Skip to contentSkip to conversation

perf lock contention: Handle error in a single place

PATCH Applied Namhyung Kim 2 participants 4 messages Aug 30, 2024

Subsystems (2)
lore
Share & tools
lore

Namhyung Kim

author maintainer
#0 lore raw

PATCH 1/3 perf lock contention: Handle error in a single place

It has some duplicate codes to do the same job. Let's add a label and goto there to handle errors in a single place.

---
 tools/perf/util/bpf_skel/lock_contention.bpf.c | 16 ++++------------
 1 file changed, 4 insertions(+), 12 deletions(-)
tools/perf/util/bpf_skel/lock_contention.bpf.c +4 -12
index d931a898c434..e8a6f6463019 100644
@@ -439,11 +439,8 @@ int contention_end(u64 *ctx)
 
 	duration = bpf_ktime_get_ns() - pelem->timestamp;
 	if ((__s64)duration < 0) {
-		pelem->lock = 0;
-		if (need_delete)
-			bpf_map_delete_elem(&tstamp, &pid);
 		__sync_fetch_and_add(&time_fail, 1);
-		return 0;
+		goto out;
 	}
 
 	switch (aggr_mode) {
@@ -477,11 +474,8 @@ int contention_end(u64 *ctx)
 	data = bpf_map_lookup_elem(&lock_stat, &key);
 	if (!data) {
 		if (data_map_full) {
-			pelem->lock = 0;
-			if (need_delete)
-				bpf_map_delete_elem(&tstamp, &pid);
 			__sync_fetch_and_add(&data_fail, 1);
-			return 0;
+			goto out;
 		}
 
 		struct contention_data first = {
@@ -502,10 +496,7 @@ int contention_end(u64 *ctx)
 				data_map_full = 1;
 			__sync_fetch_and_add(&data_fail, 1);
 		}
-		pelem->lock = 0;
-		if (need_delete)
-			bpf_map_delete_elem(&tstamp, &pid);
-		return 0;
+		goto out;
 	}
 
 	__sync_fetch_and_add(&data->total_time, duration);
@@ -517,6 +508,7 @@ int contention_end(u64 *ctx)
 	if (data->min_time > duration)
 		data->min_time = duration;
 
+out:
 	pelem->lock = 0;
 	if (need_delete)
 		bpf_map_delete_elem(&tstamp, &pid);
Signature · 1 line
2.46.0.469.g59c65b2a67-goog
author maintainer
#1 lore raw

PATCH 2/3 perf lock contention: Simplify spinlock check

The LCB_F_SPIN bit is used for spinlock, rwlock and optimistic spinning in mutex. In get_tstamp_elem() it needs to check spinlock and rwlock only. As mutex sets the LCB_F_MUTEX, it can check those two bits and reduce the number of operations.

---
 tools/perf/util/bpf_skel/lock_contention.bpf.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
tools/perf/util/bpf_skel/lock_contention.bpf.c +1 -2
index e8a6f6463019..4b7237e178bd 100644
@@ -323,8 +323,7 @@ static inline struct tstamp_data *get_tstamp_elem(__u32 flags)
 	struct tstamp_data *pelem;
 
 	/* Use per-cpu array map for spinlock and rwlock */
-	if (flags == (LCB_F_SPIN | LCB_F_READ) || flags == LCB_F_SPIN ||
-	    flags == (LCB_F_SPIN | LCB_F_WRITE)) {
+	if ((flags & (LCB_F_SPIN | LCB_F_MUTEX)) == LCB_F_SPIN) {
 		__u32 idx = 0;
 
 		pelem = bpf_map_lookup_elem(&tstamp_cpu, &idx);
Signature · 1 line
2.46.0.469.g59c65b2a67-goog
author maintainer
#2 lore raw

PATCH 3/3 perf lock contention: Do not fail EEXIST for update

In reply to Namhyung Kim

When it updates the lock stat for the first time, it needs to create an element in the BPF hash map. But if there's a concurrent thread waiting for the same lock (like for rwsem or rwlock), it might race with the thread and possibly failed to update with -EEXIST. In that case, it can lookup the map again and put the data there instead of failing.

---
 tools/perf/util/bpf_skel/lock_contention.bpf.c | 7 +++++++
 1 file changed, 7 insertions(+)
tools/perf/util/bpf_skel/lock_contention.bpf.c +7 -0
index 4b7237e178bd..52a876b42699 100644
@@ -491,6 +491,12 @@ int contention_end(u64 *ctx)
 
 		err = bpf_map_update_elem(&lock_stat, &key, &first, BPF_NOEXIST);
 		if (err < 0) {
+			if (err == -EEXIST) {
+				/* it lost the race, try to get it again */
+				data = bpf_map_lookup_elem(&lock_stat, &key);
+				if (data != NULL)
+					goto found;
+			}
 			if (err == -E2BIG)
 				data_map_full = 1;
 			__sync_fetch_and_add(&data_fail, 1);
@@ -498,6 +504,7 @@ int contention_end(u64 *ctx)
 		goto out;
 	}
 
+found:
 	__sync_fetch_and_add(&data->total_time, duration);
 	__sync_fetch_and_add(&data->count, 1);
 
Signature · 1 line
2.46.0.469.g59c65b2a67-goog
maintainer Applied
#3 lore raw
In reply to Namhyung Kim
Quoted 2 lines · Namhyung Kim

It has some duplicate codes to do the same job. Let's add a label and goto there to handle errors in a single place.

Thanks, applied to perf-tools-next,

- Arnaldo

Quoted 57 lines
---
 tools/perf/util/bpf_skel/lock_contention.bpf.c | 16 ++++------------
 1 file changed, 4 insertions(+), 12 deletions(-)
tools/perf/util/bpf_skel/lock_contention.bpf.c +4 -12
index d931a898c434..e8a6f6463019 100644
@@ -439,11 +439,8 @@ int contention_end(u64 *ctx)
 
 	duration = bpf_ktime_get_ns() - pelem->timestamp;
 	if ((__s64)duration < 0) {
-		pelem->lock = 0;
-		if (need_delete)
-			bpf_map_delete_elem(&tstamp, &pid);
 		__sync_fetch_and_add(&time_fail, 1);
-		return 0;
+		goto out;
 	}
 
 	switch (aggr_mode) {
@@ -477,11 +474,8 @@ int contention_end(u64 *ctx)
 	data = bpf_map_lookup_elem(&lock_stat, &key);
 	if (!data) {
 		if (data_map_full) {
-			pelem->lock = 0;
-			if (need_delete)
-				bpf_map_delete_elem(&tstamp, &pid);
 			__sync_fetch_and_add(&data_fail, 1);
-			return 0;
+			goto out;
 		}
 
 		struct contention_data first = {
@@ -502,10 +496,7 @@ int contention_end(u64 *ctx)
 				data_map_full = 1;
 			__sync_fetch_and_add(&data_fail, 1);
 		}
-		pelem->lock = 0;
-		if (need_delete)
-			bpf_map_delete_elem(&tstamp, &pid);
-		return 0;
+		goto out;
 	}
 
 	__sync_fetch_and_add(&data->total_time, duration);
@@ -517,6 +508,7 @@ int contention_end(u64 *ctx)
 	if (data->min_time > duration)
 		data->min_time = duration;
 
+out:
 	pelem->lock = 0;
 	if (need_delete)
 		bpf_map_delete_elem(&tstamp, &pid);
Signature · 1 line
2.46.0.469.g59c65b2a67-goog