Re: [PATCH] KVM: x86/pmu: Add Intel PMU supported fixed counters bit mask
From: Like Xu
Date: Tue Mar 28 2023 - 06:27:32 EST
On 27/3/2023 10:58 pm, Jim Mattson wrote:
On Mon, Mar 27, 2023 at 12:47 AM Like Xu <like.xu.linux@xxxxxxxxx> wrote:
On 25/3/2023 7:19 am, Jim Mattson wrote:
On Tue, Mar 21, 2023 at 4:28 AM Like Xu <like.xu.linux@xxxxxxxxx> wrote:
From: Like Xu <likexu@xxxxxxxxxxx>
Per Intel SDM, fixed-function performance counter 'i' is supported if:
FxCtr[i]_is_supported := ECX[i] || (EDX[4:0] > i);
which means that the KVM user space can use EDX to limit the number of
fixed counters and at the same time, using ECX to enable part of other
KVM supported fixed counters.
Add a bitmap (instead of always checking the vcpu's CPUIDs) to keep track
of the guest available fixed counters and perform the semantic checks.
Signed-off-by: Like Xu <likexu@xxxxxxxxxxx>
---
arch/x86/include/asm/kvm_host.h | 2 ++
arch/x86/kvm/pmu.h | 8 +++++
arch/x86/kvm/vmx/pmu_intel.c | 53 +++++++++++++++++++++------------
3 files changed, 44 insertions(+), 19 deletions(-)
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index a45de1118a42..14689e583127 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -565,6 +565,8 @@ struct kvm_pmu {
*/
bool need_cleanup;
+ DECLARE_BITMAP(supported_fixed_pmc_idx, KVM_PMC_MAX_FIXED);
+
/*
* The total number of programmed perf_events and it helps to avoid
* redundant check before cleanup if guest don't use vPMU at all.
diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h
index be62c16f2265..9f4504e5e9d5 100644
--- a/arch/x86/kvm/pmu.h
+++ b/arch/x86/kvm/pmu.h
@@ -111,6 +111,11 @@ static inline struct kvm_pmc *get_gp_pmc(struct kvm_pmu *pmu, u32 msr,
return NULL;
}
+static inline bool fixed_ctr_is_supported(struct kvm_pmu *pmu, unsigned int idx)
+{
+ return test_bit(idx, pmu->supported_fixed_pmc_idx);
+}
+
/* returns fixed PMC with the specified MSR */
static inline struct kvm_pmc *get_fixed_pmc(struct kvm_pmu *pmu, u32 msr)
{
@@ -120,6 +125,9 @@ static inline struct kvm_pmc *get_fixed_pmc(struct kvm_pmu *pmu, u32 msr)
u32 index = array_index_nospec(msr - base,
pmu->nr_arch_fixed_counters);
+ if (!fixed_ctr_is_supported(pmu, index))
+ return NULL;
+
return &pmu->fixed_counters[index];
}
diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
index e8a3be0b9df9..12f4b2fe7756 100644
--- a/arch/x86/kvm/vmx/pmu_intel.c
+++ b/arch/x86/kvm/vmx/pmu_intel.c
@@ -43,13 +43,16 @@ static int fixed_pmc_events[] = {1, 0, 7};
static void reprogram_fixed_counters(struct kvm_pmu *pmu, u64 data)
{
struct kvm_pmc *pmc;
- u8 old_fixed_ctr_ctrl = pmu->fixed_ctr_ctrl;
+ u8 new_ctrl, old_ctrl, old_fixed_ctr_ctrl = pmu->fixed_ctr_ctrl;
int i;
pmu->fixed_ctr_ctrl = data;
for (i = 0; i < pmu->nr_arch_fixed_counters; i++) {
- u8 new_ctrl = fixed_ctrl_field(data, i);
- u8 old_ctrl = fixed_ctrl_field(old_fixed_ctr_ctrl, i);
+ if (!fixed_ctr_is_supported(pmu, i))
+ continue;
+
+ new_ctrl = fixed_ctrl_field(data, i);
+ old_ctrl = fixed_ctrl_field(old_fixed_ctr_ctrl, i);
if (old_ctrl == new_ctrl)
continue;
@@ -125,6 +128,9 @@ static bool intel_is_valid_rdpmc_ecx(struct kvm_vcpu *vcpu, unsigned int idx)
idx &= ~(3u << 30);
+ if (fixed && !fixed_ctr_is_supported(pmu, idx))
+ return false;
+
return fixed ? idx < pmu->nr_arch_fixed_counters
: idx < pmu->nr_arch_gp_counters;
}
@@ -145,7 +151,7 @@ static struct kvm_pmc *intel_rdpmc_ecx_to_pmc(struct kvm_vcpu *vcpu,
counters = pmu->gp_counters;
num_counters = pmu->nr_arch_gp_counters;
}
- if (idx >= num_counters)
+ if (idx >= num_counters || (fixed && !fixed_ctr_is_supported(pmu, idx)))
return NULL;
*mask &= pmu->counter_bitmask[fixed ? KVM_PMC_FIXED : KVM_PMC_GP];
return &counters[array_index_nospec(idx, num_counters)];
@@ -500,6 +506,9 @@ static void setup_fixed_pmc_eventsel(struct kvm_pmu *pmu)
int i;
for (i = 0; i < pmu->nr_arch_fixed_counters; i++) {
+ if (!fixed_ctr_is_supported(pmu, i))
+ continue;
+
pmc = &pmu->fixed_counters[i];
event = fixed_pmc_events[array_index_nospec(i, size)];
pmc->eventsel = (intel_arch_events[event].unit_mask << 8) |
@@ -520,6 +529,7 @@ static void intel_pmu_refresh(struct kvm_vcpu *vcpu)
pmu->nr_arch_gp_counters = 0;
pmu->nr_arch_fixed_counters = 0;
+ bitmap_zero(pmu->supported_fixed_pmc_idx, KVM_PMC_MAX_FIXED);
pmu->counter_bitmask[KVM_PMC_GP] = 0;
pmu->counter_bitmask[KVM_PMC_FIXED] = 0;
pmu->version = 0;
@@ -551,13 +561,24 @@ static void intel_pmu_refresh(struct kvm_vcpu *vcpu)
pmu->available_event_types = ~entry->ebx &
((1ull << eax.split.mask_length) - 1);
- if (pmu->version == 1) {
- pmu->nr_arch_fixed_counters = 0;
- } else {
+ counter_mask = ~(BIT_ULL(pmu->nr_arch_gp_counters) - 1);
+ bitmap_set(pmu->all_valid_pmc_idx, 0, pmu->nr_arch_gp_counters);
+
+ if (pmu->version > 1) {
pmu->nr_arch_fixed_counters =
- min3(ARRAY_SIZE(fixed_pmc_events),
- (size_t) edx.split.num_counters_fixed,
- (size_t)kvm_pmu_cap.num_counters_fixed);
+ min_t(int, ARRAY_SIZE(fixed_pmc_events),
+ kvm_pmu_cap.num_counters_fixed);
+ for (i = 0; i < pmu->nr_arch_fixed_counters; i++) {
+ /* FxCtr[i]_is_supported := CPUID.0xA.ECX[i] || (EDX[4:0] > i) */
This is true only when pmu->version >= 5.
This is true in for "Version 5" section, but not mentioned in the CPUID.0xA section.
I would argue that this is a deliberate omission for the instruction implementation,
as it does use the word "version>1" in the near CPUID.0xA.EDX section.
Do you have any evidence to support such an argument? The CPUID field
in question was not defined prior to PMU version 5. (Does anyone from
Intel want to chime in?)
For virtualised use, this feature offers a kind of flexibility as users can
enable part of
the fixed counters, don't you think? Or maybe you're more looking forward to the
patch set that raises the vPMU version number from 2 to 5, that part of the code
was already in my tree some years ago.
I would not be surprised if a guest OS checked for PMU version 5
before consulting the CPUID fields defined in PMU version 5. Does
Linux even consult the fixed counter bitmask field today?
Yes, this is how host perf developer do it:
if (version >= 5)
x86_pmu.num_counters_fixed = fls(fixed_mask);
based on real fresh hardware (always marked as the latest version).
However, our KVM players can construct different valid CPUIDs, as long as the
hardware is capable,
to emulate some vPMU devices that match the CPUID semantics but do not exist in
the real world.
In the virtualisation world, use cases like "version 2 + fixed ctrs bit mask"
are perfectly possible
and should work as expected. One more case, if the forth fixed counter or more
is enabled in your guest for top-down feature and you may still find the guest's
pmu version number is stuck at 2.
This naturally does not occur in real hardware but no CPUID semantics here are
broken.
As I'm sure you've noticed, the logical relationship between CPUID.0xA.ECX and
PMU version 5
is necessary but not sufficient. Version 5 mush has fixed counters bit mask but
the reverse is not true.
From the end user's point of view, destroying the flexibility of vHW
combinations is a design failure.
So I think we can implement this feature in guest version 2, what do you think ?
I'd love to see KVM virtualize PMU version 5!
Great, I've got you and my plan will cover it.
From the SDM, volume 3, section 20.2.5 Architectural Performance
Monitoring Version 5:
With Architectural Performance Monitoring Version 5, register
CPUID.0AH.ECX indicates Fixed Counter enumeration. It is a bit mask
which enumerates the supported Fixed Counters in a processor. If bit
'i' is set, it implies that Fixed Counter 'i' is supported. Software
is recommended to use the following logic to check if a Fixed Counter
is supported on a given processor: FxCtr[i]_is_supported := ECX[i] ||
(EDX[4:0] > i);
Prior to PMU version 5, all fixed counters from 0 through <number of
fixed counters - 1> are supported.
+ if (!(entry->ecx & BIT_ULL(i) ||
+ edx.split.num_counters_fixed > i))
+ continue;
+
+ set_bit(i, pmu->supported_fixed_pmc_idx);
+ set_bit(INTEL_PMC_MAX_GENERIC + i, pmu->all_valid_pmc_idx);
+ pmu->fixed_ctr_ctrl_mask &= ~(0xbull << (i * 4));
+ counter_mask &= ~BIT_ULL(INTEL_PMC_MAX_GENERIC + i);
+ }
edx.split.bit_width_fixed = min_t(int, edx.split.bit_width_fixed,
kvm_pmu_cap.bit_width_fixed);
pmu->counter_bitmask[KVM_PMC_FIXED] =
@@ -565,10 +586,6 @@ static void intel_pmu_refresh(struct kvm_vcpu *vcpu)
setup_fixed_pmc_eventsel(pmu);
}
- for (i = 0; i < pmu->nr_arch_fixed_counters; i++)
- pmu->fixed_ctr_ctrl_mask &= ~(0xbull << (i * 4));
- counter_mask = ~(((1ull << pmu->nr_arch_gp_counters) - 1) |
- (((1ull << pmu->nr_arch_fixed_counters) - 1) << INTEL_PMC_IDX_FIXED));
pmu->global_ctrl_mask = counter_mask;
pmu->global_ovf_ctrl_mask = pmu->global_ctrl_mask
& ~(MSR_CORE_PERF_GLOBAL_OVF_CTRL_OVF_BUF |
@@ -585,11 +602,6 @@ static void intel_pmu_refresh(struct kvm_vcpu *vcpu)
pmu->raw_event_mask |= (HSW_IN_TX|HSW_IN_TX_CHECKPOINTED);
}
- bitmap_set(pmu->all_valid_pmc_idx,
- 0, pmu->nr_arch_gp_counters);
- bitmap_set(pmu->all_valid_pmc_idx,
- INTEL_PMC_MAX_GENERIC, pmu->nr_arch_fixed_counters);
-
perf_capabilities = vcpu_get_perf_capabilities(vcpu);
if (cpuid_model_is_consistent(vcpu) &&
(perf_capabilities & PMU_CAP_LBR_FMT))
@@ -605,6 +617,9 @@ static void intel_pmu_refresh(struct kvm_vcpu *vcpu)
pmu->pebs_enable_mask = counter_mask;
pmu->reserved_bits &= ~ICL_EVENTSEL_ADAPTIVE;
for (i = 0; i < pmu->nr_arch_fixed_counters; i++) {
+ if (!fixed_ctr_is_supported(pmu, i))
+ continue;
+
pmu->fixed_ctr_ctrl_mask &=
~(1ULL << (INTEL_PMC_IDX_FIXED + i * 4));
}
base-commit: d8708b80fa0e6e21bc0c9e7276ad0bccef73b6e7
--
2.40.0