Skip to contentSkip to conversation

RISC-V: KVM: change get_reg/set_reg error codes

PATCH Superseded Daniel Henrique Barboza 1 participant 12 messages Aug 3, 2023

Subsystems (4)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v3 cover RISC-V: KVM: change get_reg/set_reg error codes

Hi,

This version has changes in the document patch, as suggested by Andrew in v2. It also has a new patch (patch 9) that handles error code changes in vcpu_vector.c.

Patches rebased on top of kvm_riscv_queue.

Changes from v2:
- patch 9 (new):
- change kvm error codes for vector registers
- patch 10 (former 9):
- rewrite EBUSY doc to mention that the error code indicates that it
is not allowed to change the reg val after the vcpu started.
- v2 link: https://lore.kernel.org/kvm/20230801222629.210929-1-dbarboza at ventanamicro.com/

Andrew Jones (1):
RISC-V: KVM: Improve vector save/restore errors

Daniel Henrique Barboza (9):
RISC-V: KVM: return ENOENT in *_one_reg() when reg is unknown
RISC-V: KVM: use ENOENT in *_one_reg() when extension is unavailable
RISC-V: KVM: do not EOPNOTSUPP in set_one_reg() zicbo(m|z)
RISC-V: KVM: do not EOPNOTSUPP in set KVM_REG_RISCV_TIMER_REG
RISC-V: KVM: use EBUSY when !vcpu->arch.ran_atleast_once
RISC-V: KVM: avoid EBUSY when writing same ISA val
RISC-V: KVM: avoid EBUSY when writing the same machine ID val
RISC-V: KVM: avoid EBUSY when writing the same isa_ext val
docs: kvm: riscv: document EBUSY in KVM_SET_ONE_REG

 Documentation/virt/kvm/api.rst |  2 +
 arch/riscv/kvm/aia.c           |  4 +-
 arch/riscv/kvm/vcpu_fp.c       | 12 +++---
 arch/riscv/kvm/vcpu_onereg.c   | 68 +++++++++++++++++++++++-----------
 arch/riscv/kvm/vcpu_sbi.c      | 16 ++++----
 arch/riscv/kvm/vcpu_timer.c    | 11 +++---
 arch/riscv/kvm/vcpu_vector.c   | 60 ++++++++++++++++--------------
 7 files changed, 104 insertions(+), 69 deletions(-)
Signature · 1 line
2.41.0
#1 lore raw

PATCH v3 1/10 RISC-V: KVM: return ENOENT in *_one_reg() when reg is unknown

get_one_reg() and set_one_reg() are returning EINVAL errors for almost
everything: if a reg doesn't exist, if a reg ID is malformatted, if the associated CPU extension that implements the reg isn't present in the host, and for set_one_reg() if the value being written is invalid.

This isn't wrong according to the existing KVM API docs (EINVAL can be used when there's no such register) but adding more ENOENT instances will make easier for userspace to understand what went wrong.

Existing userspaces can be affected by this error code change. We checked a few. As of current upstream code, crosvm doesn't check for any particular errno code when using kvm_(get|set)_one_reg(). Neither does QEMU. rust-vmm doesn't have kvm-riscv support yet. Thus we have a good chance of changing these error codes now while the KVM RISC-V ecosystem is still new, minimizing user impact.

Change all get_one_reg() and set_one_reg() implementations to return -ENOENT at all "no such register" cases.

---
 arch/riscv/kvm/aia.c         |  4 ++--
 arch/riscv/kvm/vcpu_fp.c     | 12 ++++++------
 arch/riscv/kvm/vcpu_onereg.c | 30 +++++++++++++++---------------
 arch/riscv/kvm/vcpu_sbi.c    | 16 +++++++++-------
 arch/riscv/kvm/vcpu_timer.c  |  8 ++++----
 5 files changed, 36 insertions(+), 34 deletions(-)
arch/riscv/kvm/aia.c +2 -2
index 585a3b42c52c..74bb27440527 100644
@@ -176,7 +176,7 @@ int kvm_riscv_vcpu_aia_get_csr(struct kvm_vcpu *vcpu,
 	struct kvm_vcpu_aia_csr *csr = &vcpu->arch.aia_context.guest_csr;
 
 	if (reg_num >= sizeof(struct kvm_riscv_aia_csr) / sizeof(unsigned long))
-		return -EINVAL;
+		return -ENOENT;
 
 	*out_val = 0;
 	if (kvm_riscv_aia_available())
@@ -192,7 +192,7 @@ int kvm_riscv_vcpu_aia_set_csr(struct kvm_vcpu *vcpu,
 	struct kvm_vcpu_aia_csr *csr = &vcpu->arch.aia_context.guest_csr;
 
 	if (reg_num >= sizeof(struct kvm_riscv_aia_csr) / sizeof(unsigned long))
-		return -EINVAL;
+		return -ENOENT;
 
 	if (kvm_riscv_aia_available()) {
 		((unsigned long *)csr)[reg_num] = val;
arch/riscv/kvm/vcpu_fp.c +6 -6
index 9d8cbc42057a..08ba48a395aa 100644
@@ -96,7 +96,7 @@ int kvm_riscv_vcpu_get_reg_fp(struct kvm_vcpu *vcpu,
 			  reg_num <= KVM_REG_RISCV_FP_F_REG(f[31]))
 			reg_val = &cntx->fp.f.f[reg_num];
 		else
-			return -EINVAL;
+			return -ENOENT;
 	} else if ((rtype == KVM_REG_RISCV_FP_D) &&
 		   riscv_isa_extension_available(vcpu->arch.isa, d)) {
 		if (reg_num == KVM_REG_RISCV_FP_D_REG(fcsr)) {
@@ -109,9 +109,9 @@ int kvm_riscv_vcpu_get_reg_fp(struct kvm_vcpu *vcpu,
 				return -EINVAL;
 			reg_val = &cntx->fp.d.f[reg_num];
 		} else
-			return -EINVAL;
+			return -ENOENT;
 	} else
-		return -EINVAL;
+		return -ENOENT;
 
 	if (copy_to_user(uaddr, reg_val, KVM_REG_SIZE(reg->id)))
 		return -EFAULT;
@@ -141,7 +141,7 @@ int kvm_riscv_vcpu_set_reg_fp(struct kvm_vcpu *vcpu,
 			  reg_num <= KVM_REG_RISCV_FP_F_REG(f[31]))
 			reg_val = &cntx->fp.f.f[reg_num];
 		else
-			return -EINVAL;
+			return -ENOENT;
 	} else if ((rtype == KVM_REG_RISCV_FP_D) &&
 		   riscv_isa_extension_available(vcpu->arch.isa, d)) {
 		if (reg_num == KVM_REG_RISCV_FP_D_REG(fcsr)) {
@@ -154,9 +154,9 @@ int kvm_riscv_vcpu_set_reg_fp(struct kvm_vcpu *vcpu,
 				return -EINVAL;
 			reg_val = &cntx->fp.d.f[reg_num];
 		} else
-			return -EINVAL;
+			return -ENOENT;
 	} else
-		return -EINVAL;
+		return -ENOENT;
 
 	if (copy_from_user(reg_val, uaddr, KVM_REG_SIZE(reg->id)))
 		return -EFAULT;
arch/riscv/kvm/vcpu_onereg.c +15 -15
index 85773e858120..65607f80f8db 100644
@@ -156,7 +156,7 @@ static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu,
 		reg_val = satp_mode >> SATP_MODE_SHIFT;
 		break;
 	default:
-		return -EINVAL;
+		return -ENOENT;
 	}
 
 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
@@ -242,7 +242,7 @@ static int kvm_riscv_vcpu_set_reg_config(struct kvm_vcpu *vcpu,
 			return -EINVAL;
 		break;
 	default:
-		return -EINVAL;
+		return -ENOENT;
 	}
 
 	return 0;
@@ -262,7 +262,7 @@ static int kvm_riscv_vcpu_get_reg_core(struct kvm_vcpu *vcpu,
 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
 		return -EINVAL;
 	if (reg_num >= sizeof(struct kvm_riscv_core) / sizeof(unsigned long))
-		return -EINVAL;
+		return -ENOENT;
 
 	if (reg_num == KVM_REG_RISCV_CORE_REG(regs.pc))
 		reg_val = cntx->sepc;
@@ -273,7 +273,7 @@ static int kvm_riscv_vcpu_get_reg_core(struct kvm_vcpu *vcpu,
 		reg_val = (cntx->sstatus & SR_SPP) ?
 				KVM_RISCV_MODE_S : KVM_RISCV_MODE_U;
 	else
-		return -EINVAL;
+		return -ENOENT;
 
 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
 		return -EFAULT;
@@ -295,7 +295,7 @@ static int kvm_riscv_vcpu_set_reg_core(struct kvm_vcpu *vcpu,
 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
 		return -EINVAL;
 	if (reg_num >= sizeof(struct kvm_riscv_core) / sizeof(unsigned long))
-		return -EINVAL;
+		return -ENOENT;
 
 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
 		return -EFAULT;
@@ -311,7 +311,7 @@ static int kvm_riscv_vcpu_set_reg_core(struct kvm_vcpu *vcpu,
 		else
 			cntx->sstatus &= ~SR_SPP;
 	} else
-		return -EINVAL;
+		return -ENOENT;
 
 	return 0;
 }
@@ -323,7 +323,7 @@ static int kvm_riscv_vcpu_general_get_csr(struct kvm_vcpu *vcpu,
 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
 
 	if (reg_num >= sizeof(struct kvm_riscv_csr) / sizeof(unsigned long))
-		return -EINVAL;
+		return -ENOENT;
 
 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip)) {
 		kvm_riscv_vcpu_flush_interrupts(vcpu);
@@ -342,7 +342,7 @@ static int kvm_riscv_vcpu_general_set_csr(struct kvm_vcpu *vcpu,
 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
 
 	if (reg_num >= sizeof(struct kvm_riscv_csr) / sizeof(unsigned long))
-		return -EINVAL;
+		return -ENOENT;
 
 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip)) {
 		reg_val &= VSIP_VALID_MASK;
@@ -381,7 +381,7 @@ static int kvm_riscv_vcpu_get_reg_csr(struct kvm_vcpu *vcpu,
 		rc = kvm_riscv_vcpu_aia_get_csr(vcpu, reg_num, &reg_val);
 		break;
 	default:
-		rc = -EINVAL;
+		rc = -ENOENT;
 		break;
 	}
 	if (rc)
@@ -420,7 +420,7 @@ static int kvm_riscv_vcpu_set_reg_csr(struct kvm_vcpu *vcpu,
 		rc = kvm_riscv_vcpu_aia_set_csr(vcpu, reg_num, reg_val);
 		break;
 	default:
-		rc = -EINVAL;
+		rc = -ENOENT;
 		break;
 	}
 	if (rc)
@@ -437,7 +437,7 @@ static int riscv_vcpu_get_isa_ext_single(struct kvm_vcpu *vcpu,
 
 	if (reg_num >= KVM_RISCV_ISA_EXT_MAX ||
 	    reg_num >= ARRAY_SIZE(kvm_isa_ext_arr))
-		return -EINVAL;
+		return -ENOENT;
 
 	*reg_val = 0;
 	host_isa_ext = kvm_isa_ext_arr[reg_num];
@@ -455,7 +455,7 @@ static int riscv_vcpu_set_isa_ext_single(struct kvm_vcpu *vcpu,
 
 	if (reg_num >= KVM_RISCV_ISA_EXT_MAX ||
 	    reg_num >= ARRAY_SIZE(kvm_isa_ext_arr))
-		return -EINVAL;
+		return -ENOENT;
 
 	host_isa_ext = kvm_isa_ext_arr[reg_num];
 	if (!__riscv_isa_extension_available(NULL, host_isa_ext))
@@ -554,7 +554,7 @@ static int kvm_riscv_vcpu_get_reg_isa_ext(struct kvm_vcpu *vcpu,
 			reg_val = ~reg_val;
 		break;
 	default:
-		rc = -EINVAL;
+		rc = -ENOENT;
 	}
 	if (rc)
 		return rc;
@@ -592,7 +592,7 @@ static int kvm_riscv_vcpu_set_reg_isa_ext(struct kvm_vcpu *vcpu,
 	case KVM_REG_RISCV_SBI_MULTI_DIS:
 		return riscv_vcpu_set_isa_ext_multi(vcpu, reg_num, reg_val, false);
 	default:
-		return -EINVAL;
+		return -ENOENT;
 	}
 
 	return 0;
@@ -659,5 +659,5 @@ int kvm_riscv_vcpu_get_reg(struct kvm_vcpu *vcpu,
 		break;
 	}
 
-	return -EINVAL;
+	return -ENOENT;
 }
arch/riscv/kvm/vcpu_sbi.c +9 -7
index 7b46e04fb667..9cd97091c723 100644
@@ -140,8 +140,10 @@ static int riscv_vcpu_set_sbi_ext_single(struct kvm_vcpu *vcpu,
 	const struct kvm_riscv_sbi_extension_entry *sext = NULL;
 	struct kvm_vcpu_sbi_context *scontext = &vcpu->arch.sbi_context;
 
-	if (reg_num >= KVM_RISCV_SBI_EXT_MAX ||
-	    (reg_val != 1 && reg_val != 0))
+	if (reg_num >= KVM_RISCV_SBI_EXT_MAX)
+		return -ENOENT;
+
+	if (reg_val != 1 && reg_val != 0)
 		return -EINVAL;
 
 	for (i = 0; i < ARRAY_SIZE(sbi_ext); i++) {
@@ -175,7 +177,7 @@ static int riscv_vcpu_get_sbi_ext_single(struct kvm_vcpu *vcpu,
 	struct kvm_vcpu_sbi_context *scontext = &vcpu->arch.sbi_context;
 
 	if (reg_num >= KVM_RISCV_SBI_EXT_MAX)
-		return -EINVAL;
+		return -ENOENT;
 
 	for (i = 0; i < ARRAY_SIZE(sbi_ext); i++) {
 		if (sbi_ext[i].ext_idx == reg_num) {
@@ -206,7 +208,7 @@ static int riscv_vcpu_set_sbi_ext_multi(struct kvm_vcpu *vcpu,
 	unsigned long i, ext_id;
 
 	if (reg_num > KVM_REG_RISCV_SBI_MULTI_REG_LAST)
-		return -EINVAL;
+		return -ENOENT;
 
 	for_each_set_bit(i, &reg_val, BITS_PER_LONG) {
 		ext_id = i + reg_num * BITS_PER_LONG;
@@ -226,7 +228,7 @@ static int riscv_vcpu_get_sbi_ext_multi(struct kvm_vcpu *vcpu,
 	unsigned long i, ext_id, ext_val;
 
 	if (reg_num > KVM_REG_RISCV_SBI_MULTI_REG_LAST)
-		return -EINVAL;
+		return -ENOENT;
 
 	for (i = 0; i < BITS_PER_LONG; i++) {
 		ext_id = i + reg_num * BITS_PER_LONG;
@@ -272,7 +274,7 @@ int kvm_riscv_vcpu_set_reg_sbi_ext(struct kvm_vcpu *vcpu,
 	case KVM_REG_RISCV_SBI_MULTI_DIS:
 		return riscv_vcpu_set_sbi_ext_multi(vcpu, reg_num, reg_val, false);
 	default:
-		return -EINVAL;
+		return -ENOENT;
 	}
 
 	return 0;
@@ -307,7 +309,7 @@ int kvm_riscv_vcpu_get_reg_sbi_ext(struct kvm_vcpu *vcpu,
 			reg_val = ~reg_val;
 		break;
 	default:
-		rc = -EINVAL;
+		rc = -ENOENT;
 	}
 	if (rc)
 		return rc;
arch/riscv/kvm/vcpu_timer.c +4 -4
index 3ac2ff6a65da..527d269cafff 100644
@@ -170,7 +170,7 @@ int kvm_riscv_vcpu_get_reg_timer(struct kvm_vcpu *vcpu,
 	if (KVM_REG_SIZE(reg->id) != sizeof(u64))
 		return -EINVAL;
 	if (reg_num >= sizeof(struct kvm_riscv_timer) / sizeof(u64))
-		return -EINVAL;
+		return -ENOENT;
 
 	switch (reg_num) {
 	case KVM_REG_RISCV_TIMER_REG(frequency):
@@ -187,7 +187,7 @@ int kvm_riscv_vcpu_get_reg_timer(struct kvm_vcpu *vcpu,
 					  KVM_RISCV_TIMER_STATE_OFF;
 		break;
 	default:
-		return -EINVAL;
+		return -ENOENT;
 	}
 
 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
@@ -211,7 +211,7 @@ int kvm_riscv_vcpu_set_reg_timer(struct kvm_vcpu *vcpu,
 	if (KVM_REG_SIZE(reg->id) != sizeof(u64))
 		return -EINVAL;
 	if (reg_num >= sizeof(struct kvm_riscv_timer) / sizeof(u64))
-		return -EINVAL;
+		return -ENOENT;
 
 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
 		return -EFAULT;
@@ -233,7 +233,7 @@ int kvm_riscv_vcpu_set_reg_timer(struct kvm_vcpu *vcpu,
 			ret = kvm_riscv_vcpu_timer_cancel(t);
 		break;
 	default:
-		ret = -EINVAL;
+		ret = -ENOENT;
 		break;
 	}
 
Signature · 1 line
2.41.0
#2 lore raw

PATCH v3 2/10 RISC-V: KVM: use ENOENT in *_one_reg() when extension is unavailable

In reply to Daniel Henrique Barboza

Following a similar logic as the previous patch let's minimize the EINVAL usage in *_one_reg() APIs by using ENOENT when an extension that implements the reg is not available.

For consistency we're also replacing an EOPNOTSUPP instance that should be an ENOENT since it's an "extension is not available" error.

---
 arch/riscv/kvm/vcpu_onereg.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
arch/riscv/kvm/vcpu_onereg.c +3 -3
index 65607f80f8db..546f75930d63 100644
@@ -135,12 +135,12 @@ static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu,
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(zicbom_block_size):
 		if (!riscv_isa_extension_available(vcpu->arch.isa, ZICBOM))
-			return -EINVAL;
+			return -ENOENT;
 		reg_val = riscv_cbom_block_size;
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(zicboz_block_size):
 		if (!riscv_isa_extension_available(vcpu->arch.isa, ZICBOZ))
-			return -EINVAL;
+			return -ENOENT;
 		reg_val = riscv_cboz_block_size;
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(mvendorid):
@@ -459,7 +459,7 @@ static int riscv_vcpu_set_isa_ext_single(struct kvm_vcpu *vcpu,
 
 	host_isa_ext = kvm_isa_ext_arr[reg_num];
 	if (!__riscv_isa_extension_available(NULL, host_isa_ext))
-		return	-EOPNOTSUPP;
+		return -ENOENT;
 
 	if (!vcpu->arch.ran_atleast_once) {
 		/*
Signature · 1 line
2.41.0
#3 lore raw

PATCH v3 3/10 RISC-V: KVM: do not EOPNOTSUPP in set_one_reg() zicbo(m|z)

In reply to Daniel Henrique Barboza

zicbom_block_size and zicboz_block_size have a peculiar API: they can be read via get_one_reg() but any write will return a EOPNOTSUPP.

It makes sense to return a 'not supported' error since both values can't be changed, but as far as userspace goes they're regs that are throwing the same EOPNOTSUPP error even if they were read beforehand via get_one_reg(), even if the same read value is being written back. EOPNOTSUPP is also returned even if ZICBOM/ZICBOZ aren't enabled in the host.

Change both to work more like their counterparts in get_one_reg() and return -ENOENT if their respective extensions aren't available. After that, check if the userspace is written a valid value (i.e. the host value). Throw an -EINVAL if that's not case, let it slide otherwise.

This allows both regs to be read/written by userspace in a 'lazy' manner, as long as the userspace doesn't change the reg vals.

---
 arch/riscv/kvm/vcpu_onereg.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)
arch/riscv/kvm/vcpu_onereg.c +10 -2
index 546f75930d63..49d5676928e4 100644
@@ -216,9 +216,17 @@ static int kvm_riscv_vcpu_set_reg_config(struct kvm_vcpu *vcpu,
 		}
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(zicbom_block_size):
-		return -EOPNOTSUPP;
+		if (!riscv_isa_extension_available(vcpu->arch.isa, ZICBOM))
+			return -ENOENT;
+		if (reg_val != riscv_cbom_block_size)
+			return -EINVAL;
+		break;
 	case KVM_REG_RISCV_CONFIG_REG(zicboz_block_size):
-		return -EOPNOTSUPP;
+		if (!riscv_isa_extension_available(vcpu->arch.isa, ZICBOZ))
+			return -ENOENT;
+		if (reg_val != riscv_cboz_block_size)
+			return -EINVAL;
+		break;
 	case KVM_REG_RISCV_CONFIG_REG(mvendorid):
 		if (!vcpu->arch.ran_atleast_once)
 			vcpu->arch.mvendorid = reg_val;
Signature · 1 line
2.41.0
#4 lore raw

PATCH v3 4/10 RISC-V: KVM: do not EOPNOTSUPP in set KVM_REG_RISCV_TIMER_REG

In reply to Daniel Henrique Barboza

The KVM_REG_RISCV_TIMER_REG can be read via get_one_reg(). But trying to write anything in this reg via set_one_reg() results in an EOPNOTSUPP.

Change the API to behave like cbom_block_size: instead of always erroring out with EOPNOTSUPP, allow userspace to write the same value (riscv_timebase) back, throwing an EINVAL if a different value is attempted.

---
 arch/riscv/kvm/vcpu_timer.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
arch/riscv/kvm/vcpu_timer.c +2 -1
index 527d269cafff..75486b25ac45 100644
@@ -218,7 +218,8 @@ int kvm_riscv_vcpu_set_reg_timer(struct kvm_vcpu *vcpu,
 
 	switch (reg_num) {
 	case KVM_REG_RISCV_TIMER_REG(frequency):
-		ret = -EOPNOTSUPP;
+		if (reg_val != riscv_timebase)
+			return -EINVAL;
 		break;
 	case KVM_REG_RISCV_TIMER_REG(time):
 		gt->time_delta = reg_val - get_cycles64();
Signature · 1 line
2.41.0
#5 lore raw

PATCH v3 5/10 RISC-V: KVM: use EBUSY when !vcpu->arch.ran_atleast_once

In reply to Daniel Henrique Barboza

vcpu_set_reg_config() and vcpu_set_reg_isa_ext() is throwing an EOPNOTSUPP error when !vcpu->arch.ran_atleast_once. In similar cases we're throwing an EBUSY error, like in mvendorid/marchid/mimpid set_reg().

EOPNOTSUPP has a conotation of finality. EBUSY is more adequate in this case since its a condition/error related to the vcpu lifecycle.

Change these EOPNOTSUPP instances to EBUSY.

---
 arch/riscv/kvm/vcpu_onereg.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/riscv/kvm/vcpu_onereg.c +2 -2
index 49d5676928e4..0cf25c18b582 100644
@@ -212,7 +212,7 @@ static int kvm_riscv_vcpu_set_reg_config(struct kvm_vcpu *vcpu,
 			vcpu->arch.isa[0] = reg_val;
 			kvm_riscv_vcpu_fp_reset(vcpu);
 		} else {
-			return -EOPNOTSUPP;
+			return -EBUSY;
 		}
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(zicbom_block_size):
@@ -484,7 +484,7 @@ static int riscv_vcpu_set_isa_ext_single(struct kvm_vcpu *vcpu,
 			return -EINVAL;
 		kvm_riscv_vcpu_fp_reset(vcpu);
 	} else {
-		return -EOPNOTSUPP;
+		return -EBUSY;
 	}
 
 	return 0;
Signature · 1 line
2.41.0
#6 lore raw

PATCH v3 6/10 RISC-V: KVM: avoid EBUSY when writing same ISA val

In reply to Daniel Henrique Barboza

kvm_riscv_vcpu_set_reg_config() will return -EBUSY if the ISA config reg is being written after the VCPU ran at least once.

The same restriction isn't placed in kvm_riscv_vcpu_get_reg_config(), so there's a chance that we'll -EBUSY out on an ISA config reg write even if the userspace intended no changes to it.

We'll allow the same form of 'lazy writing' that registers such as zicbom/zicboz_block_size supports: avoid erroring out if userspace made no changes to the ISA config reg.

---
 arch/riscv/kvm/vcpu_onereg.c | 7 +++++++
 1 file changed, 7 insertions(+)
arch/riscv/kvm/vcpu_onereg.c +7 -0
index 0cf25c18b582..e752e2dca8ed 100644
@@ -190,6 +190,13 @@ static int kvm_riscv_vcpu_set_reg_config(struct kvm_vcpu *vcpu,
 		if (fls(reg_val) >= RISCV_ISA_EXT_BASE)
 			return -EINVAL;
 
+		/*
+		 * Return early (i.e. do nothing) if reg_val is the same
+		 * value retrievable via kvm_riscv_vcpu_get_reg_config().
+		 */
+		if (reg_val == (vcpu->arch.isa[0] & KVM_RISCV_BASE_ISA_MASK))
+			break;
+
 		if (!vcpu->arch.ran_atleast_once) {
 			/* Ignore the enable/disable request for certain extensions */
 			for (i = 0; i < RISCV_ISA_EXT_BASE; i++) {
Signature · 1 line
2.41.0
#7 lore raw

PATCH v3 7/10 RISC-V: KVM: avoid EBUSY when writing the same machine ID val

In reply to Daniel Henrique Barboza

Right now we do not allow any write in mvendorid/marchid/mimpid if the vcpu already started, preventing these regs to be changed.

However, if userspace doesn't change them, an alternative is to consider the reg write a no-op and avoid erroring out altogether. Userpace can then be oblivious about KVM internals if no changes were intended in the first place.

Allow the same form of 'lazy writing' that registers such as zicbom/zicboz_block_size supports: avoid erroring out if userspace makes no changes in mvendorid/marchid/mimpid during reg write.

---
 arch/riscv/kvm/vcpu_onereg.c | 6 ++++++
 1 file changed, 6 insertions(+)
arch/riscv/kvm/vcpu_onereg.c +6 -0
index e752e2dca8ed..818900f30859 100644
@@ -235,18 +235,24 @@ static int kvm_riscv_vcpu_set_reg_config(struct kvm_vcpu *vcpu,
 			return -EINVAL;
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(mvendorid):
+		if (reg_val == vcpu->arch.mvendorid)
+			break;
 		if (!vcpu->arch.ran_atleast_once)
 			vcpu->arch.mvendorid = reg_val;
 		else
 			return -EBUSY;
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(marchid):
+		if (reg_val == vcpu->arch.marchid)
+			break;
 		if (!vcpu->arch.ran_atleast_once)
 			vcpu->arch.marchid = reg_val;
 		else
 			return -EBUSY;
 		break;
 	case KVM_REG_RISCV_CONFIG_REG(mimpid):
+		if (reg_val == vcpu->arch.mimpid)
+			break;
 		if (!vcpu->arch.ran_atleast_once)
 			vcpu->arch.mimpid = reg_val;
 		else
Signature · 1 line
2.41.0
#8 lore raw

PATCH v3 8/10 RISC-V: KVM: avoid EBUSY when writing the same isa_ext val

In reply to Daniel Henrique Barboza

riscv_vcpu_set_isa_ext_single() will prevent any write of isa_ext regs if the vcpu already started spinning.

But if there's no extension state (enabled/disabled) made by the userspace, there's no need to -EBUSY out - we can treat the operation as a no-op.

zicbom/zicboz_block_size, ISA config reg and mvendorid/march/mimpid already works in a more permissive manner w.r.t userspace writes being a no-op, so let's do the same with isa_ext writes.

---
 arch/riscv/kvm/vcpu_onereg.c | 3 +++
 1 file changed, 3 insertions(+)
arch/riscv/kvm/vcpu_onereg.c +3 -0
index 818900f30859..fcfe2049effd 100644
@@ -482,6 +482,9 @@ static int riscv_vcpu_set_isa_ext_single(struct kvm_vcpu *vcpu,
 	if (!__riscv_isa_extension_available(NULL, host_isa_ext))
 		return -ENOENT;
 
+	if (reg_val == test_bit(host_isa_ext, vcpu->arch.isa))
+		return 0;
+
 	if (!vcpu->arch.ran_atleast_once) {
 		/*
 		 * All multi-letter extension and a few single letter
Signature · 1 line
2.41.0
#9 lore raw

PATCH v3 9/10 RISC-V: KVM: Improve vector save/restore errors

In reply to Daniel Henrique Barboza

From: Andrew Jones <[email protected]>

kvm_riscv_vcpu_(get/set)_reg_vector() now returns ENOENT if V is not available, EINVAL if reg type is not of VECTOR type, and any error that might be thrown by kvm_riscv_vcpu_vreg_addr().

---
 arch/riscv/kvm/vcpu_vector.c | 60 ++++++++++++++++++++----------------
 1 file changed, 33 insertions(+), 27 deletions(-)
arch/riscv/kvm/vcpu_vector.c +33 -27
index edd2eecbddc2..39c5bceb4d1b 100644
@@ -91,44 +91,44 @@ void kvm_riscv_vcpu_free_vector_context(struct kvm_vcpu *vcpu)
 }
 #endif
 
-static void *kvm_riscv_vcpu_vreg_addr(struct kvm_vcpu *vcpu,
+static int kvm_riscv_vcpu_vreg_addr(struct kvm_vcpu *vcpu,
 				      unsigned long reg_num,
-				      size_t reg_size)
+				      size_t reg_size,
+				      void **reg_val)
 {
 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
-	void *reg_val;
 	size_t vlenb = riscv_v_vsize / 32;
 
 	if (reg_num < KVM_REG_RISCV_VECTOR_REG(0)) {
 		if (reg_size != sizeof(unsigned long))
-			return NULL;
+			return -EINVAL;
 		switch (reg_num) {
 		case KVM_REG_RISCV_VECTOR_CSR_REG(vstart):
-			reg_val = &cntx->vector.vstart;
+			*reg_val = &cntx->vector.vstart;
 			break;
 		case KVM_REG_RISCV_VECTOR_CSR_REG(vl):
-			reg_val = &cntx->vector.vl;
+			*reg_val = &cntx->vector.vl;
 			break;
 		case KVM_REG_RISCV_VECTOR_CSR_REG(vtype):
-			reg_val = &cntx->vector.vtype;
+			*reg_val = &cntx->vector.vtype;
 			break;
 		case KVM_REG_RISCV_VECTOR_CSR_REG(vcsr):
-			reg_val = &cntx->vector.vcsr;
+			*reg_val = &cntx->vector.vcsr;
 			break;
 		case KVM_REG_RISCV_VECTOR_CSR_REG(datap):
 		default:
-			return NULL;
+			return -ENOENT;
 		}
 	} else if (reg_num <= KVM_REG_RISCV_VECTOR_REG(31)) {
 		if (reg_size != vlenb)
-			return NULL;
-		reg_val = cntx->vector.datap
+			return -EINVAL;
+		*reg_val = cntx->vector.datap
 			  + (reg_num - KVM_REG_RISCV_VECTOR_REG(0)) * vlenb;
 	} else {
-		return NULL;
+		return -ENOENT;
 	}
 
-	return reg_val;
+	return 0;
 }
 
 int kvm_riscv_vcpu_get_reg_vector(struct kvm_vcpu *vcpu,
@@ -141,17 +141,20 @@ int kvm_riscv_vcpu_get_reg_vector(struct kvm_vcpu *vcpu,
 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
 					    KVM_REG_SIZE_MASK |
 					    rtype);
-	void *reg_val = NULL;
 	size_t reg_size = KVM_REG_SIZE(reg->id);
+	void *reg_val;
+	int rc;
 
-	if (rtype == KVM_REG_RISCV_VECTOR &&
-	    riscv_isa_extension_available(isa, v)) {
-		reg_val = kvm_riscv_vcpu_vreg_addr(vcpu, reg_num, reg_size);
-	}
-
-	if (!reg_val)
+	if (rtype != KVM_REG_RISCV_VECTOR)
 		return -EINVAL;
 
+	if (!riscv_isa_extension_available(isa, v))
+		return -ENOENT;
+
+	rc = kvm_riscv_vcpu_vreg_addr(vcpu, reg_num, reg_size, &reg_val);
+	if (rc)
+		return rc;
+
 	if (copy_to_user(uaddr, reg_val, reg_size))
 		return -EFAULT;
 
@@ -168,17 +171,20 @@ int kvm_riscv_vcpu_set_reg_vector(struct kvm_vcpu *vcpu,
 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
 					    KVM_REG_SIZE_MASK |
 					    rtype);
-	void *reg_val = NULL;
 	size_t reg_size = KVM_REG_SIZE(reg->id);
+	void *reg_val;
+	int rc;
 
-	if (rtype == KVM_REG_RISCV_VECTOR &&
-	    riscv_isa_extension_available(isa, v)) {
-		reg_val = kvm_riscv_vcpu_vreg_addr(vcpu, reg_num, reg_size);
-	}
-
-	if (!reg_val)
+	if (rtype != KVM_REG_RISCV_VECTOR)
 		return -EINVAL;
 
+	if (!riscv_isa_extension_available(isa, v))
+		return -ENOENT;
+
+	rc = kvm_riscv_vcpu_vreg_addr(vcpu, reg_num, reg_size, &reg_val);
+	if (rc)
+		return rc;
+
 	if (copy_from_user(reg_val, uaddr, reg_size))
 		return -EFAULT;
 
Signature · 1 line
2.41.0
#10 lore raw

PATCH v3 10/10 docs: kvm: riscv: document EBUSY in KVM_SET_ONE_REG

In reply to Daniel Henrique Barboza

The EBUSY errno is being used for KVM_SET_ONE_REG as a way to tell userspace that a given reg can't be changed after the vcpu started.

---
 Documentation/virt/kvm/api.rst | 2 ++
 1 file changed, 2 insertions(+)
Documentation/virt/kvm/api.rst +2 -0
index c0ddd3035462..3249fb56cc69 100644
@@ -2259,6 +2259,8 @@ Errors:
   EINVAL   invalid register ID, or no such register or used with VMs in
            protected virtualization mode on s390
   EPERM    (arm64) register access not allowed before vcpu finalization
+  EBUSY    (riscv) changing register value not allowed after the vcpu
+           has run at least once
   ======   ============================================================
 
 (These error codes are indicative only: do not rely on a specific error
Signature · 1 line
2.41.0
#11 lore raw
In reply to Daniel Henrique Barboza
Quoted 5 lines · Daniel Henrique Barboza

Hi,

This version has changes in the document patch, as suggested by Andrew in v2. It also has a new patch (patch 9) that handles error code changes in vcpu_vector.c.

I forgot to include a diff that Andrew mentioned in his v2 review:

arch/riscv/kvm/vcpu_onereg.c +3 -3
index c88b0c7f7f01..6ca90c04ba61 100644
@@ -506,7 +506,7 @@ static int riscv_vcpu_get_isa_ext_multi(struct kvm_vcpu *vcpu,
         unsigned long i, ext_id, ext_val;
 
         if (reg_num > KVM_REG_RISCV_ISA_MULTI_REG_LAST)
-               return -EINVAL;
+               return -ENOENT;
 
         for (i = 0; i < BITS_PER_LONG; i++) {
                 ext_id = i + reg_num * BITS_PER_LONG;
@@ -529,7 +529,7 @@ static int riscv_vcpu_set_isa_ext_multi(struct kvm_vcpu *vcpu,
         unsigned long i, ext_id;
 
         if (reg_num > KVM_REG_RISCV_ISA_MULTI_REG_LAST)
-               return -EINVAL;
+               return -ENOENT;
 
         for_each_set_bit(i, &reg_val, BITS_PER_LONG) {
                 ext_id = i + reg_num * BITS_PER_LONG;
@@ -644,7 +644,7 @@ int kvm_riscv_vcpu_set_reg(struct kvm_vcpu *vcpu,
                 break;
         }
 
-       return -EINVAL;
+       return -ENOENT;
  }
 
  int kvm_riscv_vcpu_get_reg(struct kvm_vcpu *vcpu,

I'll send a v4 including it. Thanks,

Daniel

Quoted 34 lines

Patches rebased on top of kvm_riscv_queue.

Changes from v2:
- patch 9 (new):
- change kvm error codes for vector registers
- patch 10 (former 9):
- rewrite EBUSY doc to mention that the error code indicates that it
is not allowed to change the reg val after the vcpu started.
- v2 link: https://lore.kernel.org/kvm/20230801222629.210929-1-dbarboza at ventanamicro.com/

Andrew Jones (1):
RISC-V: KVM: Improve vector save/restore errors

Daniel Henrique Barboza (9):
RISC-V: KVM: return ENOENT in *_one_reg() when reg is unknown
RISC-V: KVM: use ENOENT in *_one_reg() when extension is unavailable
RISC-V: KVM: do not EOPNOTSUPP in set_one_reg() zicbo(m|z)
RISC-V: KVM: do not EOPNOTSUPP in set KVM_REG_RISCV_TIMER_REG
RISC-V: KVM: use EBUSY when !vcpu->arch.ran_atleast_once
RISC-V: KVM: avoid EBUSY when writing same ISA val
RISC-V: KVM: avoid EBUSY when writing the same machine ID val
RISC-V: KVM: avoid EBUSY when writing the same isa_ext val
docs: kvm: riscv: document EBUSY in KVM_SET_ONE_REG

  Documentation/virt/kvm/api.rst |  2 +
  arch/riscv/kvm/aia.c           |  4 +-
  arch/riscv/kvm/vcpu_fp.c       | 12 +++---
  arch/riscv/kvm/vcpu_onereg.c   | 68 +++++++++++++++++++++++-----------
  arch/riscv/kvm/vcpu_sbi.c      | 16 ++++----
  arch/riscv/kvm/vcpu_timer.c    | 11 +++---
  arch/riscv/kvm/vcpu_vector.c   | 60 ++++++++++++++++--------------
  7 files changed, 104 insertions(+), 69 deletions(-)