Skip to contentSkip to conversation

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

PATCH Superseded Daniel Henrique Barboza 3 participants 13 messages Jul 31, 2023 – Nov 13, 2023

Subsystems (4)

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

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

Hi,

This work was motivated by the changes made in QEMU in commit df817297d7 ("target/riscv: update multi-letter extension KVM properties") where it was required to use EINVAL to try to assess whether a given extension is available in the host.

It turns out that the RISC-V KVM module makes regular use of the EIVAL error code in all kvm_get_one_reg() and kvm_set_one_reg() callbacks, which is not ideal. For example, ENOENT is a better error code for the case where a given register does not exist. While at it I decided to take a look at other error codes from these callbacks, comparing them with how other archs (ARM) handles the same error types, and changed some of the EOPNOTSUPP instances to either ENOENT or EBUSY.

I am aware that changing error codes can be problematic to existing userspaces. I took a look and no other major userspace (checked crosvm, rust-vmm, QEMU, kvmtool), aside from QEMU now checking for EIVAL (and we can't change that because of backwards compat for that particular case we're covering), will be impacted by this kind of change since they're all checking for "return < 0 then ..." instead of doing specific error handling based on the error value. This means that we're still in good time to make this kind of change while we're still in the initial steps of the RISC-V KVM ecossystem support.

Patch 3 happens to also change the behavior of the set_reg() for zicbom/zicboz block sizes. Instead of always erroring out we'll check if userspace is writing the same value that the host uses. In this case, allow the write to succeed (i.e. do nothing). This avoids the situation in which userspace reads cbom_block_size, find out that it's set to X, and then can't write the same value back to the register. It's a QOL improvement to allow userspace to be lazier when reading/writing regs. A similar change was also made in patch 4.

Patches are based on top of riscv_kvm_queue.

Daniel Henrique Barboza (6):
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
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   | 52 ++++++++++++++++++++--------------
 arch/riscv/kvm/vcpu_sbi.c      | 16 ++++++-----
 arch/riscv/kvm/vcpu_timer.c    | 11 +++----
 6 files changed, 55 insertions(+), 42 deletions(-)
Signature · 1 line
2.41.0
#1 lore raw

PATCH 1/6 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 0dc2c2cecb45..ba63522be060 100644
@@ -153,7 +153,7 @@ static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu,
 		reg_val = vcpu->arch.mimpid;
 		break;
 	default:
-		return -EINVAL;
+		return -ENOENT;
 	}
 
 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
@@ -235,7 +235,7 @@ static int kvm_riscv_vcpu_set_reg_config(struct kvm_vcpu *vcpu,
 			return -EBUSY;
 		break;
 	default:
-		return -EINVAL;
+		return -ENOENT;
 	}
 
 	return 0;
@@ -255,7 +255,7 @@ 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;
@@ -266,7 +266,7 @@ 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;
@@ -288,7 +288,7 @@ 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;
@@ -304,7 +304,7 @@ static int kvm_riscv_vcpu_set_reg_core(struct kvm_vcpu *vcpu,
 		else
 			cntx->sstatus &= ~SR_SPP;
 	} else
-		return -EINVAL;
+		return -ENOENT;
 
 	return 0;
 }
@@ -316,7 +316,7 @@ 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);
@@ -335,7 +335,7 @@ 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;
@@ -374,7 +374,7 @@ 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)
@@ -413,7 +413,7 @@ 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)
@@ -430,7 +430,7 @@ 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];
@@ -448,7 +448,7 @@ 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))
@@ -547,7 +547,7 @@ 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;
@@ -585,7 +585,7 @@ 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;
@@ -652,5 +652,5 @@ 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

Re: [PATCH 1/6] RISC-V: KVM: return ENOENT in *_one_reg() when reg is unknown

Quoted 4 lines · Daniel Henrique Barboza

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.

That may break qemu:

$ qemu-system-riscv64 -cpu rv64 -machine virt,accel=kvm      
qemu-system-riscv64: Unable to read ISA_EXT KVM register ssaia, error -1
Signature · 3 lines
Andreas Schwab, SUSE Labs, schwab at suse.de
GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE  1748 E4D4 88E3 0EEA B9D7
"And now for something completely different."
#3 lore raw

Re: [PATCH 1/6] RISC-V: KVM: return ENOENT in *_one_reg() when reg is unknown

Quoted 11 lines · Andreas Schwab
Quoted 4 lines · Daniel Henrique Barboza

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.

That may break qemu:

$ qemu-system-riscv64 -cpu rv64 -machine virt,accel=kvm
qemu-system-riscv64: Unable to read ISA_EXT KVM register ssaia, error -1

Which QEMU version are you using? This is a problem that was fixed upstream by:

commit 608bdebb6075b757e5505f6bbc60c45a54a1390b

Author: Daniel Henrique Barboza <[email protected]>
Date:   Tue Oct 3 10:21:48 2023 -0300

target/riscv/kvm: support KVM_GET_REG_LIST

If you're getting this error with upstream QEMU let me know and I'll take a look there.

Thanks,

Daniel

Quoted 1 line
#4 lore raw

Re: [PATCH 1/6] RISC-V: KVM: return ENOENT in *_one_reg() when reg is unknown

Quoted 1 line · Daniel Henrique Barboza

Which QEMU version are you using?

The very latest release, both host and guest.

Signature · 3 lines
Andreas Schwab, SUSE Labs, schwab at suse.de
GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE  1748 E4D4 88E3 0EEA B9D7
"And now for something completely different."
#5 lore raw

Re: [PATCH 1/6] RISC-V: KVM: return ENOENT in *_one_reg() when reg is unknown

Quoted 5 lines · Andreas Schwab
Quoted 1 line · Daniel Henrique Barboza

Which QEMU version are you using?

The very latest release, both host and guest.

If by "latest release" you mean kernel 6.6 and QEMU 8.1, this combination is broken ATM.

Back in 8.1 QEMU was checking for EINVAL to make an educated guess of whether an extension was unavailable but the register exists, versus if the register was alien to KVM at all (qemu commit f7a69fa6e6). This turned out to be a mistake because EINVAL was being thrown for all sorts of errors, and QEMU would be wiser to just error out in all errors like other VMMs were doing (including kvmtool).

These were considerations made when proposing this KVM side change in the cover letter. Other VMMs would be unaffected by it, and QEMU would need changes to adapt to the new error codes. QEMU 8.2 is already adapted. It's not ideal, but it's better to take a hit now while the RISC-V ecosystem is still new and make things tidy for the future.

And now that I'm thinking more about it, I'll push a QEMU change in 8.1-stable to alleviate the issue for 8.1. My apologies, I should've thought about it earlier ...

Thanks,

Daniel

Quoted 1 line
#6 lore raw

PATCH 2/6 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 ba63522be060..e630a68e4f27 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):
@@ -452,7 +452,7 @@ 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
#7 lore raw

PATCH 3/6 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 e630a68e4f27..42bf01ab6a8f 100644
@@ -213,9 +213,17 @@ 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
#8 lore raw

PATCH 4/6 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
#9 lore raw

PATCH 5/6 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 42bf01ab6a8f..07ce747620f9 100644
@@ -209,7 +209,7 @@ 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):
@@ -477,7 +477,7 @@ 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
#10 lore raw

Re: [PATCH 5/6] RISC-V: KVM: use EBUSY when !vcpu->arch.ran_atleast_once

Quoted 30 lines · 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 +1 -1
index 42bf01ab6a8f..07ce747620f9 100644
@@ -209,7 +209,7 @@ 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):

arch/riscv/kvm/vcpu_onereg.c static int riscv_vcpu_set_isa_ext_single(struct kvm_vcpu *vcpu, · line 481

Quoted 6 lines
arch/riscv/kvm/vcpu_onereg.c +1 -1
@@ -477,7 +477,7 @@ 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;
 	}

I think we should allow these ran_atleast_once type of registers to be written when the value matches, as we now do for the other registers. EBUSY should still be returned when the value doesn't match, though.

Thanks,
drew

#11 lore raw

PATCH 6/6 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 written after the vcpu started.

---
 Documentation/virt/kvm/api.rst | 2 ++
 1 file changed, 2 insertions(+)
Documentation/virt/kvm/api.rst +2 -0
index c0ddd3035462..229e7cc091c8 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) register access 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
#12 lore raw
In reply to Daniel Henrique Barboza
Quoted 15 lines · Daniel Henrique Barboza

Hi,

This work was motivated by the changes made in QEMU in commit df817297d7 ("target/riscv: update multi-letter extension KVM properties") where it was required to use EINVAL to try to assess whether a given extension is available in the host.

It turns out that the RISC-V KVM module makes regular use of the EIVAL error code in all kvm_get_one_reg() and kvm_set_one_reg() callbacks, which is not ideal. For example, ENOENT is a better error code for the case where a given register does not exist. While at it I decided to take a look at other error codes from these callbacks, comparing them with how other archs (ARM) handles the same error types, and changed some of the EOPNOTSUPP instances to either ENOENT or EBUSY.

Quoted 5 lines

I am aware that changing error codes can be problematic to existing userspaces. I took a look and no other major userspace (checked crosvm, rust-vmm, QEMU, kvmtool), aside from QEMU now checking for EIVAL (and we can't change that because of backwards compat for that particular case we're covering),

Merging this series just prior to some other API change, like adding the get-reg-list ioctl[1], will allow QEMU to eventually drop the EINVAL handling. This is because when QEMU sees the get-reg-list ioctl it will know that get/set-reg-list has the updated API. At some point when QEMU no longer wants to support a KVM which doesn't have get-reg-list, then the EINVAL handling can be dropped.

And, once KVM has get-reg-list, then QEMU won't want to probe for registers with get-one-reg anyway. Instead, it'll get the list once and then check that each register it would have probed is in the list.

[1] https://lore.kernel.org/lkml/cover.1690273969.git.haibo1.xu at intel.com/T/

Quoted 6 lines

will be impacted by this kind of change since they're
all checking for "return < 0 then ..." instead of doing specific error handling based on the error value. This means that we're still in good time to make this kind of change while we're still in the initial steps of the RISC-V KVM ecossystem support.

Quoted 7 lines

Patch 3 happens to also change the behavior of the set_reg() for zicbom/zicboz block sizes. Instead of always erroring out we'll check if userspace is writing the same value that the host uses. In this case, allow the write to succeed (i.e. do nothing). This avoids the situation in which userspace reads cbom_block_size, find out that it's set to X, and then can't write the same value back to the register. It's a QOL improvement to allow userspace to be lazier when reading/writing regs.

Being able to write back the same value will greatly simplify save/restore for QEMU, particularly when get-reg-list is available. QEMU would then get the list of registers and, for save, it'll iterate the list blindly, calling get-one-reg on each, storing each value. Then, for restore, it'll blindly iterate again, writing each saved value back with set-one-reg.

Thanks,
drew

Quoted 24 lines

A
similar change was also made in patch 4.

Patches are based on top of riscv_kvm_queue.

Daniel Henrique Barboza (6):
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
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   | 52 ++++++++++++++++++++--------------
 arch/riscv/kvm/vcpu_sbi.c      | 16 ++++++-----
 arch/riscv/kvm/vcpu_timer.c    | 11 +++----
 6 files changed, 55 insertions(+), 42 deletions(-)
Signature · 1 line
2.41.0