perf tools: Introduce an abstraction for AUX Area and Instruction Tracing

I noticed it tho because of this:

util/session.c: In function ‘skipn’:
util/session.c:279:28: error: comparison between signed and unsigned integer expressions [-Werror=sign-compare]

   ret = read(fd, buf, MIN(n, sizeof(buf)));
                            ^
util/session.c:279:47: error: signed and unsigned type in conditional
expression [-Werror=sign-compare]
   ret = read(fd, buf, MIN(n, sizeof(buf)));
                                               ^

cc1: all warnings being treated as errors make[3]: *** [/tmp/build/perf/util/session.o] Error 1 make[3]: *** Waiting for unfinished jobs....

Fixing it to be:

    ret = read(fd, buf, min(n, (off_t)sizeof(buf)));

Ok?

Em Thu, Apr 09, 2015 at 06:53:43PM +0300, Adrian Hunter escreveu:

tools/perf/util/session.c static int process_id_index_stub(struct perf_tool *tool __maybe_unused, · line 285

Quoted 13 lines

tools/perf/util/session.c +13 -0

+static int skipn(int fd, off_t n)
+{
+	char buf[4096];
+	ssize_t ret;
+
+	while (n > 0) {
+		ret = read(fd, buf, MIN(n, sizeof(buf)));
+		if (ret <= 0)
+			return ret;
+		n -= ret;
+	}
+
+	return 0;

We have min() in tools/perf/util/include/linux/kernel.h, and that is what is being used elsewhere in tools/perf/:

[acme@zoo linux]$ find tools/perf/ -name "*.[ch]" | xargs grep '\<min(' | wc -l
20
[acme@zoo linux]$ find tools/perf/ -name "*.[ch]" | xargs grep '\<max(' | wc -l
8
[acme@zoo linux]$

I noticed it tho because of this:

util/session.c: In function ‘skipn’:
util/session.c:279:28: error: comparison between signed and unsigned integer expressions [-Werror=sign-compare]

   ret = read(fd, buf, MIN(n, sizeof(buf)));
                            ^
util/session.c:279:47: error: signed and unsigned type in conditional
expression [-Werror=sign-compare]
   ret = read(fd, buf, MIN(n, sizeof(buf)));
                                               ^

cc1: all warnings being treated as errors make[3]: *** [/tmp/build/perf/util/session.o] Error 1 make[3]: *** Waiting for unfinished jobs....

Fixing it to be:

    ret = read(fd, buf, min(n, (off_t)sizeof(buf)));

Ok?

Add support for reading from the AUX area tracing mmap and synthesizing AUX area tracing events.

This patch introduces an abstraction for recording AUX area data. Recording is initialized
by auxtrace_record__init() which is a weak function to be implemented by the architecture to provide recording callbacks. Recording is mainly handled by auxtrace_mmap__read() and perf_event__synthesize_auxtrace() but there are callbacks for miscellaneous needs including validating and processing user options, populating private data in auxtrace_info_event, and freeing the structure when finished.

Signed-off-by Adrian Hunter <[email protected]>

Acked-by Jiri Olsa <[email protected]>

---
 tools/perf/perf.h          |   2 +
 tools/perf/util/auxtrace.c | 176 +++++++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/auxtrace.h |  56 ++++++++++++++-
 tools/perf/util/record.c   |  11 ++-
 4 files changed, 243 insertions(+), 2 deletions(-)

tools/perf/perf.h +2 -0

index e14bb63..5042093 100644
@@ -54,8 +54,10 @@ struct record_opts {
 	bool	     period;
 	bool	     sample_intr_regs;
 	bool	     running_time;
+	bool	     full_auxtrace;
 	unsigned int freq;
 	unsigned int mmap_pages;
+	unsigned int auxtrace_mmap_pages;
 	unsigned int user_freq;
 	u64          branch_stack;
 	u64	     default_interval;

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

index dedb646..2cafea2 100644
@@ -23,6 +23,10 @@
 #include <linux/bitops.h>
 #include <linux/log2.h>
 
+#include <stdlib.h>
+#include <string.h>
+#include <errno.h>
+
 #include "../perf.h"
 #include "util.h"
 #include "evlist.h"

@@ -31,6 +35,9 @@
 #include "asm/bug.h"
 #include "auxtrace.h"
 
+#include "event.h"
+#include "debug.h"
+
 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm,
 			struct auxtrace_mmap_params *mp,
 			void *userpg, int fd)

tools/perf/util/auxtrace.c +61 -0

+
+static int auxtrace_not_supported(void)
+{
+	pr_err("AUX area tracing is not supported on this architecture\n");
+	return -EINVAL;
+}
+
+int auxtrace_record__info_fill(struct auxtrace_record *itr,
+			       struct perf_session *session,
+			       struct auxtrace_info_event *auxtrace_info,
+			       size_t priv_size)
+{
+	if (itr)
+		return itr->info_fill(itr, session, auxtrace_info, priv_size);
+	return auxtrace_not_supported();
+}
+
+void auxtrace_record__free(struct auxtrace_record *itr)
+{
+	if (itr)
+		itr->free(itr);
+}
+
+int auxtrace_record__options(struct auxtrace_record *itr,
+			     struct perf_evlist *evlist,
+			     struct record_opts *opts)
+{
+	if (itr)
+		return itr->recording_options(itr, evlist, opts);
+	return 0;
+}
+
+u64 auxtrace_record__reference(struct auxtrace_record *itr)
+{
+	if (itr)
+		return itr->reference(itr);
+	return 0;
+}
+
+struct auxtrace_record *__weak
+auxtrace_record__init(struct perf_evlist *evlist __maybe_unused, int *err)
+{
+	*err = 0;
+	return NULL;
+}
+
+int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr,
+					 struct perf_tool *tool,
+					 struct perf_session *session,
+					 perf_event__handler_t process)
+{
+	union perf_event *ev;
+	size_t priv_size;
+	int err;
+
+	pr_debug2("Synthesizing auxtrace information\n");
+	priv_size = auxtrace_record__info_priv_size(itr);
+	ev = zalloc(sizeof(struct auxtrace_info_event) + priv_size);
+	if (!ev)
+		return -ENOMEM;
+

tools/perf/util/auxtrace.c +89 -0

+}
+
+int auxtrace_mmap__read(struct auxtrace_mmap *mm, struct auxtrace_record *itr,
+			struct perf_tool *tool, process_auxtrace_t fn)
+{
+	u64 head = auxtrace_mmap__read_head(mm);
+	u64 old = mm->prev, offset, ref;
+	unsigned char *data = mm->base;
+	size_t size, head_off, old_off, len1, len2, padding;
+	union perf_event ev;
+	void *data1, *data2;
+
+	if (old == head)
+		return 0;
+
+	pr_debug3("auxtrace idx %d old %#"PRIx64" head %#"PRIx64" diff %#"PRIx64"\n",
+		  mm->idx, old, head, head - old);
+
+	if (mm->mask) {
+		head_off = head & mm->mask;
+		old_off = old & mm->mask;
+	} else {
+		head_off = head % mm->len;
+		old_off = old % mm->len;
+	}
+
+	if (head_off > old_off)
+		size = head_off - old_off;
+	else
+		size = mm->len - (old_off - head_off);
+
+	ref = auxtrace_record__reference(itr);
+
+	if (head > old || size <= head || mm->mask) {
+		offset = head - size;
+	} else {
+		/*
+		 * When the buffer size is not a power of 2, 'head' wraps at the
+		 * highest multiple of the buffer size, so we have to subtract
+		 * the remainder here.
+		 */
+		u64 rem = (0ULL - mm->len) % mm->len;
+
+		offset = head - size - rem;
+	}
+
+	if (size > head_off) {
+		len1 = size - head_off;
+		data1 = &data[mm->len - len1];
+		len2 = head_off;
+		data2 = &data[0];
+	} else {
+		len1 = size;
+		data1 = &data[head_off - len1];
+		len2 = 0;
+		data2 = NULL;
+	}
+
+	/* padding must be written by fn() e.g. record__process_auxtrace() */
+	padding = size & 7;
+	if (padding)
+		padding = 8 - padding;
+
+	memset(&ev, 0, sizeof(ev));
+	ev.auxtrace.header.type = PERF_RECORD_AUXTRACE;
+	ev.auxtrace.header.size = sizeof(ev.auxtrace);
+	ev.auxtrace.size = size + padding;
+	ev.auxtrace.offset = offset;
+	ev.auxtrace.reference = ref;
+	ev.auxtrace.idx = mm->idx;
+	ev.auxtrace.tid = mm->tid;
+	ev.auxtrace.cpu = mm->cpu;
+
+	if (fn(tool, &ev, data1, len1, data2, len2))
+		return -1;
+
+	mm->prev = head;
+
+	auxtrace_mmap__write_tail(mm, head);
+	if (itr->read_finish) {
+		int err;
+
+		err = itr->read_finish(itr, mm->idx);
+		if (err < 0)
+			return err;
+	}
+
+	return 1;
+}

tools/perf/util/auxtrace.h +55 -1

index 2071b36..7ab4850 100644
@@ -18,13 +18,18 @@
 
 #include <sys/types.h>
 #include <stdbool.h>
-
+#include <stddef.h>
 #include <linux/perf_event.h>
 #include <linux/types.h>
 
 #include "../perf.h"
 
+union perf_event;
+struct perf_session;
 struct perf_evlist;
+struct perf_tool;
+struct record_opts;
+struct auxtrace_info_event;
 
 /**
  * struct auxtrace_mmap - records an mmap of the auxtrace buffer.

@@ -70,6 +75,29 @@ struct auxtrace_mmap_params {
 	int		cpu;
 };
 
+/**
+ * struct auxtrace_record - callbacks for recording AUX area data.
+ * @recording_options: validate and process recording options
+ * @info_priv_size: return the size of the private data in auxtrace_info_event
+ * @info_fill: fill-in the private data in auxtrace_info_event
+ * @free: free this auxtrace record structure
+ * @reference: provide a 64-bit reference number for auxtrace_event
+ * @read_finish: called after reading from an auxtrace mmap
+ */
+struct auxtrace_record {
+	int (*recording_options)(struct auxtrace_record *itr,
+				 struct perf_evlist *evlist,
+				 struct record_opts *opts);
+	size_t (*info_priv_size)(struct auxtrace_record *itr);
+	int (*info_fill)(struct auxtrace_record *itr,
+			 struct perf_session *session,
+			 struct auxtrace_info_event *auxtrace_info,
+			 size_t priv_size);
+	void (*free)(struct auxtrace_record *itr);
+	u64 (*reference)(struct auxtrace_record *itr);
+	int (*read_finish)(struct auxtrace_record *itr, int idx);
+};
+
 static inline u64 auxtrace_mmap__read_head(struct auxtrace_mmap *mm)
 {
 	struct perf_event_mmap_page *pc = mm->userpg;

@@ -114,4 +142,30 @@ void auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp,
 				   struct perf_evlist *evlist, int idx,
 				   bool per_cpu);
 
+typedef int (*process_auxtrace_t)(struct perf_tool *tool,
+				  union perf_event *event, void *data1,
+				  size_t len1, void *data2, size_t len2);
+
+int auxtrace_mmap__read(struct auxtrace_mmap *mm, struct auxtrace_record *itr,
+			struct perf_tool *tool, process_auxtrace_t fn);
+
+struct auxtrace_record *auxtrace_record__init(struct perf_evlist *evlist,
+					      int *err);
+
+int auxtrace_record__options(struct auxtrace_record *itr,
+			     struct perf_evlist *evlist,
+			     struct record_opts *opts);
+size_t auxtrace_record__info_priv_size(struct auxtrace_record *itr);
+int auxtrace_record__info_fill(struct auxtrace_record *itr,
+			       struct perf_session *session,
+			       struct auxtrace_info_event *auxtrace_info,
+			       size_t priv_size);
+void auxtrace_record__free(struct auxtrace_record *itr);
+u64 auxtrace_record__reference(struct auxtrace_record *itr);
+
+int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr,
+					 struct perf_tool *tool,
+					 struct perf_session *session,
+					 perf_event__handler_t process);
+
 #endif

tools/perf/util/record.c +10 -1

index 8acd0df..0ccfa49 100644
@@ -119,7 +119,16 @@ void perf_evlist__config(struct perf_evlist *evlist, struct record_opts *opts)
 			evsel->attr.comm_exec = 1;
 	}
 
-	if (evlist->nr_entries > 1) {
+	if (opts->full_auxtrace) {
+		/*
+		 * Need to be able to synthesize and parse selected events with
+		 * arbitrary sample types, which requires always being able to
+		 * match the id.
+		 */
+		use_sample_identifier = perf_can_sample_identifier();
+		evlist__for_each(evlist, evsel)
+			perf_evsel__set_sample_id(evsel, use_sample_identifier);
+	} else if (evlist->nr_entries > 1) {
 		struct perf_evsel *first = perf_evlist__first(evlist);
 
 		evlist__for_each(evlist, evsel) {

Signature · 1 line

1.9.1

Hook into session processing so that AUX area decoding can synthesize events transparently to the tools.

The advantages of transparent decoding are that tools can be used directly with perf.data files containing AUX area tracing data, which is easier for the user and more efficient than having a separate decoding tool.

This will work as follows:
1. Tools will feed auxtrace events to the decoder using perf_tool->auxtrace() (support for that still to come).
2. The decoder can process side-band events as needed due to the auxtrace->process_event() hook.
3. The decoder can deliver synthesized events into the event stream using perf_session__deliver_synth_event().

Note the expectation is that decoding will work on data that is time-ordered with respect to the per-cpu or per-thread contexts that were recorded.

Signed-off-by Adrian Hunter <[email protected]>

---
 tools/perf/util/auxtrace.h | 55 ++++++++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/session.c  | 54 ++++++++++++++++++++++++++++++++++++---------
 tools/perf/util/session.h  |  3 +++
 3 files changed, 102 insertions(+), 10 deletions(-)

tools/perf/util/auxtrace.h +55 -0

index 7ab4850..199fc27 100644
@@ -23,6 +23,7 @@
 #include <linux/types.h>
 
 #include "../perf.h"
+#include "session.h"
 
 union perf_event;
 struct perf_session;

@@ -32,6 +33,24 @@ struct record_opts;
 struct auxtrace_info_event;
 
 /**
+ * struct auxtrace - session callbacks to allow AUX area data decoding.
+ * @process_event: lets the decoder see all session events
+ * @flush_events: process any remaining data
+ * @free_events: free resources associated with event processing
+ * @free: free resources associated with the session
+ */
+struct auxtrace {
+	int (*process_event)(struct perf_session *session,
+			     union perf_event *event,
+			     struct perf_sample *sample,
+			     struct perf_tool *tool);
+	int (*flush_events)(struct perf_session *session,
+			    struct perf_tool *tool);
+	void (*free_events)(struct perf_session *session);
+	void (*free)(struct perf_session *session);
+};
+
+/**
  * struct auxtrace_mmap - records an mmap of the auxtrace buffer.
  * @base: address of mapped area
  * @userpg: pointer to buffer's perf_event_mmap_page

@@ -168,4 +187,40 @@ int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr,
 					 struct perf_session *session,
 					 perf_event__handler_t process);
 
+static inline int auxtrace__process_event(struct perf_session *session,
+					  union perf_event *event,
+					  struct perf_sample *sample,
+					  struct perf_tool *tool)
+{
+	if (!session->auxtrace)
+		return 0;
+
+	return session->auxtrace->process_event(session, event, sample, tool);
+}
+
+static inline int auxtrace__flush_events(struct perf_session *session,
+					 struct perf_tool *tool)
+{
+	if (!session->auxtrace)
+		return 0;
+
+	return session->auxtrace->flush_events(session, tool);
+}
+
+static inline void auxtrace__free_events(struct perf_session *session)
+{
+	if (!session->auxtrace)
+		return;
+
+	return session->auxtrace->free_events(session);
+}
+
+static inline void auxtrace__free(struct perf_session *session)
+{
+	if (!session->auxtrace)
+		return;
+
+	return session->auxtrace->free(session);
+}
+
 #endif

tools/perf/util/session.c +44 -10

index 18e4034..cf9d678 100644
@@ -15,12 +15,13 @@
 #include "cpumap.h"
 #include "perf_regs.h"
 #include "asm/bug.h"
+#include "auxtrace.h"
 
-static int machines__deliver_event(struct machines *machines,
-				   struct perf_evlist *evlist,
-				   union perf_event *event,
-				   struct perf_sample *sample,
-				   struct perf_tool *tool, u64 file_offset);
+static int perf_session__deliver_event(struct perf_session *session,
+				       union perf_event *event,
+				       struct perf_sample *sample,
+				       struct perf_tool *tool,
+				       u64 file_offset);
 
 static int perf_session__open(struct perf_session *session)
 {

@@ -105,8 +106,8 @@ static int ordered_events__deliver_event(struct ordered_events *oe,
 		return ret;
 	}
 
-	return machines__deliver_event(&session->machines, session->evlist, event->event,
-				       &sample, session->tool, event->file_offset);
+	return perf_session__deliver_event(session, event->event, &sample,
+					   session->tool, event->file_offset);
 }
 
 struct perf_session *perf_session__new(struct perf_data_file *file,

@@ -185,6 +186,7 @@ static void perf_session_env__delete(struct perf_session_env *env)
 
 void perf_session__delete(struct perf_session *session)
 {
+	auxtrace__free(session);
 	perf_session__destroy_kernel_maps(session);
 	perf_session__delete_threads(session);
 	perf_session_env__delete(&session->header.env);

@@ -1030,6 +1032,24 @@ static int machines__deliver_event(struct machines *machines,
 	}
 }
 
+static int perf_session__deliver_event(struct perf_session *session,
+				       union perf_event *event,
+				       struct perf_sample *sample,
+				       struct perf_tool *tool,
+				       u64 file_offset)
+{
+	int ret;
+
+	ret = auxtrace__process_event(session, event, sample, tool);
+	if (ret < 0)
+		return ret;
+	if (ret > 0)
+		return 0;
+
+	return machines__deliver_event(&session->machines, session->evlist,
+				       event, sample, tool, file_offset);
+}
+
 static s64 perf_session__process_user_event(struct perf_session *session,
 					    union perf_event *event,
 					    u64 file_offset)

@@ -1190,8 +1210,8 @@ static s64 perf_session__process_event(struct perf_session *session,
 			return ret;
 	}
 
-	return machines__deliver_event(&session->machines, evlist, event,
-				       &sample, tool, file_offset);
+	return perf_session__deliver_event(session, event, &sample, tool,
+					   file_offset);
 }
 
 void perf_event_header__bswap(struct perf_event_header *hdr)

@@ -1350,10 +1370,14 @@ more:
 done:
 	/* do the final flush for ordered samples */
 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
+	if (err)
+		goto out_err;
+	err = auxtrace__flush_events(session, tool);
 out_err:
 	free(buf);
 	perf_session__warn_about_errors(session);
 	ordered_events__free(&session->ordered_events);
+	auxtrace__free_events(session);
 	return err;
 }
 

@@ -1496,10 +1520,14 @@ more:
 out:
 	/* do the final flush for ordered samples */
 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
+	if (err)
+		goto out_err;
+	err = auxtrace__flush_events(session, tool);
 out_err:
 	ui_progress__finish();
 	perf_session__warn_about_errors(session);
 	ordered_events__free(&session->ordered_events);
+	auxtrace__free_events(session);
 	session->one_mmap = false;
 	return err;
 }

@@ -1582,7 +1610,13 @@ size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp
 
 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
 {
-	size_t ret = fprintf(fp, "Aggregated stats:\n");
+	size_t ret;
+	const char *msg = "";
+
+	if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
+		msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
+
+	ret = fprintf(fp, "Aggregated stats:%s\n", msg);
 
 	ret += events_stats__fprintf(&session->evlist->stats, fp);
 	return ret;

tools/perf/util/session.h +3 -0

index d5fa7b791..8a69d3b 100644
@@ -15,10 +15,13 @@
 struct ip_callchain;
 struct thread;
 
+struct auxtrace;
+
 struct perf_session {
 	struct perf_header	header;
 	struct machines		machines;
 	struct perf_evlist	*evlist;
+	struct auxtrace		*auxtrace;
 	struct trace_event	tevent;
 	bool			repipe;
 	bool			one_mmap;

Signature · 1 line

1.9.1

Em Thu, Apr 09, 2015 at 06:53:48PM +0300, Adrian Hunter escreveu:

Quoted 8 lines

Hook into session processing so that AUX area decoding can synthesize events transparently to the tools.

The advantages of transparent decoding are that tools can be used directly with perf.data files containing AUX area tracing data, which is easier for the user and more efficient than having a separate decoding tool.

Quoted 7 lines

This will work as follows:
1. Tools will feed auxtrace events to the decoder using perf_tool->auxtrace() (support for that still to come).
2. The decoder can process side-band events as needed due to the auxtrace->process_event() hook.
3. The decoder can deliver synthesized events into the event stream using perf_session__deliver_synth_event().

So, here this clashes with work I'm doing to untie ordered_events from perf_session, undoes it even, trying to decipher if this is really needed...

- Arnaldo

Quoted 232 lines

Note the expectation is that decoding will work on data that is time-ordered with respect to the per-cpu or per-thread contexts that were recorded.

Signed-off-by Adrian Hunter <[email protected]>

---
 tools/perf/util/auxtrace.h | 55 ++++++++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/session.c  | 54 ++++++++++++++++++++++++++++++++++++---------
 tools/perf/util/session.h  |  3 +++
 3 files changed, 102 insertions(+), 10 deletions(-)

tools/perf/util/auxtrace.h +55 -0

index 7ab4850..199fc27 100644
@@ -23,6 +23,7 @@
 #include <linux/types.h>
 
 #include "../perf.h"
+#include "session.h"
 
 union perf_event;
 struct perf_session;

@@ -32,6 +33,24 @@ struct record_opts;
 struct auxtrace_info_event;
 
 /**
+ * struct auxtrace - session callbacks to allow AUX area data decoding.
+ * @process_event: lets the decoder see all session events
+ * @flush_events: process any remaining data
+ * @free_events: free resources associated with event processing
+ * @free: free resources associated with the session
+ */
+struct auxtrace {
+	int (*process_event)(struct perf_session *session,
+			     union perf_event *event,
+			     struct perf_sample *sample,
+			     struct perf_tool *tool);
+	int (*flush_events)(struct perf_session *session,
+			    struct perf_tool *tool);
+	void (*free_events)(struct perf_session *session);
+	void (*free)(struct perf_session *session);
+};
+
+/**
  * struct auxtrace_mmap - records an mmap of the auxtrace buffer.
  * @base: address of mapped area
  * @userpg: pointer to buffer's perf_event_mmap_page

@@ -168,4 +187,40 @@ int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr,
 					 struct perf_session *session,
 					 perf_event__handler_t process);
 
+static inline int auxtrace__process_event(struct perf_session *session,
+					  union perf_event *event,
+					  struct perf_sample *sample,
+					  struct perf_tool *tool)
+{
+	if (!session->auxtrace)
+		return 0;
+
+	return session->auxtrace->process_event(session, event, sample, tool);
+}
+
+static inline int auxtrace__flush_events(struct perf_session *session,
+					 struct perf_tool *tool)
+{
+	if (!session->auxtrace)
+		return 0;
+
+	return session->auxtrace->flush_events(session, tool);
+}
+
+static inline void auxtrace__free_events(struct perf_session *session)
+{
+	if (!session->auxtrace)
+		return;
+
+	return session->auxtrace->free_events(session);
+}
+
+static inline void auxtrace__free(struct perf_session *session)
+{
+	if (!session->auxtrace)
+		return;
+
+	return session->auxtrace->free(session);
+}
+
 #endif

tools/perf/util/session.c +44 -10

index 18e4034..cf9d678 100644
@@ -15,12 +15,13 @@
 #include "cpumap.h"
 #include "perf_regs.h"
 #include "asm/bug.h"
+#include "auxtrace.h"
 
-static int machines__deliver_event(struct machines *machines,
-				   struct perf_evlist *evlist,
-				   union perf_event *event,
-				   struct perf_sample *sample,
-				   struct perf_tool *tool, u64 file_offset);
+static int perf_session__deliver_event(struct perf_session *session,
+				       union perf_event *event,
+				       struct perf_sample *sample,
+				       struct perf_tool *tool,
+				       u64 file_offset);
 
 static int perf_session__open(struct perf_session *session)
 {

@@ -105,8 +106,8 @@ static int ordered_events__deliver_event(struct ordered_events *oe,
 		return ret;
 	}
 
-	return machines__deliver_event(&session->machines, session->evlist, event->event,
-				       &sample, session->tool, event->file_offset);
+	return perf_session__deliver_event(session, event->event, &sample,
+					   session->tool, event->file_offset);
 }
 
 struct perf_session *perf_session__new(struct perf_data_file *file,

@@ -185,6 +186,7 @@ static void perf_session_env__delete(struct perf_session_env *env)
 
 void perf_session__delete(struct perf_session *session)
 {
+	auxtrace__free(session);
 	perf_session__destroy_kernel_maps(session);
 	perf_session__delete_threads(session);
 	perf_session_env__delete(&session->header.env);

@@ -1030,6 +1032,24 @@ static int machines__deliver_event(struct machines *machines,
 	}
 }
 
+static int perf_session__deliver_event(struct perf_session *session,
+				       union perf_event *event,
+				       struct perf_sample *sample,
+				       struct perf_tool *tool,
+				       u64 file_offset)
+{
+	int ret;
+
+	ret = auxtrace__process_event(session, event, sample, tool);
+	if (ret < 0)
+		return ret;
+	if (ret > 0)
+		return 0;
+
+	return machines__deliver_event(&session->machines, session->evlist,
+				       event, sample, tool, file_offset);
+}
+
 static s64 perf_session__process_user_event(struct perf_session *session,
 					    union perf_event *event,
 					    u64 file_offset)

@@ -1190,8 +1210,8 @@ static s64 perf_session__process_event(struct perf_session *session,
 			return ret;
 	}
 
-	return machines__deliver_event(&session->machines, evlist, event,
-				       &sample, tool, file_offset);
+	return perf_session__deliver_event(session, event, &sample, tool,
+					   file_offset);
 }
 
 void perf_event_header__bswap(struct perf_event_header *hdr)

@@ -1350,10 +1370,14 @@ more:
 done:
 	/* do the final flush for ordered samples */
 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
+	if (err)
+		goto out_err;
+	err = auxtrace__flush_events(session, tool);
 out_err:
 	free(buf);
 	perf_session__warn_about_errors(session);
 	ordered_events__free(&session->ordered_events);
+	auxtrace__free_events(session);
 	return err;
 }
 

@@ -1496,10 +1520,14 @@ more:
 out:
 	/* do the final flush for ordered samples */
 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
+	if (err)
+		goto out_err;
+	err = auxtrace__flush_events(session, tool);
 out_err:
 	ui_progress__finish();
 	perf_session__warn_about_errors(session);
 	ordered_events__free(&session->ordered_events);
+	auxtrace__free_events(session);
 	session->one_mmap = false;
 	return err;
 }

@@ -1582,7 +1610,13 @@ size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp
 
 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
 {
-	size_t ret = fprintf(fp, "Aggregated stats:\n");
+	size_t ret;
+	const char *msg = "";
+
+	if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
+		msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
+
+	ret = fprintf(fp, "Aggregated stats:%s\n", msg);
 
 	ret += events_stats__fprintf(&session->evlist->stats, fp);
 	return ret;

tools/perf/util/session.h +3 -0

index d5fa7b791..8a69d3b 100644
@@ -15,10 +15,13 @@
 struct ip_callchain;
 struct thread;
 
+struct auxtrace;
+
 struct perf_session {
 	struct perf_header	header;
 	struct machines		machines;
 	struct perf_evlist	*evlist;
+	struct auxtrace		*auxtrace;
 	struct trace_event	tevent;
 	bool			repipe;
 	bool			one_mmap;

Signature · 1 line

1.9.1

It is assumed that AUX area decoding will synthesize events for consumption by other tools. At this time, the main use of AUX area tracing will be to capture instruction trace (aka processor trace) data. The nature of instruction tracing suggests the initial inclusion of options for "instructions" and "branches" events, but more could be added as needed.

+++ b/tools/perf/util/auxtrace.c

tools/perf/util/auxtrace.c +29 -0

@@ -37,6 +37,7 @@
 
 #include "event.h"
 #include "debug.h"
+#include "parse-options.h"
 
 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm,
 			struct auxtrace_mmap_params *mp,

@@ -200,6 +201,136 @@ out_free:
 	return err;
 }
 
+#define PERF_ITRACE_DEFAULT_PERIOD_TYPE		PERF_ITRACE_PERIOD_NANOSECS
+#define PERF_ITRACE_DEFAULT_PERIOD		100000
+#define PERF_ITRACE_DEFAULT_CALLCHAIN_SZ	16
+#define PERF_ITRACE_MAX_CALLCHAIN_SZ		1024
+
+void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts)
+{
+	synth_opts->instructions = true;
+	synth_opts->branches = true;
+	synth_opts->errors = true;
+	synth_opts->period_type = PERF_ITRACE_DEFAULT_PERIOD_TYPE;
+	synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
+	synth_opts->callchain_sz = PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
+}
+
+int itrace_parse_synth_opts(const struct option *opt, const char *str,
+			    int unset)
+{
+	struct itrace_synth_opts *synth_opts = opt->value;
+	const char *p;
+	char *endptr;
+
+	synth_opts->set = true;
+
+	if (unset) {
+		synth_opts->dont_decode = true;
+		return 0;
+	}

tools/perf/util/auxtrace.c +92 -0

+			while (*p == ' ' || *p == ',')
+				p += 1;
+			if (isdigit(*p)) {
+				synth_opts->period = strtoull(p, &endptr, 10);
+				p = endptr;
+				while (*p == ' ' || *p == ',')
+					p += 1;
+				switch (*p++) {
+				case 'i':
+					synth_opts->period_type =
+						PERF_ITRACE_PERIOD_INSTRUCTIONS;
+					break;
+				case 't':
+					synth_opts->period_type =
+						PERF_ITRACE_PERIOD_TICKS;
+					break;
+				case 'm':
+					synth_opts->period *= 1000;
+					/* Fall through */
+				case 'u':
+					synth_opts->period *= 1000;
+					/* Fall through */
+				case 'n':
+					if (*p++ != 's')
+						goto out_err;
+					synth_opts->period_type =
+						PERF_ITRACE_PERIOD_NANOSECS;
+					break;
+				case '\0':
+					goto out;
+				default:
+					goto out_err;
+				}
+			}
+			break;
+		case 'b':
+			synth_opts->branches = true;
+			break;
+		case 'e':
+			synth_opts->errors = true;
+			break;
+		case 'd':
+			synth_opts->log = true;
+			break;
+		case 'c':
+			synth_opts->branches = true;
+			synth_opts->calls = true;
+			break;
+		case 'r':
+			synth_opts->branches = true;
+			synth_opts->returns = true;
+			break;
+		case 'g':
+			synth_opts->instructions = true;
+			synth_opts->callchain = true;
+			synth_opts->callchain_sz =
+					PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
+			while (*p == ' ' || *p == ',')
+				p += 1;
+			if (isdigit(*p)) {
+				unsigned int val;
+
+				val = strtoul(p, &endptr, 10);
+				p = endptr;
+				if (!val || val > PERF_ITRACE_MAX_CALLCHAIN_SZ)
+					goto out_err;
+				synth_opts->callchain_sz = val;
+			}
+			break;
+		case ' ':
+		case ',':
+			break;
+		default:
+			goto out_err;
+		}
+	}
+out:
+	if (synth_opts->instructions) {
+		if (!synth_opts->period_type)
+			synth_opts->period_type =
+					PERF_ITRACE_DEFAULT_PERIOD_TYPE;
+		if (!synth_opts->period)
+			synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
+	}
+
+	return 0;
+
+out_err:
+	pr_err("Bad Instruction Tracing options '%s'\n", str);
+	return -EINVAL;
+}
+
 int auxtrace_mmap__read(struct auxtrace_mmap *mm, struct auxtrace_record *itr,
 			struct perf_tool *tool, process_auxtrace_t fn)
 {

tools/perf/util/auxtrace.h +43 -0

index 199fc27..6355315 100644
@@ -29,9 +29,49 @@ union perf_event;
 struct perf_session;
 struct perf_evlist;
 struct perf_tool;
+struct option;
 struct record_opts;
 struct auxtrace_info_event;
 
+enum itrace_period_type {
+	PERF_ITRACE_PERIOD_INSTRUCTIONS,
+	PERF_ITRACE_PERIOD_TICKS,
+	PERF_ITRACE_PERIOD_NANOSECS,
+};
+
+/**
+ * struct itrace_synth_opts - AUX area tracing synthesis options.
+ * @set: indicates whether or not options have been set
+ * @inject: indicates the event (not just the sample) must be fully synthesized
+ *          because 'perf inject' will write it out
+ * @instructions: whether to synthesize 'instructions' events
+ * @branches: whether to synthesize 'branches' events
+ * @errors: whether to synthesize decoder error events
+ * @dont_decode: whether to skip decoding entirely
+ * @log: write a decoding log
+ * @calls: limit branch samples to calls (can be combined with @returns)
+ * @returns: limit branch samples to returns (can be combined with @calls)
+ * @callchain: add callchain to 'instructions' events
+ * @callchain_sz: maximum callchain size
+ * @period: 'instructions' events period
+ * @period_type: 'instructions' events period type
+ */
+struct itrace_synth_opts {
+	bool			set;
+	bool			inject;
+	bool			instructions;
+	bool			branches;
+	bool			errors;
+	bool			dont_decode;
+	bool			log;
+	bool			calls;
+	bool			returns;
+	bool			callchain;
+	unsigned int		callchain_sz;
+	unsigned long long	period;
+	enum itrace_period_type	period_type;
+};
+
 /**
  * struct auxtrace - session callbacks to allow AUX area data decoding.
  * @process_event: lets the decoder see all session events

@@ -186,6 +226,9 @@ int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr,
 					 struct perf_tool *tool,
 					 struct perf_session *session,
 					 perf_event__handler_t process);
+int itrace_parse_synth_opts(const struct option *opt, const char *str,
+			    int unset);
+void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts);
 
 static inline int auxtrace__process_event(struct perf_session *session,
 					  union perf_event *event,

tools/perf/util/session.h +2 -0

index 8a69d3b..9ed5135 100644
@@ -16,12 +16,14 @@ struct ip_callchain;
 struct thread;
 
 struct auxtrace;
+struct itrace_synth_opts;
 
 struct perf_session {
 	struct perf_header	header;
 	struct machines		machines;
 	struct perf_evlist	*evlist;
 	struct auxtrace		*auxtrace;
+	struct itrace_synth_opts *itrace_synth_opts;
 	struct trace_event	tevent;
 	bool			repipe;
 	bool			one_mmap;

Signature · 1 line

1.9.1

Em Thu, Apr 09, 2015 at 06:53:49PM +0300, Adrian Hunter escreveu:

Quoted 53 lines

It is assumed that AUX area decoding will synthesize events for consumption by other tools. At this time, the main use of AUX area tracing will be to capture instruction trace (aka processor trace) data. The nature of instruction tracing suggests the initial inclusion of options for "instructions" and "branches" events, but more could be added as needed.

+++ b/tools/perf/util/auxtrace.c

tools/perf/util/auxtrace.c +34 -0

@@ -37,6 +37,7 @@
 
 #include "event.h"
 #include "debug.h"
+#include "parse-options.h"
 
 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm,
 			struct auxtrace_mmap_params *mp,

@@ -200,6 +201,136 @@ out_free:
 	return err;
 }
 
+#define PERF_ITRACE_DEFAULT_PERIOD_TYPE		PERF_ITRACE_PERIOD_NANOSECS
+#define PERF_ITRACE_DEFAULT_PERIOD		100000
+#define PERF_ITRACE_DEFAULT_CALLCHAIN_SZ	16
+#define PERF_ITRACE_MAX_CALLCHAIN_SZ		1024
+
+void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts)
+{
+	synth_opts->instructions = true;
+	synth_opts->branches = true;
+	synth_opts->errors = true;
+	synth_opts->period_type = PERF_ITRACE_DEFAULT_PERIOD_TYPE;
+	synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
+	synth_opts->callchain_sz = PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
+}
+
+int itrace_parse_synth_opts(const struct option *opt, const char *str,
+			    int unset)
+{
+	struct itrace_synth_opts *synth_opts = opt->value;
+	const char *p;
+	char *endptr;
+
+	synth_opts->set = true;
+
+	if (unset) {
+		synth_opts->dont_decode = true;
+		return 0;
+	}
+
+	if (!str) {
+		itrace_synth_opts__set_default(synth_opts);
+		return 0;
+	}

tools/perf/util/auxtrace.c out_free: · line 241

Quoted 5 lines

tools/perf/util/auxtrace.c +5 -0

+
+	for (p = str; *p;) {
+		switch (*p++) {
+		case 'i':
+			synth_opts->instructions = true;

Is this documented anywhere? I.e. examples for setting up this synth ops?

Quoted 55 lines

Em Thu, Apr 09, 2015 at 06:53:49PM +0300, Adrian Hunter escreveu:

Quoted 53 lines

It is assumed that AUX area decoding will synthesize events for consumption by other tools. At this time, the main use of AUX area tracing will be to capture instruction trace (aka processor trace) data. The nature of instruction tracing suggests the initial inclusion of options for "instructions" and "branches" events, but more could be added as needed.

+++ b/tools/perf/util/auxtrace.c

tools/perf/util/auxtrace.c +34 -0

@@ -37,6 +37,7 @@
 
 #include "event.h"
 #include "debug.h"
+#include "parse-options.h"
 
 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm,
 			struct auxtrace_mmap_params *mp,

@@ -200,6 +201,136 @@ out_free:
 	return err;
 }
 
+#define PERF_ITRACE_DEFAULT_PERIOD_TYPE		PERF_ITRACE_PERIOD_NANOSECS
+#define PERF_ITRACE_DEFAULT_PERIOD		100000
+#define PERF_ITRACE_DEFAULT_CALLCHAIN_SZ	16
+#define PERF_ITRACE_MAX_CALLCHAIN_SZ		1024
+
+void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts)
+{
+	synth_opts->instructions = true;
+	synth_opts->branches = true;
+	synth_opts->errors = true;
+	synth_opts->period_type = PERF_ITRACE_DEFAULT_PERIOD_TYPE;
+	synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
+	synth_opts->callchain_sz = PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
+}
+
+int itrace_parse_synth_opts(const struct option *opt, const char *str,
+			    int unset)
+{
+	struct itrace_synth_opts *synth_opts = opt->value;
+	const char *p;
+	char *endptr;
+
+	synth_opts->set = true;
+
+	if (unset) {
+		synth_opts->dont_decode = true;
+		return 0;
+	}
+
+	if (!str) {
+		itrace_synth_opts__set_default(synth_opts);
+		return 0;
+	}

Quoted 8 lines

tools/perf/util/auxtrace.c out_free: · line 241

Quoted 5 lines

tools/perf/util/auxtrace.c +5 -0

+
+	for (p = str; *p;) {
+		switch (*p++) {
+		case 'i':
+			synth_opts->instructions = true;

Is this documented anywhere? I.e. examples for setting up this synth ops?

They are documented for each tool that uses them:
Documentation/perf-script.txt
Documentation/perf-record.txt
and also in:
Documentation/intel-pt.txt