On 08/15/2016 05:09 AM, Andrew Jones wrote:
> On Sat, Aug 13, 2016 at 12:52:50AM -0500, Wei Huang wrote:
>> This patch adds a pmu=[on/off] option to enable/disable host vPMU
>> support in guest VM. There are several reasons to justify this option.
>> First, host-backed vPMU can be problematic for cross-migration between
>> different SoC as perf counters are architecture-dependent. It is more
>> flexible to have an option to turn it on/off. Secondly this option
>> matches the "pmu" option as supported in libvirt tool.
>>
>> Note that, like "has_el3", the "pmu" option is only made available on
>> CPUs that support host-backed vPMU. They include:
>> * cortex-a53 + kvm
>> * cortex-a57 + kvm
>> * host + kvm
>> This option is removed in other configs where it doesn't make sense
>> (e.g. cortex-a57 + TCG); and the default pmu support is off. This patch
>
> Some of the PMU registers are already emulated with TCG. While it
> doesn't make much sense to use perf counters in a TCG guest for
> measuring anything, it may have merit for emulation completeness
> and debug. With that in mind, then I think we need to add the PMU
> feature to all processors that support it, and thus the 'has_pmu'
The TCG case you mentioned was one reason I changed the naming. The
existing "has_pmu" variable was introduced recently and it didn't cover
the case of TCG. Reading 5c0a3819, it is only turned on when KVM is
enabled and PMU_V3_extension is available. If we want to extend the
meaning of "has_pmu" for all PMU cases (i.e. KVM + TCG), we probably
need to address cases of has_pmu=ON + TCG=ON.
> name is better. As the feature is optional, I agree it does make
> sense that it's off by default, and only available with the pmu=on
> property.
>
> I think there are more processors than just A53 and A57.
host-backed vPMU is only available on A53/A57/host. But TCG supports
more than A53/A57.
>
>> has been tested under both DT/ACPI modes.
>>
>> Signed-off-by: Wei Huang <w...@redhat.com>
>> ---
>> hw/arm/virt-acpi-build.c | 2 +-
>> hw/arm/virt.c | 2 +-
>> target-arm/cpu.c | 13 +++++++++++++
>> target-arm/cpu.h | 3 ++-
>> target-arm/cpu64.c | 6 ++++++
>> target-arm/kvm64.c | 10 +++++-----
>> 6 files changed, 28 insertions(+), 8 deletions(-)
>>
>> diff --git a/hw/arm/virt-acpi-build.c b/hw/arm/virt-acpi-build.c
>> index 28fc59c..22fb9d9 100644
>> --- a/hw/arm/virt-acpi-build.c
>> +++ b/hw/arm/virt-acpi-build.c
>> @@ -540,7 +540,7 @@ build_madt(GArray *table_data, BIOSLinker *linker,
>> VirtGuestInfo *guest_info)
>> gicc->uid = i;
>> gicc->flags = cpu_to_le32(ACPI_GICC_ENABLED);
>>
>> - if (armcpu->has_pmu) {
>> + if (armcpu->has_host_pmu) {
>> gicc->performance_interrupt = cpu_to_le32(PPI(VIRTUAL_PMU_IRQ));
>> }
>> }
>> diff --git a/hw/arm/virt.c b/hw/arm/virt.c
>> index a193b5a..4975f38 100644
>> --- a/hw/arm/virt.c
>> +++ b/hw/arm/virt.c
>> @@ -477,7 +477,7 @@ static void fdt_add_pmu_nodes(const VirtBoardInfo *vbi,
>> int gictype)
>>
>> CPU_FOREACH(cpu) {
>> armcpu = ARM_CPU(cpu);
>> - if (!armcpu->has_pmu ||
>> + if (!armcpu->has_host_pmu ||
>> !kvm_arm_pmu_create(cpu, PPI(VIRTUAL_PMU_IRQ))) {
>> return;
>> }
>> diff --git a/target-arm/cpu.c b/target-arm/cpu.c
>> index ce8b8f4..a527128 100644
>> --- a/target-arm/cpu.c
>> +++ b/target-arm/cpu.c
>> @@ -509,6 +509,10 @@ static Property arm_cpu_rvbar_property =
>> static Property arm_cpu_has_el3_property =
>> DEFINE_PROP_BOOL("has_el3", ARMCPU, has_el3, true);
>>
>> +/* use property name "pmu" to match with other archs and libvirt */
>> +static Property arm_cpu_has_host_pmu_property =
>> + DEFINE_PROP_BOOL("pmu", ARMCPU, has_host_pmu, false);
>> +
>> static Property arm_cpu_has_mpu_property =
>> DEFINE_PROP_BOOL("has-mpu", ARMCPU, has_mpu, true);
>>
>> @@ -552,6 +556,11 @@ static void arm_cpu_post_init(Object *obj)
>> #endif
>> }
>>
>> + if (arm_feature(&cpu->env, ARM_FEATURE_HOST_PMU)) {
>> + qdev_property_add_static(DEVICE(obj),
>> &arm_cpu_has_host_pmu_property,
>> + &error_abort);
>> + }
>> +
>> if (arm_feature(&cpu->env, ARM_FEATURE_MPU)) {
>> qdev_property_add_static(DEVICE(obj), &arm_cpu_has_mpu_property,
>> &error_abort);
>> @@ -648,6 +657,10 @@ static void arm_cpu_realizefn(DeviceState *dev, Error
>> **errp)
>> cpu->id_aa64pfr0 &= ~0xf000;
>> }
>>
>> + if (!cpu->has_host_pmu) {
>> + unset_feature(env, ARM_FEATURE_HOST_PMU);
>> + }
>
> I think this is the right approach. But we should add the feature to all
> processors that can have it, and then, after checking the property's
> value, we remove it when it's not desired.
>
>> +
>> if (!arm_feature(env, ARM_FEATURE_EL2)) {
>> /* Disable the hypervisor feature bits in the processor feature
>> * registers if we don't have EL2. These are id_pfr1[15:12] and
>> diff --git a/target-arm/cpu.h b/target-arm/cpu.h
>> index 76d824d..f3f6d3f 100644
>> --- a/target-arm/cpu.h
>> +++ b/target-arm/cpu.h
>> @@ -580,7 +580,7 @@ struct ARMCPU {
>> /* CPU has security extension */
>> bool has_el3;
>> /* CPU has PMU (Performance Monitor Unit) */
>> - bool has_pmu;
>> + bool has_host_pmu;
>>
>> /* CPU has memory protection unit */
>> bool has_mpu;
>> @@ -1129,6 +1129,7 @@ enum arm_features {
>> ARM_FEATURE_V8_SHA256, /* implements SHA256 part of v8 Crypto
>> Extensions */
>> ARM_FEATURE_V8_PMULL, /* implements PMULL part of v8 Crypto Extensions
>> */
>> ARM_FEATURE_THUMB_DSP, /* DSP insns supported in the Thumb encodings */
>> + ARM_FEATURE_HOST_PMU, /* PMU supported by host */
>> };
>>
>> static inline int arm_feature(CPUARMState *env, int feature)
>> diff --git a/target-arm/cpu64.c b/target-arm/cpu64.c
>> index 1635deb..19e8127 100644
>> --- a/target-arm/cpu64.c
>> +++ b/target-arm/cpu64.c
>> @@ -111,6 +111,9 @@ static void aarch64_a57_initfn(Object *obj)
>> set_feature(&cpu->env, ARM_FEATURE_V8_PMULL);
>> set_feature(&cpu->env, ARM_FEATURE_CRC);
>> set_feature(&cpu->env, ARM_FEATURE_EL3);
>> + if (kvm_enabled()) {
>> + set_feature(&cpu->env, ARM_FEATURE_HOST_PMU);
>> + }
>
> We shouldn't need the if (kvm_enabled()) here.
>
>> cpu->kvm_target = QEMU_KVM_ARM_TARGET_CORTEX_A57;
>> cpu->midr = 0x411fd070;
>> cpu->revidr = 0x00000000;
>> @@ -166,6 +169,9 @@ static void aarch64_a53_initfn(Object *obj)
>> set_feature(&cpu->env, ARM_FEATURE_V8_PMULL);
>> set_feature(&cpu->env, ARM_FEATURE_CRC);
>> set_feature(&cpu->env, ARM_FEATURE_EL3);
>> + if (kvm_enabled()) {
>> + set_feature(&cpu->env, ARM_FEATURE_HOST_PMU);
>> + }
>
> Or here.
>
>> cpu->kvm_target = QEMU_KVM_ARM_TARGET_CORTEX_A53;
>> cpu->midr = 0x410fd034;
>> cpu->revidr = 0x00000000;
>> diff --git a/target-arm/kvm64.c b/target-arm/kvm64.c
>> index 5faa76c..588e5e3 100644
>> --- a/target-arm/kvm64.c
>> +++ b/target-arm/kvm64.c
>> @@ -469,6 +469,7 @@ bool kvm_arm_get_host_cpu_features(ARMHostCPUClass *ahcc)
>> set_feature(&features, ARM_FEATURE_VFP4);
>> set_feature(&features, ARM_FEATURE_NEON);
>> set_feature(&features, ARM_FEATURE_AARCH64);
>> + set_feature(&features, ARM_FEATURE_HOST_PMU);
>>
>> ahcc->features = features;
>>
>> @@ -501,11 +502,10 @@ int kvm_arch_init_vcpu(CPUState *cs)
>> if (!arm_feature(&cpu->env, ARM_FEATURE_AARCH64)) {
>> cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_EL1_32BIT;
>> }
>> - if (kvm_irqchip_in_kernel() &&
>> - kvm_check_extension(cs->kvm_state, KVM_CAP_ARM_PMU_V3)) {
>> - cpu->has_pmu = true;
>> - cpu->kvm_init_features[0] |= 1 << KVM_ARM_VCPU_PMU_V3;
>> - }
>> + /* enable vPMU based on KVM mode, hw capability, and user setting */
>> + cpu->has_host_pmu &= kvm_irqchip_in_kernel() &&
>> + kvm_check_extension(cs->kvm_state, KVM_CAP_ARM_PMU_V3);
>> + cpu->kvm_init_features[0] |= cpu->has_host_pmu << KVM_ARM_VCPU_PMU_V3;
>
> Hmm... so now if the user doesn't specify pmu=on (off by default) then we
> silently disable it (I think that's a 2.6 compat issue), and if they do
^^^^^^^^^^^^^ True. I will see if I can create a patch backward
compatible with 2.6.
> specify pmu=on, but KVM doesn't support it, then it's silently disabled,
> which isn't nice. I think QEMU should error out in that case, explaining
> that the user specified incompatible options, i.e. they selected accel=kvm
> and pmu=on, but their KVM version (or cpu model) doesn't support PMU
> emulation.
True. This can be fixed.
>
>>
>> /* Do KVM_ARM_VCPU_INIT ioctl */
>> ret = kvm_arm_vcpu_init(cs);
>> --
>> 2.4.11
>>
>>
>
> Thanks,
> drew
>
