Skip to contentSkip to conversation

perf stat: Fix NULL vs IS_ERR() checking in check_per_pkg

PATCH In review Miaoqian Lin 2 participants 7 messages Dec 12, 2021 – Dec 13, 2021 perf events

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 perf stat: Fix NULL vs IS_ERR() checking in check_per_pkg

The hashmap__new() function does not return NULL on errors. It returns ERR_PTR(-ENOMEM). Using IS_ERR() to check the return value to fix this.

---
 tools/perf/util/stat.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/util/stat.c +2 -2
index 09ea334586f2..a77052680087 100644
@@ -311,8 +311,8 @@ static int check_per_pkg(struct evsel *counter,
 
 	if (!mask) {
 		mask = hashmap__new(pkg_id_hash, pkg_id_equal, NULL);
-		if (!mask)
-			return -ENOMEM;
+		if (IS_ERR(mask))
+			return PTR_ERR(mask);
 
 		counter->per_pkg_mask = mask;
 	}
Signature · 1 line
2.17.1
#1 lore raw

Hi Miaoqian,

Quoted 14 lines · Miaoqian Lin

The hashmap__new() function does not return NULL on errors. It returns ERR_PTR(-ENOMEM). Using IS_ERR() to check the return value to fix this.

---
 tools/perf/util/stat.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/util/stat.c +0 -0
index 09ea334586f2..a77052680087 100644

tools/perf/util/stat.c static int check_per_pkg(struct evsel *counter, · line 315

Quoted 7 lines
tools/perf/util/stat.c +2 -2
@@ -311,8 +311,8 @@ static int check_per_pkg(struct evsel *counter,
 
 	if (!mask) {
 		mask = hashmap__new(pkg_id_hash, pkg_id_equal, NULL);
-		if (!mask)
-			return -ENOMEM;
+		if (IS_ERR(mask))
+			return PTR_ERR(mask);

I see that callers to "ids__new" are also missing these checks. Did you consider patching those also?

Thanks,
German

Quoted 3 lines
tools/perf/util/stat.c +0 -0
 		counter->per_pkg_mask = mask;
 	}
#2 lore raw
Quoted 18 lines · German Gomez

Hi Miaoqian,

Quoted 15 lines · Miaoqian Lin

The hashmap__new() function does not return NULL on errors. It returns ERR_PTR(-ENOMEM). Using IS_ERR() to check the return value to fix this.

---
 tools/perf/util/stat.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/util/stat.c +0 -0
index 09ea334586f2..a77052680087 100644
@@ -311,8 +311,8 @@ static int check_per_pkg(struct evsel *counter,
 
Quoted 6 lines

tools/perf/util/stat.c static int check_per_pkg(struct evsel *counter, · line 315

Quoted 6 lines
tools/perf/util/stat.c +2 -2
 	if (!mask) {
 		mask = hashmap__new(pkg_id_hash, pkg_id_equal, NULL);
-		if (!mask)
-			return -ENOMEM;
+		if (IS_ERR(mask))
+			return PTR_ERR(mask);

Also (sorry, I haven't tried/tested the patch yet) is this return necessary? To me it seems it's still ok to keep the "return -ENOMEM".

Quoted 9 lines

I see that callers to "ids__new" are also missing these checks. Did you consider patching those also?

Thanks,
German

Quoted 3 lines
tools/perf/util/stat.c +0 -0
 		counter->per_pkg_mask = mask;
 	}
#3 lore raw

PATCH 1/1 perf expr: Fix return value of ids__new

In reply to German Gomez

callers of ids__new() function only do NULL checking for the return value. ids__new() calles hashmap__new(), which may return ERR_PTR(-ENOMEM). Instead of changing the checking one-by-one. return NULL instead of ERR_PTR(-ENOMEM) to keep consistent.

---
 tools/perf/util/expr.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
tools/perf/util/expr.c +7 -1
index 1d532b9fed29..aabdc112300c 100644
@@ -65,7 +65,13 @@ static bool key_equal(const void *key1, const void *key2,
 
 struct hashmap *ids__new(void)
 {
-	return hashmap__new(key_hash, key_equal, NULL);
+	struct hashmap *hash;
+
+	hash = hashmap__new(key_hash, key_equal, NULL);
+	if (IS_ERR(hash))
+		return NULL;
+	else
+		return hash;
 }
 
 void ids__free(struct hashmap *ids)
Signature · 1 line
2.17.1
Tested-by Reviewed-by on v1 1/1
#4 lore raw

Re: [PATCH] perf expr: Fix return value of ids__new

Hi Miaoqian,

Fails to build due to missing import: "#import <linux/err.h>".

Could you please verify?

Other than that, it looks good to me. I shared the testing below:

Quoted 20 lines · Miaoqian Lin

callers of ids__new() function only do NULL checking for the return value. ids__new() calles hashmap__new(), which may return ERR_PTR(-ENOMEM). Instead of changing the checking one-by-one. return NULL instead of ERR_PTR(-ENOMEM) to keep consistent.

---
 tools/perf/util/expr.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
tools/perf/util/expr.c +0 -1
index 1d532b9fed29..aabdc112300c 100644
@@ -65,7 +65,13 @@ static bool key_equal(const void *key1, const void *key2,
 
 struct hashmap *ids__new(void)
 {
-	return hashmap__new(key_hash, key_equal, NULL);

tools/perf/util/expr.c static bool key_equal(const void *key1, const void *key2, · line 77

Quoted 10 lines
tools/perf/util/expr.c +7 -0
+	struct hashmap *hash;
+
+	hash = hashmap__new(key_hash, key_equal, NULL);
+	if (IS_ERR(hash))
+		return NULL;
+	else
+		return hash;
 }
 
 void ids__free(struct hashmap *ids)

Before this patch, perf-test was segfaulting instead of a graceful fail. I think this could have been an issue in the perf tool as well.

(I forced hashmap__new in "tools/perf/util/hashmap.c" to always return the error for the purposes of the test).

  $ make DEBUG=1 NO_LIBBPF=1 # builds with tools/perf/util/hashmap.c
  $ ./perf test 7 -v
  Couldn't bump rlimit(MEMLOCK), failures may take place when creating BPF maps, etc
   7: Simple expression parser                                        :
  --- start ---
  test child forked, pid 1953536
  perf: Segmentation fault
  Obtained 16 stack frames.
  ./perf(dump_stack+0x31) [0x559222b3ae48]
  ./perf(sighandler_dump_stack+0x33) [0x559222b3af30]
  /lib/x86_64-linux-gnu/libc.so.6(+0x4620f) [0x7ff81df6220f]
  ./perf(hashmap__size+0x23) [0x559222bf9f47]
  ./perf(ids__union+0x5f) [0x559222be1f0e]
  ./perf(+0x2d9ba8) [0x559222ac7ba8]
  ./perf(+0x2da223) [0x559222ac8223]
  ./perf(+0x2a5fe9) [0x559222a93fe9]
  ./perf(+0x2a6119) [0x559222a94119]
  ./perf(+0x2a6de7) [0x559222a94de7]
  ./perf(cmd_test+0x25f) [0x559222a95686]
  ./perf(+0x2e4d25) [0x559222ad2d25]
  ./perf(+0x2e4faa) [0x559222ad2faa]
  ./perf(+0x2e50fd) [0x559222ad30fd]
  ./perf(main+0x29d) [0x559222ad3500]
  /lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xf2) [0x7ff81df430b2]
  test child interrupted
  ---- end ----
  Simple expression parser: FAILED!

After the patch:

  $ make DEBUG=1 NO_LIBBPF=1
  $ ./perf test 7 -v
  Couldn't bump rlimit(MEMLOCK), failures may take place when creating BPF maps, etc
   7: Simple expression parser                                        :
  --- start ---
  test child forked, pid 1960026
  FAILED tests/expr.c:16 ids__new
  FAILED tests/expr.c:16 ids__new
  FAILED tests/expr.c:73 ids_union (-1 != 0)
  test child finished with -1
  ---- end ----

  Simple expression parser: FAILED!

so with the missing import fixex:

#5 lore raw

Re: [PATCH] perf expr: Fix return value of ids__new

Quoted 20 lines · German Gomez

[...]

Quoted 17 lines · Miaoqian Lin

callers of ids__new() function only do NULL checking for the return value. ids__new() calles hashmap__new(), which may return ERR_PTR(-ENOMEM). Instead of changing the checking one-by-one. return NULL instead of ERR_PTR(-ENOMEM) to keep consistent.

---
 tools/perf/util/expr.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
tools/perf/util/expr.c +0 -0
index 1d532b9fed29..aabdc112300c 100644
@@ -65,7 +65,13 @@ static bool key_equal(const void *key1, const void *key2,
 
Quoted 10 lines

tools/perf/util/expr.c static bool key_equal(const void *key1, const void *key2, · line 74

Quoted 10 lines
tools/perf/util/expr.c +7 -1
 struct hashmap *ids__new(void)
 {
-	return hashmap__new(key_hash, key_equal, NULL);
+	struct hashmap *hash;
+
+	hash = hashmap__new(key_hash, key_equal, NULL);
+	if (IS_ERR(hash))
+		return NULL;
+	else
+		return hash;

Small nitpick but can remove the else branch and simply return at the end? Your patch still compiles for me but I think most other functions do that.

Quoted 1 line
Quoted 1 line

[...]

#6 lore raw

Re: [PATCH] perf expr: Fix return value of ids__new

In reply to Miaoqian Lin

This message appeared in my client as a separate thread, but it seems to be a reply to a message from a different thread:

"In-Reply-To: <[email protected]>"

Quoted 30 lines · Miaoqian Lin

callers of ids__new() function only do NULL checking for the return value. ids__new() calles hashmap__new(), which may return ERR_PTR(-ENOMEM). Instead of changing the checking one-by-one. return NULL instead of ERR_PTR(-ENOMEM) to keep consistent.

---
 tools/perf/util/expr.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
tools/perf/util/expr.c +7 -1
index 1d532b9fed29..aabdc112300c 100644
@@ -65,7 +65,13 @@ static bool key_equal(const void *key1, const void *key2,
 
 struct hashmap *ids__new(void)
 {
-	return hashmap__new(key_hash, key_equal, NULL);
+	struct hashmap *hash;
+
+	hash = hashmap__new(key_hash, key_equal, NULL);
+	if (IS_ERR(hash))
+		return NULL;
+	else
+		return hash;
 }
 
 void ids__free(struct hashmap *ids)