[PATCH 1/4] libtracefs: Iterate over raw events in sorted order

["Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com>]

Changed the logic of tracefs_iterate_raw_events(): instead of iterating
in CPU order, walk through the events from all CPU buffers and read the
oldest first.
Fixed the problem with cpu number in the record, passed to the callback
- set the real CPU number instead of 0.

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
---
 src/tracefs-events.c | 246 +++++++++++++++++++++++++++----------------
 1 file changed, 156 insertions(+), 90 deletions(-)

diff --git a/src/tracefs-events.c b/src/tracefs-events.c
index 825f916..4095665 100644
--- a/src/tracefs-events.c
+++ b/src/tracefs-events.c
@@ -51,120 +51,130 @@ page_to_kbuf(struct tep_handle *tep, void *page, int size)
 	return kbuf;
 }
 
-static int read_kbuf_record(struct kbuffer *kbuf, struct tep_record *record)
+struct cpu_iterate {
+	struct tep_record record;
+	struct tep_event *event;
+	struct kbuffer *kbuf;
+	void *page;
+	int psize;
+	int rsize;
+	int cpu;
+	int fd;
+};
+
+static int read_kbuf_record(struct cpu_iterate *cpu)
 {
 	unsigned long long ts;
 	void *ptr;
 
-	ptr = kbuffer_read_event(kbuf, &ts);
-	if (!ptr || !record)
+	if (!cpu || !cpu->kbuf)
+		return -1;
+	ptr = kbuffer_read_event(cpu->kbuf, &ts);
+	if (!ptr)
 		return -1;
 
-	memset(record, 0, sizeof(*record));
-	record->ts = ts;
-	record->size = kbuffer_event_size(kbuf);
-	record->record_size = kbuffer_curr_size(kbuf);
-	record->cpu = 0;
-	record->data = ptr;
-	record->ref_count = 1;
+	memset(&(cpu->record), 0, sizeof(cpu->record));
+	cpu->record.ts = ts;
+	cpu->record.size = kbuffer_event_size(cpu->kbuf);
+	cpu->record.record_size = kbuffer_curr_size(cpu->kbuf);
+	cpu->record.cpu = cpu->cpu;
+	cpu->record.data = ptr;
+	cpu->record.ref_count = 1;
 
-	kbuffer_next_event(kbuf, NULL);
+	kbuffer_next_event(cpu->kbuf, NULL);
 
 	return 0;
 }
 
-static int
-get_events_in_page(struct tep_handle *tep, void *page,
-		   int size, int cpu,
-		   int (*callback)(struct tep_event *,
-				   struct tep_record *,
-				   int, void *),
-		   void *callback_context)
+int read_next_page(struct tep_handle *tep, struct cpu_iterate *cpu)
 {
-	struct tep_record record;
-	struct tep_event *event;
-	struct kbuffer *kbuf;
-	int id, cnt = 0;
+	cpu->rsize = read(cpu->fd, cpu->page, cpu->psize);
+	if (cpu->rsize <= 0)
+		return -1;
+
+	cpu->kbuf = page_to_kbuf(tep, cpu->page, cpu->rsize);
+	if (!cpu->kbuf)
+		return -1;
+
+	return 0;
+}
+
+int read_next_record(struct tep_handle *tep, struct cpu_iterate *cpu)
+{
+	int id;
+
+	do {
+		while (!read_kbuf_record(cpu)) {
+			id = tep_data_type(tep, &(cpu->record));
+			cpu->event = tep_find_event(tep, id);
+			if (cpu->event)
+				return 0;
+		}
+	} while (!read_next_page(tep, cpu));
+
+	return -1;
+}
+
+static int read_cpu_pages(struct tep_handle *tep, struct cpu_iterate *cpus, int count,
+			  int (*callback)(struct tep_event *,
+					  struct tep_record *,
+					  int, void *),
+			  void *callback_context)
+{
+	bool has_data = false;
 	int ret;
+	int i, j;
 
-	if (size <= 0)
-		return 0;
+	for (i = 0; i < count; i++) {
+		ret = read_next_record(tep, cpus + i);
+		if (!ret)
+			has_data = true;
+	}
 
-	kbuf = page_to_kbuf(tep, page, size);
-	if (!kbuf)
-		return 0;
-
-	ret = read_kbuf_record(kbuf, &record);
-	while (!ret) {
-		id = tep_data_type(tep, &record);
-		event = tep_find_event(tep, id);
-		if (event) {
-			cnt++;
+	while (has_data) {
+		j = count;
+		for (i = 0; i < count; i++) {
+			if (!cpus[i].event)
+				continue;
+			if (j == count || cpus[j].record.ts > cpus[i].record.ts)
+				j = i;
+		}
+		if (j < count) {
 			if (callback &&
-			    callback(event, &record, cpu, callback_context))
+			    callback(cpus[j].event, &(cpus[j].record), cpus[j].cpu, callback_context))
 				break;
+			cpus[j].event = NULL;
+			read_next_record(tep, cpus + j);
+		} else {
+			has_data = false;
 		}
-		ret = read_kbuf_record(kbuf, &record);
 	}
 
-	kbuffer_free(kbuf);
-
-	return cnt;
+	return 0;
 }
 
-/*
- * tracefs_iterate_raw_events - Iterate through events in trace_pipe_raw,
- *				per CPU trace buffers
- * @tep: a handle to the trace event parser context
- * @instance: ftrace instance, can be NULL for the top instance
- * @cpus: Iterate only through the buffers of CPUs, set in the mask.
- *	  If NULL, iterate through all CPUs.
- * @cpu_size: size of @cpus set
- * @callback: A user function, called for each record from the file
- * @callback_context: A custom context, passed to the user callback function
- *
- * If the @callback returns non-zero, the iteration stops - in that case all
- * records from the current page will be lost from future reads
- *
- * Returns -1 in case of an error, or 0 otherwise
- */
-int tracefs_iterate_raw_events(struct tep_handle *tep,
-				struct tracefs_instance *instance,
-				cpu_set_t *cpus, int cpu_size,
-				int (*callback)(struct tep_event *,
-						struct tep_record *,
-						int, void *),
-				void *callback_context)
+static int open_cpu_fies(struct tracefs_instance *instance, cpu_set_t *cpus,
+			 int cpu_size, struct cpu_iterate **all_cpus, int *count)
 {
+	struct cpu_iterate *tmp;
 	unsigned int p_size;
 	struct dirent *dent;
 	char file[PATH_MAX];
-	void *page = NULL;
 	struct stat st;
+	int ret = -1;
+	int fd = -1;
 	char *path;
 	DIR *dir;
-	int ret;
 	int cpu;
-	int fd;
-	int r;
+	int i = 0;
 
-	if (!tep || !callback)
-		return -1;
-
-	p_size = getpagesize();
 	path = tracefs_instance_get_file(instance, "per_cpu");
 	if (!path)
 		return -1;
 	dir = opendir(path);
-	if (!dir) {
-		ret = -1;
-		goto error;
-	}
-	page = malloc(p_size);
-	if (!page) {
-		ret = -1;
-		goto error;
-	}
+	if (!dir)
+		goto out;
+	p_size = getpagesize();
 	while ((dent = readdir(dir))) {
 		const char *name = dent->d_name;
 
@@ -174,32 +184,88 @@ int tracefs_iterate_raw_events(struct tep_handle *tep,
 		if (cpus && !CPU_ISSET_S(cpu, cpu_size, cpus))
 			continue;
 		sprintf(file, "%s/%s", path, name);
-		ret = stat(file, &st);
-		if (ret < 0 || !S_ISDIR(st.st_mode))
+		if (stat(file, &st) < 0 || !S_ISDIR(st.st_mode))
 			continue;
 
 		sprintf(file, "%s/%s/trace_pipe_raw", path, name);
 		fd = open(file, O_RDONLY | O_NONBLOCK);
 		if (fd < 0)
 			continue;
-		do {
-			r = read(fd, page, p_size);
-			if (r > 0)
-				get_events_in_page(tep, page, r, cpu,
-						   callback, callback_context);
-		} while (r > 0);
-		close(fd);
+		tmp = realloc(*all_cpus, (i + 1) * sizeof(struct cpu_iterate));
+		if (!tmp) {
+			close(fd);
+			goto out;
+		}
+		memset(tmp + i, 0, sizeof(struct cpu_iterate));
+		tmp[i].fd = fd;
+		tmp[i].cpu = cpu;
+		tmp[i].page =  malloc(p_size);
+		tmp[i].psize = p_size;
+		*all_cpus = tmp;
+		*count = i + 1;
+		if (!tmp[i++].page)
+			goto out;
 	}
+
 	ret = 0;
 
-error:
+out:
 	if (dir)
 		closedir(dir);
-	free(page);
 	tracefs_put_tracing_file(path);
 	return ret;
 }
 
+/*
+ * tracefs_iterate_raw_events - Iterate through events in trace_pipe_raw,
+ *				per CPU trace buffers
+ * @tep: a handle to the trace event parser context
+ * @instance: ftrace instance, can be NULL for the top instance
+ * @cpus: Iterate only through the buffers of CPUs, set in the mask.
+ *	  If NULL, iterate through all CPUs.
+ * @cpu_size: size of @cpus set
+ * @callback: A user function, called for each record from the file
+ * @callback_context: A custom context, passed to the user callback function
+ *
+ * If the @callback returns non-zero, the iteration stops - in that case all
+ * records from the current page will be lost from future reads
+ * The events are iterated in sorted order, oldest first.
+ *
+ * Returns -1 in case of an error, or 0 otherwise
+ */
+int tracefs_iterate_raw_events(struct tep_handle *tep,
+				struct tracefs_instance *instance,
+				cpu_set_t *cpus, int cpu_size,
+				int (*callback)(struct tep_event *,
+						struct tep_record *,
+						int, void *),
+				void *callback_context)
+{
+	struct cpu_iterate *all_cpus = NULL;
+	int count = 0;
+	int ret;
+	int i;
+
+	if (!tep || !callback)
+		return -1;
+
+	ret = open_cpu_fies(instance, cpus, cpu_size, &all_cpus, &count);
+	if (ret < 0)
+		goto out;
+	ret = read_cpu_pages(tep, all_cpus, count, callback, callback_context);
+
+out:
+	if (all_cpus) {
+		for (i = 0; i < count; i++) {
+			close(all_cpus[i].fd);
+			free(all_cpus[i].page);
+		}
+		free(all_cpus);
+	}
+
+	return ret;
+}
+
 static char **add_list_string(char **list, const char *name, int len)
 {
 	if (!list)
-- 
2.30.2