perf: fix perf.data endianness detection

The current version of perf detects whether or not the perf.data file is written in a different endianness using the attr_size field in the header of the file. This field represents sizeof(struct perf_event) as known to perf record. If the size does not match, then perf tries the byte-swapped version. If they match, then the tool assumes a different endianness.

The issue with the approach is that it assumes the size of perf_event_attr always has to match between the file and the tool. However, the kernel API is designed to make it possible to extend perf_event_attr for new features. Consequently, it is not possible to use attr_size to detect endianness.

This patch takes another approach by using the magic number written at the beginning of the perf.data file to detect endianness mismatch. The magic number is an eight-byte signature. The patch introduces a new value for this signature. The key difference is that the signature is written differently in the file depending on the endianness. Thus, by comparing the signature from the file with the tool's own signature it is possible to detect if the endianness matches. The new signature is "PERFILE2".

Backward compatiblity with existing perf.data file is ensured.

Signed-off-by Stephane Eranian <[email protected]>

---

tools/perf/util/header.c +58 -17

index b6c1ad1..d824502 100644
@@ -57,9 +57,20 @@ char *perf_header__find_event(u64 id)
 	return NULL;
 }
 
-static const char *__perf_magic = "PERFFILE";
-
-#define PERF_MAGIC	(*(u64 *)__perf_magic)
+/*
+ * magic2 = "PERFILE2"
+ * must be a numerical value to let the endianness
+ * determine the memory layout. That way we are able
+ * to detect endianness when reading the perf.data file
+ * back.
+ *
+ * we check for legacy (PERFFILE) format.
+ */
+static const char *__perf_magic1 = "PERFFILE";
+static const u64 __perf_magic2    = 0x32454c4946524550ULL;
+static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
+
+#define PERF_MAGIC	(__perf_magic2)
 
 struct perf_file_attr {
 	struct perf_event_attr	attr;

@@ -595,25 +606,57 @@ out_free:
 	return err;
 }
 
+static int check_magic_endian(u64 *magic, struct perf_file_header *header,
+			      struct perf_header *ph)
+{
+	int ret;
+
+	/* check for legacy format */
+	ret = memcmp(magic, __perf_magic1, sizeof(*magic));
+	if (ret == 0) {
+		pr_debug("legacy perf.data format\n");
+		if (!header)
+			return -1;
+
+		if (header->attr_size != sizeof(struct perf_file_attr)) {
+			u64 attr_size = bswap_64(header->attr_size);
+
+			if (attr_size != sizeof(struct perf_file_attr))
+				return -1;
+
+			mem_bswap_64(header, offsetof(struct perf_file_header,
+						adds_features));
+			ph->needs_swap = true;
+		}
+		return 0;
+	}
+
+	/* check for our magic (same endianness) */
+	if (*magic == __perf_magic2)
+		return 0;
+
+	/* check for our magic (opposite endianness) */
+	if (*magic != __perf_magic2_sw)
+		return -1;
+
+	ph->needs_swap = true;
+
+	return 0;
+}
+
 int perf_file_header__read(struct perf_file_header *header,
 			   struct perf_header *ph, int fd)
 {
+	int ret;
+
 	lseek(fd, 0, SEEK_SET);
 
-	if (readn(fd, header, sizeof(*header)) <= 0 ||
-	    memcmp(&header->magic, __perf_magic, sizeof(header->magic)))
+	ret = readn(fd, header, sizeof(*header));
+	if (ret <= 0)
 		return -1;
 
-	if (header->attr_size != sizeof(struct perf_file_attr)) {
-		u64 attr_size = bswap_64(header->attr_size);
-
-		if (attr_size != sizeof(struct perf_file_attr))
-			return -1;
-
-		mem_bswap_64(header, offsetof(struct perf_file_header,
-					    adds_features));
-		ph->needs_swap = true;
-	}
+	if (check_magic_endian(&header->magic, header, ph) < 0)
+		return -1;
 
 	if (header->size != sizeof(*header)) {
 		/* Support the previous format */

@@ -824,8 +867,13 @@ static int perf_file_header__read_pipe(struct perf_pipe_file_header *header,
 				       struct perf_header *ph, int fd,
 				       bool repipe)
 {
-	if (readn(fd, header, sizeof(*header)) <= 0 ||
-	    memcmp(&header->magic, __perf_magic, sizeof(header->magic)))

Quoted 135 lines · Stephane Eranian

The current version of perf detects whether or not the perf.data file is written in a different endianness using the attr_size field in the header of the file. This field represents sizeof(struct perf_event) as known to perf record. If the size does not match, then perf tries the byte-swapped version. If they match, then the tool assumes a different endianness.

The issue with the approach is that it assumes the size of perf_event_attr always has to match between the file and the tool. However, the kernel API is designed to make it possible to extend perf_event_attr for new features. Consequently, it is not possible to use attr_size to detect endianness.

This patch takes another approach by using the magic number written at the beginning of the perf.data file to detect endianness mismatch. The magic number is an eight-byte signature. The patch introduces a new value for this signature. The key difference is that the signature is written differently in the file depending on the endianness. Thus, by comparing the signature from the file with the tool's own signature it is possible to detect if the endianness matches. The new signature is "PERFILE2".

Backward compatiblity with existing perf.data file is ensured.

Signed-off-by Stephane Eranian <[email protected]>

---

tools/perf/util/header.c +58 -17

index b6c1ad1..d824502 100644
@@ -57,9 +57,20 @@ char *perf_header__find_event(u64 id)
       return NULL;
  }
 
-static const char *__perf_magic = "PERFFILE";
-
-#define PERF_MAGIC   (*(u64 *)__perf_magic)
+/*
+ * magic2 = "PERFILE2"
+ * must be a numerical value to let the endianness
+ * determine the memory layout. That way we are able
+ * to detect endianness when reading the perf.data file
+ * back.
+ *
+ * we check for legacy (PERFFILE) format.
+ */
+static const char *__perf_magic1 = "PERFFILE";
+static const u64 __perf_magic2    = 0x32454c4946524550ULL;
+static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
+
+#define PERF_MAGIC   (__perf_magic2)
 
  struct perf_file_attr {
       struct perf_event_attr  attr;

@@ -595,25 +606,57 @@ out_free:
       return err;
  }
 
+static int check_magic_endian(u64 *magic, struct perf_file_header *header,
+                           struct perf_header *ph)
+{
+     int ret;
+
+     /* check for legacy format */
+     ret = memcmp(magic, __perf_magic1, sizeof(*magic));
+     if (ret == 0) {
+             pr_debug("legacy perf.data format\n");
+             if (!header)
+                     return -1;
+
+             if (header->attr_size != sizeof(struct perf_file_attr)) {
+                     u64 attr_size = bswap_64(header->attr_size);
+
+                     if (attr_size != sizeof(struct perf_file_attr))
+                             return -1;
+
+                     mem_bswap_64(header, offsetof(struct perf_file_header,
+                                             adds_features));
+                     ph->needs_swap = true;
+             }
+             return 0;
+     }
+
+     /* check for our magic (same endianness) */
+     if (*magic == __perf_magic2)
+             return 0;
+
+     /* check for our magic (opposite endianness) */
+     if (*magic != __perf_magic2_sw)
+             return -1;
+
+     ph->needs_swap = true;
+
+     return 0;
+}
+
  int perf_file_header__read(struct perf_file_header *header,
                          struct perf_header *ph, int fd)
  {
+     int ret;
+
       lseek(fd, 0, SEEK_SET);
 
-     if (readn(fd, header, sizeof(*header)) <= 0 ||
-         memcmp(&header->magic, __perf_magic, sizeof(header->magic)))
+     ret = readn(fd, header, sizeof(*header));
+     if (ret <= 0)
               return -1;
 
-     if (header->attr_size != sizeof(struct perf_file_attr)) {
-             u64 attr_size = bswap_64(header->attr_size);
-
-             if (attr_size != sizeof(struct perf_file_attr))
-                     return -1;
-
-             mem_bswap_64(header, offsetof(struct perf_file_header,
-                                         adds_features));
-             ph->needs_swap = true;
-     }
+     if (check_magic_endian(&header->magic, header, ph) < 0)
+             return -1;
 
       if (header->size != sizeof(*header)) {
               /* Support the previous format */

@@ -824,8 +867,13 @@ static int perf_file_header__read_pipe(struct perf_pipe_file_header *header,
                                      struct perf_header *ph, int fd,
                                      bool repipe)
  {
-     if (readn(fd, header, sizeof(*header)) <= 0 ||
-         memcmp(&header->magic, __perf_magic, sizeof(header->magic)))

tools/perf/util/header.c static int perf_file_header__read_pipe(struct perf_pipe_file_header *header, · line 879

Quoted 10 lines

tools/perf/util/header.c +7 -0

+     int ret;
+
+     ret = readn(fd, header, sizeof(*header));
+     if (ret <= 0)
+             return -1;
+
+      if (check_magic_endian(&header->magic, NULL, ph) < 0)
               return -1;
 
       if (repipe && do_write(STDOUT_FILENO, header, sizeof(*header)) < 0)

tip + this patch

new PPC data file created and then the data file is analyzed on x86

$ perf script -i /tmp/perf-new-ppc.data
 incompatible file format

on x86:

$ od -c /tmp/perf-new-ppc.data | head -1
0000000   2   E   L   I   F   R   E   P  \0  \0  \0  \0  \0  \0  \0   h

So it is using the new magic.

Same file on ppc:

# od -c /tmp/perf-new-ppc.data | head -1
0000000   2   E   L   I   F   R   E   P  \0  \0  \0  \0  \0  \0  \0   h

Quoted 26 lines

Quoted 26 lines

tools/perf/util/header.c +26 -0

+static int check_magic_endian(u64 *magic, struct perf_file_header *header,
+                           struct perf_header *ph)
+{
+     int ret;
+
+     /* check for legacy format */
+     ret = memcmp(magic, __perf_magic1, sizeof(*magic));
+     if (ret == 0) {
+             pr_debug("legacy perf.data format\n");
+             if (!header)
+                     return -1;
+
+             if (header->attr_size != sizeof(struct perf_file_attr)) {
+                     u64 attr_size = bswap_64(header->attr_size);
+
+                     if (attr_size != sizeof(struct perf_file_attr))
+                             return -1;
+
+                     mem_bswap_64(header, offsetof(struct perf_file_header,
+                                             adds_features));
+                     ph->needs_swap = true;
+             }
+             return 0;
+     }
+
+     /* check for our magic (same endianness) */

Quoted 8 lines

tools/perf/util/header.c out_free: · line 642

Quoted 8 lines

tools/perf/util/header.c +8 -0

+     if (*magic == __perf_magic2)
+             return 0;
+
+     /* check for our magic (opposite endianness) */
+     if (*magic != __perf_magic2_sw)
+             return -1;
+
+     ph->needs_swap = true;

Quoted 17 lines

Quoted 17 lines

tools/perf/util/header.c +8 -2

+
+     return 0;
+}
+
 int perf_file_header__read(struct perf_file_header *header,
                         struct perf_header *ph, int fd)
 {
+     int ret;
+
      lseek(fd, 0, SEEK_SET);
 
-     if (readn(fd, header, sizeof(*header)) <= 0 ||
-         memcmp(&header->magic, __perf_magic, sizeof(header->magic)))
+     ret = readn(fd, header, sizeof(*header));
+     if (ret <= 0)
              return -1;

Quoted 8 lines

tools/perf/util/header.c out_free: · line 614

Quoted 8 lines

tools/perf/util/header.c +0 -8

-     if (header->attr_size != sizeof(struct perf_file_attr)) {
-             u64 attr_size = bswap_64(header->attr_size);
-
-             if (attr_size != sizeof(struct perf_file_attr))
-                     return -1;
-
-             mem_bswap_64(header, offsetof(struct perf_file_header,
-                                         adds_features));