Skip to contentSkip to conversation

RISC-V: KVM: Guest Debug Support - Software Breakpoint Part

PATCH Superseded Chao Du 4 participants 17 messages Mar 1, 2024

Subsystems (4)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Chao Du

author
#0 lore raw

v2 cover RISC-V: KVM: Guest Debug Support - Software Breakpoint Part

This series implements the ?KVM Guset Debug? feature on RISC-V. This is an existing feature which is already supported by some other arches. It allows us to debug a RISC-V KVM guest from GDB in host side.

As the first stage, the software breakpoints (ebreak instruction) is implemented. HW breakpoints support will come later after a synthetically consideration with the SBI debug trigger extension.

A selftest case was added in this series. Manual test was done on QEMU RISC-V hypervisor emulator. (add '-s' to enable the gdbserver in QEMU)

This series is based on Linux 6.8-rc6 and also available at:
https://github.com/Du-Chao/kvm-riscv/tree/guest_debug_sw_v2

The matched QEMU is available at:
https://github.com/Du-Chao/qemu/tree/riscv_gd_sw

Please note that if Paolo's change
https://lore.kernel.org/kvm/20240131233056.10845-8-pbonzini at redhat.com/ is adopted, then we can keep 'arch/riscv/include/uapi/asm/kvm.h' untouched.

Changes from v1->v2:
- Rebased on Linux 6.8-rc6.
- Maintain a hedeleg in "struct kvm_vcpu_config" for each VCPU.
- Update the HEDELEG csr in kvm_arch_vcpu_load().

Changes from RFC->v1:
- Rebased on Linux 6.8-rc2.
- Merge PATCH1 and PATCH2 into one patch.
- kselftest case added.

v1 link:
https://lore.kernel.org/kvm/20240206074931.22930-1-duchao at eswincomputing.com RFC link:
https://lore.kernel.org/kvm/20231221095002.7404-1-duchao at eswincomputing.com

Chao Du (3):
RISC-V: KVM: Implement kvm_arch_vcpu_ioctl_set_guest_debug()
RISC-V: KVM: Handle breakpoint exits for VCPU
RISC-V: KVM: selftests: Add breakpoints test support

 arch/riscv/include/asm/kvm_host.h             | 17 +++++++
 arch/riscv/include/uapi/asm/kvm.h             |  1 +
 arch/riscv/kvm/main.c                         | 18 +------
 arch/riscv/kvm/vcpu.c                         | 15 +++++-
 arch/riscv/kvm/vcpu_exit.c                    |  4 ++
 arch/riscv/kvm/vm.c                           |  1 +
 tools/testing/selftests/kvm/Makefile          |  1 +
 .../testing/selftests/kvm/riscv/breakpoints.c | 49 +++++++++++++++++++
 8 files changed, 88 insertions(+), 18 deletions(-)
 create mode 100644 tools/testing/selftests/kvm/riscv/breakpoints.c

--
2.17.1

author
#1 lore raw

PATCH v2 1/3 RISC-V: KVM: Implement kvm_arch_vcpu_ioctl_set_guest_debug()

kvm_vm_ioctl_check_extension(): Return 1 if KVM_CAP_SET_GUEST_DEBUG is been checked.

kvm_arch_vcpu_ioctl_set_guest_debug(): Update the guest_debug flags from userspace accordingly. Route the breakpoint exceptions to HS mode if the VCPU is being debugged by userspace, by clearing the corresponding bit in hedeleg. Write the actual CSR in kvm_arch_vcpu_load().

---
 arch/riscv/include/asm/kvm_host.h | 17 +++++++++++++++++
 arch/riscv/include/uapi/asm/kvm.h |  1 +
 arch/riscv/kvm/main.c             | 18 ++----------------
 arch/riscv/kvm/vcpu.c             | 15 +++++++++++++--
 arch/riscv/kvm/vm.c               |  1 +
 5 files changed, 34 insertions(+), 18 deletions(-)
arch/riscv/include/asm/kvm_host.h +17 -0
index 484d04a92fa6..9ee3f03ba5d1 100644
@@ -43,6 +43,22 @@
 	KVM_ARCH_REQ_FLAGS(5, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
 #define KVM_REQ_STEAL_UPDATE		KVM_ARCH_REQ(6)
 
+#define KVM_HEDELEG_DEFAULT		((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+					 (_AC(1, UL) << EXC_BREAKPOINT) | \
+					 (_AC(1, UL) << EXC_SYSCALL) | \
+					 (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+					 (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+					 (_AC(1, UL) << EXC_STORE_PAGE_FAULT))
+#define KVM_HEDELEG_GUEST_DEBUG		((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+					 (_AC(1, UL) << EXC_SYSCALL) | \
+					 (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+					 (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+					 (_AC(1, UL) << EXC_STORE_PAGE_FAULT))
+
+#define KVM_HIDELEG_DEFAULT		((_AC(1, UL) << IRQ_VS_SOFT) | \
+					 (_AC(1, UL) << IRQ_VS_TIMER) | \
+					 (_AC(1, UL) << IRQ_VS_EXT))
+
 enum kvm_riscv_hfence_type {
 	KVM_RISCV_HFENCE_UNKNOWN = 0,
 	KVM_RISCV_HFENCE_GVMA_VMID_GPA,
@@ -169,6 +185,7 @@ struct kvm_vcpu_csr {
 struct kvm_vcpu_config {
 	u64 henvcfg;
 	u64 hstateen0;
+	unsigned long hedeleg;
 };
 
 struct kvm_vcpu_smstateen_csr {
arch/riscv/include/uapi/asm/kvm.h +1 -0
index 7499e88a947c..39f4f4b9dede 100644
@@ -17,6 +17,7 @@
 
 #define __KVM_HAVE_IRQ_LINE
 #define __KVM_HAVE_READONLY_MEM
+#define __KVM_HAVE_GUEST_DEBUG
 
 #define KVM_COALESCED_MMIO_PAGE_OFFSET 1
 
arch/riscv/kvm/main.c +2 -16
index 225a435d9c9a..bab2ec34cd87 100644
@@ -22,22 +22,8 @@ long kvm_arch_dev_ioctl(struct file *filp,
 
 int kvm_arch_hardware_enable(void)
 {
-	unsigned long hideleg, hedeleg;
-
-	hedeleg = 0;
-	hedeleg |= (1UL << EXC_INST_MISALIGNED);
-	hedeleg |= (1UL << EXC_BREAKPOINT);
-	hedeleg |= (1UL << EXC_SYSCALL);
-	hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
-	hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
-	hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
-	csr_write(CSR_HEDELEG, hedeleg);
-
-	hideleg = 0;
-	hideleg |= (1UL << IRQ_VS_SOFT);
-	hideleg |= (1UL << IRQ_VS_TIMER);
-	hideleg |= (1UL << IRQ_VS_EXT);
-	csr_write(CSR_HIDELEG, hideleg);
+	csr_write(CSR_HEDELEG, KVM_HEDELEG_DEFAULT);
+	csr_write(CSR_HIDELEG, KVM_HIDELEG_DEFAULT);
 
 	/* VS should access only the time counter directly. Everything else should trap */
 	csr_write(CSR_HCOUNTEREN, 0x02);
arch/riscv/kvm/vcpu.c +13 -2
index b5ca9f2e98ac..242076c2227f 100644
@@ -475,8 +475,15 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 					struct kvm_guest_debug *dbg)
 {
-	/* TODO; To be implemented later. */
-	return -EINVAL;
+	if (dbg->control & KVM_GUESTDBG_ENABLE) {
+		vcpu->guest_debug = dbg->control;
+		vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+	} else {
+		vcpu->guest_debug = 0;
+		vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+	}
+
+	return 0;
 }
 
 static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
@@ -505,6 +512,9 @@ static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
 		if (riscv_isa_extension_available(isa, SMSTATEEN))
 			cfg->hstateen0 |= SMSTATEEN0_SSTATEEN0;
 	}
+
+	if (!vcpu->guest_debug)
+		cfg->hedeleg = KVM_HEDELEG_DEFAULT;
 }
 
 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
@@ -519,6 +529,7 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 	csr_write(CSR_VSEPC, csr->vsepc);
 	csr_write(CSR_VSCAUSE, csr->vscause);
 	csr_write(CSR_VSTVAL, csr->vstval);
+	csr_write(CSR_HEDELEG, cfg->hedeleg);
 	csr_write(CSR_HVIP, csr->hvip);
 	csr_write(CSR_VSATP, csr->vsatp);
 	csr_write(CSR_HENVCFG, cfg->henvcfg);
arch/riscv/kvm/vm.c +1 -0
index ce58bc48e5b8..7396b8654f45 100644
@@ -186,6 +186,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 	case KVM_CAP_READONLY_MEM:
 	case KVM_CAP_MP_STATE:
 	case KVM_CAP_IMMEDIATE_EXIT:
+	case KVM_CAP_SET_GUEST_DEBUG:
 		r = 1;
 		break;
 	case KVM_CAP_NR_VCPUS:
Signature · 1 line
2.17.1
maintainer
#2 lore raw

Re: [PATCH v2 1/3] RISC-V: KVM: Implement kvm_arch_vcpu_ioctl_set_guest_debug()

Quoted 24 lines · On Fri, Mar 1, 2024 at 7:08?AM Chao Du <[email protected]> wrote:

kvm_vm_ioctl_check_extension(): Return 1 if KVM_CAP_SET_GUEST_DEBUG is been checked.

kvm_arch_vcpu_ioctl_set_guest_debug(): Update the guest_debug flags from userspace accordingly. Route the breakpoint exceptions to HS mode if the VCPU is being debugged by userspace, by clearing the corresponding bit in hedeleg. Write the actual CSR in kvm_arch_vcpu_load().

---
 arch/riscv/include/asm/kvm_host.h | 17 +++++++++++++++++
 arch/riscv/include/uapi/asm/kvm.h |  1 +
 arch/riscv/kvm/main.c             | 18 ++----------------
 arch/riscv/kvm/vcpu.c             | 15 +++++++++++++--
 arch/riscv/kvm/vm.c               |  1 +
 5 files changed, 34 insertions(+), 18 deletions(-)
arch/riscv/include/asm/kvm_host.h +0 -0
index 484d04a92fa6..9ee3f03ba5d1 100644

arch/riscv/include/asm/kvm_host.h line 51

Quoted 9 lines
arch/riscv/include/asm/kvm_host.h +6 -0
@@ -43,6 +43,22 @@
        KVM_ARCH_REQ_FLAGS(5, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
 #define KVM_REQ_STEAL_UPDATE           KVM_ARCH_REQ(6)
 
+#define KVM_HEDELEG_DEFAULT            ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_BREAKPOINT) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

Use BIT(xyz) here. For example: BIT(EXC_INST_MISALIGNED)

Also, BIT(EXC_BREAKPOINT) should not be part of KVM_HEDELEG_DEFAULT.

arch/riscv/include/asm/kvm_host.h line 56

Quoted 5 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+#define KVM_HEDELEG_GUEST_DEBUG                ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

No need for KVM_HEDELEG_GUEST_DEBUG, see below.

arch/riscv/include/asm/kvm_host.h line 61

Quoted 5 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+
+#define KVM_HIDELEG_DEFAULT            ((_AC(1, UL) << IRQ_VS_SOFT) | \
+                                        (_AC(1, UL) << IRQ_VS_TIMER) | \
+                                        (_AC(1, UL) << IRQ_VS_EXT))
+

Same as above, use BIT(xyz) here.

Quoted 57 lines
arch/riscv/include/asm/kvm_host.h +1 -0
 enum kvm_riscv_hfence_type {
        KVM_RISCV_HFENCE_UNKNOWN = 0,
        KVM_RISCV_HFENCE_GVMA_VMID_GPA,
@@ -169,6 +185,7 @@ struct kvm_vcpu_csr {
 struct kvm_vcpu_config {
        u64 henvcfg;
        u64 hstateen0;
+       unsigned long hedeleg;
 };
 
 struct kvm_vcpu_smstateen_csr {
arch/riscv/include/uapi/asm/kvm.h +1 -0
index 7499e88a947c..39f4f4b9dede 100644
@@ -17,6 +17,7 @@
 
 #define __KVM_HAVE_IRQ_LINE
 #define __KVM_HAVE_READONLY_MEM
+#define __KVM_HAVE_GUEST_DEBUG
 
 #define KVM_COALESCED_MMIO_PAGE_OFFSET 1
 
arch/riscv/kvm/main.c +2 -16
index 225a435d9c9a..bab2ec34cd87 100644
@@ -22,22 +22,8 @@ long kvm_arch_dev_ioctl(struct file *filp,
 
 int kvm_arch_hardware_enable(void)
 {
-       unsigned long hideleg, hedeleg;
-
-       hedeleg = 0;
-       hedeleg |= (1UL << EXC_INST_MISALIGNED);
-       hedeleg |= (1UL << EXC_BREAKPOINT);
-       hedeleg |= (1UL << EXC_SYSCALL);
-       hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
-       csr_write(CSR_HEDELEG, hedeleg);
-
-       hideleg = 0;
-       hideleg |= (1UL << IRQ_VS_SOFT);
-       hideleg |= (1UL << IRQ_VS_TIMER);
-       hideleg |= (1UL << IRQ_VS_EXT);
-       csr_write(CSR_HIDELEG, hideleg);
+       csr_write(CSR_HEDELEG, KVM_HEDELEG_DEFAULT);
+       csr_write(CSR_HIDELEG, KVM_HIDELEG_DEFAULT);
 
        /* VS should access only the time counter directly. Everything else should trap */
        csr_write(CSR_HCOUNTEREN, 0x02);
arch/riscv/kvm/vcpu.c +0 -0
index b5ca9f2e98ac..242076c2227f 100644

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 479

Quoted 5 lines
arch/riscv/kvm/vcpu.c +0 -2
@@ -475,8 +475,15 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
                                        struct kvm_guest_debug *dbg)
 {
-       /* TODO; To be implemented later. */
-       return -EINVAL;

if (vcpu->arch.ran_atleast_once)

        return -EBUSY;

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 484

Quoted 7 lines
arch/riscv/kvm/vcpu.c +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+       }

Don't update vcpu->arch.cfg.hedeleg here since it should be only done in kvm_riscv_vcpu_setup_config().

arch/riscv/kvm/vcpu.c static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu) · line 517

Quoted 12 lines
arch/riscv/kvm/vcpu.c +5 -0
+
+       return 0;
 }
 
 static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
@@ -505,6 +512,9 @@ static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
                if (riscv_isa_extension_available(isa, SMSTATEEN))
                        cfg->hstateen0 |= SMSTATEEN0_SSTATEEN0;
        }
+
+       if (!vcpu->guest_debug)
+               cfg->hedeleg = KVM_HEDELEG_DEFAULT;

This should be:

cfg->hedeleg = KVM_HEDELEG_DEFAULT;
if (vcpu->guest_debug)

        cfg->hedeleg |= BIT(EXC_BREAKPOINT);
Quoted 24 lines
arch/riscv/kvm/vcpu.c +1 -0
 }
 
 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
@@ -519,6 +529,7 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
        csr_write(CSR_VSEPC, csr->vsepc);
        csr_write(CSR_VSCAUSE, csr->vscause);
        csr_write(CSR_VSTVAL, csr->vstval);
+       csr_write(CSR_HEDELEG, cfg->hedeleg);
        csr_write(CSR_HVIP, csr->hvip);
        csr_write(CSR_VSATP, csr->vsatp);
        csr_write(CSR_HENVCFG, cfg->henvcfg);
arch/riscv/kvm/vm.c +1 -1
index ce58bc48e5b8..7396b8654f45 100644
@@ -186,6 +186,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
        case KVM_CAP_READONLY_MEM:
        case KVM_CAP_MP_STATE:
        case KVM_CAP_IMMEDIATE_EXIT:
+       case KVM_CAP_SET_GUEST_DEBUG:
                r = 1;
                break;
        case KVM_CAP_NR_VCPUS:
--
Quoted 2 lines

2.17.1

Regards,
Anup

author
#3 lore raw

Re: [PATCH v2 1/3] RISC-V: KVM: Implement kvm_arch_vcpu_ioctl_set_guest_debug()

Quoted 29 lines · Anup Patel
Quoted 27 lines · On Fri, Mar 1, 2024 at 7:08?AM Chao Du <[email protected]> wrote:

kvm_vm_ioctl_check_extension(): Return 1 if KVM_CAP_SET_GUEST_DEBUG is been checked.

kvm_arch_vcpu_ioctl_set_guest_debug(): Update the guest_debug flags from userspace accordingly. Route the breakpoint exceptions to HS mode if the VCPU is being debugged by userspace, by clearing the corresponding bit in hedeleg. Write the actual CSR in kvm_arch_vcpu_load().

---
 arch/riscv/include/asm/kvm_host.h | 17 +++++++++++++++++
 arch/riscv/include/uapi/asm/kvm.h |  1 +
 arch/riscv/kvm/main.c             | 18 ++----------------
 arch/riscv/kvm/vcpu.c             | 15 +++++++++++++--
 arch/riscv/kvm/vm.c               |  1 +
 5 files changed, 34 insertions(+), 18 deletions(-)
arch/riscv/include/asm/kvm_host.h +0 -0
index 484d04a92fa6..9ee3f03ba5d1 100644
@@ -43,6 +43,22 @@
        KVM_ARCH_REQ_FLAGS(5, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
 #define KVM_REQ_STEAL_UPDATE           KVM_ARCH_REQ(6)
 
Quoted 8 lines

arch/riscv/include/asm/kvm_host.h line 51

Quoted 6 lines
arch/riscv/include/asm/kvm_host.h +6 -0
+#define KVM_HEDELEG_DEFAULT            ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_BREAKPOINT) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

Use BIT(xyz) here. For example: BIT(EXC_INST_MISALIGNED)

Thanks, I will use BIT() instead in next revision.

Quoted 3 lines

Also, BIT(EXC_BREAKPOINT) should not be part of KVM_HEDELEG_DEFAULT.

I think the bit EXC_BREAKPOINT should be set by default, like what you already did in kvm_arch_hardware_enable(). Then the VS could get the ebreak and handle it accordingly.

If the guest_debug is enabled, ebreak instructions are inserted by the userspace(QEMU). So KVM should 'intercept' the ebreak and exit to QEMU. Bit EXC_BREAKPOINT should be cleared in this case.

Quoted 81 lines

arch/riscv/include/asm/kvm_host.h line 56

Quoted 6 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+#define KVM_HEDELEG_GUEST_DEBUG                ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

No need for KVM_HEDELEG_GUEST_DEBUG, see below.

arch/riscv/include/asm/kvm_host.h line 61

Quoted 5 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+
+#define KVM_HIDELEG_DEFAULT            ((_AC(1, UL) << IRQ_VS_SOFT) | \
+                                        (_AC(1, UL) << IRQ_VS_TIMER) | \
+                                        (_AC(1, UL) << IRQ_VS_EXT))
+

Same as above, use BIT(xyz) here.

Quoted 57 lines
arch/riscv/include/asm/kvm_host.h +1 -0
 enum kvm_riscv_hfence_type {
        KVM_RISCV_HFENCE_UNKNOWN = 0,
        KVM_RISCV_HFENCE_GVMA_VMID_GPA,
@@ -169,6 +185,7 @@ struct kvm_vcpu_csr {
 struct kvm_vcpu_config {
        u64 henvcfg;
        u64 hstateen0;
+       unsigned long hedeleg;
 };
 
 struct kvm_vcpu_smstateen_csr {
arch/riscv/include/uapi/asm/kvm.h +1 -0
index 7499e88a947c..39f4f4b9dede 100644
@@ -17,6 +17,7 @@
 
 #define __KVM_HAVE_IRQ_LINE
 #define __KVM_HAVE_READONLY_MEM
+#define __KVM_HAVE_GUEST_DEBUG
 
 #define KVM_COALESCED_MMIO_PAGE_OFFSET 1
 
arch/riscv/kvm/main.c +2 -16
index 225a435d9c9a..bab2ec34cd87 100644
@@ -22,22 +22,8 @@ long kvm_arch_dev_ioctl(struct file *filp,
 
 int kvm_arch_hardware_enable(void)
 {
-       unsigned long hideleg, hedeleg;
-
-       hedeleg = 0;
-       hedeleg |= (1UL << EXC_INST_MISALIGNED);
-       hedeleg |= (1UL << EXC_BREAKPOINT);
-       hedeleg |= (1UL << EXC_SYSCALL);
-       hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
-       csr_write(CSR_HEDELEG, hedeleg);
-
-       hideleg = 0;
-       hideleg |= (1UL << IRQ_VS_SOFT);
-       hideleg |= (1UL << IRQ_VS_TIMER);
-       hideleg |= (1UL << IRQ_VS_EXT);
-       csr_write(CSR_HIDELEG, hideleg);
+       csr_write(CSR_HEDELEG, KVM_HEDELEG_DEFAULT);
+       csr_write(CSR_HIDELEG, KVM_HIDELEG_DEFAULT);
 
        /* VS should access only the time counter directly. Everything else should trap */
        csr_write(CSR_HCOUNTEREN, 0x02);
arch/riscv/kvm/vcpu.c +0 -0
index b5ca9f2e98ac..242076c2227f 100644

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 479

Quoted 5 lines
arch/riscv/kvm/vcpu.c +0 -2
@@ -475,8 +475,15 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
                                        struct kvm_guest_debug *dbg)
 {
-       /* TODO; To be implemented later. */
-       return -EINVAL;

if (vcpu->arch.ran_atleast_once)

Quoted 1 line
        return -EBUSY;

If we enabled the guest_debug in QEMU side, then the KVM_SET_GUEST_DEBUG ioctl will come before the first KVM_RUN. This will always cause an ERROR.

Quoted 28 lines

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 484

Quoted 9 lines
arch/riscv/kvm/vcpu.c +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+       }

Don't update vcpu->arch.cfg.hedeleg here since it should be only done in kvm_riscv_vcpu_setup_config().

arch/riscv/kvm/vcpu.c static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu) · line 517

Quoted 12 lines
arch/riscv/kvm/vcpu.c +5 -0
+
+       return 0;
 }
 
 static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
@@ -505,6 +512,9 @@ static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
                if (riscv_isa_extension_available(isa, SMSTATEEN))
                        cfg->hstateen0 |= SMSTATEEN0_SSTATEEN0;
        }
+
+       if (!vcpu->guest_debug)
+               cfg->hedeleg = KVM_HEDELEG_DEFAULT;

This should be:

Quoted 3 lines

cfg->hedeleg = KVM_HEDELEG_DEFAULT;
if (vcpu->guest_debug)

        cfg->hedeleg |= BIT(EXC_BREAKPOINT);

Like above, here the logic should be:

cfg->hedeleg = KVM_HEDELEG_DEFAULT; // with BIT(EXC_BREAKPOINT) if (vcpu->guest_debug)

        cfg->hedeleg &= ~BIT(EXC_BREAKPOINT);

Another approach is:
initialize the cfg->hedeleg as KVM_HEDELEG_DEFAULT during kvm_arch_vcpu_create(). Besides that, only update the cfg->hedeleg in kvm_arch_vcpu_ioctl_set_guest_debug().

Quoted 22 lines
Quoted 22 lines
arch/riscv/kvm/vcpu.c +1 -0
 }
 
 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
@@ -519,6 +529,7 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
        csr_write(CSR_VSEPC, csr->vsepc);
        csr_write(CSR_VSCAUSE, csr->vscause);
        csr_write(CSR_VSTVAL, csr->vstval);
+       csr_write(CSR_HEDELEG, cfg->hedeleg);
        csr_write(CSR_HVIP, csr->hvip);
        csr_write(CSR_VSATP, csr->vsatp);
        csr_write(CSR_HENVCFG, cfg->henvcfg);
arch/riscv/kvm/vm.c +1 -0
index ce58bc48e5b8..7396b8654f45 100644
@@ -186,6 +186,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
        case KVM_CAP_READONLY_MEM:
        case KVM_CAP_MP_STATE:
        case KVM_CAP_IMMEDIATE_EXIT:
+       case KVM_CAP_SET_GUEST_DEBUG:
                r = 1;
Quoted 8 lines
Quoted 5 lines
arch/riscv/kvm/vm.c +0 -1
                break;
        case KVM_CAP_NR_VCPUS:
--

2.17.1

Regards,
Anup

Thanks,
Chao

maintainer
#4 lore raw

Re: [PATCH v2 1/3] RISC-V: KVM: Implement kvm_arch_vcpu_ioctl_set_guest_debug()

Quoted 46 lines · On Fri, Mar 1, 2024 at 12:05?PM Chao Du <[email protected]> wrote:
Quoted 29 lines · Anup Patel
Quoted 27 lines · On Fri, Mar 1, 2024 at 7:08?AM Chao Du <[email protected]> wrote:

kvm_vm_ioctl_check_extension(): Return 1 if KVM_CAP_SET_GUEST_DEBUG is been checked.

kvm_arch_vcpu_ioctl_set_guest_debug(): Update the guest_debug flags from userspace accordingly. Route the breakpoint exceptions to HS mode if the VCPU is being debugged by userspace, by clearing the corresponding bit in hedeleg. Write the actual CSR in kvm_arch_vcpu_load().

---
 arch/riscv/include/asm/kvm_host.h | 17 +++++++++++++++++
 arch/riscv/include/uapi/asm/kvm.h |  1 +
 arch/riscv/kvm/main.c             | 18 ++----------------
 arch/riscv/kvm/vcpu.c             | 15 +++++++++++++--
 arch/riscv/kvm/vm.c               |  1 +
 5 files changed, 34 insertions(+), 18 deletions(-)
arch/riscv/include/asm/kvm_host.h +0 -0
index 484d04a92fa6..9ee3f03ba5d1 100644
@@ -43,6 +43,22 @@
        KVM_ARCH_REQ_FLAGS(5, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
 #define KVM_REQ_STEAL_UPDATE           KVM_ARCH_REQ(6)
 
Quoted 8 lines

arch/riscv/include/asm/kvm_host.h line 51

Quoted 6 lines
arch/riscv/include/asm/kvm_host.h +6 -0
+#define KVM_HEDELEG_DEFAULT            ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_BREAKPOINT) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

Use BIT(xyz) here. For example: BIT(EXC_INST_MISALIGNED)

Thanks, I will use BIT() instead in next revision.

Quoted 3 lines

Also, BIT(EXC_BREAKPOINT) should not be part of KVM_HEDELEG_DEFAULT.

Quoted 7 lines

I think the bit EXC_BREAKPOINT should be set by default, like what you already did in kvm_arch_hardware_enable(). Then the VS could get the ebreak and handle it accordingly.

If the guest_debug is enabled, ebreak instructions are inserted by the userspace(QEMU). So KVM should 'intercept' the ebreak and exit to QEMU. Bit EXC_BREAKPOINT should be cleared in this case.

If EXC_BREAKPOINT is delegated by default then it is not consistent with vcpu->guest_debug which is not enabled by default.

Quoted 84 lines
Quoted 82 lines

arch/riscv/include/asm/kvm_host.h line 56

Quoted 7 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+#define KVM_HEDELEG_GUEST_DEBUG                ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

No need for KVM_HEDELEG_GUEST_DEBUG, see below.

arch/riscv/include/asm/kvm_host.h line 61

Quoted 5 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+
+#define KVM_HIDELEG_DEFAULT            ((_AC(1, UL) << IRQ_VS_SOFT) | \
+                                        (_AC(1, UL) << IRQ_VS_TIMER) | \
+                                        (_AC(1, UL) << IRQ_VS_EXT))
+

Same as above, use BIT(xyz) here.

Quoted 57 lines
arch/riscv/include/asm/kvm_host.h +1 -0
 enum kvm_riscv_hfence_type {
        KVM_RISCV_HFENCE_UNKNOWN = 0,
        KVM_RISCV_HFENCE_GVMA_VMID_GPA,
@@ -169,6 +185,7 @@ struct kvm_vcpu_csr {
 struct kvm_vcpu_config {
        u64 henvcfg;
        u64 hstateen0;
+       unsigned long hedeleg;
 };
 
 struct kvm_vcpu_smstateen_csr {
arch/riscv/include/uapi/asm/kvm.h +1 -0
index 7499e88a947c..39f4f4b9dede 100644
@@ -17,6 +17,7 @@
 
 #define __KVM_HAVE_IRQ_LINE
 #define __KVM_HAVE_READONLY_MEM
+#define __KVM_HAVE_GUEST_DEBUG
 
 #define KVM_COALESCED_MMIO_PAGE_OFFSET 1
 
arch/riscv/kvm/main.c +2 -16
index 225a435d9c9a..bab2ec34cd87 100644
@@ -22,22 +22,8 @@ long kvm_arch_dev_ioctl(struct file *filp,
 
 int kvm_arch_hardware_enable(void)
 {
-       unsigned long hideleg, hedeleg;
-
-       hedeleg = 0;
-       hedeleg |= (1UL << EXC_INST_MISALIGNED);
-       hedeleg |= (1UL << EXC_BREAKPOINT);
-       hedeleg |= (1UL << EXC_SYSCALL);
-       hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
-       csr_write(CSR_HEDELEG, hedeleg);
-
-       hideleg = 0;
-       hideleg |= (1UL << IRQ_VS_SOFT);
-       hideleg |= (1UL << IRQ_VS_TIMER);
-       hideleg |= (1UL << IRQ_VS_EXT);
-       csr_write(CSR_HIDELEG, hideleg);
+       csr_write(CSR_HEDELEG, KVM_HEDELEG_DEFAULT);
+       csr_write(CSR_HIDELEG, KVM_HIDELEG_DEFAULT);
 
        /* VS should access only the time counter directly. Everything else should trap */
        csr_write(CSR_HCOUNTEREN, 0x02);
arch/riscv/kvm/vcpu.c +0 -0
index b5ca9f2e98ac..242076c2227f 100644

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 479

Quoted 5 lines
arch/riscv/kvm/vcpu.c +0 -2
@@ -475,8 +475,15 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
                                        struct kvm_guest_debug *dbg)
 {
-       /* TODO; To be implemented later. */
-       return -EINVAL;

if (vcpu->arch.ran_atleast_once)

Quoted 1 line
        return -EBUSY;
Quoted 2 lines

If we enabled the guest_debug in QEMU side, then the KVM_SET_GUEST_DEBUG ioctl will come before the first KVM_RUN. This will always cause an ERROR.

The check ensures that KVM user space can only enable/disable guest debug before the VCPU is run. I don't see why this would fail for QEMU.

Quoted 35 lines
Quoted 29 lines

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 484

Quoted 10 lines
arch/riscv/kvm/vcpu.c +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+       }

Don't update vcpu->arch.cfg.hedeleg here since it should be only done in kvm_riscv_vcpu_setup_config().

arch/riscv/kvm/vcpu.c static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu) · line 517

Quoted 12 lines
arch/riscv/kvm/vcpu.c +5 -0
+
+       return 0;
 }
 
 static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
@@ -505,6 +512,9 @@ static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
                if (riscv_isa_extension_available(isa, SMSTATEEN))
                        cfg->hstateen0 |= SMSTATEEN0_SSTATEEN0;
        }
+
+       if (!vcpu->guest_debug)
+               cfg->hedeleg = KVM_HEDELEG_DEFAULT;

This should be:

Quoted 3 lines

cfg->hedeleg = KVM_HEDELEG_DEFAULT;
if (vcpu->guest_debug)

        cfg->hedeleg |= BIT(EXC_BREAKPOINT);

Like above, here the logic should be:

Quoted 7 lines

cfg->hedeleg = KVM_HEDELEG_DEFAULT; // with BIT(EXC_BREAKPOINT) if (vcpu->guest_debug)

        cfg->hedeleg &= ~BIT(EXC_BREAKPOINT);

Another approach is:
initialize the cfg->hedeleg as KVM_HEDELEG_DEFAULT during kvm_arch_vcpu_create(). Besides that, only update the cfg->hedeleg in kvm_arch_vcpu_ioctl_set_guest_debug().

I disagree. We should handle hedeleg just like we handle henvcfg.

Quoted 32 lines
Quoted 23 lines
Quoted 23 lines
arch/riscv/kvm/vcpu.c +1 -0
 }
 
 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
@@ -519,6 +529,7 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
        csr_write(CSR_VSEPC, csr->vsepc);
        csr_write(CSR_VSCAUSE, csr->vscause);
        csr_write(CSR_VSTVAL, csr->vstval);
+       csr_write(CSR_HEDELEG, cfg->hedeleg);
        csr_write(CSR_HVIP, csr->hvip);
        csr_write(CSR_VSATP, csr->vsatp);
        csr_write(CSR_HENVCFG, cfg->henvcfg);
arch/riscv/kvm/vm.c +1 -0
index ce58bc48e5b8..7396b8654f45 100644
@@ -186,6 +186,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
        case KVM_CAP_READONLY_MEM:
        case KVM_CAP_MP_STATE:
        case KVM_CAP_IMMEDIATE_EXIT:
+       case KVM_CAP_SET_GUEST_DEBUG:
                r = 1;
Quoted 8 lines
Quoted 5 lines
arch/riscv/kvm/vm.c +0 -1
                break;
        case KVM_CAP_NR_VCPUS:
--

2.17.1

Regards,
Anup

Quoted 2 lines

Thanks,
Chao

Regards,
Anup

author
#5 lore raw

Re: [PATCH v2 1/3] RISC-V: KVM: Implement kvm_arch_vcpu_ioctl_set_guest_debug()

Quoted 52 lines · Anup Patel
Quoted 50 lines · On Fri, Mar 1, 2024 at 12:05?PM Chao Du <[email protected]> wrote:
Quoted 29 lines · Anup Patel
Quoted 27 lines · On Fri, Mar 1, 2024 at 7:08?AM Chao Du <[email protected]> wrote:

kvm_vm_ioctl_check_extension(): Return 1 if KVM_CAP_SET_GUEST_DEBUG is been checked.

kvm_arch_vcpu_ioctl_set_guest_debug(): Update the guest_debug flags from userspace accordingly. Route the breakpoint exceptions to HS mode if the VCPU is being debugged by userspace, by clearing the corresponding bit in hedeleg. Write the actual CSR in kvm_arch_vcpu_load().

---
 arch/riscv/include/asm/kvm_host.h | 17 +++++++++++++++++
 arch/riscv/include/uapi/asm/kvm.h |  1 +
 arch/riscv/kvm/main.c             | 18 ++----------------
 arch/riscv/kvm/vcpu.c             | 15 +++++++++++++--
 arch/riscv/kvm/vm.c               |  1 +
 5 files changed, 34 insertions(+), 18 deletions(-)
arch/riscv/include/asm/kvm_host.h +0 -0
index 484d04a92fa6..9ee3f03ba5d1 100644
@@ -43,6 +43,22 @@
        KVM_ARCH_REQ_FLAGS(5, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
 #define KVM_REQ_STEAL_UPDATE           KVM_ARCH_REQ(6)
 
Quoted 8 lines

arch/riscv/include/asm/kvm_host.h line 51

Quoted 6 lines
arch/riscv/include/asm/kvm_host.h +6 -0
+#define KVM_HEDELEG_DEFAULT            ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_BREAKPOINT) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

Use BIT(xyz) here. For example: BIT(EXC_INST_MISALIGNED)

Thanks, I will use BIT() instead in next revision.

Quoted 3 lines

Also, BIT(EXC_BREAKPOINT) should not be part of KVM_HEDELEG_DEFAULT.

I think the bit EXC_BREAKPOINT should be set by default, like what you already did in kvm_arch_hardware_enable(). Then the VS could get the ebreak and handle it accordingly.

Quoted 6 lines
Quoted 3 lines

If the guest_debug is enabled, ebreak instructions are inserted by the userspace(QEMU). So KVM should 'intercept' the ebreak and exit to QEMU. Bit EXC_BREAKPOINT should be cleared in this case.

If EXC_BREAKPOINT is delegated by default then it is not consistent with vcpu->guest_debug which is not enabled by default.

To enable the guest_debug corresponding to NOT delegate the EXC_BREAKPOINT. They are somehow 'opposite'.

This 'kvm_guest_debug' feature is different from "debug in the guest". The later requires the delegation of EXC_BREAKPOINT.
The former does not.

Quoted 85 lines
Quoted 85 lines
Quoted 84 lines

arch/riscv/include/asm/kvm_host.h line 56

Quoted 8 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+#define KVM_HEDELEG_GUEST_DEBUG                ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

No need for KVM_HEDELEG_GUEST_DEBUG, see below.

arch/riscv/include/asm/kvm_host.h line 61

Quoted 5 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+
+#define KVM_HIDELEG_DEFAULT            ((_AC(1, UL) << IRQ_VS_SOFT) | \
+                                        (_AC(1, UL) << IRQ_VS_TIMER) | \
+                                        (_AC(1, UL) << IRQ_VS_EXT))
+

Same as above, use BIT(xyz) here.

Quoted 57 lines
arch/riscv/include/asm/kvm_host.h +1 -0
 enum kvm_riscv_hfence_type {
        KVM_RISCV_HFENCE_UNKNOWN = 0,
        KVM_RISCV_HFENCE_GVMA_VMID_GPA,
@@ -169,6 +185,7 @@ struct kvm_vcpu_csr {
 struct kvm_vcpu_config {
        u64 henvcfg;
        u64 hstateen0;
+       unsigned long hedeleg;
 };
 
 struct kvm_vcpu_smstateen_csr {
arch/riscv/include/uapi/asm/kvm.h +1 -0
index 7499e88a947c..39f4f4b9dede 100644
@@ -17,6 +17,7 @@
 
 #define __KVM_HAVE_IRQ_LINE
 #define __KVM_HAVE_READONLY_MEM
+#define __KVM_HAVE_GUEST_DEBUG
 
 #define KVM_COALESCED_MMIO_PAGE_OFFSET 1
 
arch/riscv/kvm/main.c +2 -16
index 225a435d9c9a..bab2ec34cd87 100644
@@ -22,22 +22,8 @@ long kvm_arch_dev_ioctl(struct file *filp,
 
 int kvm_arch_hardware_enable(void)
 {
-       unsigned long hideleg, hedeleg;
-
-       hedeleg = 0;
-       hedeleg |= (1UL << EXC_INST_MISALIGNED);
-       hedeleg |= (1UL << EXC_BREAKPOINT);
-       hedeleg |= (1UL << EXC_SYSCALL);
-       hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
-       csr_write(CSR_HEDELEG, hedeleg);
-
-       hideleg = 0;
-       hideleg |= (1UL << IRQ_VS_SOFT);
-       hideleg |= (1UL << IRQ_VS_TIMER);
-       hideleg |= (1UL << IRQ_VS_EXT);
-       csr_write(CSR_HIDELEG, hideleg);
+       csr_write(CSR_HEDELEG, KVM_HEDELEG_DEFAULT);
+       csr_write(CSR_HIDELEG, KVM_HIDELEG_DEFAULT);
 
        /* VS should access only the time counter directly. Everything else should trap */
        csr_write(CSR_HCOUNTEREN, 0x02);
arch/riscv/kvm/vcpu.c +0 -0
index b5ca9f2e98ac..242076c2227f 100644

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 479

Quoted 5 lines
arch/riscv/kvm/vcpu.c +0 -2
@@ -475,8 +475,15 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
                                        struct kvm_guest_debug *dbg)
 {
-       /* TODO; To be implemented later. */
-       return -EINVAL;

if (vcpu->arch.ran_atleast_once)

        return -EBUSY;
Quoted 6 lines
Quoted 2 lines

If we enabled the guest_debug in QEMU side, then the KVM_SET_GUEST_DEBUG ioctl will come before the first KVM_RUN. This will always cause an ERROR.

The check ensures that KVM user space can only enable/disable guest debug before the VCPU is run. I don't see why this would fail for QEMU.

In the current implementation of GDB and QEMU, the userspace will enable/disable guest_debug frequently during the debugging (almost every step). The sequence should like:

KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
...

Quoted 40 lines
Quoted 40 lines
Quoted 30 lines

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 484

Quoted 11 lines
arch/riscv/kvm/vcpu.c +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+       }

Don't update vcpu->arch.cfg.hedeleg here since it should be only done in kvm_riscv_vcpu_setup_config().

arch/riscv/kvm/vcpu.c static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu) · line 517

Quoted 12 lines
arch/riscv/kvm/vcpu.c +5 -0
+
+       return 0;
 }
 
 static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
@@ -505,6 +512,9 @@ static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
                if (riscv_isa_extension_available(isa, SMSTATEEN))
                        cfg->hstateen0 |= SMSTATEEN0_SSTATEEN0;
        }
+
+       if (!vcpu->guest_debug)
+               cfg->hedeleg = KVM_HEDELEG_DEFAULT;

This should be:

Quoted 3 lines

cfg->hedeleg = KVM_HEDELEG_DEFAULT;
if (vcpu->guest_debug)

        cfg->hedeleg |= BIT(EXC_BREAKPOINT);

Like above, here the logic should be:

cfg->hedeleg = KVM_HEDELEG_DEFAULT; // with BIT(EXC_BREAKPOINT) if (vcpu->guest_debug)

        cfg->hedeleg &= ~BIT(EXC_BREAKPOINT);
Quoted 5 lines
Quoted 3 lines

Another approach is:
initialize the cfg->hedeleg as KVM_HEDELEG_DEFAULT during kvm_arch_vcpu_create(). Besides that, only update the cfg->hedeleg in kvm_arch_vcpu_ioctl_set_guest_debug().

I disagree. We should handle hedeleg just like we handle henvcfg.

OK, let's only update the cfg->hedeleg in kvm_riscv_vcpu_setup_config().

Quoted 33 lines
Quoted 33 lines
Quoted 32 lines
Quoted 27 lines
arch/riscv/kvm/vcpu.c +1 -0
 }
 
 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
@@ -519,6 +529,7 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
        csr_write(CSR_VSEPC, csr->vsepc);
        csr_write(CSR_VSCAUSE, csr->vscause);
        csr_write(CSR_VSTVAL, csr->vstval);
+       csr_write(CSR_HEDELEG, cfg->hedeleg);
        csr_write(CSR_HVIP, csr->hvip);
        csr_write(CSR_VSATP, csr->vsatp);
        csr_write(CSR_HENVCFG, cfg->henvcfg);
arch/riscv/kvm/vm.c +1 -1
index ce58bc48e5b8..7396b8654f45 100644
@@ -186,6 +186,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
        case KVM_CAP_READONLY_MEM:
        case KVM_CAP_MP_STATE:
        case KVM_CAP_IMMEDIATE_EXIT:
+       case KVM_CAP_SET_GUEST_DEBUG:
                r = 1;
                break;
        case KVM_CAP_NR_VCPUS:
--
Quoted 2 lines

2.17.1

Regards,
Anup

Quoted 5 lines
Quoted 2 lines

Thanks,
Chao

Regards,
Anup

Thanks,
Chao

#6 lore raw

Re: [PATCH v2 1/3] RISC-V: KVM: Implement kvm_arch_vcpu_ioctl_set_guest_debug()

Quoted 61 lines · On Fri, Mar 1, 2024 at 12:57?PM Chao Du <[email protected]> wrote:
Quoted 52 lines · Anup Patel
Quoted 50 lines · On Fri, Mar 1, 2024 at 12:05?PM Chao Du <[email protected]> wrote:
Quoted 29 lines · Anup Patel
Quoted 27 lines · On Fri, Mar 1, 2024 at 7:08?AM Chao Du <[email protected]> wrote:

kvm_vm_ioctl_check_extension(): Return 1 if KVM_CAP_SET_GUEST_DEBUG is been checked.

kvm_arch_vcpu_ioctl_set_guest_debug(): Update the guest_debug flags from userspace accordingly. Route the breakpoint exceptions to HS mode if the VCPU is being debugged by userspace, by clearing the corresponding bit in hedeleg. Write the actual CSR in kvm_arch_vcpu_load().

---
 arch/riscv/include/asm/kvm_host.h | 17 +++++++++++++++++
 arch/riscv/include/uapi/asm/kvm.h |  1 +
 arch/riscv/kvm/main.c             | 18 ++----------------
 arch/riscv/kvm/vcpu.c             | 15 +++++++++++++--
 arch/riscv/kvm/vm.c               |  1 +
 5 files changed, 34 insertions(+), 18 deletions(-)
arch/riscv/include/asm/kvm_host.h +0 -0
index 484d04a92fa6..9ee3f03ba5d1 100644
@@ -43,6 +43,22 @@
        KVM_ARCH_REQ_FLAGS(5, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
 #define KVM_REQ_STEAL_UPDATE           KVM_ARCH_REQ(6)
 
Quoted 8 lines

arch/riscv/include/asm/kvm_host.h line 51

Quoted 6 lines
arch/riscv/include/asm/kvm_host.h +6 -0
+#define KVM_HEDELEG_DEFAULT            ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_BREAKPOINT) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

Use BIT(xyz) here. For example: BIT(EXC_INST_MISALIGNED)

Thanks, I will use BIT() instead in next revision.

Quoted 3 lines

Also, BIT(EXC_BREAKPOINT) should not be part of KVM_HEDELEG_DEFAULT.

I think the bit EXC_BREAKPOINT should be set by default, like what you already did in kvm_arch_hardware_enable(). Then the VS could get the ebreak and handle it accordingly.

Quoted 6 lines
Quoted 3 lines

If the guest_debug is enabled, ebreak instructions are inserted by the userspace(QEMU). So KVM should 'intercept' the ebreak and exit to QEMU. Bit EXC_BREAKPOINT should be cleared in this case.

If EXC_BREAKPOINT is delegated by default then it is not consistent with vcpu->guest_debug which is not enabled by default.

Quoted 6 lines

To enable the guest_debug corresponding to NOT delegate the EXC_BREAKPOINT. They are somehow 'opposite'.

This 'kvm_guest_debug' feature is different from "debug in the guest". The later requires the delegation of EXC_BREAKPOINT.
The former does not.

In which case your below code is totally misleading.

+ if (dbg->control & KVM_GUESTDBG_ENABLE) {
+ vcpu->guest_debug = dbg->control;
+ vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+ } else {
+ vcpu->guest_debug = 0;
+ vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+ }

This should have been:

+ if (dbg->control & KVM_GUESTDBG_ENABLE) {
+ vcpu->guest_debug = dbg->control;
+ vcpu->arch.cfg.hedeleg &= ~BIT(EXC_BREAKPOINT);
+ } else {
+ vcpu->guest_debug = 0;
+ vcpu->arch.cfg.hedeleg |= BIT(EXC_BREAKPOINT);
+ }

Quoted 97 lines
Quoted 86 lines
Quoted 86 lines
Quoted 85 lines

arch/riscv/include/asm/kvm_host.h line 56

Quoted 9 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+#define KVM_HEDELEG_GUEST_DEBUG                ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

No need for KVM_HEDELEG_GUEST_DEBUG, see below.

arch/riscv/include/asm/kvm_host.h line 61

Quoted 5 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+
+#define KVM_HIDELEG_DEFAULT            ((_AC(1, UL) << IRQ_VS_SOFT) | \
+                                        (_AC(1, UL) << IRQ_VS_TIMER) | \
+                                        (_AC(1, UL) << IRQ_VS_EXT))
+

Same as above, use BIT(xyz) here.

Quoted 57 lines
arch/riscv/include/asm/kvm_host.h +1 -0
 enum kvm_riscv_hfence_type {
        KVM_RISCV_HFENCE_UNKNOWN = 0,
        KVM_RISCV_HFENCE_GVMA_VMID_GPA,
@@ -169,6 +185,7 @@ struct kvm_vcpu_csr {
 struct kvm_vcpu_config {
        u64 henvcfg;
        u64 hstateen0;
+       unsigned long hedeleg;
 };
 
 struct kvm_vcpu_smstateen_csr {
arch/riscv/include/uapi/asm/kvm.h +1 -0
index 7499e88a947c..39f4f4b9dede 100644
@@ -17,6 +17,7 @@
 
 #define __KVM_HAVE_IRQ_LINE
 #define __KVM_HAVE_READONLY_MEM
+#define __KVM_HAVE_GUEST_DEBUG
 
 #define KVM_COALESCED_MMIO_PAGE_OFFSET 1
 
arch/riscv/kvm/main.c +2 -16
index 225a435d9c9a..bab2ec34cd87 100644
@@ -22,22 +22,8 @@ long kvm_arch_dev_ioctl(struct file *filp,
 
 int kvm_arch_hardware_enable(void)
 {
-       unsigned long hideleg, hedeleg;
-
-       hedeleg = 0;
-       hedeleg |= (1UL << EXC_INST_MISALIGNED);
-       hedeleg |= (1UL << EXC_BREAKPOINT);
-       hedeleg |= (1UL << EXC_SYSCALL);
-       hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
-       csr_write(CSR_HEDELEG, hedeleg);
-
-       hideleg = 0;
-       hideleg |= (1UL << IRQ_VS_SOFT);
-       hideleg |= (1UL << IRQ_VS_TIMER);
-       hideleg |= (1UL << IRQ_VS_EXT);
-       csr_write(CSR_HIDELEG, hideleg);
+       csr_write(CSR_HEDELEG, KVM_HEDELEG_DEFAULT);
+       csr_write(CSR_HIDELEG, KVM_HIDELEG_DEFAULT);
 
        /* VS should access only the time counter directly. Everything else should trap */
        csr_write(CSR_HCOUNTEREN, 0x02);
arch/riscv/kvm/vcpu.c +0 -0
index b5ca9f2e98ac..242076c2227f 100644

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 479

Quoted 5 lines
arch/riscv/kvm/vcpu.c +0 -2
@@ -475,8 +475,15 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
                                        struct kvm_guest_debug *dbg)
 {
-       /* TODO; To be implemented later. */
-       return -EINVAL;

if (vcpu->arch.ran_atleast_once)

        return -EBUSY;
Quoted 6 lines
Quoted 2 lines

If we enabled the guest_debug in QEMU side, then the KVM_SET_GUEST_DEBUG ioctl will come before the first KVM_RUN. This will always cause an ERROR.

The check ensures that KVM user space can only enable/disable guest debug before the VCPU is run. I don't see why this would fail for QEMU.

In the current implementation of GDB and QEMU, the userspace will enable/disable guest_debug frequently during the debugging (almost every step). The sequence should like:

Quoted 10 lines

KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
...

Fair enough, no need to check "ran_atleast_once"

Quoted 88 lines
Quoted 41 lines
Quoted 41 lines
Quoted 31 lines

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 484

Quoted 12 lines
arch/riscv/kvm/vcpu.c +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+       }

Don't update vcpu->arch.cfg.hedeleg here since it should be only done in kvm_riscv_vcpu_setup_config().

arch/riscv/kvm/vcpu.c static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu) · line 517

Quoted 12 lines
arch/riscv/kvm/vcpu.c +5 -0
+
+       return 0;
 }
 
 static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
@@ -505,6 +512,9 @@ static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
                if (riscv_isa_extension_available(isa, SMSTATEEN))
                        cfg->hstateen0 |= SMSTATEEN0_SSTATEEN0;
        }
+
+       if (!vcpu->guest_debug)
+               cfg->hedeleg = KVM_HEDELEG_DEFAULT;

This should be:

Quoted 3 lines

cfg->hedeleg = KVM_HEDELEG_DEFAULT;
if (vcpu->guest_debug)

        cfg->hedeleg |= BIT(EXC_BREAKPOINT);

Like above, here the logic should be:

cfg->hedeleg = KVM_HEDELEG_DEFAULT; // with BIT(EXC_BREAKPOINT) if (vcpu->guest_debug)

        cfg->hedeleg &= ~BIT(EXC_BREAKPOINT);
Quoted 5 lines
Quoted 3 lines

Another approach is:
initialize the cfg->hedeleg as KVM_HEDELEG_DEFAULT during kvm_arch_vcpu_create(). Besides that, only update the cfg->hedeleg in kvm_arch_vcpu_ioctl_set_guest_debug().

I disagree. We should handle hedeleg just like we handle henvcfg.

OK, let's only update the cfg->hedeleg in kvm_riscv_vcpu_setup_config().

Quoted 33 lines
Quoted 33 lines
Quoted 32 lines
Quoted 27 lines
arch/riscv/kvm/vcpu.c +1 -0
 }
 
 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
@@ -519,6 +529,7 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
        csr_write(CSR_VSEPC, csr->vsepc);
        csr_write(CSR_VSCAUSE, csr->vscause);
        csr_write(CSR_VSTVAL, csr->vstval);
+       csr_write(CSR_HEDELEG, cfg->hedeleg);
        csr_write(CSR_HVIP, csr->hvip);
        csr_write(CSR_VSATP, csr->vsatp);
        csr_write(CSR_HENVCFG, cfg->henvcfg);
arch/riscv/kvm/vm.c +1 -1
index ce58bc48e5b8..7396b8654f45 100644
@@ -186,6 +186,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
        case KVM_CAP_READONLY_MEM:
        case KVM_CAP_MP_STATE:
        case KVM_CAP_IMMEDIATE_EXIT:
+       case KVM_CAP_SET_GUEST_DEBUG:
                r = 1;
                break;
        case KVM_CAP_NR_VCPUS:
--
Quoted 2 lines

2.17.1

Regards,
Anup

Quoted 5 lines
Quoted 2 lines

Thanks,
Chao

Regards,
Anup

Quoted 6 lines

Thanks,
Chao
--
kvm-riscv mailing list
kvm-riscv at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kvm-riscv

Regards,
Anup

author
#7 lore raw

Re: [PATCH v2 1/3] RISC-V: KVM: Implement kvm_arch_vcpu_ioctl_set_guest_debug()

Thanks Anup.
Let me try to summarize the changes in next revision:

1. Use BIT().
2. In kvm_arch_vcpu_ioctl_set_guest_debug(), do below things:

       if (dbg->control & KVM_GUESTDBG_ENABLE) {
               vcpu->guest_debug = dbg->control;
       } else {
               vcpu->guest_debug = 0;
       }
3. In kvm_riscv_vcpu_setup_config(), do below things:
       cfg->hedeleg = KVM_HEDELEG_DEFAULT; // with BIT(EXC_BREAKPOINT)
       if (vcpu->guest_debug)
               cfg->hedeleg &= ~BIT(EXC_BREAKPOINT);

Will prepare a PATCH v3 accordingly.

Regards,
Chao

Quoted 297 lines · Anup Patel
Quoted 61 lines · On Fri, Mar 1, 2024 at 12:57?PM Chao Du <[email protected]> wrote:
Quoted 52 lines · Anup Patel
Quoted 50 lines · On Fri, Mar 1, 2024 at 12:05?PM Chao Du <[email protected]> wrote:
Quoted 29 lines · Anup Patel
Quoted 27 lines · On Fri, Mar 1, 2024 at 7:08?AM Chao Du <[email protected]> wrote:

kvm_vm_ioctl_check_extension(): Return 1 if KVM_CAP_SET_GUEST_DEBUG is been checked.

kvm_arch_vcpu_ioctl_set_guest_debug(): Update the guest_debug flags from userspace accordingly. Route the breakpoint exceptions to HS mode if the VCPU is being debugged by userspace, by clearing the corresponding bit in hedeleg. Write the actual CSR in kvm_arch_vcpu_load().

---
 arch/riscv/include/asm/kvm_host.h | 17 +++++++++++++++++
 arch/riscv/include/uapi/asm/kvm.h |  1 +
 arch/riscv/kvm/main.c             | 18 ++----------------
 arch/riscv/kvm/vcpu.c             | 15 +++++++++++++--
 arch/riscv/kvm/vm.c               |  1 +
 5 files changed, 34 insertions(+), 18 deletions(-)
arch/riscv/include/asm/kvm_host.h +0 -0
index 484d04a92fa6..9ee3f03ba5d1 100644
@@ -43,6 +43,22 @@
        KVM_ARCH_REQ_FLAGS(5, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
 #define KVM_REQ_STEAL_UPDATE           KVM_ARCH_REQ(6)
 
Quoted 8 lines

arch/riscv/include/asm/kvm_host.h line 51

Quoted 6 lines
arch/riscv/include/asm/kvm_host.h +6 -0
+#define KVM_HEDELEG_DEFAULT            ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_BREAKPOINT) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

Use BIT(xyz) here. For example: BIT(EXC_INST_MISALIGNED)

Thanks, I will use BIT() instead in next revision.

Quoted 3 lines

Also, BIT(EXC_BREAKPOINT) should not be part of KVM_HEDELEG_DEFAULT.

I think the bit EXC_BREAKPOINT should be set by default, like what you already did in kvm_arch_hardware_enable(). Then the VS could get the ebreak and handle it accordingly.

Quoted 6 lines
Quoted 3 lines

If the guest_debug is enabled, ebreak instructions are inserted by the userspace(QEMU). So KVM should 'intercept' the ebreak and exit to QEMU. Bit EXC_BREAKPOINT should be cleared in this case.

If EXC_BREAKPOINT is delegated by default then it is not consistent with vcpu->guest_debug which is not enabled by default.

Quoted 6 lines

To enable the guest_debug corresponding to NOT delegate the EXC_BREAKPOINT. They are somehow 'opposite'.

This 'kvm_guest_debug' feature is different from "debug in the guest". The later requires the delegation of EXC_BREAKPOINT.
The former does not.

In which case your below code is totally misleading.

arch/riscv/kvm/vcpu.c +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+       }

This should have been:

arch/riscv/kvm/vcpu.c +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg &= ~BIT(EXC_BREAKPOINT);
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg |= BIT(EXC_BREAKPOINT);
+       }
Quoted 98 lines
Quoted 87 lines
Quoted 87 lines
Quoted 86 lines

arch/riscv/include/asm/kvm_host.h line 56

Quoted 10 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+#define KVM_HEDELEG_GUEST_DEBUG                ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

No need for KVM_HEDELEG_GUEST_DEBUG, see below.

arch/riscv/include/asm/kvm_host.h line 61

Quoted 5 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+
+#define KVM_HIDELEG_DEFAULT            ((_AC(1, UL) << IRQ_VS_SOFT) | \
+                                        (_AC(1, UL) << IRQ_VS_TIMER) | \
+                                        (_AC(1, UL) << IRQ_VS_EXT))
+

Same as above, use BIT(xyz) here.

Quoted 57 lines
arch/riscv/include/asm/kvm_host.h +1 -0
 enum kvm_riscv_hfence_type {
        KVM_RISCV_HFENCE_UNKNOWN = 0,
        KVM_RISCV_HFENCE_GVMA_VMID_GPA,
@@ -169,6 +185,7 @@ struct kvm_vcpu_csr {
 struct kvm_vcpu_config {
        u64 henvcfg;
        u64 hstateen0;
+       unsigned long hedeleg;
 };
 
 struct kvm_vcpu_smstateen_csr {
arch/riscv/include/uapi/asm/kvm.h +1 -0
index 7499e88a947c..39f4f4b9dede 100644
@@ -17,6 +17,7 @@
 
 #define __KVM_HAVE_IRQ_LINE
 #define __KVM_HAVE_READONLY_MEM
+#define __KVM_HAVE_GUEST_DEBUG
 
 #define KVM_COALESCED_MMIO_PAGE_OFFSET 1
 
arch/riscv/kvm/main.c +2 -16
index 225a435d9c9a..bab2ec34cd87 100644
@@ -22,22 +22,8 @@ long kvm_arch_dev_ioctl(struct file *filp,
 
 int kvm_arch_hardware_enable(void)
 {
-       unsigned long hideleg, hedeleg;
-
-       hedeleg = 0;
-       hedeleg |= (1UL << EXC_INST_MISALIGNED);
-       hedeleg |= (1UL << EXC_BREAKPOINT);
-       hedeleg |= (1UL << EXC_SYSCALL);
-       hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
-       csr_write(CSR_HEDELEG, hedeleg);
-
-       hideleg = 0;
-       hideleg |= (1UL << IRQ_VS_SOFT);
-       hideleg |= (1UL << IRQ_VS_TIMER);
-       hideleg |= (1UL << IRQ_VS_EXT);
-       csr_write(CSR_HIDELEG, hideleg);
+       csr_write(CSR_HEDELEG, KVM_HEDELEG_DEFAULT);
+       csr_write(CSR_HIDELEG, KVM_HIDELEG_DEFAULT);
 
        /* VS should access only the time counter directly. Everything else should trap */
        csr_write(CSR_HCOUNTEREN, 0x02);
arch/riscv/kvm/vcpu.c +0 -0
index b5ca9f2e98ac..242076c2227f 100644

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 479

Quoted 5 lines
arch/riscv/kvm/vcpu.c +0 -2
@@ -475,8 +475,15 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
                                        struct kvm_guest_debug *dbg)
 {
-       /* TODO; To be implemented later. */
-       return -EINVAL;

if (vcpu->arch.ran_atleast_once)

        return -EBUSY;
Quoted 6 lines
Quoted 2 lines

If we enabled the guest_debug in QEMU side, then the KVM_SET_GUEST_DEBUG ioctl will come before the first KVM_RUN. This will always cause an ERROR.

The check ensures that KVM user space can only enable/disable guest debug before the VCPU is run. I don't see why this would fail for QEMU.

In the current implementation of GDB and QEMU, the userspace will enable/disable guest_debug frequently during the debugging (almost every step). The sequence should like:

Quoted 10 lines

KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
...

Fair enough, no need to check "ran_atleast_once"

Quoted 88 lines
Quoted 41 lines
Quoted 41 lines
Quoted 31 lines

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 484

Quoted 12 lines
arch/riscv/kvm/vcpu.c +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+       }

Don't update vcpu->arch.cfg.hedeleg here since it should be only done in kvm_riscv_vcpu_setup_config().

arch/riscv/kvm/vcpu.c static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu) · line 517

Quoted 12 lines
arch/riscv/kvm/vcpu.c +5 -0
+
+       return 0;
 }
 
 static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
@@ -505,6 +512,9 @@ static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
                if (riscv_isa_extension_available(isa, SMSTATEEN))
                        cfg->hstateen0 |= SMSTATEEN0_SSTATEEN0;
        }
+
+       if (!vcpu->guest_debug)
+               cfg->hedeleg = KVM_HEDELEG_DEFAULT;

This should be:

Quoted 3 lines

cfg->hedeleg = KVM_HEDELEG_DEFAULT;
if (vcpu->guest_debug)

        cfg->hedeleg |= BIT(EXC_BREAKPOINT);

Like above, here the logic should be:

cfg->hedeleg = KVM_HEDELEG_DEFAULT; // with BIT(EXC_BREAKPOINT) if (vcpu->guest_debug)

        cfg->hedeleg &= ~BIT(EXC_BREAKPOINT);
Quoted 5 lines
Quoted 3 lines

Another approach is:
initialize the cfg->hedeleg as KVM_HEDELEG_DEFAULT during kvm_arch_vcpu_create(). Besides that, only update the cfg->hedeleg in kvm_arch_vcpu_ioctl_set_guest_debug().

I disagree. We should handle hedeleg just like we handle henvcfg.

OK, let's only update the cfg->hedeleg in kvm_riscv_vcpu_setup_config().

Quoted 33 lines
Quoted 33 lines
Quoted 32 lines
Quoted 27 lines
arch/riscv/kvm/vcpu.c +1 -0
 }
 
 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
@@ -519,6 +529,7 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
        csr_write(CSR_VSEPC, csr->vsepc);
        csr_write(CSR_VSCAUSE, csr->vscause);
        csr_write(CSR_VSTVAL, csr->vstval);
+       csr_write(CSR_HEDELEG, cfg->hedeleg);
        csr_write(CSR_HVIP, csr->hvip);
        csr_write(CSR_VSATP, csr->vsatp);
        csr_write(CSR_HENVCFG, cfg->henvcfg);
arch/riscv/kvm/vm.c +1 -1
index ce58bc48e5b8..7396b8654f45 100644
@@ -186,6 +186,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
        case KVM_CAP_READONLY_MEM:
        case KVM_CAP_MP_STATE:
        case KVM_CAP_IMMEDIATE_EXIT:
+       case KVM_CAP_SET_GUEST_DEBUG:
                r = 1;
                break;
        case KVM_CAP_NR_VCPUS:
--
Quoted 2 lines

2.17.1

Regards,
Anup

Quoted 5 lines
Quoted 2 lines

Thanks,
Chao

Regards,
Anup

Quoted 6 lines

Thanks,
Chao
--
kvm-riscv mailing list
kvm-riscv at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kvm-riscv

Regards,
Anup

maintainer
#8 lore raw

Re: [PATCH v2 1/3] RISC-V: KVM: Implement kvm_arch_vcpu_ioctl_set_guest_debug()

Quoted 11 lines · On Fri, Mar 1, 2024 at 1:46?PM Chao Du <[email protected]> wrote:

Thanks Anup.
Let me try to summarize the changes in next revision:

1. Use BIT().
2. In kvm_arch_vcpu_ioctl_set_guest_debug(), do below things:

       if (dbg->control & KVM_GUESTDBG_ENABLE) {
               vcpu->guest_debug = dbg->control;
       } else {
               vcpu->guest_debug = 0;
       }

Since, kvm_arch_vcpu_ioctl_set_guest_debug() is called multiple times at runtime, this should be:

        if (dbg->control & KVM_GUESTDBG_ENABLE) {
                vcpu->guest_debug = dbg->control;
                vcpu->arch.cfg.hedeleg &= ~BIT(EXC_BREAKPOINT);
        } else {
                vcpu->guest_debug = 0;
                vcpu->arch.cfg.hedeleg |= BIT(EXC_BREAKPOINT);
        }
Quoted 6 lines

3. In kvm_riscv_vcpu_setup_config(), do below things:

       cfg->hedeleg = KVM_HEDELEG_DEFAULT; // with BIT(EXC_BREAKPOINT)
       if (vcpu->guest_debug)
               cfg->hedeleg &= ~BIT(EXC_BREAKPOINT);

Will prepare a PATCH v3 accordingly.

Yes, please send v3.

Thanks,
Anup

Quoted 298 lines · Anup Patel
Quoted 297 lines · On Fri, Mar 1, 2024 at 12:57?PM Chao Du <[email protected]> wrote:
Quoted 61 lines · Anup Patel
Quoted 52 lines · On Fri, Mar 1, 2024 at 12:05?PM Chao Du <[email protected]> wrote:
Quoted 50 lines · Anup Patel
Quoted 29 lines · On Fri, Mar 1, 2024 at 7:08?AM Chao Du <[email protected]> wrote:
Quoted 27 lines

kvm_vm_ioctl_check_extension(): Return 1 if KVM_CAP_SET_GUEST_DEBUG is been checked.

kvm_arch_vcpu_ioctl_set_guest_debug(): Update the guest_debug flags from userspace accordingly. Route the breakpoint exceptions to HS mode if the VCPU is being debugged by userspace, by clearing the corresponding bit in hedeleg. Write the actual CSR in kvm_arch_vcpu_load().

---
 arch/riscv/include/asm/kvm_host.h | 17 +++++++++++++++++
 arch/riscv/include/uapi/asm/kvm.h |  1 +
 arch/riscv/kvm/main.c             | 18 ++----------------
 arch/riscv/kvm/vcpu.c             | 15 +++++++++++++--
 arch/riscv/kvm/vm.c               |  1 +
 5 files changed, 34 insertions(+), 18 deletions(-)
arch/riscv/include/asm/kvm_host.h +0 -0
index 484d04a92fa6..9ee3f03ba5d1 100644
@@ -43,6 +43,22 @@
        KVM_ARCH_REQ_FLAGS(5, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
 #define KVM_REQ_STEAL_UPDATE           KVM_ARCH_REQ(6)
 
Quoted 8 lines

arch/riscv/include/asm/kvm_host.h line 51

Quoted 6 lines
arch/riscv/include/asm/kvm_host.h +6 -0
+#define KVM_HEDELEG_DEFAULT            ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_BREAKPOINT) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

Use BIT(xyz) here. For example: BIT(EXC_INST_MISALIGNED)

Thanks, I will use BIT() instead in next revision.

Quoted 3 lines

Also, BIT(EXC_BREAKPOINT) should not be part of KVM_HEDELEG_DEFAULT.

I think the bit EXC_BREAKPOINT should be set by default, like what you already did in kvm_arch_hardware_enable(). Then the VS could get the ebreak and handle it accordingly.

Quoted 6 lines
Quoted 3 lines

If the guest_debug is enabled, ebreak instructions are inserted by the userspace(QEMU). So KVM should 'intercept' the ebreak and exit to QEMU. Bit EXC_BREAKPOINT should be cleared in this case.

If EXC_BREAKPOINT is delegated by default then it is not consistent with vcpu->guest_debug which is not enabled by default.

Quoted 6 lines

To enable the guest_debug corresponding to NOT delegate the EXC_BREAKPOINT. They are somehow 'opposite'.

This 'kvm_guest_debug' feature is different from "debug in the guest". The later requires the delegation of EXC_BREAKPOINT.
The former does not.

In which case your below code is totally misleading.

(quoted diff) +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+       }

This should have been:

(quoted diff) +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg &= ~BIT(EXC_BREAKPOINT);
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg |= BIT(EXC_BREAKPOINT);
+       }
Quoted 98 lines
Quoted 87 lines
Quoted 87 lines
Quoted 86 lines

arch/riscv/include/asm/kvm_host.h line 56

Quoted 10 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+#define KVM_HEDELEG_GUEST_DEBUG                ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

No need for KVM_HEDELEG_GUEST_DEBUG, see below.

arch/riscv/include/asm/kvm_host.h line 61

Quoted 5 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+
+#define KVM_HIDELEG_DEFAULT            ((_AC(1, UL) << IRQ_VS_SOFT) | \
+                                        (_AC(1, UL) << IRQ_VS_TIMER) | \
+                                        (_AC(1, UL) << IRQ_VS_EXT))
+

Same as above, use BIT(xyz) here.

Quoted 57 lines
arch/riscv/include/asm/kvm_host.h +1 -0
 enum kvm_riscv_hfence_type {
        KVM_RISCV_HFENCE_UNKNOWN = 0,
        KVM_RISCV_HFENCE_GVMA_VMID_GPA,
@@ -169,6 +185,7 @@ struct kvm_vcpu_csr {
 struct kvm_vcpu_config {
        u64 henvcfg;
        u64 hstateen0;
+       unsigned long hedeleg;
 };
 
 struct kvm_vcpu_smstateen_csr {
arch/riscv/include/uapi/asm/kvm.h +1 -0
index 7499e88a947c..39f4f4b9dede 100644
@@ -17,6 +17,7 @@
 
 #define __KVM_HAVE_IRQ_LINE
 #define __KVM_HAVE_READONLY_MEM
+#define __KVM_HAVE_GUEST_DEBUG
 
 #define KVM_COALESCED_MMIO_PAGE_OFFSET 1
 
arch/riscv/kvm/main.c +2 -16
index 225a435d9c9a..bab2ec34cd87 100644
@@ -22,22 +22,8 @@ long kvm_arch_dev_ioctl(struct file *filp,
 
 int kvm_arch_hardware_enable(void)
 {
-       unsigned long hideleg, hedeleg;
-
-       hedeleg = 0;
-       hedeleg |= (1UL << EXC_INST_MISALIGNED);
-       hedeleg |= (1UL << EXC_BREAKPOINT);
-       hedeleg |= (1UL << EXC_SYSCALL);
-       hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
-       csr_write(CSR_HEDELEG, hedeleg);
-
-       hideleg = 0;
-       hideleg |= (1UL << IRQ_VS_SOFT);
-       hideleg |= (1UL << IRQ_VS_TIMER);
-       hideleg |= (1UL << IRQ_VS_EXT);
-       csr_write(CSR_HIDELEG, hideleg);
+       csr_write(CSR_HEDELEG, KVM_HEDELEG_DEFAULT);
+       csr_write(CSR_HIDELEG, KVM_HIDELEG_DEFAULT);
 
        /* VS should access only the time counter directly. Everything else should trap */
        csr_write(CSR_HCOUNTEREN, 0x02);
arch/riscv/kvm/vcpu.c +0 -0
index b5ca9f2e98ac..242076c2227f 100644

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 479

Quoted 5 lines
arch/riscv/kvm/vcpu.c +0 -2
@@ -475,8 +475,15 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
                                        struct kvm_guest_debug *dbg)
 {
-       /* TODO; To be implemented later. */
-       return -EINVAL;

if (vcpu->arch.ran_atleast_once)

        return -EBUSY;
Quoted 6 lines
Quoted 2 lines

If we enabled the guest_debug in QEMU side, then the KVM_SET_GUEST_DEBUG ioctl will come before the first KVM_RUN. This will always cause an ERROR.

The check ensures that KVM user space can only enable/disable guest debug before the VCPU is run. I don't see why this would fail for QEMU.

In the current implementation of GDB and QEMU, the userspace will enable/disable guest_debug frequently during the debugging (almost every step). The sequence should like:

Quoted 10 lines

KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
...

Fair enough, no need to check "ran_atleast_once"

Quoted 88 lines
Quoted 41 lines
Quoted 41 lines
Quoted 31 lines

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 484

Quoted 12 lines
arch/riscv/kvm/vcpu.c +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+       }

Don't update vcpu->arch.cfg.hedeleg here since it should be only done in kvm_riscv_vcpu_setup_config().

arch/riscv/kvm/vcpu.c static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu) · line 517

Quoted 12 lines
arch/riscv/kvm/vcpu.c +5 -0
+
+       return 0;
 }
 
 static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
@@ -505,6 +512,9 @@ static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
                if (riscv_isa_extension_available(isa, SMSTATEEN))
                        cfg->hstateen0 |= SMSTATEEN0_SSTATEEN0;
        }
+
+       if (!vcpu->guest_debug)
+               cfg->hedeleg = KVM_HEDELEG_DEFAULT;

This should be:

Quoted 3 lines

cfg->hedeleg = KVM_HEDELEG_DEFAULT;
if (vcpu->guest_debug)

        cfg->hedeleg |= BIT(EXC_BREAKPOINT);

Like above, here the logic should be:

cfg->hedeleg = KVM_HEDELEG_DEFAULT; // with BIT(EXC_BREAKPOINT) if (vcpu->guest_debug)

        cfg->hedeleg &= ~BIT(EXC_BREAKPOINT);
Quoted 5 lines
Quoted 3 lines

Another approach is:
initialize the cfg->hedeleg as KVM_HEDELEG_DEFAULT during kvm_arch_vcpu_create(). Besides that, only update the cfg->hedeleg in kvm_arch_vcpu_ioctl_set_guest_debug().

I disagree. We should handle hedeleg just like we handle henvcfg.

OK, let's only update the cfg->hedeleg in kvm_riscv_vcpu_setup_config().

Quoted 33 lines
Quoted 33 lines
Quoted 32 lines
Quoted 27 lines
arch/riscv/kvm/vcpu.c +1 -0
 }
 
 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
@@ -519,6 +529,7 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
        csr_write(CSR_VSEPC, csr->vsepc);
        csr_write(CSR_VSCAUSE, csr->vscause);
        csr_write(CSR_VSTVAL, csr->vstval);
+       csr_write(CSR_HEDELEG, cfg->hedeleg);
        csr_write(CSR_HVIP, csr->hvip);
        csr_write(CSR_VSATP, csr->vsatp);
        csr_write(CSR_HENVCFG, cfg->henvcfg);
arch/riscv/kvm/vm.c +1 -1
index ce58bc48e5b8..7396b8654f45 100644
@@ -186,6 +186,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
        case KVM_CAP_READONLY_MEM:
        case KVM_CAP_MP_STATE:
        case KVM_CAP_IMMEDIATE_EXIT:
+       case KVM_CAP_SET_GUEST_DEBUG:
                r = 1;
                break;
        case KVM_CAP_NR_VCPUS:
--
Quoted 2 lines

2.17.1

Regards,
Anup

Quoted 5 lines
Quoted 2 lines

Thanks,
Chao

Regards,
Anup

Quoted 6 lines

Thanks,
Chao
--
kvm-riscv mailing list
kvm-riscv at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kvm-riscv

Regards,
Anup

author
#9 lore raw

Re: [PATCH v2 1/3] RISC-V: KVM: Implement kvm_arch_vcpu_ioctl_set_guest_debug()

Quoted 17 lines · Anup Patel
Quoted 11 lines · On Fri, Mar 1, 2024 at 1:46?PM Chao Du <[email protected]> wrote:

Thanks Anup.
Let me try to summarize the changes in next revision:

1. Use BIT().
2. In kvm_arch_vcpu_ioctl_set_guest_debug(), do below things:

       if (dbg->control & KVM_GUESTDBG_ENABLE) {
               vcpu->guest_debug = dbg->control;
       } else {
               vcpu->guest_debug = 0;
       }

Since, kvm_arch_vcpu_ioctl_set_guest_debug() is called multiple times at runtime, this should be:

Quoted 14 lines
        if (dbg->control & KVM_GUESTDBG_ENABLE) {
                vcpu->guest_debug = dbg->control;
                vcpu->arch.cfg.hedeleg &= ~BIT(EXC_BREAKPOINT);
        } else {
                vcpu->guest_debug = 0;
                vcpu->arch.cfg.hedeleg |= BIT(EXC_BREAKPOINT);
        }
Quoted 6 lines

3. In kvm_riscv_vcpu_setup_config(), do below things:

       cfg->hedeleg = KVM_HEDELEG_DEFAULT; // with BIT(EXC_BREAKPOINT)
       if (vcpu->guest_debug)
               cfg->hedeleg &= ~BIT(EXC_BREAKPOINT);

Will prepare a PATCH v3 accordingly.

I thought it over, maybe we should do this:

- In kvm_arch_vcpu_ioctl_set_guest_debug():

       if (dbg->control & KVM_GUESTDBG_ENABLE) {
               vcpu->guest_debug = dbg->control;
               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT & (~BIT(EXC_BREAKPOINT));
       } else {
               vcpu->guest_debug = 0;
               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
       }
- In kvm_riscv_vcpu_setup_config():
       if (!vcpu->guest_debug)
               cfg->hedeleg = KVM_HEDELEG_DEFAULT;

which is similar with my original changes, but clear the bit EXC_BREAKPOINT explicitly.

Could you please confirm ?

Quoted 304 lines

Yes, please send v3.

Thanks,
Anup

Quoted 298 lines · Anup Patel
Quoted 297 lines · On Fri, Mar 1, 2024 at 12:57?PM Chao Du <[email protected]> wrote:
Quoted 61 lines · Anup Patel
Quoted 52 lines · On Fri, Mar 1, 2024 at 12:05?PM Chao Du <[email protected]> wrote:
Quoted 50 lines · Anup Patel
Quoted 29 lines · On Fri, Mar 1, 2024 at 7:08?AM Chao Du <[email protected]> wrote:
Quoted 27 lines

kvm_vm_ioctl_check_extension(): Return 1 if KVM_CAP_SET_GUEST_DEBUG is been checked.

kvm_arch_vcpu_ioctl_set_guest_debug(): Update the guest_debug flags from userspace accordingly. Route the breakpoint exceptions to HS mode if the VCPU is being debugged by userspace, by clearing the corresponding bit in hedeleg. Write the actual CSR in kvm_arch_vcpu_load().

---
 arch/riscv/include/asm/kvm_host.h | 17 +++++++++++++++++
 arch/riscv/include/uapi/asm/kvm.h |  1 +
 arch/riscv/kvm/main.c             | 18 ++----------------
 arch/riscv/kvm/vcpu.c             | 15 +++++++++++++--
 arch/riscv/kvm/vm.c               |  1 +
 5 files changed, 34 insertions(+), 18 deletions(-)
arch/riscv/include/asm/kvm_host.h +0 -0
index 484d04a92fa6..9ee3f03ba5d1 100644
@@ -43,6 +43,22 @@
        KVM_ARCH_REQ_FLAGS(5, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
 #define KVM_REQ_STEAL_UPDATE           KVM_ARCH_REQ(6)
 
Quoted 8 lines

arch/riscv/include/asm/kvm_host.h line 51

Quoted 6 lines
arch/riscv/include/asm/kvm_host.h +6 -0
+#define KVM_HEDELEG_DEFAULT            ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_BREAKPOINT) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

Use BIT(xyz) here. For example: BIT(EXC_INST_MISALIGNED)

Thanks, I will use BIT() instead in next revision.

Quoted 3 lines

Also, BIT(EXC_BREAKPOINT) should not be part of KVM_HEDELEG_DEFAULT.

I think the bit EXC_BREAKPOINT should be set by default, like what you already did in kvm_arch_hardware_enable(). Then the VS could get the ebreak and handle it accordingly.

Quoted 6 lines
Quoted 3 lines

If the guest_debug is enabled, ebreak instructions are inserted by the userspace(QEMU). So KVM should 'intercept' the ebreak and exit to QEMU. Bit EXC_BREAKPOINT should be cleared in this case.

If EXC_BREAKPOINT is delegated by default then it is not consistent with vcpu->guest_debug which is not enabled by default.

Quoted 6 lines

To enable the guest_debug corresponding to NOT delegate the EXC_BREAKPOINT. They are somehow 'opposite'.

This 'kvm_guest_debug' feature is different from "debug in the guest". The later requires the delegation of EXC_BREAKPOINT.
The former does not.

In which case your below code is totally misleading.

(quoted diff) +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+       }

This should have been:

(quoted diff) +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg &= ~BIT(EXC_BREAKPOINT);
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg |= BIT(EXC_BREAKPOINT);
+       }
Quoted 98 lines
Quoted 87 lines
Quoted 87 lines
Quoted 86 lines

arch/riscv/include/asm/kvm_host.h line 56

Quoted 10 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+#define KVM_HEDELEG_GUEST_DEBUG                ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

No need for KVM_HEDELEG_GUEST_DEBUG, see below.

arch/riscv/include/asm/kvm_host.h line 61

Quoted 5 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+
+#define KVM_HIDELEG_DEFAULT            ((_AC(1, UL) << IRQ_VS_SOFT) | \
+                                        (_AC(1, UL) << IRQ_VS_TIMER) | \
+                                        (_AC(1, UL) << IRQ_VS_EXT))
+

Same as above, use BIT(xyz) here.

Quoted 57 lines
arch/riscv/include/asm/kvm_host.h +1 -0
 enum kvm_riscv_hfence_type {
        KVM_RISCV_HFENCE_UNKNOWN = 0,
        KVM_RISCV_HFENCE_GVMA_VMID_GPA,
@@ -169,6 +185,7 @@ struct kvm_vcpu_csr {
 struct kvm_vcpu_config {
        u64 henvcfg;
        u64 hstateen0;
+       unsigned long hedeleg;
 };
 
 struct kvm_vcpu_smstateen_csr {
arch/riscv/include/uapi/asm/kvm.h +1 -0
index 7499e88a947c..39f4f4b9dede 100644
@@ -17,6 +17,7 @@
 
 #define __KVM_HAVE_IRQ_LINE
 #define __KVM_HAVE_READONLY_MEM
+#define __KVM_HAVE_GUEST_DEBUG
 
 #define KVM_COALESCED_MMIO_PAGE_OFFSET 1
 
arch/riscv/kvm/main.c +2 -16
index 225a435d9c9a..bab2ec34cd87 100644
@@ -22,22 +22,8 @@ long kvm_arch_dev_ioctl(struct file *filp,
 
 int kvm_arch_hardware_enable(void)
 {
-       unsigned long hideleg, hedeleg;
-
-       hedeleg = 0;
-       hedeleg |= (1UL << EXC_INST_MISALIGNED);
-       hedeleg |= (1UL << EXC_BREAKPOINT);
-       hedeleg |= (1UL << EXC_SYSCALL);
-       hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
-       csr_write(CSR_HEDELEG, hedeleg);
-
-       hideleg = 0;
-       hideleg |= (1UL << IRQ_VS_SOFT);
-       hideleg |= (1UL << IRQ_VS_TIMER);
-       hideleg |= (1UL << IRQ_VS_EXT);
-       csr_write(CSR_HIDELEG, hideleg);
+       csr_write(CSR_HEDELEG, KVM_HEDELEG_DEFAULT);
+       csr_write(CSR_HIDELEG, KVM_HIDELEG_DEFAULT);
 
        /* VS should access only the time counter directly. Everything else should trap */
        csr_write(CSR_HCOUNTEREN, 0x02);
arch/riscv/kvm/vcpu.c +0 -0
index b5ca9f2e98ac..242076c2227f 100644

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 479

Quoted 5 lines
arch/riscv/kvm/vcpu.c +0 -2
@@ -475,8 +475,15 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
                                        struct kvm_guest_debug *dbg)
 {
-       /* TODO; To be implemented later. */
-       return -EINVAL;

if (vcpu->arch.ran_atleast_once)

        return -EBUSY;
Quoted 6 lines
Quoted 2 lines

If we enabled the guest_debug in QEMU side, then the KVM_SET_GUEST_DEBUG ioctl will come before the first KVM_RUN. This will always cause an ERROR.

The check ensures that KVM user space can only enable/disable guest debug before the VCPU is run. I don't see why this would fail for QEMU.

In the current implementation of GDB and QEMU, the userspace will enable/disable guest_debug frequently during the debugging (almost every step). The sequence should like:

Quoted 10 lines

KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
...

Fair enough, no need to check "ran_atleast_once"

Quoted 88 lines
Quoted 41 lines
Quoted 41 lines
Quoted 31 lines

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 484

Quoted 12 lines
arch/riscv/kvm/vcpu.c +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+       }

Don't update vcpu->arch.cfg.hedeleg here since it should be only done in kvm_riscv_vcpu_setup_config().

arch/riscv/kvm/vcpu.c static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu) · line 517

Quoted 12 lines
arch/riscv/kvm/vcpu.c +5 -0
+
+       return 0;
 }
 
 static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
@@ -505,6 +512,9 @@ static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
                if (riscv_isa_extension_available(isa, SMSTATEEN))
                        cfg->hstateen0 |= SMSTATEEN0_SSTATEEN0;
        }
+
+       if (!vcpu->guest_debug)
+               cfg->hedeleg = KVM_HEDELEG_DEFAULT;

This should be:

Quoted 3 lines

cfg->hedeleg = KVM_HEDELEG_DEFAULT;
if (vcpu->guest_debug)

        cfg->hedeleg |= BIT(EXC_BREAKPOINT);

Like above, here the logic should be:

cfg->hedeleg = KVM_HEDELEG_DEFAULT; // with BIT(EXC_BREAKPOINT) if (vcpu->guest_debug)

        cfg->hedeleg &= ~BIT(EXC_BREAKPOINT);
Quoted 5 lines
Quoted 3 lines

Another approach is:
initialize the cfg->hedeleg as KVM_HEDELEG_DEFAULT during kvm_arch_vcpu_create(). Besides that, only update the cfg->hedeleg in kvm_arch_vcpu_ioctl_set_guest_debug().

I disagree. We should handle hedeleg just like we handle henvcfg.

OK, let's only update the cfg->hedeleg in kvm_riscv_vcpu_setup_config().

Quoted 33 lines
Quoted 33 lines
Quoted 32 lines
Quoted 27 lines
arch/riscv/kvm/vcpu.c +1 -0
 }
 
 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
@@ -519,6 +529,7 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
        csr_write(CSR_VSEPC, csr->vsepc);
        csr_write(CSR_VSCAUSE, csr->vscause);
        csr_write(CSR_VSTVAL, csr->vstval);
+       csr_write(CSR_HEDELEG, cfg->hedeleg);
        csr_write(CSR_HVIP, csr->hvip);
        csr_write(CSR_VSATP, csr->vsatp);
        csr_write(CSR_HENVCFG, cfg->henvcfg);
arch/riscv/kvm/vm.c +1 -1
index ce58bc48e5b8..7396b8654f45 100644
@@ -186,6 +186,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
        case KVM_CAP_READONLY_MEM:
        case KVM_CAP_MP_STATE:
        case KVM_CAP_IMMEDIATE_EXIT:
+       case KVM_CAP_SET_GUEST_DEBUG:
                r = 1;
                break;
        case KVM_CAP_NR_VCPUS:
--
Quoted 2 lines

2.17.1

Regards,
Anup

Quoted 5 lines
Quoted 2 lines

Thanks,
Chao

Regards,
Anup

Quoted 6 lines

Thanks,
Chao
--
kvm-riscv mailing list
kvm-riscv at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kvm-riscv

Regards,
Anup

#10 lore raw

Re: [PATCH v2 1/3] RISC-V: KVM: Implement kvm_arch_vcpu_ioctl_set_guest_debug()

Quoted 48 lines · On Fri, Mar 1, 2024 at 2:53?PM Chao Du <[email protected]> wrote:
Quoted 17 lines · Anup Patel
Quoted 11 lines · On Fri, Mar 1, 2024 at 1:46?PM Chao Du <[email protected]> wrote:

Thanks Anup.
Let me try to summarize the changes in next revision:

1. Use BIT().
2. In kvm_arch_vcpu_ioctl_set_guest_debug(), do below things:

       if (dbg->control & KVM_GUESTDBG_ENABLE) {
               vcpu->guest_debug = dbg->control;
       } else {
               vcpu->guest_debug = 0;
       }

Since, kvm_arch_vcpu_ioctl_set_guest_debug() is called multiple times at runtime, this should be:

Quoted 14 lines
        if (dbg->control & KVM_GUESTDBG_ENABLE) {
                vcpu->guest_debug = dbg->control;
                vcpu->arch.cfg.hedeleg &= ~BIT(EXC_BREAKPOINT);
        } else {
                vcpu->guest_debug = 0;
                vcpu->arch.cfg.hedeleg |= BIT(EXC_BREAKPOINT);
        }
Quoted 6 lines

3. In kvm_riscv_vcpu_setup_config(), do below things:

       cfg->hedeleg = KVM_HEDELEG_DEFAULT; // with BIT(EXC_BREAKPOINT)
       if (vcpu->guest_debug)
               cfg->hedeleg &= ~BIT(EXC_BREAKPOINT);

Will prepare a PATCH v3 accordingly.

I thought it over, maybe we should do this:

- In kvm_arch_vcpu_ioctl_set_guest_debug():

       if (dbg->control & KVM_GUESTDBG_ENABLE) {
               vcpu->guest_debug = dbg->control;
               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT & (~BIT(EXC_BREAKPOINT));
       } else {
               vcpu->guest_debug = 0;
               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
       }
- In kvm_riscv_vcpu_setup_config():
       if (!vcpu->guest_debug)
               cfg->hedeleg = KVM_HEDELEG_DEFAULT;
Quoted 3 lines

which is similar with my original changes, but clear the bit EXC_BREAKPOINT explicitly.

Could you please confirm ?

I suggest going with my previous suggestion.

Regards,
Anup

Quoted 305 lines
Quoted 305 lines

Yes, please send v3.

Thanks,
Anup

Quoted 298 lines · Anup Patel
Quoted 297 lines · On Fri, Mar 1, 2024 at 12:57?PM Chao Du <[email protected]> wrote:
Quoted 61 lines · Anup Patel
Quoted 52 lines · On Fri, Mar 1, 2024 at 12:05?PM Chao Du <[email protected]> wrote:
Quoted 50 lines · Anup Patel
Quoted 29 lines · On Fri, Mar 1, 2024 at 7:08?AM Chao Du <[email protected]> wrote:
Quoted 27 lines

kvm_vm_ioctl_check_extension(): Return 1 if KVM_CAP_SET_GUEST_DEBUG is been checked.

kvm_arch_vcpu_ioctl_set_guest_debug(): Update the guest_debug flags from userspace accordingly. Route the breakpoint exceptions to HS mode if the VCPU is being debugged by userspace, by clearing the corresponding bit in hedeleg. Write the actual CSR in kvm_arch_vcpu_load().

---
 arch/riscv/include/asm/kvm_host.h | 17 +++++++++++++++++
 arch/riscv/include/uapi/asm/kvm.h |  1 +
 arch/riscv/kvm/main.c             | 18 ++----------------
 arch/riscv/kvm/vcpu.c             | 15 +++++++++++++--
 arch/riscv/kvm/vm.c               |  1 +
 5 files changed, 34 insertions(+), 18 deletions(-)
arch/riscv/include/asm/kvm_host.h +0 -0
index 484d04a92fa6..9ee3f03ba5d1 100644
@@ -43,6 +43,22 @@
        KVM_ARCH_REQ_FLAGS(5, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
 #define KVM_REQ_STEAL_UPDATE           KVM_ARCH_REQ(6)
 
Quoted 8 lines

arch/riscv/include/asm/kvm_host.h line 51

Quoted 6 lines
arch/riscv/include/asm/kvm_host.h +6 -0
+#define KVM_HEDELEG_DEFAULT            ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_BREAKPOINT) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

Use BIT(xyz) here. For example: BIT(EXC_INST_MISALIGNED)

Thanks, I will use BIT() instead in next revision.

Quoted 3 lines

Also, BIT(EXC_BREAKPOINT) should not be part of KVM_HEDELEG_DEFAULT.

I think the bit EXC_BREAKPOINT should be set by default, like what you already did in kvm_arch_hardware_enable(). Then the VS could get the ebreak and handle it accordingly.

Quoted 6 lines
Quoted 3 lines

If the guest_debug is enabled, ebreak instructions are inserted by the userspace(QEMU). So KVM should 'intercept' the ebreak and exit to QEMU. Bit EXC_BREAKPOINT should be cleared in this case.

If EXC_BREAKPOINT is delegated by default then it is not consistent with vcpu->guest_debug which is not enabled by default.

Quoted 6 lines

To enable the guest_debug corresponding to NOT delegate the EXC_BREAKPOINT. They are somehow 'opposite'.

This 'kvm_guest_debug' feature is different from "debug in the guest". The later requires the delegation of EXC_BREAKPOINT.
The former does not.

In which case your below code is totally misleading.

(quoted diff) +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+       }

This should have been:

(quoted diff) +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg &= ~BIT(EXC_BREAKPOINT);
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg |= BIT(EXC_BREAKPOINT);
+       }
Quoted 98 lines
Quoted 87 lines
Quoted 87 lines
Quoted 86 lines

arch/riscv/include/asm/kvm_host.h line 56

Quoted 10 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+#define KVM_HEDELEG_GUEST_DEBUG                ((_AC(1, UL) << EXC_INST_MISALIGNED) | \
+                                        (_AC(1, UL) << EXC_SYSCALL) | \
+                                        (_AC(1, UL) << EXC_INST_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_LOAD_PAGE_FAULT) | \
+                                        (_AC(1, UL) << EXC_STORE_PAGE_FAULT))

No need for KVM_HEDELEG_GUEST_DEBUG, see below.

arch/riscv/include/asm/kvm_host.h line 61

Quoted 5 lines
arch/riscv/include/asm/kvm_host.h +5 -0
+
+#define KVM_HIDELEG_DEFAULT            ((_AC(1, UL) << IRQ_VS_SOFT) | \
+                                        (_AC(1, UL) << IRQ_VS_TIMER) | \
+                                        (_AC(1, UL) << IRQ_VS_EXT))
+

Same as above, use BIT(xyz) here.

Quoted 57 lines
arch/riscv/include/asm/kvm_host.h +1 -0
 enum kvm_riscv_hfence_type {
        KVM_RISCV_HFENCE_UNKNOWN = 0,
        KVM_RISCV_HFENCE_GVMA_VMID_GPA,
@@ -169,6 +185,7 @@ struct kvm_vcpu_csr {
 struct kvm_vcpu_config {
        u64 henvcfg;
        u64 hstateen0;
+       unsigned long hedeleg;
 };
 
 struct kvm_vcpu_smstateen_csr {
arch/riscv/include/uapi/asm/kvm.h +1 -0
index 7499e88a947c..39f4f4b9dede 100644
@@ -17,6 +17,7 @@
 
 #define __KVM_HAVE_IRQ_LINE
 #define __KVM_HAVE_READONLY_MEM
+#define __KVM_HAVE_GUEST_DEBUG
 
 #define KVM_COALESCED_MMIO_PAGE_OFFSET 1
 
arch/riscv/kvm/main.c +2 -16
index 225a435d9c9a..bab2ec34cd87 100644
@@ -22,22 +22,8 @@ long kvm_arch_dev_ioctl(struct file *filp,
 
 int kvm_arch_hardware_enable(void)
 {
-       unsigned long hideleg, hedeleg;
-
-       hedeleg = 0;
-       hedeleg |= (1UL << EXC_INST_MISALIGNED);
-       hedeleg |= (1UL << EXC_BREAKPOINT);
-       hedeleg |= (1UL << EXC_SYSCALL);
-       hedeleg |= (1UL << EXC_INST_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_LOAD_PAGE_FAULT);
-       hedeleg |= (1UL << EXC_STORE_PAGE_FAULT);
-       csr_write(CSR_HEDELEG, hedeleg);
-
-       hideleg = 0;
-       hideleg |= (1UL << IRQ_VS_SOFT);
-       hideleg |= (1UL << IRQ_VS_TIMER);
-       hideleg |= (1UL << IRQ_VS_EXT);
-       csr_write(CSR_HIDELEG, hideleg);
+       csr_write(CSR_HEDELEG, KVM_HEDELEG_DEFAULT);
+       csr_write(CSR_HIDELEG, KVM_HIDELEG_DEFAULT);
 
        /* VS should access only the time counter directly. Everything else should trap */
        csr_write(CSR_HCOUNTEREN, 0x02);
arch/riscv/kvm/vcpu.c +0 -0
index b5ca9f2e98ac..242076c2227f 100644

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 479

Quoted 5 lines
arch/riscv/kvm/vcpu.c +0 -2
@@ -475,8 +475,15 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
                                        struct kvm_guest_debug *dbg)
 {
-       /* TODO; To be implemented later. */
-       return -EINVAL;

if (vcpu->arch.ran_atleast_once)

        return -EBUSY;
Quoted 6 lines
Quoted 2 lines

If we enabled the guest_debug in QEMU side, then the KVM_SET_GUEST_DEBUG ioctl will come before the first KVM_RUN. This will always cause an ERROR.

The check ensures that KVM user space can only enable/disable guest debug before the VCPU is run. I don't see why this would fail for QEMU.

In the current implementation of GDB and QEMU, the userspace will enable/disable guest_debug frequently during the debugging (almost every step). The sequence should like:

Quoted 10 lines

KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
KVM_SET_GUEST_DEBUG enable
KVM_RUN
KVM_SET_GUEST_DEBUG disable
...

Fair enough, no need to check "ran_atleast_once"

Quoted 88 lines
Quoted 41 lines
Quoted 41 lines
Quoted 31 lines

arch/riscv/kvm/vcpu.c int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, · line 484

Quoted 12 lines
arch/riscv/kvm/vcpu.c +7 -0
+       if (dbg->control & KVM_GUESTDBG_ENABLE) {
+               vcpu->guest_debug = dbg->control;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_GUEST_DEBUG;
+       } else {
+               vcpu->guest_debug = 0;
+               vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT;
+       }

Don't update vcpu->arch.cfg.hedeleg here since it should be only done in kvm_riscv_vcpu_setup_config().

arch/riscv/kvm/vcpu.c static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu) · line 517

Quoted 12 lines
arch/riscv/kvm/vcpu.c +5 -0
+
+       return 0;
 }
 
 static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
@@ -505,6 +512,9 @@ static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu)
                if (riscv_isa_extension_available(isa, SMSTATEEN))
                        cfg->hstateen0 |= SMSTATEEN0_SSTATEEN0;
        }
+
+       if (!vcpu->guest_debug)
+               cfg->hedeleg = KVM_HEDELEG_DEFAULT;

This should be:

Quoted 3 lines

cfg->hedeleg = KVM_HEDELEG_DEFAULT;
if (vcpu->guest_debug)

        cfg->hedeleg |= BIT(EXC_BREAKPOINT);

Like above, here the logic should be:

cfg->hedeleg = KVM_HEDELEG_DEFAULT; // with BIT(EXC_BREAKPOINT) if (vcpu->guest_debug)

        cfg->hedeleg &= ~BIT(EXC_BREAKPOINT);
Quoted 5 lines
Quoted 3 lines

Another approach is:
initialize the cfg->hedeleg as KVM_HEDELEG_DEFAULT during kvm_arch_vcpu_create(). Besides that, only update the cfg->hedeleg in kvm_arch_vcpu_ioctl_set_guest_debug().

I disagree. We should handle hedeleg just like we handle henvcfg.

OK, let's only update the cfg->hedeleg in kvm_riscv_vcpu_setup_config().

Quoted 33 lines
Quoted 33 lines
Quoted 32 lines
Quoted 27 lines
arch/riscv/kvm/vcpu.c +1 -0
 }
 
 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
@@ -519,6 +529,7 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
        csr_write(CSR_VSEPC, csr->vsepc);
        csr_write(CSR_VSCAUSE, csr->vscause);
        csr_write(CSR_VSTVAL, csr->vstval);
+       csr_write(CSR_HEDELEG, cfg->hedeleg);
        csr_write(CSR_HVIP, csr->hvip);
        csr_write(CSR_VSATP, csr->vsatp);
        csr_write(CSR_HENVCFG, cfg->henvcfg);
arch/riscv/kvm/vm.c +1 -1
index ce58bc48e5b8..7396b8654f45 100644
@@ -186,6 +186,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
        case KVM_CAP_READONLY_MEM:
        case KVM_CAP_MP_STATE:
        case KVM_CAP_IMMEDIATE_EXIT:
+       case KVM_CAP_SET_GUEST_DEBUG:
                r = 1;
                break;
        case KVM_CAP_NR_VCPUS:
--
Quoted 2 lines

2.17.1

Regards,
Anup

Quoted 5 lines
Quoted 2 lines

Thanks,
Chao

Regards,
Anup

Quoted 6 lines

Thanks,
Chao
--
kvm-riscv mailing list
kvm-riscv at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kvm-riscv

Regards,
Anup

author
#11 lore raw

PATCH v2 2/3 RISC-V: KVM: Handle breakpoint exits for VCPU

In reply to Chao Du

Exit to userspace for breakpoint traps. Set the exit_reason as KVM_EXIT_DEBUG before exit.

---
 arch/riscv/kvm/vcpu_exit.c | 4 ++++
 1 file changed, 4 insertions(+)
arch/riscv/kvm/vcpu_exit.c +4 -0
index 2415722c01b8..5761f95abb60 100644
@@ -204,6 +204,10 @@ int kvm_riscv_vcpu_exit(struct kvm_vcpu *vcpu, struct kvm_run *run,
 		if (vcpu->arch.guest_context.hstatus & HSTATUS_SPV)
 			ret = kvm_riscv_vcpu_sbi_ecall(vcpu, run);
 		break;
+	case EXC_BREAKPOINT:
+		run->exit_reason = KVM_EXIT_DEBUG;
+		ret = 0;
+		break;
 	default:
 		break;
 	}
Signature · 1 line
2.17.1
maintainer Reviewed-by on v2 2/3
#12 lore raw

Re: [PATCH v2 2/3] RISC-V: KVM: Handle breakpoint exits for VCPU

Quoted 5 lines · On Fri, Mar 1, 2024 at 7:08?AM Chao Du <[email protected]> wrote:

Exit to userspace for breakpoint traps. Set the exit_reason as KVM_EXIT_DEBUG before exit.

LGTM.

Thanks,
Anup

Quoted 22 lines
---
 arch/riscv/kvm/vcpu_exit.c | 4 ++++
 1 file changed, 4 insertions(+)
arch/riscv/kvm/vcpu_exit.c +4 -1
index 2415722c01b8..5761f95abb60 100644
@@ -204,6 +204,10 @@ int kvm_riscv_vcpu_exit(struct kvm_vcpu *vcpu, struct kvm_run *run,
                if (vcpu->arch.guest_context.hstatus & HSTATUS_SPV)
                        ret = kvm_riscv_vcpu_sbi_ecall(vcpu, run);
                break;
+       case EXC_BREAKPOINT:
+               run->exit_reason = KVM_EXIT_DEBUG;
+               ret = 0;
+               break;
        default:
                break;
        }
--

2.17.1

author
#13 lore raw

PATCH v2 3/3 RISC-V: KVM: selftests: Add breakpoints test support

In reply to Chao Du

Initial support for RISC-V KVM breakpoint test. Check the exit reason and the PC when guest debug is enabled.

---
 tools/testing/selftests/kvm/Makefile          |  1 +
 .../testing/selftests/kvm/riscv/breakpoints.c | 49 +++++++++++++++++++
 2 files changed, 50 insertions(+)
 create mode 100644 tools/testing/selftests/kvm/riscv/breakpoints.c
tools/testing/selftests/kvm/Makefile +1 -0
index 492e937fab00..5f9048a740b0 100644
@@ -184,6 +184,7 @@ TEST_GEN_PROGS_s390x += rseq_test
 TEST_GEN_PROGS_s390x += set_memory_region_test
 TEST_GEN_PROGS_s390x += kvm_binary_stats_test
 
+TEST_GEN_PROGS_riscv += riscv/breakpoints
 TEST_GEN_PROGS_riscv += demand_paging_test
 TEST_GEN_PROGS_riscv += dirty_log_test
 TEST_GEN_PROGS_riscv += get-reg-list
tools/testing/selftests/kvm/riscv/breakpoints.c new file +49 -0

49 lines not on this page · raw message

Signature · 1 line
2.17.1
maintainer
#14 lore raw

Re: [PATCH v2 3/3] RISC-V: KVM: selftests: Add breakpoints test support

This is actually a ebreak test so subject should be:
"RISC-V: KVM: selftests: Add ebreak test support"

Quoted 3 lines · On Fri, Mar 1, 2024 at 7:08?AM Chao Du <[email protected]> wrote:

Initial support for RISC-V KVM breakpoint test. Check the exit reason and the PC when guest debug is enabled.

s/breakpoint/ebreak/

Quoted 5 lines
---
 tools/testing/selftests/kvm/Makefile          |  1 +
 .../testing/selftests/kvm/riscv/breakpoints.c | 49 +++++++++++++++++++

Rename riscv/breakpoints.c to riscv/ebreak_test.c

Quoted 21 lines
 2 files changed, 50 insertions(+)
 create mode 100644 tools/testing/selftests/kvm/riscv/breakpoints.c
tools/testing/selftests/kvm/Makefile +1 -0
index 492e937fab00..5f9048a740b0 100644
@@ -184,6 +184,7 @@ TEST_GEN_PROGS_s390x += rseq_test
 TEST_GEN_PROGS_s390x += set_memory_region_test
 TEST_GEN_PROGS_s390x += kvm_binary_stats_test
 
+TEST_GEN_PROGS_riscv += riscv/breakpoints
 TEST_GEN_PROGS_riscv += demand_paging_test
 TEST_GEN_PROGS_riscv += dirty_log_test
 TEST_GEN_PROGS_riscv += get-reg-list
tools/testing/selftests/kvm/riscv/breakpoints.c new file +0 -0
new file mode 100644
index 000000000000..be2d94837c83

tools/testing/selftests/kvm/riscv/breakpoints.c line 3

Quoted 3 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +3 -0
@@ -0,0 +1,49 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * RISC-V KVM breakpoint tests.

s/breakpoint tests/ebreak test/

Quoted 27 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +27 -0
+ *
+ * Copyright 2024 Beijing ESWIN Computing Technology Co., Ltd.
+ *
+ */
+#include "kvm_util.h"
+
+#define PC(v) ((uint64_t)&(v))
+
+extern unsigned char sw_bp;
+
+static void guest_code(void)
+{
+       asm volatile("sw_bp: ebreak");
+       asm volatile("nop");
+       asm volatile("nop");
+       asm volatile("nop");
+
+       GUEST_DONE();
+}
+
+int main(void)
+{
+       struct kvm_vm *vm;
+       struct kvm_vcpu *vcpu;
+       struct kvm_guest_debug debug;
+       uint64_t pc;
+

tools/testing/selftests/kvm/riscv/breakpoints.c line 40

Quoted 10 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +10 -0
+       TEST_REQUIRE(kvm_has_cap(KVM_CAP_SET_GUEST_DEBUG));
+
+       vm = vm_create_with_one_vcpu(&vcpu, guest_code);
+
+       memset(&debug, 0, sizeof(debug));
+       debug.control = KVM_GUESTDBG_ENABLE;
+       vcpu_guest_debug_set(vcpu, &debug);
+       vcpu_run(vcpu);
+
+       TEST_ASSERT_KVM_EXIT_REASON(vcpu, KVM_EXIT_DEBUG);

This is only testing the case where KVM_GUESTDBG_ENABLE is set.

This test should be enhanced to:
1) First test "ebreak" executed by Guest when KVM_GUESTDBG_ENABLE is not set. Here, we would expect main() to receive KVM_EXIT_DEBUG.
2) Second test "ebreak" executed by Guest when KVM_GUESTDBG_ENABLE is set. Here, we would expect main() will not receive KVM_EXIT_DEBUG.

Quoted 10 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +9 -1
+
+       vcpu_get_reg(vcpu, RISCV_CORE_REG(regs.pc), &pc);
+
+       TEST_ASSERT_EQ(pc, PC(sw_bp));
+
+       kvm_vm_free(vm);
+
+       return 0;
+}
--
Quoted 2 lines

2.17.1

Regards,
Anup

#15 lore raw

Re: [PATCH v2 3/3] RISC-V: KVM: selftests: Add breakpoints test support

In reply to Chao Du
Quoted 37 lines · Chao Du

Initial support for RISC-V KVM breakpoint test. Check the exit reason and the PC when guest debug is enabled.

---
 tools/testing/selftests/kvm/Makefile          |  1 +
 .../testing/selftests/kvm/riscv/breakpoints.c | 49 +++++++++++++++++++
 2 files changed, 50 insertions(+)
 create mode 100644 tools/testing/selftests/kvm/riscv/breakpoints.c
tools/testing/selftests/kvm/Makefile +1 -0
index 492e937fab00..5f9048a740b0 100644
@@ -184,6 +184,7 @@ TEST_GEN_PROGS_s390x += rseq_test
 TEST_GEN_PROGS_s390x += set_memory_region_test
 TEST_GEN_PROGS_s390x += kvm_binary_stats_test
 
+TEST_GEN_PROGS_riscv += riscv/breakpoints
 TEST_GEN_PROGS_riscv += demand_paging_test
 TEST_GEN_PROGS_riscv += dirty_log_test
 TEST_GEN_PROGS_riscv += get-reg-list
tools/testing/selftests/kvm/riscv/breakpoints.c new file +9 -0
new file mode 100644
index 000000000000..be2d94837c83
@@ -0,0 +1,49 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * RISC-V KVM breakpoint tests.
+ *
+ * Copyright 2024 Beijing ESWIN Computing Technology Co., Ltd.
+ *
+ */
+#include "kvm_util.h"
+

tools/testing/selftests/kvm/riscv/breakpoints.c line 19

Quoted 10 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +10 -0
+#define PC(v) ((uint64_t)&(v))
+
+extern unsigned char sw_bp;
+
+static void guest_code(void)
+{
+	asm volatile("sw_bp: ebreak");
+	asm volatile("nop");
+	asm volatile("nop");
+	asm volatile("nop");

What are the nops for? And, since they're all in their own asm()'s the compiler could be inserting instructions between them and also the ebreak above. If we need three nops immediately following the ebreak then we need to put everything in one asm()

asm volatile(
"sw_bp: ebreak\n"
" nop\n"
" nop\n"
" nop\n");

Quoted 7 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +7 -0
+
+	GUEST_DONE();
+}
+
+int main(void)
+{
+	struct kvm_vm *vm;

tools/testing/selftests/kvm/riscv/breakpoints.c line 36

Quoted 10 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +10 -0
+	struct kvm_vcpu *vcpu;
+	struct kvm_guest_debug debug;
+	uint64_t pc;
+
+	TEST_REQUIRE(kvm_has_cap(KVM_CAP_SET_GUEST_DEBUG));
+
+	vm = vm_create_with_one_vcpu(&vcpu, guest_code);
+
+	memset(&debug, 0, sizeof(debug));
+	debug.control = KVM_GUESTDBG_ENABLE;

nit: The above two lines can be removed if we initialize debug as

  struct kvm_guest_debug debug = {
     .control = KVM_GUESTDBG_ENABLE,
  };

tools/testing/selftests/kvm/riscv/breakpoints.c line 40

Quoted 4 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +4 -0
+	vcpu_guest_debug_set(vcpu, &debug);
+	vcpu_run(vcpu);
+
+	TEST_ASSERT_KVM_EXIT_REASON(vcpu, KVM_EXIT_DEBUG);

As Anup pointed out, we need to also ensure that without making the KVM_SET_GUEST_DEBUG ioctl call we get the expected behavior. You can use GUEST_SYNC() in the guest code to prove that it was able to issue an ebreak without exiting to the VMM.

Quoted 12 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +9 -0
+
+	vcpu_get_reg(vcpu, RISCV_CORE_REG(regs.pc), &pc);
+
+	TEST_ASSERT_EQ(pc, PC(sw_bp));
+
+	kvm_vm_free(vm);
+
+	return 0;
+}
Signature · 1 line
2.17.1

Thanks,
drew

author
#16 lore raw

Re: [PATCH v2 3/3] RISC-V: KVM: selftests: Add breakpoints test support

Quoted 55 lines · Andrew Jones
Quoted 37 lines · Chao Du

Initial support for RISC-V KVM breakpoint test. Check the exit reason and the PC when guest debug is enabled.

---
 tools/testing/selftests/kvm/Makefile          |  1 +
 .../testing/selftests/kvm/riscv/breakpoints.c | 49 +++++++++++++++++++
 2 files changed, 50 insertions(+)
 create mode 100644 tools/testing/selftests/kvm/riscv/breakpoints.c
tools/testing/selftests/kvm/Makefile +1 -0
index 492e937fab00..5f9048a740b0 100644
@@ -184,6 +184,7 @@ TEST_GEN_PROGS_s390x += rseq_test
 TEST_GEN_PROGS_s390x += set_memory_region_test
 TEST_GEN_PROGS_s390x += kvm_binary_stats_test
 
+TEST_GEN_PROGS_riscv += riscv/breakpoints
 TEST_GEN_PROGS_riscv += demand_paging_test
 TEST_GEN_PROGS_riscv += dirty_log_test
 TEST_GEN_PROGS_riscv += get-reg-list
tools/testing/selftests/kvm/riscv/breakpoints.c new file +9 -0
new file mode 100644
index 000000000000..be2d94837c83
@@ -0,0 +1,49 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * RISC-V KVM breakpoint tests.
+ *
+ * Copyright 2024 Beijing ESWIN Computing Technology Co., Ltd.
+ *
+ */
+#include "kvm_util.h"
+

tools/testing/selftests/kvm/riscv/breakpoints.c line 19

Quoted 10 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +10 -0
+#define PC(v) ((uint64_t)&(v))
+
+extern unsigned char sw_bp;
+
+static void guest_code(void)
+{
+	asm volatile("sw_bp: ebreak");
+	asm volatile("nop");
+	asm volatile("nop");
+	asm volatile("nop");

What are the nops for? And, since they're all in their own asm()'s the compiler could be inserting instructions between them and also the ebreak above. If we need three nops immediately following the ebreak then we need to put everything in one asm()

Quoted 6 lines

asm volatile(
"sw_bp: ebreak\n"
" nop\n"
" nop\n"
" nop\n");

Actually the nops have no special purpose, I will remove and just keep the ebreak here.

Quoted 20 lines
Quoted 7 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +7 -0
+
+	GUEST_DONE();
+}
+
+int main(void)
+{
+	struct kvm_vm *vm;

tools/testing/selftests/kvm/riscv/breakpoints.c line 36

Quoted 10 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +10 -0
+	struct kvm_vcpu *vcpu;
+	struct kvm_guest_debug debug;
+	uint64_t pc;
+
+	TEST_REQUIRE(kvm_has_cap(KVM_CAP_SET_GUEST_DEBUG));
+
+	vm = vm_create_with_one_vcpu(&vcpu, guest_code);
+
+	memset(&debug, 0, sizeof(debug));
+	debug.control = KVM_GUESTDBG_ENABLE;

nit: The above two lines can be removed if we initialize debug as

Quoted 4 lines
  struct kvm_guest_debug debug = {
     .control = KVM_GUESTDBG_ENABLE,
  };

Yeah, I will do so in the next patch.

Quoted 10 lines

tools/testing/selftests/kvm/riscv/breakpoints.c line 40

Quoted 4 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +4 -0
+	vcpu_guest_debug_set(vcpu, &debug);
+	vcpu_run(vcpu);
+
+	TEST_ASSERT_KVM_EXIT_REASON(vcpu, KVM_EXIT_DEBUG);

As Anup pointed out, we need to also ensure that without making the KVM_SET_GUEST_DEBUG ioctl call we get the expected behavior. You can use GUEST_SYNC() in the guest code to prove that it was able to issue an ebreak without exiting to the VMM.

Sure, will cover that case also.

Quoted 15 lines
Quoted 12 lines
tools/testing/selftests/kvm/riscv/breakpoints.c new file +9 -0
+
+	vcpu_get_reg(vcpu, RISCV_CORE_REG(regs.pc), &pc);
+
+	TEST_ASSERT_EQ(pc, PC(sw_bp));
+
+	kvm_vm_free(vm);
+
+	return 0;
+}
Signature · 1 line
2.17.1

Thanks,
drew