[PATCH v5 07/10] perf: simplify and rename visit_groups_merge

From: Ian Rogers
Date: Fri Dec 06 2019 - 18:16:24 EST


To enable a future caching optimization, pass in whether
visit_groups_merge is operating on pinned or flexible groups. The
is_pinned argument makes the func argument redundant, rename the
function to ctx_groups_sched_in as it just schedules pinned or flexible
groups in. Compute the cpu and groups arguments locally to reduce the
argument list size. Remove sched_in_data as it repeats arguments already
passed in. Merge pinned_sched_in and flexible_sched_in and use the
pinned argument to determine the active list.

Signed-off-by: Ian Rogers <irogers@xxxxxxxxxx>
---
kernel/events/core.c | 149 ++++++++++++++-----------------------------
1 file changed, 49 insertions(+), 100 deletions(-)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 20e08d0c1cb9..3da9cc1ebc2d 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -2133,7 +2133,6 @@ static void perf_group_detach(struct perf_event *event)

if (!RB_EMPTY_NODE(&event->group_node)) {
add_event_to_groups(sibling, event->ctx);
-
if (sibling->state == PERF_EVENT_STATE_ACTIVE) {
struct list_head *list = sibling->attr.pinned ?
&ctx->pinned_active : &ctx->flexible_active;
@@ -2456,6 +2455,8 @@ event_sched_in(struct perf_event *event,
{
int ret = 0;

+ WARN_ON_ONCE(event->ctx != ctx);
+
lockdep_assert_held(&ctx->lock);

if (event->state <= PERF_EVENT_STATE_OFF)
@@ -3524,10 +3525,42 @@ static void __heap_add(struct min_heap *heap, struct perf_event *event)
}
}

-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)
+static int merge_sched_in(struct perf_event_context *ctx,
+ struct perf_cpu_context *cpuctx,
+ struct perf_event *event,
+ bool is_pinned,
+ int *can_add_hw)
+{
+ WARN_ON_ONCE(event->ctx != ctx);
+
+ if (event->state <= PERF_EVENT_STATE_OFF)
+ return 0;
+
+ if (!event_filter_match(event))
+ return 0;
+
+ if (group_can_go_on(event, cpuctx, 1)) {
+ if (!group_sched_in(event, cpuctx, ctx)) {
+ list_add_tail(&event->active_list, is_pinned
+ ? &ctx->pinned_active
+ : &ctx->flexible_active);
+ }
+ }
+
+ if (event->state == PERF_EVENT_STATE_INACTIVE) {
+ if (is_pinned)
+ perf_event_set_state(event, PERF_EVENT_STATE_ERROR);
+
+ *can_add_hw = 0;
+ ctx->rotate_necessary = 1;
+ }
+
+ return 0;
+}
+
+static int ctx_groups_sched_in(struct perf_event_context *ctx,
+ struct perf_cpu_context *cpuctx,
+ bool is_pinned)
{
#ifdef CONFIG_CGROUP_PERF
struct cgroup_subsys_state *css = NULL;
@@ -3537,9 +3570,13 @@ static noinline int visit_groups_merge(struct perf_cpu_context *cpuctx,
struct min_heap event_heap;
struct perf_event **evt;
struct perf_event *next;
- int ret;
+ int ret, can_add_hw = 1;
+ int cpu = smp_processor_id();
+ struct perf_event_groups *groups = is_pinned
+ ? &ctx->pinned_groups
+ : &ctx->flexible_groups;

- if (cpuctx) {
+ if (ctx == &cpuctx->ctx) {
event_heap = (struct min_heap){
.data = cpuctx->itr_storage,
.size = 0,
@@ -3576,7 +3613,8 @@ static noinline int visit_groups_merge(struct perf_cpu_context *cpuctx,
min_heapify_all(&event_heap, &perf_min_heap);

while (event_heap.size) {
- ret = func(*evt, data);
+ ret = merge_sched_in(ctx, cpuctx, *evt, is_pinned, &can_add_hw);
+
if (ret)
return ret;

@@ -3590,96 +3628,6 @@ static noinline int visit_groups_merge(struct perf_cpu_context *cpuctx,
return 0;
}

-struct sched_in_data {
- struct perf_event_context *ctx;
- struct perf_cpu_context *cpuctx;
- int can_add_hw;
-};
-
-static int pinned_sched_in(struct perf_event *event, void *data)
-{
- struct sched_in_data *sid = data;
-
- if (event->state <= PERF_EVENT_STATE_OFF)
- return 0;
-
- if (!event_filter_match(event))
- return 0;
-
- if (group_can_go_on(event, sid->cpuctx, sid->can_add_hw)) {
- if (!group_sched_in(event, sid->cpuctx, sid->ctx))
- list_add_tail(&event->active_list, &sid->ctx->pinned_active);
- }
-
- /*
- * If this pinned group hasn't been scheduled,
- * put it in error state.
- */
- if (event->state == PERF_EVENT_STATE_INACTIVE)
- perf_event_set_state(event, PERF_EVENT_STATE_ERROR);
-
- return 0;
-}
-
-static int flexible_sched_in(struct perf_event *event, void *data)
-{
- struct sched_in_data *sid = data;
-
- if (event->state <= PERF_EVENT_STATE_OFF)
- return 0;
-
- if (!event_filter_match(event))
- return 0;
-
- if (group_can_go_on(event, sid->cpuctx, sid->can_add_hw)) {
- int ret = group_sched_in(event, sid->cpuctx, sid->ctx);
- if (ret) {
- sid->can_add_hw = 0;
- sid->ctx->rotate_necessary = 1;
- return 0;
- }
- list_add_tail(&event->active_list, &sid->ctx->flexible_active);
- }
-
- return 0;
-}
-
-static void
-ctx_pinned_sched_in(struct perf_event_context *ctx,
- struct perf_cpu_context *cpuctx)
-{
- struct sched_in_data sid = {
- .ctx = ctx,
- .cpuctx = cpuctx,
- .can_add_hw = 1,
- };
-
- if (ctx != &cpuctx->ctx)
- cpuctx = NULL;
-
- visit_groups_merge(cpuctx, &ctx->pinned_groups,
- smp_processor_id(),
- pinned_sched_in, &sid);
-}
-
-static void
-ctx_flexible_sched_in(struct perf_event_context *ctx,
- struct perf_cpu_context *cpuctx)
-{
- struct sched_in_data sid = {
- .ctx = ctx,
- .cpuctx = cpuctx,
- .can_add_hw = 1,
- };
-
- if (ctx != &cpuctx->ctx)
- cpuctx = NULL;
-
- visit_groups_merge(cpuctx, &ctx->flexible_groups,
- smp_processor_id(),
- flexible_sched_in, &sid);
-}
-
static void
ctx_sched_in(struct perf_event_context *ctx,
struct perf_cpu_context *cpuctx,
@@ -3716,11 +3664,12 @@ ctx_sched_in(struct perf_event_context *ctx,
* in order to give them the best chance of going on.
*/
if (is_active & EVENT_PINNED)
- ctx_pinned_sched_in(ctx, cpuctx);
+ ctx_groups_sched_in(ctx, cpuctx, /*is_pinned=*/true);
+

/* Then walk through the lower prio flexible groups */
if (is_active & EVENT_FLEXIBLE)
- ctx_flexible_sched_in(ctx, cpuctx);
+ ctx_groups_sched_in(ctx, cpuctx, /*is_pinned=*/false);
}

static void cpu_ctx_sched_in(struct perf_cpu_context *cpuctx,
--
2.24.0.393.g34dc348eaf-goog