Re: [PATCH 2/3] target/i386: Call KVM_CAP_PMU_CAPABILITY iotcl to enable/disable PMU

["Mi, Dapeng" <dapeng1.mi@linux.intel.com>]

On 3/26/2025 2:46 PM, Dongli Zhang wrote:
> Hi Dapeng,
>
> PATCH 1-4 from the below patchset are already reviewed. (PATCH 5-10 are for PMU
> registers reset).
>
> https://lore.kernel.org/all/20250302220112.17653-1-dongli.zhang@oracle.com/
>
> They require only trivial modification. i.e.:
>
> https://github.com/finallyjustice/patchset/tree/master/qemu-amd-pmu-mid/v03
>
> Therefore, since PATCH 5-10 are for another topic, any chance if I re-send 1-4
> as a prerequisite for the patch to explicitly call KVM_CAP_PMU_CAPABILITY?

any option is fine for me, spiting it into two separated ones or still keep
in a whole patch series. I would rebase this this patchset on top of your
v3 patches.


>
> In addition, I have a silly question. Can mediated vPMU coexist with legacy
> perf-based vPMU, that is, something like tdp and tdp_mmu? Or the legacy
> perf-based vPMU is going to be purged from the most recent kernel?

No, they can't. As long as mediated vPMU is enabled, it would totally
replace the legacy perf-based vPMU. The legacy perf-based vPMU code would
still be kept in the kernel in near future, but the long-term target is to
totally remove the perf-based vPMU once mediated vPMU is mature.


>
> If they can coexist, how about add property to QEMU control between
> legacy/modern? i.e. by default use legacy and change to modern as default in the
> future once the feature is stable.
>
> Thank you very much!
>
> Dongli Zhang
>
> On 3/24/25 5:37 AM, Dapeng Mi wrote:
>> After introducing mediated vPMU, mediated vPMU must be enabled by
>> explicitly calling KVM_CAP_PMU_CAPABILITY to enable. Thus call
>> KVM_CAP_PMU_CAPABILITY to enable/disable PMU base on user configuration.
>>
>> Suggested-by: Zhao Liu <zhao1.liu@intel.com>
>> Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
>> ---
>>  target/i386/kvm/kvm.c | 17 +++++++++++++++++
>>  1 file changed, 17 insertions(+)
>>
>> diff --git a/target/i386/kvm/kvm.c b/target/i386/kvm/kvm.c
>> index f41e190fb8..d3e6984844 100644
>> --- a/target/i386/kvm/kvm.c
>> +++ b/target/i386/kvm/kvm.c
>> @@ -2051,8 +2051,25 @@ full:
>>      abort();
>>  }
>>  
>> +static bool pmu_cap_set = false;
>>  int kvm_arch_pre_create_vcpu(CPUState *cpu, Error **errp)
>>  {
>> +    KVMState *s = kvm_state;
>> +    X86CPU *x86_cpu = X86_CPU(cpu);
>> +
>> +    if (!pmu_cap_set && kvm_check_extension(s, KVM_CAP_PMU_CAPABILITY)) {
>> +        int r = kvm_vm_enable_cap(s, KVM_CAP_PMU_CAPABILITY, 0,
>> +                                  KVM_PMU_CAP_DISABLE & !x86_cpu->enable_pmu);
>> +        if (r < 0) {
>> +            error_report("kvm: Failed to %s pmu cap: %s",
>> +                         x86_cpu->enable_pmu ? "enable" : "disable",
>> +                         strerror(-r));
>> +            return r;
>> +        }
>> +
>> +        pmu_cap_set = true;
>> +    }
>> +
>>      return 0;
>>  }
>>