dev-util/perf/files/4.14-Fix-hugepage-text.patch

diff --git a/tools/perf/util/event.c b/tools/perf/util/event.c
index fc690fecbfd66..1fe8728cf5349 100644
--- a/tools/perf/util/event.c
+++ b/tools/perf/util/event.c
@@ -316,6 +316,80 @@ static int perf_event__synthesize_fork(struct perf_tool *tool,
 	return 0;
 }
 
+/*
+ * We use transparent hugepage technique to map hot text section
+ * to impove the performance. After the mapping, the map of chrome
+ * looks like below
+ * 7f9ba2626000-7f9ba2800000 r-xp 00000000 fe:00 49961 /opt/google/chrome/chrome
+ * 7f9ba2800000-7f9ba4600000 r-xp 00000000 00:00 0
+ * 7f9ba4600000-7f9ba8ff9000 r-xp 01fda000 fe:00 49961 /opt/google/chrome/chrome
+ *
+ * Since the hugepage needs to be 2MB aligned, the first row is the small
+ * portion of chrome text section that is not 2MB aligned. The second row is
+ * the hot text section that mapped to the transparent hugepages. So there is
+ * no exename, offset on it. The third row is the remaining of the chrome text
+ * section. The goal of this function is to merge the three rows to make it a whole
+ * chrome binary.
+ *
+ * The idea is if whiling scanning the /proc/{pid}/maps, if
+ * we see a section that it has 'x' attribute, we look ahead two sections.
+ * For the second section, we check whether the size is mulitple of 2MB and the
+ * starting address is 2MB aligned and it does not have exename and it has
+ * 'x' attribute.
+ * For the third section, we check whether it has 'x' attribute and the execname is
+ * the same as the first section and the offset is the sum of the size of the first
+ * two sections.
+ *
+ * If all the conditons above are met, we can infer these three sections are actually
+ * from a single binary and we need to merge them.
+ */
+static int perf_event__merge_mmap_events_for_hugepage(char * second_bf,
+							char * third_bf,
+							union perf_event * event) {
+	char prot[5];
+	unsigned int ino;
+	ssize_t n;
+	char execname[PATH_MAX];
+	union perf_event second_event, third_event;
+	const int hugepage_size = 1<<21;
+
+	if (second_bf[0] == '\0' || third_bf[0] == '\0')
+		return 0;
+	/* 00400000-0040c000 r-xp 00000000 fd:01 41038  /bin/cat */
+	n = sscanf(second_bf, "%"PRIx64"-%"PRIx64" %4s %"PRIx64" %x:%x %u %s\n",
+			&second_event.mmap2.start, &second_event.mmap2.len, prot,
+			&second_event.mmap2.pgoff, &second_event.mmap2.maj,
+			&second_event.mmap2.min,
+			&ino, execname);
+
+	if (n > 7)
+		return 0;
+	if (prot[2] != 'x')
+		return 0;
+	if (second_event.mmap2.start % hugepage_size ||
+		second_event.mmap2.len % hugepage_size)
+		return 0;
+
+	strcpy(execname, "");
+	n = sscanf(third_bf, "%"PRIx64"-%"PRIx64" %4s %"PRIx64" %x:%x %u %s\n",
+			&third_event.mmap2.start, &third_event.mmap2.len, prot,
+			&third_event.mmap2.pgoff, &third_event.mmap2.maj,
+			&third_event.mmap2.min,
+			&ino, execname);
+	if (n < 7)
+		return 0;
+	if (prot[2] != 'x')
+		return 0;
+	if (strncmp(execname, event->mmap2.filename, PATH_MAX))
+		return 0;
+	if (third_event.mmap2.pgoff != event->mmap2.len +
+		second_event.mmap2.len - second_event.mmap2.start)
+		return 0;
+	event->mmap2.len += second_event.mmap2.len - second_event.mmap2.start
+				+ third_event.mmap2.len - third_event.mmap2.start;
+	return 1;
+}
+
 int perf_event__synthesize_mmap_events(struct perf_tool *tool,
 				       union perf_event *event,
 				       pid_t pid, pid_t tgid,
@@ -332,6 +406,9 @@ int perf_event__synthesize_mmap_events(struct perf_tool *tool,
 	int rc = 0;
 	const char *hugetlbfs_mnt = hugetlbfs__mountpoint();
 	int hugetlbfs_mnt_len = hugetlbfs_mnt ? strlen(hugetlbfs_mnt) : 0;
+	char second_bf[BUFSIZ];
+	char third_bf[BUFSIZ];
+	int merged = 0;
 
 	if (machine__is_default_guest(machine))
 		return 0;
@@ -348,6 +425,11 @@ int perf_event__synthesize_mmap_events(struct perf_tool *tool,
 		return -1;
 	}
 
+	if (fgets(second_bf, sizeof(second_bf), fp) == NULL)
+		second_bf[0] = '\0';
+	else if (fgets(third_bf, sizeof(third_bf), fp) == NULL)
+		third_bf[0] = '\0';
+
 	event->header.type = PERF_RECORD_MMAP2;
 	t = rdclock();
 
@@ -360,7 +442,22 @@ int perf_event__synthesize_mmap_events(struct perf_tool *tool,
 		size_t size;
 		ssize_t n;
 
-		if (fgets(bf, sizeof(bf), fp) == NULL)
+		if (merged) {
+			merged = 0;
+			if (fgets(bf, sizeof(bf), fp) == NULL)
+				bf[0] = '\0';
+			else if (fgets(second_bf, sizeof(second_bf), fp) == NULL)
+				second_bf[0] = '\0';
+			else if (fgets(third_bf, sizeof(third_bf), fp) == NULL)
+				third_bf[0] = '\0';
+		} else {
+			strncpy(bf, second_bf, sizeof(second_bf));
+			strncpy(second_bf, third_bf, sizeof(third_bf));
+			if (fgets(third_bf, sizeof(third_bf), fp) == NULL)
+				third_bf[0] = '\0';
+		}
+
+		if (bf[0] == '\0')
 			break;
 
 		if ((rdclock() - t) > timeout) {
@@ -444,6 +541,10 @@ int perf_event__synthesize_mmap_events(struct perf_tool *tool,
 		event->mmap2.pid = tgid;
 		event->mmap2.tid = pid;
 
+		if (strcmp(execname, anonstr) && prot[2] == 'x')
+			merged = perf_event__merge_mmap_events_for_hugepage(
+				second_bf, third_bf, event);
+
 		if (perf_tool__process_synth_event(tool, event, machine, process) != 0) {
 			rc = -1;
 			break;