KVM: riscv: SBI FWFT: Apply LOCK flag only on successful set
Subsystems (3)
PATCH 1/1 KVM: riscv: SBI FWFT: Apply LOCK flag only on successful set
kvm_sbi_fwft_set() applies the caller's flags to conf->flags before invoking the set() callback. If the callback returns an error, the LOCK bit persists and the feature becomes permanently locked without its value ever being changed.
Move the flags assignment after the callback so LOCK takes effect only on success.
--- arch/riscv/kvm/vcpu_sbi_fwft.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
arch/riscv/kvm/vcpu_sbi_fwft.c +4 -2
@@ -327,9 +327,11 @@ static int kvm_sbi_fwft_set(struct kvm_vcpu *vcpu, u32 feature,if (conf->flags &SBI_FWFT_SET_FLAG_LOCK )return SBI_ERR_DENIED_LOCKED ; - conf->flags = flags; + ret = conf->feature->set (vcpu, conf, false, value); +if (ret ==SBI_SUCCESS ) + conf->flags = flags; -return conf->feature->set (vcpu, conf, false, value); +return ret; } static int kvm_sbi_fwft_get(struct kvm_vcpu *vcpu, unsigned long feature,
Signature · 1 line
2.52.0
[email protected]
AI reviewThank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:
Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:
Pre-existing issues:
- [High] SBI FWFT sub-features are enabled by default if supported by the host, violating KVM API architectural rules and breaking live migration.
- [High] Modifying FWFT features via KVM_SET_ONE_REG fails to set vcpu->arch.csr_dirty, which can cause the hardware CSRs to remain out of sync with the vCPU configuration.
- [High] Hardware feature probe in try_to_set_pmm() can be preempted, leading to false negatives and host CSR corruption.
- [High] Direct hardware CSR modifications in FWFT set handlers break nested virtualization (nacl) and are vulnerable to preemption.
--
commit 599bb2245d648d414e424da52cb50ced8f5d755b
Author: SeungJu Cheon <[email protected]>
KVM: riscv: SBI FWFT: Apply LOCK flag only on successful set
kvm_sbi_fwft_set() applies the caller's flags to conf->flags before invoking the set() callback. If the callback returns an error, the LOCK bit persists and the feature becomes permanently locked without its value ever being changed.
Move the flags assignment after the callback so LOCK takes effect only on success.
Quoted 8 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +0
-0@@ -303,9 +303,11 @@ static int kvm_sbi_fwft_set(struct kvm_vcpu *vcpu, u32 feature,if (conf->flags &SBI_FWFT_SET_FLAG_LOCK )return SBI_ERR_DENIED_LOCKED ;
arch/riscv/kvm/vcpu_sbi_fwft.c static int kvm_sbi_fwft_set(struct kvm_vcpu *vcpu, u32 feature, · line 313
Quoted 10 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +4
-2- conf->flags = flags; + ret = conf->feature->set (vcpu, conf, false, value); +if (ret ==SBI_SUCCESS ) + conf->flags = flags; -return conf->feature->set (vcpu, conf, false, value); +return ret; } static int kvm_sbi_fwft_get(struct kvm_vcpu *vcpu, unsigned long feature,
While reviewing this file, I noticed a few pre-existing issues in arch/riscv/kvm/vcpu_sbi_fwft.c:
In kvm_sbi_ext_fwft_init():
if (feature->supported)
conf->supported = feature->supported(vcpu);
else
conf->supported = true;
conf->enabled = conf->supported;[Severity: High]
This isn't a bug introduced by this patch, but it appears FWFT sub-features are unconditionally enabled by default if the hardware supports them. Does this violate the KVM architectural rule that new guest-visible features must default to OFF?
Could this break live migration if a VM is moved to a host lacking support for a feature that was silently exposed?
In kvm_sbi_ext_fwft_set_reg():
case 2:
ret = conf->feature->set(vcpu, conf, true, value);
break;[Severity: High]
This is a pre-existing issue, but does modifying FWFT features via KVM_SET_ONE_REG fail to set vcpu->arch.csr_dirty?
When vcpu_load() runs on the same physical CPU, it bypasses kvm_riscv_vcpu_config_load(). Can this leave the hardware CSRs (like CSR_HEDELEG or CSR_HENVCFG) out of sync with the vCPU configuration?
In try_to_set_pmm():
csr_set(CSR_HENVCFG, value);
return (csr_read_clear(CSR_HENVCFG, ENVCFG_PMM) & ENVCFG_PMM) == value;[Severity: High]
This is a pre-existing issue, but is it safe to modify and read back the CSR here without disabling preemption?
If the thread is migrated between csr_set() and csr_read_clear(), could this result in false negatives and corrupt the original CPU's CSR_HENVCFG by persistently leaving ENVCFG_PMM bits set?
In kvm_sbi_fwft_set_misaligned_delegation():
if (!one_reg_access)
csr_set(CSR_HEDELEG, MIS_DELEG);And in kvm_sbi_fwft_set_pointer_masking_pmlen():
if (!one_reg_access)
csr_write(CSR_HENVCFG, vcpu->arch.cfg.henvcfg);[Severity: High]
This is a pre-existing issue, but do these direct hardware CSR modifications break nested virtualization? By bypassing nacl_shmem(), the L0 hypervisor might not observe L1's FWFT changes.
Also, since SBI_EXT_FWFT_SET exits are handled with preemption enabled, could preempting or migrating these calls corrupt CSR_HEDELEG or CSR_HENVCFG on a random CPU?
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
Quoted 11 lines · SeungJu Cheon
kvm_sbi_fwft_set()applies the caller's flags toconf->flagsbefore invoking theset()callback. If the callback returns an error, the LOCK bit persists and the feature becomes permanently locked without its value ever being changed.Move the flags assignment after the callback so LOCK takes effect only on success.
Fixes 6b72fd170592 ("RISC-V: KVM: add support for FWFT SBI extension")Signed-off-by SeungJu Cheon <[email protected]>
LGTM.
Queued this patch as a fix for Linux-7.2-rcX
Thanks,
Anup
Quoted 25 lines
--- arch/riscv/kvm/vcpu_sbi_fwft.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)arch/riscv/kvm/vcpu_sbi_fwft.c +4
-3@@ -327,9 +327,11 @@ static int kvm_sbi_fwft_set(struct kvm_vcpu *vcpu, u32 feature,if (conf->flags &SBI_FWFT_SET_FLAG_LOCK )return SBI_ERR_DENIED_LOCKED ; - conf->flags = flags; + ret = conf->feature->set (vcpu, conf, false, value); +if (ret ==SBI_SUCCESS ) + conf->flags = flags; -return conf->feature->set (vcpu, conf, false, value); +return ret; } static int kvm_sbi_fwft_get(struct kvm_vcpu *vcpu, unsigned long feature, --2.52.0