[bpf-next,v3,1/2] bpf: add runtime stats, max cost

Commit Message

Tonghao Zhang Dec. 19, 2022, 6:44 a.m. UTC

From: Tonghao Zhang <[e-mail]>

Now user can enable sysctl kernel.bpf_stats_enabled to fetch
run_time_ns and run_cnt. It's easy to calculate the average value.

In some case, the max cost for bpf prog invoked, are more useful:
is there a burst sysload or high cpu usage:

* If prog invoked frequently(run_cnt may be too large), run_time_ns/run_cnt
  is not ideal to indicate a bpf prog cpu burst. And syscall frequently
  may consume a lot of CPU cycles.
* This also help us to debug bpf prog, the cost is what we want?
  if not, there may be an issue in bpf prog.

This patch introduce a update stats helper.

$ bpftool --json --pretty p s
   ...
   "run_max_cost_ns": 313367

Signed-off-by: Tonghao Zhang <[e-mail]>
Cc: Alexei Starovoitov <[e-mail]>
Cc: Daniel Borkmann <[e-mail]>
Cc: Andrii Nakryiko <[e-mail]>
Cc: Martin KaFai Lau <[e-mail]>
Cc: Song Liu <[e-mail]>
Cc: Yonghong Song <[e-mail]>
Cc: John Fastabend <[e-mail]>
Cc: KP Singh <[e-mail]>
Cc: Stanislav Fomichev <[e-mail]>
Cc: Hao Luo <[e-mail]>
Cc: Jiri Olsa <[e-mail]>
Cc: Hou Tao <[e-mail]>
---
 include/linux/filter.h   | 29 ++++++++++++++++++++++-------
 include/uapi/linux/bpf.h |  1 +
 kernel/bpf/syscall.c     | 10 +++++++++-
 kernel/bpf/trampoline.c  | 10 +---------
 4 files changed, 33 insertions(+), 17 deletions(-)

Patch

diff --git a/include/linux/filter.h b/include/linux/filter.h
index bf701976056e..886b65fcd4ac 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -556,6 +556,7 @@  struct bpf_prog_stats {
 	u64_stats_t cnt;
 	u64_stats_t nsecs;
 	u64_stats_t misses;
+	u64_stats_t max_cost;
 	struct u64_stats_sync syncp;
 } __aligned(2 * sizeof(u64));
 
@@ -578,6 +579,26 @@  typedef unsigned int (*bpf_dispatcher_fn)(const void *ctx,
 					  unsigned int (*bpf_func)(const void *,
 								   const struct bpf_insn *));
 
+static inline void bpf_prog_update_stats(const struct bpf_prog *prog, u64 start)
+{
+	struct bpf_prog_stats *stats;
+	unsigned long flags;
+	u64 run_time, max_cost;
+
+	stats = this_cpu_ptr(prog->stats);
+	flags = u64_stats_update_begin_irqsave(&stats->syncp);
+
+	run_time =  sched_clock() - start;
+	u64_stats_inc(&stats->cnt);
+	u64_stats_add(&stats->nsecs, run_time);
+
+	max_cost = u64_stats_read(&stats->max_cost);
+	if (max_cost < run_time)
+		u64_stats_set(&stats->max_cost, run_time);
+
+	u64_stats_update_end_irqrestore(&stats->syncp, flags);
+}
+
 static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
 					  const void *ctx,
 					  bpf_dispatcher_fn dfunc)
@@ -586,16 +607,10 @@  static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
 
 	cant_migrate();
 	if (static_branch_unlikely(&bpf_stats_enabled_key)) {
-		struct bpf_prog_stats *stats;
 		u64 start = sched_clock();
-		unsigned long flags;
 
 		ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
-		stats = this_cpu_ptr(prog->stats);
-		flags = u64_stats_update_begin_irqsave(&stats->syncp);
-		u64_stats_inc(&stats->cnt);
-		u64_stats_add(&stats->nsecs, sched_clock() - start);
-		u64_stats_update_end_irqrestore(&stats->syncp, flags);
+		bpf_prog_update_stats(prog, start);
 	} else {
 		ret = dfunc(ctx, prog->insnsi, prog->bpf_func);
 	}
diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index 464ca3f01fe7..da4d1f2d7bc2 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h
@@ -6259,6 +6259,7 @@  struct bpf_prog_info {
 	__u32 verified_insns;
 	__u32 attach_btf_obj_id;
 	__u32 attach_btf_id;
+	__u64 run_max_cost_ns;
 } __attribute__((aligned(8)));
 
 struct bpf_map_info {
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index 64131f88c553..06439b09863d 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -2105,6 +2105,7 @@  struct bpf_prog_kstats {
 	u64 nsecs;
 	u64 cnt;
 	u64 misses;
+	u64 max_cost;
 };
 
 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
@@ -2122,12 +2123,13 @@  static void bpf_prog_get_stats(const struct bpf_prog *prog,
 			       struct bpf_prog_kstats *stats)
 {
 	u64 nsecs = 0, cnt = 0, misses = 0;
+	u64 max_cost = 0;
 	int cpu;
 
 	for_each_possible_cpu(cpu) {
 		const struct bpf_prog_stats *st;
 		unsigned int start;
-		u64 tnsecs, tcnt, tmisses;
+		u64 tnsecs, tcnt, tmisses, tmax_cost;
 
 		st = per_cpu_ptr(prog->stats, cpu);
 		do {
@@ -2135,14 +2137,17 @@  static void bpf_prog_get_stats(const struct bpf_prog *prog,
 			tnsecs = u64_stats_read(&st->nsecs);
 			tcnt = u64_stats_read(&st->cnt);
 			tmisses = u64_stats_read(&st->misses);
+			tmax_cost = u64_stats_read(&st->max_cost);
 		} while (u64_stats_fetch_retry(&st->syncp, start));
 		nsecs += tnsecs;
 		cnt += tcnt;
 		misses += tmisses;
+		max_cost = max(max_cost, tmax_cost);
 	}
 	stats->nsecs = nsecs;
 	stats->cnt = cnt;
 	stats->misses = misses;
+	stats->max_cost = max_cost;
 }
 
 #ifdef CONFIG_PROC_FS
@@ -2162,6 +2167,7 @@  static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
 		   "prog_id:\t%u\n"
 		   "run_time_ns:\t%llu\n"
 		   "run_cnt:\t%llu\n"
+		   "run_max_cost_ns:\t%llu\n"
 		   "recursion_misses:\t%llu\n"
 		   "verified_insns:\t%u\n",
 		   prog->type,
@@ -2171,6 +2177,7 @@  static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
 		   prog->aux->id,
 		   stats.nsecs,
 		   stats.cnt,
+		   stats.max_cost,
 		   stats.misses,
 		   prog->aux->verified_insns);
 }
@@ -3962,6 +3969,7 @@  static int bpf_prog_get_info_by_fd(struct file *file,
 	info.run_time_ns = stats.nsecs;
 	info.run_cnt = stats.cnt;
 	info.recursion_misses = stats.misses;
+	info.run_max_cost_ns = stats.max_cost;
 
 	info.verified_insns = prog->aux->verified_insns;
 
diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c
index d6395215b849..4ddad462562e 100644
--- a/kernel/bpf/trampoline.c
+++ b/kernel/bpf/trampoline.c
@@ -882,8 +882,6 @@  static u64 notrace __bpf_prog_enter_recur(struct bpf_prog *prog, struct bpf_tram
 static void notrace update_prog_stats(struct bpf_prog *prog,
 				      u64 start)
 {
-	struct bpf_prog_stats *stats;
-
 	if (static_branch_unlikely(&bpf_stats_enabled_key) &&
 	    /* static_key could be enabled in __bpf_prog_enter*
 	     * and disabled in __bpf_prog_exit*.
@@ -891,13 +889,7 @@  static void notrace update_prog_stats(struct bpf_prog *prog,
 	     * Hence check that 'start' is valid.
 	     */
 	    start > NO_START_TIME) {
-		unsigned long flags;
-
-		stats = this_cpu_ptr(prog->stats);
-		flags = u64_stats_update_begin_irqsave(&stats->syncp);
-		u64_stats_inc(&stats->cnt);
-		u64_stats_add(&stats->nsecs, sched_clock() - start);
-		u64_stats_update_end_irqrestore(&stats->syncp, flags);
+		bpf_prog_update_stats(prog, start);
 	}
 }