On 5/8/26 3:03 PM, Shameer Kolothum Thodi wrote: > Hi Eric, > >> -----Original Message----- >> From: Eric Auger <[email protected]> >> Sent: 03 May 2026 08:34 >> To: [email protected]; [email protected]; qemu- >> [email protected]; [email protected]; [email protected]; >> [email protected]; [email protected]; >> [email protected]; [email protected]; Shameer Kolothum Thodi >> <[email protected]>; [email protected] >> Cc: [email protected]; [email protected]; [email protected]; >> [email protected]; [email protected]; [email protected]; >> [email protected] >> Subject: [PATCH v4 11/17] arm/kvm: write back modified ID regs to KVM >> >> External email: Use caution opening links or attachments >> >> >> We want to give a chance to override the value of host ID regs. >> In a previous patch we made sure all their values could be fetched >> through kvm_get_one_reg() calls before their modification. After >> their potential modification we need to make sure we write back >> the values through kvm_set_one_reg() calls. >> >> Make sure the cpreg_list is modified with updated values and >> transfer those values back to kvm. >> >> Signed-off-by: Eric Auger <[email protected]> >> Signed-off-by: Cornelia Huck <[email protected]> >> --- >> target/arm/kvm.c | 59 >> ++++++++++++++++++++++++++++++++++++++++- >> target/arm/trace-events | 1 + >> 2 files changed, 59 insertions(+), 1 deletion(-) >> >> diff --git a/target/arm/kvm.c b/target/arm/kvm.c >> index 1a9b91bf8a..ca9a7d9439 100644 >> --- a/target/arm/kvm.c >> +++ b/target/arm/kvm.c >> @@ -275,6 +275,21 @@ static uint32_t kvm_arm_sve_get_vls(int fd) >> return vls[0] & MAKE_64BIT_MASK(0, ARM_MAX_VQ); >> } >> >> +static int kvm_feature_idx_to_idregs_idx(int kidx) >> +{ >> + int op1, crm, op2; >> + ARMSysRegs sysreg; >> + >> + op1 = kidx / 64; >> + if (op1 == 2) { >> + op1 = 3; >> + } >> + crm = (kidx % 64) / 8; >> + op2 = kidx % 8; >> + sysreg = ENCODE_ID_REG(3, op1, 0, crm, op2); >> + return get_sysreg_idx(sysreg); >> +} >> + >> static int idregs_idx_to_kvm_feature_idx(ARMIDRegisterIdx idx) >> { >> ARMSysRegs sysreg = id_register_sysreg[idx]; >> @@ -1197,6 +1212,39 @@ bool kvm_arm_cpu_post_load(ARMCPU *cpu) >> return true; >> } >> >> +static void kvm_arm_writable_idregs_to_cpreg_list(ARMCPU *cpu) >> +{ >> + if (!cpu->writable_map) { >> + return; >> + } >> + for (int i = 0; i < NR_ID_REG_MASKS; i++) { >> + uint64_t writable_mask = cpu->writable_map->regs[i]; >> + uint64_t *cpreg; >> + >> + if (writable_mask) { >> + uint64_t previous, new; >> + int idx = kvm_feature_idx_to_idregs_idx(i); >> + ARM64SysReg *sysregdesc; >> + uint32_t sysreg; >> + >> + if (idx == -1) { >> + /* sysreg writable, but we don't know it */ >> + continue; >> + } >> + sysregdesc = &arm64_id_regs[idx]; >> + sysreg = sysregdesc->sysreg; >> + cpreg = kvm_arm_get_cpreg_ptr(cpu, >> idregs_sysreg_to_kvm_reg(sysreg)); >> + previous = *cpreg; >> + new = cpu->isar.idregs[idx]; >> + if (previous != new) { >> + *cpreg = new; >> + >> trace_kvm_arm_writable_idregs_to_cpreg_list(sysregdesc->name, >> + previous, new); >> + } >> + } >> + } >> +} >> + >> void kvm_arm_reset_vcpu(ARMCPU *cpu) >> { >> int ret; >> @@ -2148,7 +2196,16 @@ int kvm_arch_init_vcpu(CPUState *cs) >> } >> cpu->mp_affinity = mpidr & ARM64_AFFINITY_MASK; >> >> - return kvm_arm_init_cpreg_list(cpu); >> + ret = kvm_arm_init_cpreg_list(cpu); >> + if (ret) { >> + return ret; >> + } >> + /* overwrite writable ID regs with their updated property values */ >> + kvm_arm_writable_idregs_to_cpreg_list(cpu); >> + >> + write_list_to_kvmstate(cpu, KVM_PUT_FULL_STATE); > KVM may hide registers based on ID register fields when features > are disabled. Updating writable ID registers after KVM_GET_REG_LIST > can result in a mismatched register list and failures when setting > registers. > > This was reported on v3: > https://lore.kernel.org/qemu-devel/[email protected]/ Effectively I failed to address Jinqian's comment. This should be fixed in next respin Thanks to both of you! Eric > > Is this fixed someway in this series? I am not sure. I have faced the > same issue on NV platforms as well. > > Thanks, > Shameer >
