The hypervisor may request the perf subsystem to schedule a time window
to directly access the LBR stack msrs for its own use. Normally, it would
create a guest LBR event with callstack mode enabled, which is scheduled
along with other ordinary LBR events on the host but in an exclusive way.
To avoid wasting a counter for the guest LBR event, the perf tracks it via
needs_guest_lbr_without_counter() and assigns it with a fake VLBR counter
with the help of new lbr_without_counter_constraint. As with the BTS event,
there is actually no hardware counter assigned for the guest LBR event.
Cc: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
Signed-off-by: Like Xu <like.xu@xxxxxxxxxxxxxxx>
---
arch/x86/events/core.c | 1 +
arch/x86/events/intel/core.c | 17 +++++++++++++++++
arch/x86/events/intel/lbr.c | 3 +++
arch/x86/events/perf_event.h | 12 ++++++++++++
arch/x86/include/asm/perf_event.h | 16 +++++++++++++++-
include/linux/perf_event.h | 7 +++++++
kernel/events/core.c | 7 -------
7 files changed, 55 insertions(+), 8 deletions(-)
diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c
index f7a259dcbb06..2405926e2dba 100644
--- a/arch/x86/events/core.c
+++ b/arch/x86/events/core.c
@@ -1104,6 +1104,7 @@ static inline void x86_assign_hw_event(struct perf_event *event,
switch (hwc->idx) {
case INTEL_PMC_IDX_FIXED_BTS:
+ case INTEL_PMC_IDX_FIXED_VLBR:
hwc->config_base = 0;
hwc->event_base = 0;
break;
diff --git a/arch/x86/events/intel/core.c b/arch/x86/events/intel/core.c
index f1439acbf7e6..fe5595275368 100644
--- a/arch/x86/events/intel/core.c
+++ b/arch/x86/events/intel/core.c
@@ -2621,6 +2621,19 @@ intel_bts_constraints(struct perf_event *event)
return NULL;
}
+/*
+ * Note, the event that satisfies this constraint will not be assigned
+ * with a normal hardware counter but a fake one like BTS event.
+ */
+static struct event_constraint *
+intel_guest_lbr_constraints(struct perf_event *event)
+{
+ if (unlikely(needs_guest_lbr_without_counter(event)))
+ return &lbr_without_counter_constraint;
+
+ return NULL;
+}
+
static int intel_alt_er(int idx, u64 config)
{
int alt_idx = idx;
@@ -2811,6 +2824,10 @@ __intel_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
{
struct event_constraint *c;
+ c = intel_guest_lbr_constraints(event);
+ if (c)
+ return c;
+
c = intel_bts_constraints(event);
if (c)
return c;
diff --git a/arch/x86/events/intel/lbr.c b/arch/x86/events/intel/lbr.c
index 6c60dcaaaf69..2fca4aff7621 100644
--- a/arch/x86/events/intel/lbr.c
+++ b/arch/x86/events/intel/lbr.c
@@ -1363,3 +1363,6 @@ int x86_perf_get_lbr(struct x86_pmu_lbr *stack)
return 0;
}
EXPORT_SYMBOL_GPL(x86_perf_get_lbr);
+
+struct event_constraint lbr_without_counter_constraint =
+ EVENT_CONSTRAINT(0, 1ULL << INTEL_PMC_IDX_FIXED_VLBR, 0);
diff --git a/arch/x86/events/perf_event.h b/arch/x86/events/perf_event.h
index 1025bc6eb04f..e6e8c626ed00 100644
--- a/arch/x86/events/perf_event.h
+++ b/arch/x86/events/perf_event.h
@@ -969,6 +969,17 @@ static inline bool intel_pmu_has_bts(struct perf_event *event)
return intel_pmu_has_bts_period(event, hwc->sample_period);
}
+/*
+ * It's safe to generate an event with attr.exclude_host set and also
+ * using LBR to profile guest for other in-kernel users because the
+ * intel_guest_lbr_constraints() makes LBR registers to be used exclusively.
+ */
+static inline bool needs_guest_lbr_without_counter(struct perf_event *event)
+{
+ return needs_branch_stack(event) && is_kernel_event(event) &&
+ event->attr.exclude_host;
+}
+
int intel_pmu_save_and_restart(struct perf_event *event);
struct event_constraint *
@@ -989,6 +1000,7 @@ void release_ds_buffers(void);
void reserve_ds_buffers(void);
extern struct event_constraint bts_constraint;
+extern struct event_constraint lbr_without_counter_constraint;
void intel_pmu_enable_bts(u64 config);
diff --git a/arch/x86/include/asm/perf_event.h b/arch/x86/include/asm/perf_event.h
index 5071515f6b0f..7be581027ebb 100644
--- a/arch/x86/include/asm/perf_event.h
+++ b/arch/x86/include/asm/perf_event.h
@@ -192,9 +192,23 @@ struct x86_pmu_capability {
#define GLOBAL_STATUS_UNC_OVF BIT_ULL(61)
#define GLOBAL_STATUS_ASIF BIT_ULL(60)
#define GLOBAL_STATUS_COUNTERS_FROZEN BIT_ULL(59)
-#define GLOBAL_STATUS_LBRS_FROZEN BIT_ULL(58)
+#define GLOBAL_STATUS_LBRS_FROZEN_BIT 58
+#define GLOBAL_STATUS_LBRS_FROZEN BIT_ULL(GLOBAL_STATUS_LBRS_FROZEN_BIT)
#define GLOBAL_STATUS_TRACE_TOPAPMI BIT_ULL(55)
+/*
+ * We model guest LBR event tracing as another fixed-mode PMC like BTS.
+ *
+ * We choose bit 58 because it's used to indicate LBR stack frozen state
+ * for architectural perfmon v4, also we unconditionally mask that bit in
+ * the handle_pmi_common(), so it'll never be set in the overflow handling.
+ *
+ * With this fake counter assigned, the guest LBR event user (such as KVM),
+ * can program the LBR registers on its own, and we don't actually do anything
+ * with then in the host context.
+ */
+#define INTEL_PMC_IDX_FIXED_VLBR GLOBAL_STATUS_LBRS_FROZEN_BIT
+
/*
* Adaptive PEBS v4
*/
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 8768a39b5258..e25930fa526c 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -1101,6 +1101,13 @@ static inline bool is_sampling_event(struct perf_event *event)
return event->attr.sample_period != 0;
}
+#define TASK_TOMBSTONE ((void *)-1L)
+
+static inline bool is_kernel_event(struct perf_event *event)
+{
+ return READ_ONCE(event->owner) == TASK_TOMBSTONE;
+}
+
/*
* Return 1 for a software event, 0 for a hardware event
*/
diff --git a/kernel/events/core.c b/kernel/events/core.c
index e1459df73043..5ea22596ede2 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -165,13 +165,6 @@ static void perf_ctx_unlock(struct perf_cpu_context *cpuctx,
raw_spin_unlock(&cpuctx->ctx.lock);
}
-#define TASK_TOMBSTONE ((void *)-1L)
-
-static bool is_kernel_event(struct perf_event *event)
-{
- return READ_ONCE(event->owner) == TASK_TOMBSTONE;
-}
-
/*
* On task ctx scheduling...
*
--
2.21.1
