[PATCH v11 08/17] KVM: x86/pmu: Reprogram PEBS event to emulate guest PEBS counter

[Like Xu <like.xu.linux@gmail.com>]

From: Like Xu <likexu@tencent.com>

When a guest counter is configured as a PEBS counter through
IA32_PEBS_ENABLE, a guest PEBS event will be reprogrammed by
configuring a non-zero precision level in the perf_event_attr.

The guest PEBS overflow PMI bit would be set in the guest
GLOBAL_STATUS MSR when PEBS facility generates a PEBS
overflow PMI based on guest IA32_DS_AREA MSR.

Even with the same counter index and the same event code and
mask, guest PEBS events will not be reused for non-PEBS events.

Originally-by: Andi Kleen <ak@linux.intel.com>
Co-developed-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Like Xu <likexu@tencent.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 arch/x86/kvm/pmu.c | 36 +++++++++++++++++++++++++++++++++---
 1 file changed, 33 insertions(+), 3 deletions(-)

diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
index a20207ee4014..8934261e9563 100644
--- a/arch/x86/kvm/pmu.c
+++ b/arch/x86/kvm/pmu.c
@@ -58,15 +58,22 @@ static void kvm_pmi_trigger_fn(struct irq_work *irq_work)
 static inline void __kvm_perf_overflow(struct kvm_pmc *pmc, bool in_pmi)
 {
 	struct kvm_pmu *pmu = pmc_to_pmu(pmc);
+	bool skip_pmi = false;
 
 	/* Ignore counters that have been reprogrammed already. */
 	if (test_and_set_bit(pmc->idx, pmu->reprogram_pmi))
 		return;
 
-	__set_bit(pmc->idx, (unsigned long *)&pmu->global_status);
+	if (pmc->perf_event && pmc->perf_event->attr.precise_ip) {
+		/* Indicate PEBS overflow PMI to guest. */
+		skip_pmi = __test_and_set_bit(GLOBAL_STATUS_BUFFER_OVF_BIT,
+					      (unsigned long *)&pmu->global_status);
+	} else {
+		__set_bit(pmc->idx, (unsigned long *)&pmu->global_status);
+	}
 	kvm_make_request(KVM_REQ_PMU, pmc->vcpu);
 
-	if (!pmc->intr)
+	if (!pmc->intr || skip_pmi)
 		return;
 
 	/*
@@ -97,6 +104,7 @@ static void pmc_reprogram_counter(struct kvm_pmc *pmc, u32 type,
 				  bool exclude_kernel, bool intr,
 				  bool in_tx, bool in_tx_cp)
 {
+	struct kvm_pmu *pmu = pmc_to_pmu(pmc);
 	struct perf_event *event;
 	struct perf_event_attr attr = {
 		.type = type,
@@ -108,6 +116,7 @@ static void pmc_reprogram_counter(struct kvm_pmc *pmc, u32 type,
 		.exclude_kernel = exclude_kernel,
 		.config = config,
 	};
+	bool pebs = test_bit(pmc->idx, (unsigned long *)&pmu->pebs_enable);
 
 	attr.sample_period = get_sample_period(pmc, pmc->counter);
 
@@ -122,6 +131,23 @@ static void pmc_reprogram_counter(struct kvm_pmc *pmc, u32 type,
 		attr.sample_period = 0;
 		attr.config |= HSW_IN_TX_CHECKPOINTED;
 	}
+	if (pebs) {
+		/*
+		 * The non-zero precision level of guest event makes the ordinary
+		 * guest event becomes a guest PEBS event and triggers the host
+		 * PEBS PMI handler to determine whether the PEBS overflow PMI
+		 * comes from the host counters or the guest.
+		 *
+		 * For most PEBS hardware events, the difference in the software
+		 * precision levels of guest and host PEBS events will not affect
+		 * the accuracy of the PEBS profiling result, because the "event IP"
+		 * in the PEBS record is calibrated on the guest side.
+		 *
+		 * On Icelake everything is fine. Other hardware (GLC+, TNT+) that
+		 * could possibly care here is unsupported and needs changes.
+		 */
+		attr.precise_ip = 1;
+	}
 
 	event = perf_event_create_kernel_counter(&attr, -1, current,
 						 kvm_perf_overflow, pmc);
@@ -135,7 +161,7 @@ static void pmc_reprogram_counter(struct kvm_pmc *pmc, u32 type,
 	pmc_to_pmu(pmc)->event_count++;
 	clear_bit(pmc->idx, pmc_to_pmu(pmc)->reprogram_pmi);
 	pmc->is_paused = false;
-	pmc->intr = intr;
+	pmc->intr = intr || pebs;
 }
 
 static void pmc_pause_counter(struct kvm_pmc *pmc)
@@ -161,6 +187,10 @@ static bool pmc_resume_counter(struct kvm_pmc *pmc)
 			      get_sample_period(pmc, pmc->counter)))
 		return false;
 
+	if (!test_bit(pmc->idx, (unsigned long *)&pmc_to_pmu(pmc)->pebs_enable) &&
+	    pmc->perf_event->attr.precise_ip)
+		return false;
+
 	/* reuse perf_event to serve as pmc_reprogram_counter() does*/
 	perf_event_enable(pmc->perf_event);
 	pmc->is_paused = false;
-- 
2.33.1