE_REG for Zicbo block sizes
--

commit 66e4542b7a02ac3b1e7a491baa6d04283e1ccc3a
Author: Yong-Xuan Wang <[email protected]>

RISC-V: KVM: fix the Zicbo[m|z|p] block sizes in GET_ONE_REG

This commit modifies the ternary operator logic used to determine the Zicbo block sizes reported to KVM userspace via GET_ONE_REG. It changes the previous behavior to return the block size or 0 based on host extension support.

arch/riscv/kvm/vcpu_onereg.c static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu, · line 53

Quoted 10 lines
arch/riscv/kvm/vcpu_onereg.c +1 -1
index 877ba823fc747..f2b1255694491 100644
@@ -50,13 +50,13 @@ static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu,
              reg_val = vcpu->arch.isa[0] & KVM_RISCV_BASE_ISA_MASK;
              break;
      case KVM_REG_RISCV_CONFIG_REG(zicbom_block_size):
-             reg_val = (kvm_riscv_isa_check_host(ZICBOM)) ? 0 : riscv_cbom_block_size;
+             reg_val = (kvm_riscv_isa_check_host(ZICBOM)) ? riscv_cbom_block_size : 0;

[Severity: High]
Does this change reverse the intended logic and cause a regression in the KVM userspace ABI?

Looking at kvm_riscv_vcpu_get_reg_config() in arch/riscv/kvm/vcpu_onereg.c, kvm_riscv_isa_check_host() returns 0 on success (supported) and -ENOENT on failure.

Because 0 evaluates to false in C, the updated ternary operator returns 0 when the extension is supported, and the block size when unsupported (-ENOENT evaluates to true).

This causes KVM userspace to receive a 0 block size for supported extensions, which can prevent VMs from starting or cause guest crashes when performing cache operations.

arch/riscv/kvm/vcpu_onereg.c static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu, · line 56

Quoted 4 lines
arch/riscv/kvm/vcpu_onereg.c +1 -1
              break;
      case KVM_REG_RISCV_CONFIG_REG(zicboz_block_size):
-             reg_val = (kvm_riscv_isa_check_host(ZICBOZ)) ? 0 : riscv_cboz_block_size;
+             reg_val = (kvm_riscv_isa_check_host(ZICBOZ)) ? riscv_cboz_block_size : 0;

[Severity: High]
Is this subject to the same logic reversal as above?

In kvm_riscv_vcpu_get_reg_config(), the ternary operator for the ZICBOZ extension similarly returns 0 when the host extension is supported.

arch/riscv/kvm/vcpu_onereg.c static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu, · line 59

Quoted 4 lines
arch/riscv/kvm/vcpu_onereg.c +1 -1
              break;
      case KVM_REG_RISCV_CONFIG_REG(zicbop_block_size):
-             reg_val = (kvm_riscv_isa_check_host(ZICBOP)) ? 0 : riscv_cbop_block_size;
+             reg_val = (kvm_riscv_isa_check_host(ZICBOP)) ? riscv_cbop_block_size : 0;

[Severity: High]
Does this also share the same logic reversal?

For the ZICBOP extension in kvm_riscv_vcpu_get_reg_config(), the ternary operator evaluates to 0 when the host supports the extension.

arch/riscv/kvm/vcpu_onereg.c static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu, · line 62

Quoted 3 lines
arch/riscv/kvm/vcpu_onereg.c +0 -0
              break;
      case KVM_REG_RISCV_CONFIG_REG(mvendorid):
              reg_val = vcpu->arch.mvendorid;

--
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1