lkml.org 
[lkml]   [2019]   [Dec]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
Subject[PATCH v5 04/10] perf: Add per perf_cpu_context min_heap storage
From
The storage required for visit_groups_merge's min heap needs to vary in
order to support more iterators, such as when multiple nested cgroups'
events are being visited. This change allows for 2 iterators and doesn't
support growth.
Based-on-work-by: Peter Zijlstra (Intel) <peterz@infradead.org>

Signed-off-by: Ian Rogers <irogers@google.com>
---
include/linux/perf_event.h | 7 +++++
kernel/events/core.c | 63 +++++++++++++++++++++++++-------------
2 files changed, 49 insertions(+), 21 deletions(-)

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 34c7c6910026..cd7d3b624655 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -850,6 +850,13 @@ struct perf_cpu_context {
int sched_cb_usage;

int online;
+ /*
+ * Per-CPU storage for iterators used in visit_groups_merge. The default
+ * storage is of size 2 to hold the CPU and any CPU event iterators.
+ */
+ int itr_storage_cap;
+ struct perf_event **itr_storage;
+ struct perf_event *itr_default[2];
};

struct perf_output_handle {
diff --git a/kernel/events/core.c b/kernel/events/core.c
index e0cc1c833408..259eca137563 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -49,7 +49,7 @@
#include <linux/sched/mm.h>
#include <linux/proc_ns.h>
#include <linux/mount.h>
-#include <linux/min_max_heap.h>
+#include <linux/min_heap.h>

#include "internal.h"

@@ -3402,13 +3402,13 @@ static void swap_ptr(void *l, void *r)
swap(*lp, *rp);
}

-static const struct min_max_heap_callbacks perf_min_heap = {
+static const struct min_heap_callbacks perf_min_heap = {
.elem_size = sizeof(struct perf_event *),
.cmp = perf_cmp_group_idx,
.swp = swap_ptr,
};

-static void __heap_add(struct min_max_heap *heap, struct perf_event *event)
+static void __heap_add(struct min_heap *heap, struct perf_event *event)
{
struct perf_event **itrs = heap->data;

@@ -3418,37 +3418,49 @@ static void __heap_add(struct min_max_heap *heap, struct perf_event *event)
}
}

-static noinline int visit_groups_merge(struct perf_event_groups *groups,
- int cpu,
+static noinline int visit_groups_merge(struct perf_cpu_context *cpuctx,
+ struct perf_event_groups *groups, int cpu,
int (*func)(struct perf_event *, void *),
void *data)
{
/* Space for per CPU and/or any CPU event iterators. */
struct perf_event *itrs[2];
- struct min_max_heap event_heap = {
- .data = itrs,
- .size = 0,
- .cap = ARRAY_SIZE(itrs),
- };
+ struct min_heap event_heap;
+ struct perf_event **evt;
struct perf_event *next;
int ret;

- __heap_add(&event_heap, perf_event_groups_first(groups, -1));
+ if (cpuctx) {
+ event_heap = (struct min_heap){
+ .data = cpuctx->itr_storage,
+ .size = 0,
+ .cap = cpuctx->itr_storage_cap,
+ };
+ } else {
+ event_heap = (struct min_heap){
+ .data = itrs,
+ .size = 0,
+ .cap = ARRAY_SIZE(itrs),
+ };
+ /* Events not within a CPU context may be on any CPU. */
+ __heap_add(&event_heap, perf_event_groups_first(groups, -1));
+ }
+ evt = event_heap.data;
+
__heap_add(&event_heap, perf_event_groups_first(groups, cpu));

- min_max_heapify_all(&event_heap, &perf_min_heap);
+ min_heapify_all(&event_heap, &perf_min_heap);

while (event_heap.size) {
- ret = func(itrs[0], data);
+ ret = func(*evt, data);
if (ret)
return ret;

- next = perf_event_groups_next(itrs[0]);
- if (next) {
- min_max_heap_pop_push(&event_heap, &next,
- &perf_min_heap);
- } else
- min_max_heap_pop(&event_heap, &perf_min_heap);
+ next = perf_event_groups_next(*evt);
+ if (next)
+ min_heap_pop_push(&event_heap, &next, &perf_min_heap);
+ else
+ min_heap_pop(&event_heap, &perf_min_heap);
}

return 0;
@@ -3518,7 +3530,10 @@ ctx_pinned_sched_in(struct perf_event_context *ctx,
.can_add_hw = 1,
};

- visit_groups_merge(&ctx->pinned_groups,
+ if (ctx != &cpuctx->ctx)
+ cpuctx = NULL;
+
+ visit_groups_merge(cpuctx, &ctx->pinned_groups,
smp_processor_id(),
pinned_sched_in, &sid);
}
@@ -3533,7 +3548,10 @@ ctx_flexible_sched_in(struct perf_event_context *ctx,
.can_add_hw = 1,
};

- visit_groups_merge(&ctx->flexible_groups,
+ if (ctx != &cpuctx->ctx)
+ cpuctx = NULL;
+
+ visit_groups_merge(cpuctx, &ctx->flexible_groups,
smp_processor_id(),
flexible_sched_in, &sid);
}
@@ -10350,6 +10368,9 @@ int perf_pmu_register(struct pmu *pmu, const char *name, int type)
cpuctx->online = cpumask_test_cpu(cpu, perf_online_mask);

__perf_mux_hrtimer_init(cpuctx, cpu);
+
+ cpuctx->itr_storage_cap = ARRAY_SIZE(cpuctx->itr_default);
+ cpuctx->itr_storage = cpuctx->itr_default;
}

got_cpu_context:
--
2.24.0.393.g34dc348eaf-goog
\
 
 \ /
  Last update: 2019-12-07 00:16    [W:0.332 / U:16.076 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site