RISC-V: KVM: Add Svadu/Zicfiss/Zicfilp FWFT support
Subsystems (4)
v2 cover RISC-V: KVM: Add Svadu/Zicfiss/Zicfilp FWFT support
This patchset based on the patchset "RISC-V: KVM: Add Zicfiss/Zicfilp support".
The patch includes all the function of original patchset from Quan:
adding the Svadu/Zicfiss/Zicfilp FWFT for the RISC-V KVM. And this patchset already passes the KVM unit test.
Changed from Quan's v1:
- https://lore.kernel.org/kvm/[email protected]/
1. rebase to the latest master.
2. Split Svadu change as separate patch
3. Fix all build error from the original series
4. Remove the Svadu check in kvm_riscv_vcpu_config_ran_once() and in favor of the FWFT.
5. Remove the LPE/SSE setting bit in kvm_riscv_vcpu_config_ran_once() (kvm_riscv_vcpu_setup_config() in original patchset)
Inochi Amaoto (7):
RISC-V: KVM: Add support for Svadu FWFT features
KVM: riscv: selftests: add Svadu FWFT extension to get-reg-list test
RISC-V: KVM: Only enable Svadu extension when Guest/VM requests
RISC-V: KVM: Allow Zicfiss/Zicfilp extensions for Guest/VM
RISC-V: KVM: Handle software-check exits for VCPU
RISC-V: KVM: Add support for control-flow integrity FWFT features
KVM: riscv: selftests: add Zicfiss/Zicfilp extension to get-reg-list
test
arch/riscv/include/asm/csr.h | 1 + arch/riscv/include/uapi/asm/kvm.h | 5 + arch/riscv/kvm/isa.c | 2 + arch/riscv/kvm/vcpu_config.c | 8 +- arch/riscv/kvm/vcpu_exit.c | 3 + arch/riscv/kvm/vcpu_sbi_fwft.c | 142 ++++++++++++++++++ .../selftests/kvm/riscv/get-reg-list.c | 22 +++ 7 files changed, 179 insertions(+), 4 deletions(-)
--
2.54.0
PATCH v2 1/7 RISC-V: KVM: Add support for Svadu FWFT features
Hardware updating of PTE A/D bits is controlled through ADUE bit in henvcfg Expose the feature only if the Svadu is supported for VS-mode. Allow the VMM to block access to the feature by disabling the ISA extension in the guest.
--- arch/riscv/include/uapi/asm/kvm.h | 3 ++ arch/riscv/kvm/vcpu_sbi_fwft.c | 72 +++++++++++++++++++++++++++++++ 2 files changed, 75 insertions(+)
arch/riscv/include/uapi/asm/kvm.h +3 -0
@@ -240,6 +240,9 @@ struct kvm_riscv_sbi_fwft_feature {struct kvm_riscv_sbi_fwft {struct kvm_riscv_sbi_fwft_feature misaligned_deleg;struct kvm_riscv_sbi_fwft_feature pointer_masking; +struct kvm_riscv_sbi_fwft_feature pte_ad_hw_updating; +struct kvm_riscv_sbi_fwft_feature landing_pad; +struct kvm_riscv_sbi_fwft_feature shadow_stack; };/* If you need to interpret the index values, here is the key: */
arch/riscv/kvm/vcpu_sbi_fwft.c +72 -0
90 lines not on this page · raw message
Signature · 1 line
2.54.0
[email protected]
AI reviewRe: [PATCH v2 1/7] RISC-V: KVM: Add support for Svadu FWFT features
Thank you for your contribution! Sashiko AI review found 5 potential issue(s) to consider: - [High] KVM_SET_ONE_REG for FWFT state registers modifies VCPU configuration in memory but fails to set…
Thank you for your contribution! Sashiko AI review found 5 potential issue(s) to consider:
- [High] KVM_SET_ONE_REG for FWFT state registers modifies VCPU configuration in memory but fails to set vcpu->arch.csr_dirty = true, leading to stale hardware CSRs.
- [High] The FWFT support check kvm_sbi_fwft_pte_ad_hw_updating_supported() incorrectly disables the toggle when both SVADU and SVADE are present, and dangerously enables it when only SVADU is present.
- [High] The 'flag' parameter in henvcfg helper functions is 'unsigned long', causing truncation of 64-bit flags like ENVCFG_ADUE on 32-bit systems and permanently corrupting guest henvcfg state during vCPU reset.
- [High] Dynamically setting the Svadu FWFT feature on RV32 fails to update the physical CPU's CSR_HENVCFGH register, leaving the hardware state stale.
- [High] Direct modification of CSR_HENVCFG bypasses Nested Acceleration (NACL) shared memory, leading to state desynchronization in nested virtualization.
--
commit df2a6571e0257cf7e8c4ca0758aefd4d6f6c3a32
Author: Inochi Amaoto <[email protected]>
RISC-V: KVM: Add support for Svadu FWFT features
This patch adds support for exposing the Svadu extension to guests via the SBI FWFT interface. It introduces helpers to get, set, and reset henvcfg flags, specifically for the ADUE bit which controls hardware updating of PTE A/D bits.
Quoted 4 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +0
-0
[ ... ]
arch/riscv/kvm/vcpu_sbi_fwft.c static bool kvm_fwft_is_defined_feature(enum sbi_fwft_feature_t feature) · line 91
Quoted 9 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +5
-0@@ -84,6 +84,42 @@ static bool kvm_fwft_is_defined_feature(enum sbi_fwft_feature_t feature)return false; } +static void kvm_sbi_fwft_env_flag_reset_helper (struct kvm_vcpu *vcpu, +unsigned long flag) +{ + vcpu->arch.cfg.henvcfg &= ~flag; +}
[Severity: High]
Does using unsigned long for the flag parameter truncate 64-bit flags like ENVCFG_ADUE on 32-bit systems?
Since ENVCFG_ADUE is in the upper 32 bits, passing it as unsigned long on CONFIG_32BIT systems would silently truncate it to 0.
When vcpu->arch.cfg.henvcfg &= ~flag evaluates with flag as 0, it results in henvcfg &= ~0UL. Because ~0UL is 0xFFFFFFFF, it zero-extends to u64, resulting in henvcfg &= 0x00000000FFFFFFFFULL.
Will this unconditionally wipe out all upper 32-bit extensions like ENVCFG_STCE and ENVCFG_PBMTE for the guest during vCPU reset?
arch/riscv/kvm/vcpu_sbi_fwft.c static bool kvm_fwft_is_defined_feature(enum sbi_fwft_feature_t feature) · line 97
Quoted 6 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +6
-0+ +static long kvm_sbi_fwft_env_flag_set_helper (struct kvm_vcpu *vcpu, +struct kvm_sbi_fwft_config *conf, +bool one_reg_access, +unsigned long value,unsigned long flag) +{
[Severity: High]
Is this helper also affected by the same unsigned long truncation issue for the flag parameter?
arch/riscv/kvm/vcpu_sbi_fwft.c static bool kvm_fwft_is_defined_feature(enum sbi_fwft_feature_t feature) · line 106
Quoted 9 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +9
-0+struct kvm_vcpu_config *cfg = &vcpu->arch.cfg; + +if (value ==0 ) + cfg->henvcfg &= ~flag; +else if (value ==1 ) + cfg->henvcfg |= flag; +else +return SBI_ERR_INVALID_PARAM ; +
[Severity: High]
When one_reg_access is true, does this omit setting vcpu->arch.csr_dirty to true?
If the VCPU is later scheduled on the same physical CPU, vcpu_load() might skip calling kvm_riscv_vcpu_config_load() due to the fast-path condition checking the csr_dirty flag.
Could this cause the guest to run with a stale hardware configuration since the updated henvcfg is never flushed to the hardware CSR_HENVCFG?
arch/riscv/kvm/vcpu_sbi_fwft.c static bool kvm_fwft_is_defined_feature(enum sbi_fwft_feature_t feature) · line 108
Quoted 2 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +2
-0+if (!one_reg_access) +csr_write (CSR_HENVCFG , vcpu->arch.cfg.henvcfg);
[Severity: High]
On 32-bit RISC-V systems, does csr_write(CSR_HENVCFG, ...) only write the lower 32 bits?
Since ENVCFG_ADUE resides in the upper 32 bits, shouldn't there be an explicit write to CSR_HENVCFGH as well? Without it, the physical CPU's configuration might not be properly synchronized for upper 32-bit flags.
[Severity: High]
Does directly modifying CSR_HENVCFG bypass Nested Acceleration (NACL)?
When KVM runs as an L1 hypervisor with NACL enabled, it must synchronize hypervisor CSRs via the shared memory region (nsh) rather than directly writing to the physical CSRs.
If ncsr_write() is not used here, could the L0 hypervisor miss the update and overwrite the CSR with a stale value from nsh on subsequent VM entries?
Quoted 3 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +3
-0+ +return SBI_SUCCESS ; +}
[ ... ]
arch/riscv/kvm/vcpu_sbi_fwft.c static long kvm_sbi_fwft_get_misaligned_delegation(struct kvm_vcpu *vcpu, · line 170
Quoted 5 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +5
-0+static bool kvm_sbi_fwft_pte_ad_hw_updating_supported (struct kvm_vcpu *vcpu) +{ +return riscv_isa_extension_available (vcpu->arch.isa,SVADU ) && + !riscv_isa_extension_available (vcpu->arch.isa,SVADE ); +}
[Severity: High]
Does this condition behave backward with respect to the SBI FWFT specification?
The specification requires that if a VM supports only Svadu, ADUE must be hardwired to 1. If it supports both Svadu and Svade, the guest can toggle the feature.
By returning true only when Svadu is present and Svade is absent, does this disable the toggle for fully capable VMs, and incorrectly allow Svadu-only VMs to set ADUE=0?
If a Svadu-only VM sets ADUE=0, could it silently corrupt its memory since the Svade software fallback is unavailable?
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
PATCH v2 2/7 KVM: riscv: selftests: add Svadu FWFT extension to get-reg-list test
In reply to Inochi AmaotoThe KVM RISC-V allows Svadu FWFT for Guest/VM so add this extension to get-reg-list test.
--- tools/testing/selftests/kvm/riscv/get-reg-list.c | 6 ++++++ 1 file changed, 6 insertions(+)
tools/testing/selftests/kvm/riscv/get-reg-list.c +6 -0
@@ -712,6 +712,9 @@ static const char *sbi_fwft_id_to_str(__u64 reg_off)case 3 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(pointer_masking.enable)" ;case 4 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(pointer_masking.flags)" ;case 5 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(pointer_masking.value)" ; +case 6 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(pte_ad_hw_updating.enable)" ; +case 7 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(pte_ad_hw_updating.flags)" ; +case 8 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(pte_ad_hw_updating.value)" ; }return strdup_printf ("KVM_REG_RISCV_SBI_FWFT | %lld /* UNKNOWN */" , reg_off); }
@@ -905,6 +908,9 @@ static __u64 sbi_fwft_regs[] = {KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pointer_masking.enable),KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pointer_masking.flags),KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pointer_masking.value), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pte_ad_hw_updating.enable), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pte_ad_hw_updating.flags), + KVM_REG_RISCV | KVM_REG_SIZE_ULONG | KVM_REG_RISCV_SBI_STATE | KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(pte_ad_hw_updating.value), }; static __u64 zicbom_regs[] = {
Signature · 1 line
2.54.0
[email protected]
AI reviewRe: [PATCH v2 2/7] KVM: riscv: selftests: add Svadu FWFT extension to get-reg-list test
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Medium] Unconditional selftest failure in get-reg-list due to dynamically-omitted registers in the…
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] Unconditional selftest failure in get-reg-list due to dynamically-omitted registers in the blessed list. --
commit 6df4c1234104e4c676015802e2b610d953076f46
Author: Inochi Amaoto <[email protected]>
KVM: riscv: selftests: add Svadu FWFT extension to get-reg-list test
This commit adds the Svadu FWFT extension registers to the get-reg-list selftest's expected register list.
Quoted 4 lines
tools/testing/selftests/kvm/riscv/get-reg-list.c +0
-0
[ ... ]
tools/testing/selftests/kvm/riscv/get-reg-list.c static __u64 sbi_fwft_regs[] = { · line 914
Quoted 8 lines
tools/testing/selftests/kvm/riscv/get-reg-list.c +3
-0@@ -905,6 +908,9 @@ static __u64 sbi_fwft_regs[] = {KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pointer_masking.enable),KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pointer_masking.flags),KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pointer_masking.value), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pte_ad_hw_updating.enable), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pte_ad_hw_updating.flags), + KVM_REG_RISCV | KVM_REG_SIZE_ULONG | KVM_REG_RISCV_SBI_STATE | KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(pte_ad_hw_updating.value), };
[Severity: Medium]
Will this cause an unconditional selftest failure when the SVADU extension is disabled?
The RISC-V get-reg-list selftest disables all ISA extensions (including SVADU) by default to ensure a clean baseline. The KVM implementation conditionally exposes these pte_ad_hw_updating registers only if the SVADU ISA extension is enabled on the vCPU.
Because the registers are added unconditionally to the expected list here, and RISC-V does not override check_supported_reg() to conditionally filter them out, KVM omitting these registers from KVM_GET_REG_LIST will cause the test framework to fail with a missing registers assertion.
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=2
PATCH v2 3/7 RISC-V: KVM: Only enable Svadu extension when Guest/VM requests
In reply to Inochi AmaotoSince the Svadu extension can be controlled by the FWFT, there is no need to enable Svadu at startup. Only enable it when it is requested.
--- arch/riscv/kvm/vcpu_config.c | 4 ---- 1 file changed, 4 deletions(-)
arch/riscv/kvm/vcpu_config.c +0 -4
@@ -53,10 +53,6 @@ void kvm_riscv_vcpu_config_ran_once(struct kvm_vcpu *vcpu)if (riscv_isa_extension_available (isa,ZICBOZ )) cfg->henvcfg |=ENVCFG_CBZE ; -if (riscv_isa_extension_available (isa,SVADU ) && - !riscv_isa_extension_available (isa,SVADE )) - cfg->henvcfg |=ENVCFG_ADUE ; -if (riscv_has_extension_unlikely (RISCV_ISA_EXT_SMSTATEEN )) { cfg->hstateen0 |=SMSTATEEN0_HSENVCFG ;if (riscv_isa_extension_available (isa,SSAIA ))
Signature · 1 line
2.54.0
[email protected]
AI reviewRe: [PATCH v2 3/7] RISC-V: KVM: Only enable Svadu extension when Guest/VM requests
Thank you for your contribution! Sashiko AI review found 6 potential issue(s) to consider:
Thank you for your contribution! Sashiko AI review found 6 potential issue(s) to consider:
Pre-existing issues:
- [Critical] The FWFT reset callback unconditionally clears ENVCFG_ADUE, breaking the required default state.
- [High] The FWFT supported callback for PTE_AD_HW_UPDATING uses inverted logic, disabling the feature when both extensions are present.
- [High] Modifying FWFT features via KVM_SET_ONE_REG updates vCPU software config but fails to set vcpu->arch.csr_dirty = true, leading to stale hardware CSR state.
- [High] Probing host hardware support for pointer masking modifies a physical host CSR (CSR_HENVCFG) with preemption enabled, risking hardware state corruption.
- [High] kvm_sbi_fwft_env_flag_set_helper fails to update the upper 32 bits of HENVCFG on 32-bit systems when handling SBI_EXT_FWFT_SET.
- [High] Direct CSR writes in FWFT handlers bypass Nested Architecture for Confidential Linux (NaCl) shared memory synchronization.
--
commit 9785e1b9e42d11cc65b0f2a69b673df176c4c687
Author: Inochi Amaoto <[email protected]>
RISC-V: KVM: Only enable Svadu extension when Guest/VM requests
This commit removes the default startup initialization of the Svadu extension in KVM, relying instead on the guest or VM requesting it through the FWFT interface.
Quoted 4 lines
arch/riscv/kvm/vcpu_config.c +0
-0
[ ... ]
arch/riscv/kvm/vcpu_config.c void kvm_riscv_vcpu_config_ran_once(struct kvm_vcpu *vcpu) · line 59
Quoted 8 lines
arch/riscv/kvm/vcpu_config.c +0
-4@@ -53,10 +53,6 @@ void kvm_riscv_vcpu_config_ran_once(struct kvm_vcpu *vcpu)if (riscv_isa_extension_available (isa,ZICBOZ )) cfg->henvcfg |=ENVCFG_CBZE ; -if (riscv_isa_extension_available (isa,SVADU ) && - !riscv_isa_extension_available (isa,SVADE )) - cfg->henvcfg |=ENVCFG_ADUE ; -
[Severity: High]
This isn't a bug introduced by this patch, but by removing this default initialization, does this leave guests without an A/D updating mechanism at boot?
If a platform has SVADU but no SVADE, hardware A/D updating is the only mechanism available. The FWFT reset handler unconditionally clears ENVCFG_ADUE:
arch/riscv/kvm/vcpu_sbi_fwft.c:kvm_sbi_fwft_reset_pte_ad_hw_updating() {
...
kvm_sbi_fwft_env_flag_reset_helper(vcpu, ENVCFG_ADUE);
...
}
[Severity: High]
This is a pre-existing issue, but does the FWFT supported callback use inverted logic?
arch/riscv/kvm/vcpu_sbi_fwft.c:kvm_sbi_fwft_pte_ad_hw_updating_supported() {
...
return riscv_isa_extension_available(vcpu->arch.isa, SVADU) && !riscv_isa_extension_available(vcpu->arch.isa, SVADE);
...
}
If a platform supports both SVADU and SVADE, it can safely toggle between hardware updating and page faults.
However, this logic returns true only when SVADU is present and SVADE is absent. Since this patch relies on FWFT as the sole way to enable SVADU, does this logic prevent it from working on platforms that can actually toggle it?
[Severity: High]
This is a pre-existing issue, but how does the code interact with the FWFT handlers updating the vCPU software config without setting the dirty flag?
When KVM_SET_ONE_REG modifies an FWFT feature:
arch/riscv/kvm/vcpu_sbi_fwft.c:kvm_sbi_fwft_env_flag_set_helper() {
...
else if (value == 1) cfg->henvcfg |= flag; ... if (!one_reg_access) csr_write(CSR_HENVCFG, vcpu->arch.cfg.henvcfg);
...
}
It updates cfg->henvcfg but doesn't set vcpu->arch.csr_dirty = true. If the vCPU is scheduled on the same physical CPU, the load fast-path skips hardware state restoration:
arch/riscv/kvm/vcpu.c:kvm_arch_vcpu_load() {
...
else if (vcpu->arch.last_exit_cpu == cpu && !vcpu->arch.csr_dirty) goto csr_restore_done;
...
}
Does this result in the hardware running with a stale CSR_HENVCFG value until the vCPU migrates?
[Severity: High]
This is a pre-existing issue, but looking at the FWFT initialization, does testing host pointer masking support risk hardware state corruption?
arch/riscv/kvm/vcpu_sbi_fwft.c:try_to_set_pmm() {
...
csr_set(CSR_HENVCFG, value); return (csr_read_clear(CSR_HENVCFG, ENVCFG_PMM) & ENVCFG_PMM) == value;
...
}
If the thread is preempted and migrated to a different physical CPU between csr_set and csr_read_clear, is the first CPU permanently left with the test bits set in its physical CSR_HENVCFG?
[Severity: High]
This isn't a bug introduced by this patch, but does the FWFT set helper fail to update the upper 32 bits of HENVCFG on 32-bit systems?
arch/riscv/kvm/vcpu_sbi_fwft.c:kvm_sbi_fwft_env_flag_set_helper() {
...
if (!one_reg_access) csr_write(CSR_HENVCFG, vcpu->arch.cfg.henvcfg);
...
}
Since ENVCFG_ADUE is located at bit 61, wouldn't the hardware state remain out-of-sync with vcpu->arch.cfg.henvcfg when a 32-bit guest executes an SBI FWFT SET call?
[Severity: High]
This is a pre-existing issue, but do direct CSR writes in the FWFT handlers bypass Nested Architecture for Confidential Linux (NaCl) shared memory?
arch/riscv/kvm/vcpu_sbi_fwft.c:kvm_sbi_fwft_env_flag_set_helper() {
...
csr_write(CSR_HENVCFG, vcpu->arch.cfg.henvcfg);
...
}
and in kvm_sbi_fwft_set_misaligned_delegation():
arch/riscv/kvm/vcpu_sbi_fwft.c:kvm_sbi_fwft_set_misaligned_delegation() {
...
csr_set(CSR_HEDELEG, MIS_DELEG);
...
}
Should these be using the ncsr_write() wrapper or nacl_csr_write() to ensure the nested hypervisor's view remains synchronized?
Quoted 3 lines
arch/riscv/kvm/vcpu_config.c +0
-0if (riscv_has_extension_unlikely (RISCV_ISA_EXT_SMSTATEEN )) { cfg->hstateen0 |=SMSTATEEN0_HSENVCFG ;if (riscv_isa_extension_available (isa,SSAIA ))
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=3
PATCH v2 4/7 RISC-V: KVM: Allow Zicfiss/Zicfilp extensions for Guest/VM
In reply to Inochi AmaotoExtend the KVM ISA extension array to allow KVM userspace to detect and enable Zicfiss/Zicfilp extensions for Guest/VM.
--- arch/riscv/include/uapi/asm/kvm.h | 2 ++ arch/riscv/kvm/isa.c | 2 ++ 2 files changed, 4 insertions(+)
arch/riscv/include/uapi/asm/kvm.h +2 -0
@@ -199,6 +199,8 @@ enum KVM_RISCV_ISA_EXT_ID {KVM_RISCV_ISA_EXT_ZCLSD ,KVM_RISCV_ISA_EXT_ZILSD ,KVM_RISCV_ISA_EXT_ZALASR , +KVM_RISCV_ISA_EXT_ZICFILP , +KVM_RISCV_ISA_EXT_ZICFISS , KVM_RISCV_ISA_EXT_MAX, };
arch/riscv/kvm/isa.c +2 -0
@@ -65,6 +65,8 @@ static const unsigned long kvm_isa_ext_arr[] = {KVM_ISA_EXT_ARR (ZICBOP ),KVM_ISA_EXT_ARR (ZICBOZ ),KVM_ISA_EXT_ARR (ZICCRSE ), +KVM_ISA_EXT_ARR (ZICFILP ), +KVM_ISA_EXT_ARR (ZICFISS ),KVM_ISA_EXT_ARR (ZICNTR ),KVM_ISA_EXT_ARR (ZICOND ), KVM_ISA_EXT_ARR(ZICSR),
Signature · 1 line
2.54.0
[email protected]
AI reviewRe: [PATCH v2 4/7] RISC-V: KVM: Allow Zicfiss/Zicfilp extensions for Guest/VM
Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider: - [Critical] Missing context switch and API support for the Zicfiss shadow stack pointer (`CSR_SSP`).…
Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [Critical] Missing context switch and API support for the Zicfiss shadow stack pointer (CSR_SSP).
- [High] Zicfilp and Zicfiss extensions are enabled by default, violating KVM subsystem guidelines. --
commit a0c518ab3b8be233eb4530a1f65b1f7970a43692
Author: Inochi Amaoto <[email protected]>
RISC-V: KVM: Allow Zicfiss/Zicfilp extensions for Guest/VM
Extend the KVM ISA extension array to allow KVM userspace to detect and enable Zicfiss/Zicfilp extensions for Guest/VM.
arch/riscv/include/uapi/asm/kvm.h enum KVM_RISCV_ISA_EXT_ID { · line 203
Quoted 10 lines
arch/riscv/include/uapi/asm/kvm.h +2
-0@@ -199,6 +199,8 @@ enum KVM_RISCV_ISA_EXT_ID {KVM_RISCV_ISA_EXT_ZCLSD ,KVM_RISCV_ISA_EXT_ZILSD ,KVM_RISCV_ISA_EXT_ZALASR , +KVM_RISCV_ISA_EXT_ZICFILP , + KVM_RISCV_ISA_EXT_ZICFISS,
[Severity: Critical]
Does this patch miss the context switch and API support for the Zicfiss shadow stack pointer (CSR_SSP)?
Since the Zicfiss extension is exposed to the guest here, shouldn't the unbanked ssp CSR (0x011) be saved for the host and restored for the guest in arch/riscv/kvm/vcpu_switch.S:__kvm_riscv_switch_to()?
Without this, if the host userspace uses shadow stacks, returning from the guest might leave the guest's ssp active, potentially pointing into guest memory.
Additionally, ssp appears to be missing from struct kvm_riscv_csr in arch/riscv/include/uapi/asm/kvm.h and struct kvm_vcpu_csr in arch/riscv/include/asm/kvm_host.h. Does this prevent saving or restoring the register during live migration?
arch/riscv/kvm/isa.c static const unsigned long kvm_isa_ext_arr[] = { · line 69
Quoted 13 lines
arch/riscv/include/uapi/asm/kvm.h +0
-0KVM_RISCV_ISA_EXT_MAX, };arch/riscv/kvm/isa.c +2
-0@@ -65,6 +65,8 @@ static const unsigned long kvm_isa_ext_arr[] = {KVM_ISA_EXT_ARR (ZICBOP ),KVM_ISA_EXT_ARR (ZICBOZ ),KVM_ISA_EXT_ARR (ZICCRSE ), +KVM_ISA_EXT_ARR (ZICFILP ), + KVM_ISA_EXT_ARR(ZICFISS),
[Severity: High]
Are these extensions being enabled by default?
Adding KVM_RISCV_ISA_EXT_ZICFILP and KVM_RISCV_ISA_EXT_ZICFISS here advertises them to the KVM capabilities, but they don't appear to be added to the exclusion list in kvm_riscv_isa_enable_allowed() in the same file.
Would this cause them to fall through to the default return true, and subsequently be enabled automatically for all new VMs if the host supports them?
If so, this might silently alter guest hardware discovery and affect backwards compatibility for live migration to older hosts.
Quoted 3 lines
arch/riscv/kvm/isa.c +0
-0KVM_ISA_EXT_ARR (ZICNTR ),KVM_ISA_EXT_ARR (ZICOND ), KVM_ISA_EXT_ARR(ZICSR),
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=4
Re: [PATCH v2 4/7] RISC-V: KVM: Allow Zicfiss/Zicfilp extensions for Guest/VM
In reply to Inochi AmaotoQuoted 12 lines · Inochi Amaoto
Extend the KVM ISA extension array to allow KVM userspace to detect and enable Zicfiss/Zicfilp extensions for Guest/VM.
Assisted-by YuanSheng:claude-4.7-opusCo-developed-by Quan Zhou <[email protected]>Signed-off-by Quan Zhou <[email protected]>Signed-off-by Inochi Amaoto <[email protected]>--- arch/riscv/include/uapi/asm/kvm.h | 2 ++ arch/riscv/kvm/isa.c | 2 ++ 2 files changed, 4 insertions(+)
I have forgot to add CSR_SSP to the context, will add in the next version. But I have a small question for it: could I reuse the generic subtype or use a new CSR subtype?
Regards,
Inochi
Quoted 34 lines
arch/riscv/include/uapi/asm/kvm.h +2
-0@@ -199,6 +199,8 @@ enum KVM_RISCV_ISA_EXT_ID {KVM_RISCV_ISA_EXT_ZCLSD ,KVM_RISCV_ISA_EXT_ZILSD ,KVM_RISCV_ISA_EXT_ZALASR , +KVM_RISCV_ISA_EXT_ZICFILP , +KVM_RISCV_ISA_EXT_ZICFISS , KVM_RISCV_ISA_EXT_MAX, };arch/riscv/kvm/isa.c +2
-0@@ -65,6 +65,8 @@ static const unsigned long kvm_isa_ext_arr[] = {KVM_ISA_EXT_ARR (ZICBOP ),KVM_ISA_EXT_ARR (ZICBOZ ),KVM_ISA_EXT_ARR (ZICCRSE ), +KVM_ISA_EXT_ARR (ZICFILP ), +KVM_ISA_EXT_ARR (ZICFISS ),KVM_ISA_EXT_ARR (ZICNTR ),KVM_ISA_EXT_ARR (ZICOND ), KVM_ISA_EXT_ARR(ZICSR),Signature · 7 lines
2.54.0 _______________________________________________ linux-riscv mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-riscv
Anup Patel
maintainerRe: [PATCH v2 4/7] RISC-V: KVM: Allow Zicfiss/Zicfilp extensions for Guest/VM
Quoted 15 lines · Inochi Amaoto
Quoted 12 lines · Inochi Amaoto
Extend the KVM ISA extension array to allow KVM userspace to detect and enable Zicfiss/Zicfilp extensions for Guest/VM.
Assisted-by YuanSheng:claude-4.7-opusCo-developed-by Quan Zhou <[email protected]>Signed-off-by Quan Zhou <[email protected]>Signed-off-by Inochi Amaoto <[email protected]>--- arch/riscv/include/uapi/asm/kvm.h | 2 ++ arch/riscv/kvm/isa.c | 2 ++ 2 files changed, 4 insertions(+)
Quoted 4 lines
I have forgot to add
CSR_SSPto the context, will add in the next version. But I have a small question for it: could I reuse the generic subtype or use a new CSR subtype?
Add another CSR subtype for Zicfiss.
Regards,
Anup
Re: [PATCH v2 4/7] RISC-V: KVM: Allow Zicfiss/Zicfilp extensions for Guest/VM
Quoted 21 lines · Anup Patel
Quoted 15 lines · Inochi Amaoto
Quoted 12 lines · Inochi Amaoto
Extend the KVM ISA extension array to allow KVM userspace to detect and enable Zicfiss/Zicfilp extensions for Guest/VM.
Assisted-by YuanSheng:claude-4.7-opusCo-developed-by Quan Zhou <[email protected]>Signed-off-by Quan Zhou <[email protected]>Signed-off-by Inochi Amaoto <[email protected]>--- arch/riscv/include/uapi/asm/kvm.h | 2 ++ arch/riscv/kvm/isa.c | 2 ++ 2 files changed, 4 insertions(+)Quoted 4 lines
I have forgot to add
CSR_SSPto the context, will add in the next version. But I have a small question for it: could I reuse the generic subtype or use a new CSR subtype?
Quoted 4 lines
Add another CSR subtype for Zicfiss.
Regards,
Anup
Thanks, I will do it.
Regards,
Inochi
PATCH v2 5/7 RISC-V: KVM: Handle software-check exits for VCPU
In reply to Inochi AmaotoThe extension Zicfiss/Zicfilp introduces a software-check exception to the privilege architecture. Delegate this exception to VS mode so the CFI violations in VU/VS could be reported.
--- arch/riscv/include/asm/csr.h | 1 + arch/riscv/kvm/vcpu_config.c | 4 ++++ arch/riscv/kvm/vcpu_exit.c | 3 +++ 3 files changed, 8 insertions(+)
arch/riscv/include/asm/csr.h +1 -0
@@ -117,6 +117,7 @@#define EXC_INST_PAGE_FAULT 12#define EXC_LOAD_PAGE_FAULT 13#define EXC_STORE_PAGE_FAULT 15 +#define EXC_SOFTWARE_CHECK 18#define EXC_INST_GUEST_PAGE_FAULT 20#define EXC_LOAD_GUEST_PAGE_FAULT 21#define EXC_VIRTUAL_INST_FAULT 22
arch/riscv/kvm/vcpu_config.c +4 -0
@@ -65,6 +65,10 @@ void kvm_riscv_vcpu_config_ran_once(struct kvm_vcpu *vcpu)if (vcpu->guest_debug) cfg->hedeleg &= ~BIT (EXC_BREAKPOINT ); + +if (riscv_isa_extension_available (isa,ZICFILP ) || +riscv_isa_extension_available (isa,ZICFISS )) + cfg->hedeleg |=BIT (EXC_SOFTWARE_CHECK ); }void kvm_riscv_vcpu_config_load (struct kvm_vcpu *vcpu)
arch/riscv/kvm/vcpu_exit.c +3 -0
@@ -243,6 +243,9 @@ int kvm_riscv_vcpu_exit(struct kvm_vcpu *vcpu, struct kvm_run *run, run->exit_reason =KVM_EXIT_DEBUG ; ret =0 ;break ; +case EXC_SOFTWARE_CHECK : + ret =vcpu_redirect (vcpu, trap); +break ;default :break ; }
Signature · 1 line
2.54.0
[email protected]
AI reviewRe: [PATCH v2 5/7] RISC-V: KVM: Handle software-check exits for VCPU
Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider: - [High] KVM fails to context-switch the Shadow Stack Pointer (CSR_SSP), leading to cross-VCPU leaks and…
Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:
- [High] KVM fails to context-switch the Shadow Stack Pointer (CSR_SSP), leading to cross-VCPU leaks and guest shadow stack corruption.
- [High] KVM fails to propagate the SPELP (Supervisor Previous Expected Landing Pad) bit when redirecting traps to the guest, breaking guest CFI tracking.
- [Medium] KVM unconditionally redirects EXC_SOFTWARE_CHECK to the guest upon VM exit, which will inject an un-enumerated exception into guests that have not negotiated the Zicfilp/Zicfiss extensions. --
commit bea0e82f22f5523f333dc995d01ccbe77f69b200
Author: Inochi Amaoto <[email protected]>
RISC-V: KVM: Handle software-check exits for VCPU
This commit enables the software-check exception for RISC-V KVM, delegating it to VS mode so CFI violations (Zicfiss/Zicfilp) can be reported to the guest.
Quoted 4 lines
arch/riscv/kvm/vcpu_config.c +0
-0
[ ... ]
arch/riscv/kvm/vcpu_config.c void kvm_riscv_vcpu_config_ran_once(struct kvm_vcpu *vcpu) · line 71
Quoted 8 lines
arch/riscv/kvm/vcpu_config.c +4
-0@@ -65,6 +65,10 @@ void kvm_riscv_vcpu_config_ran_once(struct kvm_vcpu *vcpu)if (vcpu->guest_debug) cfg->hedeleg &= ~BIT (EXC_BREAKPOINT ); + +if (riscv_isa_extension_available (isa,ZICFILP ) || +riscv_isa_extension_available (isa,ZICFISS )) + cfg->hedeleg |=BIT (EXC_SOFTWARE_CHECK );
[Severity: High]
The Zicfiss extension introduces the CSR_SSP (Shadow Stack Pointer) register, which is a user-mode CSR and isn't automatically saved or restored by the hardware on VM exits.
If KVM does not manually context-switch this register in arch/riscv/kvm/vcpu.c:kvm_riscv_vcpu_context_switch() or during VCPU load/put paths, could a new VCPU inherit the previous VCPU's CSR_SSP and leak state across VMs?
Additionally, when returning to host userspace, won't the host kernel's exception return code overwrite CSR_SSP with the host process's shadow stack pointer, leading to an immediate crash upon re-entry to the guest?
Quoted 6 lines
arch/riscv/kvm/vcpu_config.c +0
-0}arch/riscv/kvm/vcpu_exit.c +0
-0
[ ... ]
arch/riscv/kvm/vcpu_exit.c int kvm_riscv_vcpu_exit(struct kvm_vcpu *vcpu, struct kvm_run *run, · line 247
Quoted 6 lines
arch/riscv/kvm/vcpu_exit.c +2
-0@@ -243,6 +243,9 @@ int kvm_riscv_vcpu_exit(struct kvm_vcpu *vcpu, struct kvm_run *run, run->exit_reason =KVM_EXIT_DEBUG ; ret =0 ;break ; +case EXC_SOFTWARE_CHECK : + ret =vcpu_redirect (vcpu, trap);
[Severity: Medium]
This unconditionally redirects EXC_SOFTWARE_CHECK back to the guest.
If a guest has not negotiated the Zicfilp/Zicfiss extensions, KVM won't delegate EXC_SOFTWARE_CHECK via hedeleg.
If an unexpected CPU state or hardware issue triggers this exception in VS-mode and traps to HS-mode, this redirect will inject an un-enumerated exception into the guest.
Does this code need to verify riscv_isa_extension_available() before redirecting the trap?
[Severity: High]
When a trap occurs in VS/VU-mode and is taken to HS-mode, the hardware records the previous Zicfilp ELP state in the host sstatus.SPELP bit, which KVM subsequently saves into vcpu->arch.guest_context.sstatus.
When vcpu_redirect() is called here, it eventually invokes kvm_riscv_vcpu_trap_redirect(), which updates the guest's VSSTATUS register:
arch/riscv/kvm/vcpu_exit.c:kvm_riscv_vcpu_trap_redirect() {
...
unsigned long vsstatus = ncsr_read(CSR_VSSTATUS);
vsstatus &= ~SR_SPP;
if (vcpu->arch.guest_context.sstatus & SR_SPP)
vsstatus |= SR_SPP;
ncsr_write(CSR_VSSTATUS, vsstatus);
...
}Since kvm_riscv_vcpu_trap_redirect() explicitly copies SPP, SPIE, and SIE but does not copy the SPELP bit (SR_SPELP), does this cause the guest OS to lose its Zicfilp CFI tracking state across KVM-injected traps?
Quoted 4 lines
arch/riscv/kvm/vcpu_exit.c +1
-0+break ;default :break ; }
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=5
PATCH v2 6/7 RISC-V: KVM: Add support for control-flow integrity FWFT features
In reply to Inochi AmaotoControl-flow integrity is controlled through a WARL field in henvcfg. Expose the feature only if the Zicfilp/Zicfiss is supported for VS-mode. Allow the VMM to block access to the feature by disabling the ISA extension in the guest.
--- arch/riscv/kvm/vcpu_sbi_fwft.c | 70 ++++++++++++++++++++++++++++++++++ 1 file changed, 70 insertions(+)
arch/riscv/kvm/vcpu_sbi_fwft.c +70 -0
@@ -163,6 +163,58 @@ static long kvm_sbi_fwft_get_misaligned_delegation(struct kvm_vcpu *vcpu,return SBI_SUCCESS ; } +static bool kvm_sbi_fwft_landing_pad_supported (struct kvm_vcpu *vcpu) +{ +return riscv_isa_extension_available (vcpu->arch.isa,ZICFILP ); +} + +static void kvm_sbi_fwft_reset_landing_pad (struct kvm_vcpu *vcpu) +{ +kvm_sbi_fwft_env_flag_reset_helper (vcpu,ENVCFG_LPE ); +} + +static long kvm_sbi_fwft_set_landing_pad (struct kvm_vcpu *vcpu, +struct kvm_sbi_fwft_config *conf, +bool one_reg_access,unsigned long value) +{ +return kvm_sbi_fwft_env_flag_set_helper (vcpu, conf, one_reg_access, + value,ENVCFG_LPE ); +} + +static long kvm_sbi_fwft_get_landing_pad (struct kvm_vcpu *vcpu, +struct kvm_sbi_fwft_config *conf, +bool one_reg_access,unsigned long *value) +{ +return kvm_sbi_fwft_env_flag_get_helper (vcpu, conf, one_reg_access, + value,ENVCFG_LPE ); +} + +static bool kvm_sbi_fwft_shadow_stack_supported (struct kvm_vcpu *vcpu) +{ +return riscv_isa_extension_available (vcpu->arch.isa,ZICFISS ); +} + +static void kvm_sbi_fwft_reset_shadow_stack (struct kvm_vcpu *vcpu) +{ +kvm_sbi_fwft_env_flag_reset_helper (vcpu,ENVCFG_SSE ); +} + +static long kvm_sbi_fwft_set_shadow_stack (struct kvm_vcpu *vcpu, +struct kvm_sbi_fwft_config *conf, +bool one_reg_access,unsigned long value) +{ +return kvm_sbi_fwft_env_flag_set_helper (vcpu, conf, one_reg_access, + value,ENVCFG_SSE ); +} + +static long kvm_sbi_fwft_get_shadow_stack (struct kvm_vcpu *vcpu, +struct kvm_sbi_fwft_config *conf, +bool one_reg_access,unsigned long *value) +{ +return kvm_sbi_fwft_env_flag_get_helper (vcpu, conf, one_reg_access, + value,ENVCFG_SSE ); +} +static bool kvm_sbi_fwft_pte_ad_hw_updating_supported (struct kvm_vcpu *vcpu) { return riscv_isa_extension_available(vcpu->arch.isa, SVADU) &&
@@ -288,6 +340,24 @@ static const struct kvm_sbi_fwft_feature features[] = { .set = kvm_sbi_fwft_set_misaligned_delegation, .get = kvm_sbi_fwft_get_misaligned_delegation, }, + { + .id = SBI_FWFT_LANDING_PAD, + .first_reg_num = offsetof(struct kvm_riscv_sbi_fwft, landing_pad.enable) / + sizeof(unsigned long), + .supported = kvm_sbi_fwft_landing_pad_supported, + .reset = kvm_sbi_fwft_reset_landing_pad, + .set = kvm_sbi_fwft_set_landing_pad, + .get = kvm_sbi_fwft_get_landing_pad, + }, + { + .id = SBI_FWFT_SHADOW_STACK, + .first_reg_num = offsetof(struct kvm_riscv_sbi_fwft, shadow_stack.enable) / + sizeof(unsigned long), + .supported = kvm_sbi_fwft_shadow_stack_supported, + .reset = kvm_sbi_fwft_reset_shadow_stack, + .set = kvm_sbi_fwft_set_shadow_stack, + .get = kvm_sbi_fwft_get_shadow_stack, + }, { .id = SBI_FWFT_PTE_AD_HW_UPDATING, .first_reg_num = offsetof(struct kvm_riscv_sbi_fwft, pte_ad_hw_updating.enable) /
Signature · 1 line
2.54.0
[email protected]
AI reviewRe: [PATCH v2 6/7] RISC-V: KVM: Add support for control-flow integrity FWFT features
Thank you for your contribution! Sashiko AI review found 6 potential issue(s) to consider: - [High] KVM fails to context-switch the shadow stack pointer (CSR_SSP) when exposing ZICFISS to the guest.…
Thank you for your contribution! Sashiko AI review found 6 potential issue(s) to consider:
- [High] KVM fails to context-switch the shadow stack pointer (CSR_SSP) when exposing ZICFISS to the guest.
- [High] Direct write to CSR_HENVCFG bypasses nested acceleration (NaCl) shared memory synchronization.
- [High] The unsigned long flag argument in henvcfg helpers zero-extends and corrupts the upper 32 bits on RV32, and silently truncates 64-bit flags.
- [High] Guest can bypass VMM ISA restrictions to enable FWFT features due to caching at vCPU creation.
- [High] Modifying FWFT features via KVM_SET_ONE_REG updates vcpu->arch.cfg.henvcfg but fails to mark the CSR as dirty, leading to stale hardware state execution.
- [Medium] SBI FWFT flags are committed even if the feature value set operation fails.
--
commit c2d9c5584a3fada7da09588693c511ec3fdbbcf1
Author: Inochi Amaoto <[email protected]>
Subject: RISC-V: KVM: Add support for control-flow integrity FWFT features
This commit exposes the Zicfilp and Zicfiss control-flow integrity features to guests through the SBI FWFT interface. It also allows the VMM to block access to these features by disabling the corresponding ISA extensions.
Quoted 5 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +0
-0@@ -163,6 +163,58 @@ static long kvm_sbi_fwft_get_misaligned_delegation(struct kvm_vcpu *vcpu,
[ ... ]
arch/riscv/kvm/vcpu_sbi_fwft.c static long kvm_sbi_fwft_get_misaligned_delegation(struct kvm_vcpu *vcpu, · line 169
Quoted 4 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +4
-0+static bool kvm_sbi_fwft_landing_pad_supported (struct kvm_vcpu *vcpu) +{ +return riscv_isa_extension_available (vcpu->arch.isa,ZICFILP ); +}
[Severity: High]
Does checking the ISA extension at vCPU creation allow the guest to bypass VMM restrictions?
The FWFT framework caches feature availability during kvm_sbi_ext_fwft_init() which executes at vCPU creation. At that time, vcpu->arch.isa typically contains all default host-supported extensions.
If the VMM later configures the VM by disabling unsupported extensions via KVM_SET_ONE_REG, the guest can still successfully enable the hardware features because the FWFT layer permanently cached conf->supported as true.
[ ... ]
arch/riscv/kvm/vcpu_sbi_fwft.c static long kvm_sbi_fwft_get_misaligned_delegation(struct kvm_vcpu *vcpu, · line 182
Quoted 7 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +7
-0+static long kvm_sbi_fwft_set_landing_pad (struct kvm_vcpu *vcpu, +struct kvm_sbi_fwft_config *conf, +bool one_reg_access,unsigned long value) +{ +return kvm_sbi_fwft_env_flag_set_helper (vcpu, conf, one_reg_access, + value,ENVCFG_LPE ); +}
[Severity: High]
Can passing ENVCFG_LPE to this helper cause truncation or zero-extension issues?
Looking at kvm_sbi_fwft_env_flag_set_helper(), the flag parameter is typed as unsigned long. On RV32, unsigned long is 32 bits, which would truncate 64-bit flags.
Furthermore, when clearing a flag via cfg->henvcfg &= ~flag, if flag is 0, ~flag evaluates to the 32-bit 0xFFFFFFFF. Because cfg->henvcfg is a u64, the 32-bit mask is zero-extended to 0x00000000FFFFFFFFULL.
This zero-extension permanently wipes the entire upper 32 bits of henvcfg and corrupts hardware states like Svpbmt and Zicbom on 32-bit KVM hosts.
[Severity: High]
Does modifying FWFT features via KVM_SET_ONE_REG lead to stale hardware state?
When userspace modifies an FWFT feature via KVM_SET_ONE_REG, one_reg_access is true. In kvm_sbi_fwft_env_flag_set_helper():
if (!one_reg_access)
csr_write(CSR_HENVCFG, vcpu->arch.cfg.henvcfg);This updates the in-memory vcpu->arch.cfg.henvcfg but skips the immediate CSR write. However, it fails to set vcpu->arch.csr_dirty = true.
When vcpu_load() later executes, it evaluates !vcpu->arch.csr_dirty and skips kvm_riscv_vcpu_config_load(), causing the guest to run with stale hardware configurations.
[Severity: High]
Does writing directly to CSR_HENVCFG bypass nested acceleration?
When one_reg_access is false, kvm_sbi_fwft_env_flag_set_helper() directly writes the hardware CSR using csr_write(CSR_HENVCFG, ...).
When KVM runs as an L1 nested hypervisor with RISC-V nested acceleration enabled, hardware CSRs must be synchronized with the L0 hypervisor via a shared memory region.
Bypassing the ncsr_write() wrapper fails to set the dirty bit in the shared memory bitmap, meaning the L0 hypervisor remains unaware of the change and will overwrite the hardware CSR with a stale value on the next nested VM entry.
[ ... ]
arch/riscv/kvm/vcpu_sbi_fwft.c static long kvm_sbi_fwft_get_misaligned_delegation(struct kvm_vcpu *vcpu, · line 195
Quoted 4 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +4
-0+static bool kvm_sbi_fwft_shadow_stack_supported (struct kvm_vcpu *vcpu) +{ +return riscv_isa_extension_available (vcpu->arch.isa,ZICFISS ); +}
[Severity: High]
Does exposing ZICFISS without context-switching CSR_SSP leak guest state into the host?
When a guest enables and uses the shadow stack, it writes to the shadow stack pointer CSR_SSP. KVM does not currently save or restore CSR_SSP during VCPU context switches.
Since U-mode CSRs are not automatically banked by the RISC-V hardware between host and guest execution modes, the guest's shadow stack pointer remains active when returning to the host.
When the host's userspace resumes execution, it does so with a corrupted shadow stack pointer, leading to shadow stack violations and a denial-of-service crash.
[ ... ]
arch/riscv/kvm/vcpu_sbi_fwft.c static const struct kvm_sbi_fwft_feature features[] = { · line 351
Quoted 13 lines
arch/riscv/kvm/vcpu_sbi_fwft.c +9
-0@@ -288,6 +340,24 @@ static const struct kvm_sbi_fwft_feature features[] = { .set = kvm_sbi_fwft_set_misaligned_delegation, .get = kvm_sbi_fwft_get_misaligned_delegation, }, + { + .id = SBI_FWFT_LANDING_PAD, + .first_reg_num = offsetof(struct kvm_riscv_sbi_fwft, landing_pad.enable) / + sizeof(unsigned long), + .supported = kvm_sbi_fwft_landing_pad_supported, + .reset = kvm_sbi_fwft_reset_landing_pad, + .set = kvm_sbi_fwft_set_landing_pad, + .get = kvm_sbi_fwft_get_landing_pad, + },
[Severity: Medium]
Are SBI FWFT flags committed even if the set operation fails?
Looking at kvm_sbi_fwft_set(), the flags are assigned before the feature-specific set callback is executed:
conf->flags = flags;
return conf->feature->set(vcpu, conf, false, value);If the set callback fails and returns an error like SBI_ERR_INVALID_PARAM, the flags are not rolled back.
If the guest request included SBI_FWFT_SET_FLAG_LOCK, the feature becomes permanently locked into a stale state, preventing subsequent corrections.
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=6
PATCH v2 7/7 KVM: riscv: selftests: add Zicfiss/Zicfilp extension to get-reg-list test
In reply to Inochi AmaotoThe KVM RISC-V allows Zicfiss/Zicfilp extensions for Guest/VM so add these extensions to get-reg-list test.
--- tools/testing/selftests/kvm/riscv/get-reg-list.c | 16 ++++++++++++++++ 1 file changed, 16 insertions(+)
tools/testing/selftests/kvm/riscv/get-reg-list.c +16 -0
@@ -89,6 +89,8 @@ bool filter_reg(__u64 reg)case KVM_REG_RISCV_ISA_EXT |KVM_REG_RISCV_ISA_SINGLE |KVM_RISCV_ISA_EXT_ZICBOP :case KVM_REG_RISCV_ISA_EXT |KVM_REG_RISCV_ISA_SINGLE |KVM_RISCV_ISA_EXT_ZICBOZ :case KVM_REG_RISCV_ISA_EXT |KVM_REG_RISCV_ISA_SINGLE |KVM_RISCV_ISA_EXT_ZICCRSE : +case KVM_REG_RISCV_ISA_EXT |KVM_REG_RISCV_ISA_SINGLE |KVM_RISCV_ISA_EXT_ZICFILP : +case KVM_REG_RISCV_ISA_EXT |KVM_REG_RISCV_ISA_SINGLE |KVM_RISCV_ISA_EXT_ZICFISS :case KVM_REG_RISCV_ISA_EXT |KVM_REG_RISCV_ISA_SINGLE |KVM_RISCV_ISA_EXT_ZICNTR :case KVM_REG_RISCV_ISA_EXT |KVM_REG_RISCV_ISA_SINGLE |KVM_RISCV_ISA_EXT_ZICOND :case KVM_REG_RISCV_ISA_EXT |KVM_REG_RISCV_ISA_SINGLE |KVM_RISCV_ISA_EXT_ZICSR :
@@ -552,6 +554,8 @@ static const char *isa_ext_single_id_to_str(__u64 reg_off)KVM_ISA_EXT_ARR (ZICBOP ),KVM_ISA_EXT_ARR (ZICBOZ ),KVM_ISA_EXT_ARR (ZICCRSE ), +KVM_ISA_EXT_ARR (ZICFILP ), +KVM_ISA_EXT_ARR (ZICFISS ),KVM_ISA_EXT_ARR (ZICNTR ),KVM_ISA_EXT_ARR (ZICOND ), KVM_ISA_EXT_ARR(ZICSR),
@@ -715,6 +719,12 @@ static const char *sbi_fwft_id_to_str(__u64 reg_off)case 6 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(pte_ad_hw_updating.enable)" ;case 7 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(pte_ad_hw_updating.flags)" ;case 8 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(pte_ad_hw_updating.value)" ; +case 9 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(landing_pad.enable)" ; +case 10 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(landing_pad.flags)" ; +case 11 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(landing_pad.value)" ; +case 12 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(shadow_stack.enable)" ; +case 13 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(shadow_stack.flags)" ; +case 14 :return "KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(shadow_stack.value)" ; }return strdup_printf ("KVM_REG_RISCV_SBI_FWFT | %lld /* UNKNOWN */" , reg_off); }
@@ -911,6 +921,12 @@ static __u64 sbi_fwft_regs[] = {KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pte_ad_hw_updating.enable),KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pte_ad_hw_updating.flags),KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pte_ad_hw_updating.value), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (landing_pad.enable), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (landing_pad.flags), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (landing_pad.value), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (shadow_stack.enable), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (shadow_stack.flags), + KVM_REG_RISCV | KVM_REG_SIZE_ULONG | KVM_REG_RISCV_SBI_STATE | KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(shadow_stack.value), }; static __u64 zicbom_regs[] = {
Signature · 1 line
2.54.0
[email protected]
AI reviewRe: [PATCH v2 7/7] KVM: riscv: selftests: add Zicfiss/Zicfilp extension to get-reg-list test
Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
New issues:
- [Medium] The commit adds Zicfiss and Zicfilp extensions to the get-reg-list selftest but completely omits defining their vCPU test configurations, meaning the standalone tests for these extensions are never run.
Pre-existing issues:
- [High] This problem wasn't introduced by this patch, but KVM statically caches the supported state of FWFT features at vCPU initialization, allowing a guest to use Zicfiss/Zicfilp (and other) FWFT features even after the VMM explicitly disables these ISA extensions. --
commit c2918e7cd002c4f98c92703739dfdfd8015014a7
Author: Inochi Amaoto <[email protected]>
KVM: riscv: selftests: add Zicfiss/Zicfilp extension to get-reg-list test
The KVM RISC-V allows Zicfiss/Zicfilp extensions for Guest/VM so add these extensions to get-reg-list test.
Quoted 4 lines
tools/testing/selftests/kvm/riscv/get-reg-list.c +0
-0
[ ... ]
tools/testing/selftests/kvm/riscv/get-reg-list.c static __u64 sbi_fwft_regs[] = { · line 930
Quoted 11 lines
tools/testing/selftests/kvm/riscv/get-reg-list.c +6
-0@@ -911,6 +921,12 @@ static __u64 sbi_fwft_regs[] = {KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pte_ad_hw_updating.enable),KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pte_ad_hw_updating.flags),KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (pte_ad_hw_updating.value), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (landing_pad.enable), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (landing_pad.flags), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (landing_pad.value), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (shadow_stack.enable), +KVM_REG_RISCV |KVM_REG_SIZE_ULONG |KVM_REG_RISCV_SBI_STATE |KVM_REG_RISCV_SBI_FWFT |KVM_REG_RISCV_SBI_FWFT_REG (shadow_stack.flags), + KVM_REG_RISCV | KVM_REG_SIZE_ULONG | KVM_REG_RISCV_SBI_STATE | KVM_REG_RISCV_SBI_FWFT | KVM_REG_RISCV_SBI_FWFT_REG(shadow_stack.value), };
[Severity: High]
This isn't a bug introduced by this patch, but KVM statically caches the supported state of FWFT features at vCPU initialization, allowing a guest to use Zicfiss/Zicfilp (and other) FWFT features even after the VMM explicitly disables these ISA extensions.
During kvm_arch_vcpu_create(), kvm_sbi_ext_fwft_init() statically caches whether an FWFT feature is supported by calling its supported() callback (e.g. riscv_isa_extension_available(vcpu->arch.isa, ZICFILP)):
arch/riscv/kvm/vcpu_sbi_fwft.c:kvm_sbi_ext_fwft_init() {
...
if (feature->supported)
conf->supported = feature->supported(vcpu);
...
}The VMM may subsequently disable the underlying ISA extension using KVM_SET_ONE_REG, which clears the bit in vcpu->arch.isa.
However, the cached conf->supported state is never re-evaluated. When the guest later attempts to enable the feature via SBI_EXT_FWFT_SET, kvm_fwft_get_feature() reads the stale true value:
arch/riscv/kvm/vcpu_sbi_fwft.c:kvm_fwft_get_feature() {
...
if (!tconf->supported || !tconf->enabled)
return SBI_ERR_NOT_SUPPORTED;
...
}Could this allow the guest to bypass the VMM's configuration and trigger hardware states that were explicitly disabled, violating the virtualization boundary?
[Severity: Medium]
Are we missing the vcpu test configurations for the Zicfilp and Zicfiss extensions?
The get-reg-list test suite relies on the vcpu_configs[] array and explicitly declared configurations to determine which subtests to run:
tools/testing/selftests/kvm/riscv/get-reg-list.c:
struct vcpu_reg_list *vcpu_configs[] = {
&config_sbi_base,
...
};While the commit updates filter_reg() and adds the registers to sbi_fwft_regs, it omits the required KVM_ISA_EXT_SIMPLE_CONFIG() macro definitions for these extensions and fails to append &config_zicfilp and &config_zicfiss to the vcpu_configs[] array.
Will this cause the standalone validation for getting/setting these new ISA extension registers to be silently skipped, leaving a gap in test coverage?
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=7