All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH v4 0/6] Fix missing leaf-function callers when recording
@ 2021-12-15 15:11 ` German Gomez
  0 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: German Gomez, John Garry, Will Deacon, Mathieu Poirier, Leo Yan,
	Mark Rutland, Alexander Shishkin, Jiri Olsa, Namhyung Kim,
	linux-arm-kernel

(This cset applies on top of [1])

Call-graphs on ARM64 using the option "--call-grah fp" are missing the
callers of the leaf functions. See [PATCH 6/6] for a before and after
after example using this cset.

[1] https://lore.kernel.org/all/20211207180653.1147374-1-german.gomez@arm.com/

---
Changes since v3
  - Only record LR register instead of all registers in [PATCH 1/6].
  - Introduce [PATCH 5/6] to refactor the SAMPL_REG macro.
  - Fix compilation issues on different platforms.

Alexandre Truong (5):
  perf tools: record ARM64 LR register automatically
  perf tools: add a mechanism to inject stack frames
  perf tools: Refactor script__setup_sample_type()
  perf tools: enable dwarf_callchain_users on arm64
  perf tools: determine if LR is the return address

German Gomez (1):
  perf tools: Refactor SMPL_REG macro in perf_regs.h

 tools/perf/arch/arm64/util/machine.c          |  7 +++
 tools/perf/builtin-record.c                   |  8 +++
 tools/perf/builtin-report.c                   |  4 +-
 tools/perf/builtin-script.c                   | 13 +---
 tools/perf/util/Build                         |  1 +
 .../util/arm64-frame-pointer-unwind-support.c | 63 +++++++++++++++++++
 .../util/arm64-frame-pointer-unwind-support.h | 10 +++
 tools/perf/util/callchain.c                   |  9 ++-
 tools/perf/util/callchain.h                   |  4 +-
 tools/perf/util/machine.c                     | 50 ++++++++++++++-
 tools/perf/util/machine.h                     |  1 +
 tools/perf/util/perf_regs.h                   |  7 ++-
 12 files changed, 157 insertions(+), 20 deletions(-)
 create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.c
 create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.h

-- 
2.25.1


^ permalink raw reply	[flat|nested] 22+ messages in thread

* [PATCH v4 0/6] Fix missing leaf-function callers when recording
@ 2021-12-15 15:11 ` German Gomez
  0 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: German Gomez, John Garry, Will Deacon, Mathieu Poirier, Leo Yan,
	Mark Rutland, Alexander Shishkin, Jiri Olsa, Namhyung Kim,
	linux-arm-kernel

(This cset applies on top of [1])

Call-graphs on ARM64 using the option "--call-grah fp" are missing the
callers of the leaf functions. See [PATCH 6/6] for a before and after
after example using this cset.

[1] https://lore.kernel.org/all/20211207180653.1147374-1-german.gomez@arm.com/

---
Changes since v3
  - Only record LR register instead of all registers in [PATCH 1/6].
  - Introduce [PATCH 5/6] to refactor the SAMPL_REG macro.
  - Fix compilation issues on different platforms.

Alexandre Truong (5):
  perf tools: record ARM64 LR register automatically
  perf tools: add a mechanism to inject stack frames
  perf tools: Refactor script__setup_sample_type()
  perf tools: enable dwarf_callchain_users on arm64
  perf tools: determine if LR is the return address

German Gomez (1):
  perf tools: Refactor SMPL_REG macro in perf_regs.h

 tools/perf/arch/arm64/util/machine.c          |  7 +++
 tools/perf/builtin-record.c                   |  8 +++
 tools/perf/builtin-report.c                   |  4 +-
 tools/perf/builtin-script.c                   | 13 +---
 tools/perf/util/Build                         |  1 +
 .../util/arm64-frame-pointer-unwind-support.c | 63 +++++++++++++++++++
 .../util/arm64-frame-pointer-unwind-support.h | 10 +++
 tools/perf/util/callchain.c                   |  9 ++-
 tools/perf/util/callchain.h                   |  4 +-
 tools/perf/util/machine.c                     | 50 ++++++++++++++-
 tools/perf/util/machine.h                     |  1 +
 tools/perf/util/perf_regs.h                   |  7 ++-
 12 files changed, 157 insertions(+), 20 deletions(-)
 create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.c
 create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.h

-- 
2.25.1


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

^ permalink raw reply	[flat|nested] 22+ messages in thread

* [PATCH v4 1/6] perf tools: record ARM64 LR register automatically
  2021-12-15 15:11 ` German Gomez
@ 2021-12-15 15:11   ` German Gomez
  -1 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: Alexandre Truong, German Gomez, John Garry, Will Deacon,
	Mathieu Poirier, Leo Yan, Mark Rutland, Alexander Shishkin,
	Jiri Olsa, Namhyung Kim, linux-arm-kernel

From: Alexandre Truong <alexandre.truong@arm.com>

On ARM64, automatically record the link register if the frame pointer
mode is on. It will be used to do a dwarf unwind to find the caller
of the leaf frame if the frame pointer was omitted.

Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
Signed-off-by: German Gomez <german.gomez@arm.com>
---
 tools/perf/arch/arm64/util/machine.c | 7 +++++++
 tools/perf/builtin-record.c          | 8 ++++++++
 tools/perf/util/callchain.h          | 2 ++
 3 files changed, 17 insertions(+)

diff --git a/tools/perf/arch/arm64/util/machine.c b/tools/perf/arch/arm64/util/machine.c
index 7e7714290a87..d2ce31e28cd7 100644
--- a/tools/perf/arch/arm64/util/machine.c
+++ b/tools/perf/arch/arm64/util/machine.c
@@ -5,6 +5,8 @@
 #include <string.h>
 #include "debug.h"
 #include "symbol.h"
+#include "callchain.h"
+#include "record.h"
 
 /* On arm64, kernel text segment starts at high memory address,
  * for example 0xffff 0000 8xxx xxxx. Modules start at a low memory
@@ -26,3 +28,8 @@ void arch__symbols__fixup_end(struct symbol *p, struct symbol *c)
 		p->end = c->start;
 	pr_debug4("%s sym:%s end:%#" PRIx64 "\n", __func__, p->name, p->end);
 }
+
+void arch__add_leaf_frame_record_opts(struct record_opts *opts)
+{
+	opts->sample_user_regs |= sample_reg_masks[PERF_REG_ARM64_LR].mask;
+}
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 0338b813585a..6ac2160913ea 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -2267,6 +2267,10 @@ static int record__parse_mmap_pages(const struct option *opt,
 	return ret;
 }
 
+void __weak arch__add_leaf_frame_record_opts(struct record_opts *opts __maybe_unused)
+{
+}
+
 static int parse_control_option(const struct option *opt,
 				const char *str,
 				int unset __maybe_unused)
@@ -2898,6 +2902,10 @@ int cmd_record(int argc, const char **argv)
 	}
 
 	rec->opts.target.hybrid = perf_pmu__has_hybrid();
+
+	if (callchain_param.enabled && callchain_param.record_mode == CALLCHAIN_FP)
+		arch__add_leaf_frame_record_opts(&rec->opts);
+
 	err = -ENOMEM;
 	if (evlist__create_maps(rec->evlist, &rec->opts.target) < 0)
 		usage_with_options(record_usage, record_options);
diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h
index 5824134f983b..77fba053c677 100644
--- a/tools/perf/util/callchain.h
+++ b/tools/perf/util/callchain.h
@@ -280,6 +280,8 @@ static inline int arch_skip_callchain_idx(struct thread *thread __maybe_unused,
 }
 #endif
 
+void arch__add_leaf_frame_record_opts(struct record_opts *opts);
+
 char *callchain_list__sym_name(struct callchain_list *cl,
 			       char *bf, size_t bfsize, bool show_dso);
 char *callchain_node__scnprintf_value(struct callchain_node *node,
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 22+ messages in thread

* [PATCH v4 1/6] perf tools: record ARM64 LR register automatically
@ 2021-12-15 15:11   ` German Gomez
  0 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: Alexandre Truong, German Gomez, John Garry, Will Deacon,
	Mathieu Poirier, Leo Yan, Mark Rutland, Alexander Shishkin,
	Jiri Olsa, Namhyung Kim, linux-arm-kernel

From: Alexandre Truong <alexandre.truong@arm.com>

On ARM64, automatically record the link register if the frame pointer
mode is on. It will be used to do a dwarf unwind to find the caller
of the leaf frame if the frame pointer was omitted.

Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
Signed-off-by: German Gomez <german.gomez@arm.com>
---
 tools/perf/arch/arm64/util/machine.c | 7 +++++++
 tools/perf/builtin-record.c          | 8 ++++++++
 tools/perf/util/callchain.h          | 2 ++
 3 files changed, 17 insertions(+)

diff --git a/tools/perf/arch/arm64/util/machine.c b/tools/perf/arch/arm64/util/machine.c
index 7e7714290a87..d2ce31e28cd7 100644
--- a/tools/perf/arch/arm64/util/machine.c
+++ b/tools/perf/arch/arm64/util/machine.c
@@ -5,6 +5,8 @@
 #include <string.h>
 #include "debug.h"
 #include "symbol.h"
+#include "callchain.h"
+#include "record.h"
 
 /* On arm64, kernel text segment starts at high memory address,
  * for example 0xffff 0000 8xxx xxxx. Modules start at a low memory
@@ -26,3 +28,8 @@ void arch__symbols__fixup_end(struct symbol *p, struct symbol *c)
 		p->end = c->start;
 	pr_debug4("%s sym:%s end:%#" PRIx64 "\n", __func__, p->name, p->end);
 }
+
+void arch__add_leaf_frame_record_opts(struct record_opts *opts)
+{
+	opts->sample_user_regs |= sample_reg_masks[PERF_REG_ARM64_LR].mask;
+}
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 0338b813585a..6ac2160913ea 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -2267,6 +2267,10 @@ static int record__parse_mmap_pages(const struct option *opt,
 	return ret;
 }
 
+void __weak arch__add_leaf_frame_record_opts(struct record_opts *opts __maybe_unused)
+{
+}
+
 static int parse_control_option(const struct option *opt,
 				const char *str,
 				int unset __maybe_unused)
@@ -2898,6 +2902,10 @@ int cmd_record(int argc, const char **argv)
 	}
 
 	rec->opts.target.hybrid = perf_pmu__has_hybrid();
+
+	if (callchain_param.enabled && callchain_param.record_mode == CALLCHAIN_FP)
+		arch__add_leaf_frame_record_opts(&rec->opts);
+
 	err = -ENOMEM;
 	if (evlist__create_maps(rec->evlist, &rec->opts.target) < 0)
 		usage_with_options(record_usage, record_options);
diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h
index 5824134f983b..77fba053c677 100644
--- a/tools/perf/util/callchain.h
+++ b/tools/perf/util/callchain.h
@@ -280,6 +280,8 @@ static inline int arch_skip_callchain_idx(struct thread *thread __maybe_unused,
 }
 #endif
 
+void arch__add_leaf_frame_record_opts(struct record_opts *opts);
+
 char *callchain_list__sym_name(struct callchain_list *cl,
 			       char *bf, size_t bfsize, bool show_dso);
 char *callchain_node__scnprintf_value(struct callchain_node *node,
-- 
2.25.1


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

^ permalink raw reply related	[flat|nested] 22+ messages in thread

* [PATCH v4 2/6] perf tools: add a mechanism to inject stack frames
  2021-12-15 15:11 ` German Gomez
@ 2021-12-15 15:11   ` German Gomez
  -1 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: Alexandre Truong, German Gomez, John Garry, Will Deacon,
	Mathieu Poirier, Leo Yan, Mark Rutland, Alexander Shishkin,
	Jiri Olsa, Namhyung Kim, linux-arm-kernel

From: Alexandre Truong <alexandre.truong@arm.com>

Add a mechanism for platforms to inject stack frames for the leaf
frame caller if there is enough information to determine a frame
is missing from dwarf or other post processing mechanisms.

Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
Signed-off-by: German Gomez <german.gomez@arm.com>
---
 tools/perf/util/machine.c | 37 ++++++++++++++++++++++++++++++++++++-
 1 file changed, 36 insertions(+), 1 deletion(-)

diff --git a/tools/perf/util/machine.c b/tools/perf/util/machine.c
index fb8496df8432..3eddad009f78 100644
--- a/tools/perf/util/machine.c
+++ b/tools/perf/util/machine.c
@@ -2710,6 +2710,12 @@ static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
 	return err;
 }
 
+static u64 get_leaf_frame_caller(struct perf_sample *sample __maybe_unused,
+		struct thread *thread __maybe_unused, int usr_idx __maybe_unused)
+{
+	return 0;
+}
+
 static int thread__resolve_callchain_sample(struct thread *thread,
 					    struct callchain_cursor *cursor,
 					    struct evsel *evsel,
@@ -2723,9 +2729,10 @@ static int thread__resolve_callchain_sample(struct thread *thread,
 	struct ip_callchain *chain = sample->callchain;
 	int chain_nr = 0;
 	u8 cpumode = PERF_RECORD_MISC_USER;
-	int i, j, err, nr_entries;
+	int i, j, err, nr_entries, usr_idx;
 	int skip_idx = -1;
 	int first_call = 0;
+	u64 leaf_frame_caller;
 
 	if (chain)
 		chain_nr = chain->nr;
@@ -2850,6 +2857,34 @@ static int thread__resolve_callchain_sample(struct thread *thread,
 			continue;
 		}
 
+		/*
+		 * PERF_CONTEXT_USER allows us to locate where the user stack ends.
+		 * Depending on callchain_param.order and the position of PERF_CONTEXT_USER,
+		 * the index will be different in order to add the missing frame
+		 * at the right place.
+		 */
+
+		usr_idx = callchain_param.order == ORDER_CALLEE ? j-2 : j-1;
+
+		if (usr_idx >= 0 && chain->ips[usr_idx] == PERF_CONTEXT_USER) {
+
+			leaf_frame_caller = get_leaf_frame_caller(sample, thread, usr_idx);
+
+			/*
+			 * check if leaf_frame_Caller != ip to not add the same
+			 * value twice.
+			 */
+
+			if (leaf_frame_caller && leaf_frame_caller != ip) {
+
+				err = add_callchain_ip(thread, cursor, parent,
+					       root_al, &cpumode, leaf_frame_caller,
+					       false, NULL, NULL, 0);
+				if (err)
+					return (err < 0) ? err : 0;
+			}
+		}
+
 		err = add_callchain_ip(thread, cursor, parent,
 				       root_al, &cpumode, ip,
 				       false, NULL, NULL, 0);
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 22+ messages in thread

* [PATCH v4 2/6] perf tools: add a mechanism to inject stack frames
@ 2021-12-15 15:11   ` German Gomez
  0 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: Alexandre Truong, German Gomez, John Garry, Will Deacon,
	Mathieu Poirier, Leo Yan, Mark Rutland, Alexander Shishkin,
	Jiri Olsa, Namhyung Kim, linux-arm-kernel

From: Alexandre Truong <alexandre.truong@arm.com>

Add a mechanism for platforms to inject stack frames for the leaf
frame caller if there is enough information to determine a frame
is missing from dwarf or other post processing mechanisms.

Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
Signed-off-by: German Gomez <german.gomez@arm.com>
---
 tools/perf/util/machine.c | 37 ++++++++++++++++++++++++++++++++++++-
 1 file changed, 36 insertions(+), 1 deletion(-)

diff --git a/tools/perf/util/machine.c b/tools/perf/util/machine.c
index fb8496df8432..3eddad009f78 100644
--- a/tools/perf/util/machine.c
+++ b/tools/perf/util/machine.c
@@ -2710,6 +2710,12 @@ static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
 	return err;
 }
 
+static u64 get_leaf_frame_caller(struct perf_sample *sample __maybe_unused,
+		struct thread *thread __maybe_unused, int usr_idx __maybe_unused)
+{
+	return 0;
+}
+
 static int thread__resolve_callchain_sample(struct thread *thread,
 					    struct callchain_cursor *cursor,
 					    struct evsel *evsel,
@@ -2723,9 +2729,10 @@ static int thread__resolve_callchain_sample(struct thread *thread,
 	struct ip_callchain *chain = sample->callchain;
 	int chain_nr = 0;
 	u8 cpumode = PERF_RECORD_MISC_USER;
-	int i, j, err, nr_entries;
+	int i, j, err, nr_entries, usr_idx;
 	int skip_idx = -1;
 	int first_call = 0;
+	u64 leaf_frame_caller;
 
 	if (chain)
 		chain_nr = chain->nr;
@@ -2850,6 +2857,34 @@ static int thread__resolve_callchain_sample(struct thread *thread,
 			continue;
 		}
 
+		/*
+		 * PERF_CONTEXT_USER allows us to locate where the user stack ends.
+		 * Depending on callchain_param.order and the position of PERF_CONTEXT_USER,
+		 * the index will be different in order to add the missing frame
+		 * at the right place.
+		 */
+
+		usr_idx = callchain_param.order == ORDER_CALLEE ? j-2 : j-1;
+
+		if (usr_idx >= 0 && chain->ips[usr_idx] == PERF_CONTEXT_USER) {
+
+			leaf_frame_caller = get_leaf_frame_caller(sample, thread, usr_idx);
+
+			/*
+			 * check if leaf_frame_Caller != ip to not add the same
+			 * value twice.
+			 */
+
+			if (leaf_frame_caller && leaf_frame_caller != ip) {
+
+				err = add_callchain_ip(thread, cursor, parent,
+					       root_al, &cpumode, leaf_frame_caller,
+					       false, NULL, NULL, 0);
+				if (err)
+					return (err < 0) ? err : 0;
+			}
+		}
+
 		err = add_callchain_ip(thread, cursor, parent,
 				       root_al, &cpumode, ip,
 				       false, NULL, NULL, 0);
-- 
2.25.1


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

^ permalink raw reply related	[flat|nested] 22+ messages in thread

* [PATCH v4 3/6] perf tools: Refactor script__setup_sample_type()
  2021-12-15 15:11 ` German Gomez
@ 2021-12-15 15:11   ` German Gomez
  -1 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: Alexandre Truong, German Gomez, John Garry, Will Deacon,
	Mathieu Poirier, Leo Yan, Mark Rutland, Alexander Shishkin,
	Jiri Olsa, Namhyung Kim, linux-arm-kernel

From: Alexandre Truong <alexandre.truong@arm.com>

Refactoring script__setup_sample_type() by using
callchain_param_setup() to replace the duplicate code
for callchain parameter setting up.

Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
Signed-off-by: German Gomez <german.gomez@arm.com>
---
 tools/perf/builtin-script.c | 11 +----------
 1 file changed, 1 insertion(+), 10 deletions(-)

diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
index da2175d70ac9..ab7d575f97f2 100644
--- a/tools/perf/builtin-script.c
+++ b/tools/perf/builtin-script.c
@@ -3468,16 +3468,7 @@ static void script__setup_sample_type(struct perf_script *script)
 	struct perf_session *session = script->session;
 	u64 sample_type = evlist__combined_sample_type(session->evlist);
 
-	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
-		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-		    (sample_type & PERF_SAMPLE_STACK_USER)) {
-			callchain_param.record_mode = CALLCHAIN_DWARF;
-			dwarf_callchain_users = true;
-		} else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
-			callchain_param.record_mode = CALLCHAIN_LBR;
-		else
-			callchain_param.record_mode = CALLCHAIN_FP;
-	}
+	callchain_param_setup(sample_type);
 
 	if (script->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
 		pr_warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 22+ messages in thread

* [PATCH v4 3/6] perf tools: Refactor script__setup_sample_type()
@ 2021-12-15 15:11   ` German Gomez
  0 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: Alexandre Truong, German Gomez, John Garry, Will Deacon,
	Mathieu Poirier, Leo Yan, Mark Rutland, Alexander Shishkin,
	Jiri Olsa, Namhyung Kim, linux-arm-kernel

From: Alexandre Truong <alexandre.truong@arm.com>

Refactoring script__setup_sample_type() by using
callchain_param_setup() to replace the duplicate code
for callchain parameter setting up.

Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
Signed-off-by: German Gomez <german.gomez@arm.com>
---
 tools/perf/builtin-script.c | 11 +----------
 1 file changed, 1 insertion(+), 10 deletions(-)

diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
index da2175d70ac9..ab7d575f97f2 100644
--- a/tools/perf/builtin-script.c
+++ b/tools/perf/builtin-script.c
@@ -3468,16 +3468,7 @@ static void script__setup_sample_type(struct perf_script *script)
 	struct perf_session *session = script->session;
 	u64 sample_type = evlist__combined_sample_type(session->evlist);
 
-	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
-		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-		    (sample_type & PERF_SAMPLE_STACK_USER)) {
-			callchain_param.record_mode = CALLCHAIN_DWARF;
-			dwarf_callchain_users = true;
-		} else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
-			callchain_param.record_mode = CALLCHAIN_LBR;
-		else
-			callchain_param.record_mode = CALLCHAIN_FP;
-	}
+	callchain_param_setup(sample_type);
 
 	if (script->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
 		pr_warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
-- 
2.25.1


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

^ permalink raw reply related	[flat|nested] 22+ messages in thread

* [PATCH v4 4/6] perf tools: enable dwarf_callchain_users on arm64
  2021-12-15 15:11 ` German Gomez
@ 2021-12-15 15:11   ` German Gomez
  -1 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: Alexandre Truong, German Gomez, John Garry, Will Deacon,
	Mathieu Poirier, Leo Yan, Mark Rutland, Alexander Shishkin,
	Jiri Olsa, Namhyung Kim, linux-arm-kernel

From: Alexandre Truong <alexandre.truong@arm.com>

On arm64, enable dwarf_callchain_users which will be needed
to do a dwarf unwind in order to get the caller of the leaf frame.

Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
Signed-off-by: German Gomez <german.gomez@arm.com>
---
 tools/perf/builtin-report.c | 4 ++--
 tools/perf/builtin-script.c | 4 ++--
 tools/perf/util/callchain.c | 9 ++++++++-
 tools/perf/util/callchain.h | 2 +-
 4 files changed, 13 insertions(+), 6 deletions(-)

diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index 8167ebfe776a..a31ad60ba66e 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -410,7 +410,7 @@ static int report__setup_sample_type(struct report *rep)
 		}
 	}
 
-	callchain_param_setup(sample_type);
+	callchain_param_setup(sample_type, perf_env__arch(&rep->session->header.env));
 
 	if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
 		ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
@@ -1124,7 +1124,7 @@ static int process_attr(struct perf_tool *tool __maybe_unused,
 	 * on events sample_type.
 	 */
 	sample_type = evlist__combined_sample_type(*pevlist);
-	callchain_param_setup(sample_type);
+	callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env));
 	return 0;
 }
 
diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
index ab7d575f97f2..d308adfd1176 100644
--- a/tools/perf/builtin-script.c
+++ b/tools/perf/builtin-script.c
@@ -2318,7 +2318,7 @@ static int process_attr(struct perf_tool *tool, union perf_event *event,
 	 * on events sample_type.
 	 */
 	sample_type = evlist__combined_sample_type(evlist);
-	callchain_param_setup(sample_type);
+	callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env));
 
 	/* Enable fields for callchain entries */
 	if (symbol_conf.use_callchain &&
@@ -3468,7 +3468,7 @@ static void script__setup_sample_type(struct perf_script *script)
 	struct perf_session *session = script->session;
 	u64 sample_type = evlist__combined_sample_type(session->evlist);
 
-	callchain_param_setup(sample_type);
+	callchain_param_setup(sample_type, perf_env__arch(session->machines.host.env));
 
 	if (script->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
 		pr_warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
diff --git a/tools/perf/util/callchain.c b/tools/perf/util/callchain.c
index 8e2777133bd9..aaab9a674807 100644
--- a/tools/perf/util/callchain.c
+++ b/tools/perf/util/callchain.c
@@ -1600,7 +1600,7 @@ void callchain_cursor_reset(struct callchain_cursor *cursor)
 		map__zput(node->ms.map);
 }
 
-void callchain_param_setup(u64 sample_type)
+void callchain_param_setup(u64 sample_type, const char *arch)
 {
 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
@@ -1612,6 +1612,13 @@ void callchain_param_setup(u64 sample_type)
 		else
 			callchain_param.record_mode = CALLCHAIN_FP;
 	}
+
+	/*
+	 * It's possible to determine the caller of leaf frames with omitted
+	 * frame pointers on aarch64 using libunwind, so enable it.
+	 */
+	if (callchain_param.record_mode == CALLCHAIN_FP && !strcmp(arch, "arm64"))
+		dwarf_callchain_users = true;
 }
 
 static bool chain_match(struct callchain_list *base_chain,
diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h
index 77fba053c677..d95615daed73 100644
--- a/tools/perf/util/callchain.h
+++ b/tools/perf/util/callchain.h
@@ -300,7 +300,7 @@ int callchain_branch_counts(struct callchain_root *root,
 			    u64 *branch_count, u64 *predicted_count,
 			    u64 *abort_count, u64 *cycles_count);
 
-void callchain_param_setup(u64 sample_type);
+void callchain_param_setup(u64 sample_type, const char *arch);
 
 bool callchain_cnode_matched(struct callchain_node *base_cnode,
 			     struct callchain_node *pair_cnode);
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 22+ messages in thread

* [PATCH v4 4/6] perf tools: enable dwarf_callchain_users on arm64
@ 2021-12-15 15:11   ` German Gomez
  0 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: Alexandre Truong, German Gomez, John Garry, Will Deacon,
	Mathieu Poirier, Leo Yan, Mark Rutland, Alexander Shishkin,
	Jiri Olsa, Namhyung Kim, linux-arm-kernel

From: Alexandre Truong <alexandre.truong@arm.com>

On arm64, enable dwarf_callchain_users which will be needed
to do a dwarf unwind in order to get the caller of the leaf frame.

Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
Signed-off-by: German Gomez <german.gomez@arm.com>
---
 tools/perf/builtin-report.c | 4 ++--
 tools/perf/builtin-script.c | 4 ++--
 tools/perf/util/callchain.c | 9 ++++++++-
 tools/perf/util/callchain.h | 2 +-
 4 files changed, 13 insertions(+), 6 deletions(-)

diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index 8167ebfe776a..a31ad60ba66e 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -410,7 +410,7 @@ static int report__setup_sample_type(struct report *rep)
 		}
 	}
 
-	callchain_param_setup(sample_type);
+	callchain_param_setup(sample_type, perf_env__arch(&rep->session->header.env));
 
 	if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
 		ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
@@ -1124,7 +1124,7 @@ static int process_attr(struct perf_tool *tool __maybe_unused,
 	 * on events sample_type.
 	 */
 	sample_type = evlist__combined_sample_type(*pevlist);
-	callchain_param_setup(sample_type);
+	callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env));
 	return 0;
 }
 
diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
index ab7d575f97f2..d308adfd1176 100644
--- a/tools/perf/builtin-script.c
+++ b/tools/perf/builtin-script.c
@@ -2318,7 +2318,7 @@ static int process_attr(struct perf_tool *tool, union perf_event *event,
 	 * on events sample_type.
 	 */
 	sample_type = evlist__combined_sample_type(evlist);
-	callchain_param_setup(sample_type);
+	callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env));
 
 	/* Enable fields for callchain entries */
 	if (symbol_conf.use_callchain &&
@@ -3468,7 +3468,7 @@ static void script__setup_sample_type(struct perf_script *script)
 	struct perf_session *session = script->session;
 	u64 sample_type = evlist__combined_sample_type(session->evlist);
 
-	callchain_param_setup(sample_type);
+	callchain_param_setup(sample_type, perf_env__arch(session->machines.host.env));
 
 	if (script->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
 		pr_warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
diff --git a/tools/perf/util/callchain.c b/tools/perf/util/callchain.c
index 8e2777133bd9..aaab9a674807 100644
--- a/tools/perf/util/callchain.c
+++ b/tools/perf/util/callchain.c
@@ -1600,7 +1600,7 @@ void callchain_cursor_reset(struct callchain_cursor *cursor)
 		map__zput(node->ms.map);
 }
 
-void callchain_param_setup(u64 sample_type)
+void callchain_param_setup(u64 sample_type, const char *arch)
 {
 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
@@ -1612,6 +1612,13 @@ void callchain_param_setup(u64 sample_type)
 		else
 			callchain_param.record_mode = CALLCHAIN_FP;
 	}
+
+	/*
+	 * It's possible to determine the caller of leaf frames with omitted
+	 * frame pointers on aarch64 using libunwind, so enable it.
+	 */
+	if (callchain_param.record_mode == CALLCHAIN_FP && !strcmp(arch, "arm64"))
+		dwarf_callchain_users = true;
 }
 
 static bool chain_match(struct callchain_list *base_chain,
diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h
index 77fba053c677..d95615daed73 100644
--- a/tools/perf/util/callchain.h
+++ b/tools/perf/util/callchain.h
@@ -300,7 +300,7 @@ int callchain_branch_counts(struct callchain_root *root,
 			    u64 *branch_count, u64 *predicted_count,
 			    u64 *abort_count, u64 *cycles_count);
 
-void callchain_param_setup(u64 sample_type);
+void callchain_param_setup(u64 sample_type, const char *arch);
 
 bool callchain_cnode_matched(struct callchain_node *base_cnode,
 			     struct callchain_node *pair_cnode);
-- 
2.25.1


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

^ permalink raw reply related	[flat|nested] 22+ messages in thread

* [PATCH v4 5/6] perf tools: Refactor SMPL_REG macro in perf_regs.h
  2021-12-15 15:11 ` German Gomez
@ 2021-12-15 15:11   ` German Gomez
  -1 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: German Gomez, John Garry, Will Deacon, Mathieu Poirier, Leo Yan,
	Mark Rutland, Alexander Shishkin, Jiri Olsa, Namhyung Kim,
	linux-arm-kernel

Refactor the SAMPL_REG macro so that it can be used in a followup commit
to obtain the masks for ARM64 registers.

Signed-off-by: German Gomez <german.gomez@arm.com>
---
 tools/perf/util/perf_regs.h | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/tools/perf/util/perf_regs.h b/tools/perf/util/perf_regs.h
index 4e6b1299c571..ce1127af05e4 100644
--- a/tools/perf/util/perf_regs.h
+++ b/tools/perf/util/perf_regs.h
@@ -11,8 +11,11 @@ struct sample_reg {
 	const char *name;
 	uint64_t mask;
 };
-#define SMPL_REG(n, b) { .name = #n, .mask = 1ULL << (b) }
-#define SMPL_REG2(n, b) { .name = #n, .mask = 3ULL << (b) }
+
+#define SMPL_REG_MASK(b) (1ULL << (b))
+#define SMPL_REG(n, b) { .name = #n, .mask = SMPL_REG_MASK(b) }
+#define SMPL_REG2_MASK(b) (3ULL << (b))
+#define SMPL_REG2(n, b) { .name = #n, .mask = SMPL_REG2_MASK(b) }
 #define SMPL_REG_END { .name = NULL }
 
 enum {
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 22+ messages in thread

* [PATCH v4 5/6] perf tools: Refactor SMPL_REG macro in perf_regs.h
@ 2021-12-15 15:11   ` German Gomez
  0 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: German Gomez, John Garry, Will Deacon, Mathieu Poirier, Leo Yan,
	Mark Rutland, Alexander Shishkin, Jiri Olsa, Namhyung Kim,
	linux-arm-kernel

Refactor the SAMPL_REG macro so that it can be used in a followup commit
to obtain the masks for ARM64 registers.

Signed-off-by: German Gomez <german.gomez@arm.com>
---
 tools/perf/util/perf_regs.h | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/tools/perf/util/perf_regs.h b/tools/perf/util/perf_regs.h
index 4e6b1299c571..ce1127af05e4 100644
--- a/tools/perf/util/perf_regs.h
+++ b/tools/perf/util/perf_regs.h
@@ -11,8 +11,11 @@ struct sample_reg {
 	const char *name;
 	uint64_t mask;
 };
-#define SMPL_REG(n, b) { .name = #n, .mask = 1ULL << (b) }
-#define SMPL_REG2(n, b) { .name = #n, .mask = 3ULL << (b) }
+
+#define SMPL_REG_MASK(b) (1ULL << (b))
+#define SMPL_REG(n, b) { .name = #n, .mask = SMPL_REG_MASK(b) }
+#define SMPL_REG2_MASK(b) (3ULL << (b))
+#define SMPL_REG2(n, b) { .name = #n, .mask = SMPL_REG2_MASK(b) }
 #define SMPL_REG_END { .name = NULL }
 
 enum {
-- 
2.25.1


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

^ permalink raw reply related	[flat|nested] 22+ messages in thread

* [PATCH v4 6/6] perf tools: determine if LR is the return address
  2021-12-15 15:11 ` German Gomez
@ 2021-12-15 15:11   ` German Gomez
  -1 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: Alexandre Truong, German Gomez, John Garry, Will Deacon,
	Mathieu Poirier, Leo Yan, Mark Rutland, Alexander Shishkin,
	Jiri Olsa, Namhyung Kim, linux-arm-kernel

From: Alexandre Truong <alexandre.truong@arm.com>

On arm64 and frame pointer mode (e.g: perf record --callgraph fp),
use dwarf unwind info to check if the link register is the return
address in order to inject it to the frame pointer stack.

Write the following application:

	int a = 10;

	void f2(void)
	{
		for (int i = 0; i < 1000000; i++)
			a *= a;
	}

	void f1()
	{
		for (int i = 0; i < 10; i++)
			f2();
	}

	int main(void)
	{
		f1();
		return 0;
	}

with the following compilation flags:
        gcc -fno-omit-frame-pointer -fno-inline -O2

The compiler omits the frame pointer for f2 on arm. This is a problem
with any leaf call, for example an application with many different
calls to malloc() would always omit the calling frame, even if it
can be determined.

	./perf record --call-graph fp ./a.out
	./perf report

currently gives the following stack:

0xffffea52f361
_start
__libc_start_main
main
f2

After this change, perf report correctly shows f1() calling f2(),
even though it was missing from the frame pointer unwind:

	./perf report

0xffffea52f361
_start
__libc_start_main
main
f1
f2

Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
Signed-off-by: German Gomez <german.gomez@arm.com>
---
 tools/perf/util/Build                         |  1 +
 .../util/arm64-frame-pointer-unwind-support.c | 63 +++++++++++++++++++
 .../util/arm64-frame-pointer-unwind-support.h | 10 +++
 tools/perf/util/machine.c                     | 19 ++++--
 tools/perf/util/machine.h                     |  1 +
 5 files changed, 89 insertions(+), 5 deletions(-)
 create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.c
 create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.h

diff --git a/tools/perf/util/Build b/tools/perf/util/Build
index 2e5bfbb69960..03d4c647bd86 100644
--- a/tools/perf/util/Build
+++ b/tools/perf/util/Build
@@ -1,3 +1,4 @@
+perf-y += arm64-frame-pointer-unwind-support.o
 perf-y += annotate.o
 perf-y += block-info.o
 perf-y += block-range.o
diff --git a/tools/perf/util/arm64-frame-pointer-unwind-support.c b/tools/perf/util/arm64-frame-pointer-unwind-support.c
new file mode 100644
index 000000000000..4f5ecf51ed38
--- /dev/null
+++ b/tools/perf/util/arm64-frame-pointer-unwind-support.c
@@ -0,0 +1,63 @@
+// SPDX-License-Identifier: GPL-2.0
+#include "arm64-frame-pointer-unwind-support.h"
+#include "callchain.h"
+#include "event.h"
+#include "perf_regs.h" // SMPL_REG_MASK
+#include "unwind.h"
+
+#define perf_event_arm_regs perf_event_arm64_regs
+#include "../arch/arm64/include/uapi/asm/perf_regs.h"
+#undef perf_event_arm_regs
+
+struct entries {
+	u64 stack[2];
+	size_t length;
+};
+
+static bool get_leaf_frame_caller_enabled(struct perf_sample *sample)
+{
+	return callchain_param.record_mode == CALLCHAIN_FP && sample->user_regs.regs
+		&& sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_LR);
+}
+
+static int add_entry(struct unwind_entry *entry, void *arg)
+{
+	struct entries *entries = arg;
+
+	entries->stack[entries->length++] = entry->ip;
+	return 0;
+}
+
+u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thread, int usr_idx)
+{
+	int ret;
+	struct entries entries = {};
+	struct regs_dump old_regs = sample->user_regs;
+
+	if (!get_leaf_frame_caller_enabled(sample))
+		return 0;
+
+	/*
+	 * If PC and SP are not recorded, get the value of PC from the stack
+	 * and set its mask. SP is not used when doing the unwinding but it
+	 * still needs to be set to prevent failures.
+	 */
+
+	if (!(sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_PC))) {
+		sample->user_regs.cache_mask |= SMPL_REG_MASK(PERF_REG_ARM64_PC);
+		sample->user_regs.cache_regs[PERF_REG_ARM64_PC] = sample->callchain->ips[usr_idx+1];
+	}
+
+	if (!(sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_SP))) {
+		sample->user_regs.cache_mask |= SMPL_REG_MASK(PERF_REG_ARM64_SP);
+		sample->user_regs.cache_regs[PERF_REG_ARM64_SP] = 0;
+	}
+
+	ret = unwind__get_entries(add_entry, &entries, thread, sample, 2);
+	sample->user_regs = old_regs;
+
+	if (ret || entries.length != 2)
+		return ret;
+
+	return callchain_param.order == ORDER_CALLER ? entries.stack[0] : entries.stack[1];
+}
diff --git a/tools/perf/util/arm64-frame-pointer-unwind-support.h b/tools/perf/util/arm64-frame-pointer-unwind-support.h
new file mode 100644
index 000000000000..32af9ce94398
--- /dev/null
+++ b/tools/perf/util/arm64-frame-pointer-unwind-support.h
@@ -0,0 +1,10 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H
+#define __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H
+
+#include "event.h"
+#include "thread.h"
+
+u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thread, int user_idx);
+
+#endif /* __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H */
diff --git a/tools/perf/util/machine.c b/tools/perf/util/machine.c
index 3eddad009f78..a00fd6796b35 100644
--- a/tools/perf/util/machine.c
+++ b/tools/perf/util/machine.c
@@ -34,6 +34,7 @@
 #include "bpf-event.h"
 #include <internal/lib.h> // page_size
 #include "cgroup.h"
+#include "arm64-frame-pointer-unwind-support.h"
 
 #include <linux/ctype.h>
 #include <symbol/kallsyms.h>
@@ -2710,10 +2711,13 @@ static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
 	return err;
 }
 
-static u64 get_leaf_frame_caller(struct perf_sample *sample __maybe_unused,
-		struct thread *thread __maybe_unused, int usr_idx __maybe_unused)
+static u64 get_leaf_frame_caller(struct perf_sample *sample,
+		struct thread *thread, int usr_idx)
 {
-	return 0;
+	if (machine__normalize_is(thread->maps->machine, "arm64"))
+		return get_leaf_frame_caller_aarch64(sample, thread, usr_idx);
+	else
+		return 0;
 }
 
 static int thread__resolve_callchain_sample(struct thread *thread,
@@ -3114,14 +3118,19 @@ int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
 }
 
 /*
- * Compares the raw arch string. N.B. see instead perf_env__arch() if a
- * normalized arch is needed.
+ * Compares the raw arch string. N.B. see instead perf_env__arch() or
+ * machine__normalize_is() if a normalized arch is needed.
  */
 bool machine__is(struct machine *machine, const char *arch)
 {
 	return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
 }
 
+bool machine__normalize_is(struct machine *machine, const char *arch)
+{
+	return machine && !strcmp(perf_env__arch(machine->env), arch);
+}
+
 int machine__nr_cpus_avail(struct machine *machine)
 {
 	return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
diff --git a/tools/perf/util/machine.h b/tools/perf/util/machine.h
index a143087eeb47..665535153411 100644
--- a/tools/perf/util/machine.h
+++ b/tools/perf/util/machine.h
@@ -208,6 +208,7 @@ static inline bool machine__is_host(struct machine *machine)
 }
 
 bool machine__is(struct machine *machine, const char *arch);
+bool machine__normalize_is(struct machine *machine, const char *arch);
 int machine__nr_cpus_avail(struct machine *machine);
 
 struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid);
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 22+ messages in thread

* [PATCH v4 6/6] perf tools: determine if LR is the return address
@ 2021-12-15 15:11   ` German Gomez
  0 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-15 15:11 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: Alexandre Truong, German Gomez, John Garry, Will Deacon,
	Mathieu Poirier, Leo Yan, Mark Rutland, Alexander Shishkin,
	Jiri Olsa, Namhyung Kim, linux-arm-kernel

From: Alexandre Truong <alexandre.truong@arm.com>

On arm64 and frame pointer mode (e.g: perf record --callgraph fp),
use dwarf unwind info to check if the link register is the return
address in order to inject it to the frame pointer stack.

Write the following application:

	int a = 10;

	void f2(void)
	{
		for (int i = 0; i < 1000000; i++)
			a *= a;
	}

	void f1()
	{
		for (int i = 0; i < 10; i++)
			f2();
	}

	int main(void)
	{
		f1();
		return 0;
	}

with the following compilation flags:
        gcc -fno-omit-frame-pointer -fno-inline -O2

The compiler omits the frame pointer for f2 on arm. This is a problem
with any leaf call, for example an application with many different
calls to malloc() would always omit the calling frame, even if it
can be determined.

	./perf record --call-graph fp ./a.out
	./perf report

currently gives the following stack:

0xffffea52f361
_start
__libc_start_main
main
f2

After this change, perf report correctly shows f1() calling f2(),
even though it was missing from the frame pointer unwind:

	./perf report

0xffffea52f361
_start
__libc_start_main
main
f1
f2

Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
Signed-off-by: German Gomez <german.gomez@arm.com>
---
 tools/perf/util/Build                         |  1 +
 .../util/arm64-frame-pointer-unwind-support.c | 63 +++++++++++++++++++
 .../util/arm64-frame-pointer-unwind-support.h | 10 +++
 tools/perf/util/machine.c                     | 19 ++++--
 tools/perf/util/machine.h                     |  1 +
 5 files changed, 89 insertions(+), 5 deletions(-)
 create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.c
 create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.h

diff --git a/tools/perf/util/Build b/tools/perf/util/Build
index 2e5bfbb69960..03d4c647bd86 100644
--- a/tools/perf/util/Build
+++ b/tools/perf/util/Build
@@ -1,3 +1,4 @@
+perf-y += arm64-frame-pointer-unwind-support.o
 perf-y += annotate.o
 perf-y += block-info.o
 perf-y += block-range.o
diff --git a/tools/perf/util/arm64-frame-pointer-unwind-support.c b/tools/perf/util/arm64-frame-pointer-unwind-support.c
new file mode 100644
index 000000000000..4f5ecf51ed38
--- /dev/null
+++ b/tools/perf/util/arm64-frame-pointer-unwind-support.c
@@ -0,0 +1,63 @@
+// SPDX-License-Identifier: GPL-2.0
+#include "arm64-frame-pointer-unwind-support.h"
+#include "callchain.h"
+#include "event.h"
+#include "perf_regs.h" // SMPL_REG_MASK
+#include "unwind.h"
+
+#define perf_event_arm_regs perf_event_arm64_regs
+#include "../arch/arm64/include/uapi/asm/perf_regs.h"
+#undef perf_event_arm_regs
+
+struct entries {
+	u64 stack[2];
+	size_t length;
+};
+
+static bool get_leaf_frame_caller_enabled(struct perf_sample *sample)
+{
+	return callchain_param.record_mode == CALLCHAIN_FP && sample->user_regs.regs
+		&& sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_LR);
+}
+
+static int add_entry(struct unwind_entry *entry, void *arg)
+{
+	struct entries *entries = arg;
+
+	entries->stack[entries->length++] = entry->ip;
+	return 0;
+}
+
+u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thread, int usr_idx)
+{
+	int ret;
+	struct entries entries = {};
+	struct regs_dump old_regs = sample->user_regs;
+
+	if (!get_leaf_frame_caller_enabled(sample))
+		return 0;
+
+	/*
+	 * If PC and SP are not recorded, get the value of PC from the stack
+	 * and set its mask. SP is not used when doing the unwinding but it
+	 * still needs to be set to prevent failures.
+	 */
+
+	if (!(sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_PC))) {
+		sample->user_regs.cache_mask |= SMPL_REG_MASK(PERF_REG_ARM64_PC);
+		sample->user_regs.cache_regs[PERF_REG_ARM64_PC] = sample->callchain->ips[usr_idx+1];
+	}
+
+	if (!(sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_SP))) {
+		sample->user_regs.cache_mask |= SMPL_REG_MASK(PERF_REG_ARM64_SP);
+		sample->user_regs.cache_regs[PERF_REG_ARM64_SP] = 0;
+	}
+
+	ret = unwind__get_entries(add_entry, &entries, thread, sample, 2);
+	sample->user_regs = old_regs;
+
+	if (ret || entries.length != 2)
+		return ret;
+
+	return callchain_param.order == ORDER_CALLER ? entries.stack[0] : entries.stack[1];
+}
diff --git a/tools/perf/util/arm64-frame-pointer-unwind-support.h b/tools/perf/util/arm64-frame-pointer-unwind-support.h
new file mode 100644
index 000000000000..32af9ce94398
--- /dev/null
+++ b/tools/perf/util/arm64-frame-pointer-unwind-support.h
@@ -0,0 +1,10 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H
+#define __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H
+
+#include "event.h"
+#include "thread.h"
+
+u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thread, int user_idx);
+
+#endif /* __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H */
diff --git a/tools/perf/util/machine.c b/tools/perf/util/machine.c
index 3eddad009f78..a00fd6796b35 100644
--- a/tools/perf/util/machine.c
+++ b/tools/perf/util/machine.c
@@ -34,6 +34,7 @@
 #include "bpf-event.h"
 #include <internal/lib.h> // page_size
 #include "cgroup.h"
+#include "arm64-frame-pointer-unwind-support.h"
 
 #include <linux/ctype.h>
 #include <symbol/kallsyms.h>
@@ -2710,10 +2711,13 @@ static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
 	return err;
 }
 
-static u64 get_leaf_frame_caller(struct perf_sample *sample __maybe_unused,
-		struct thread *thread __maybe_unused, int usr_idx __maybe_unused)
+static u64 get_leaf_frame_caller(struct perf_sample *sample,
+		struct thread *thread, int usr_idx)
 {
-	return 0;
+	if (machine__normalize_is(thread->maps->machine, "arm64"))
+		return get_leaf_frame_caller_aarch64(sample, thread, usr_idx);
+	else
+		return 0;
 }
 
 static int thread__resolve_callchain_sample(struct thread *thread,
@@ -3114,14 +3118,19 @@ int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
 }
 
 /*
- * Compares the raw arch string. N.B. see instead perf_env__arch() if a
- * normalized arch is needed.
+ * Compares the raw arch string. N.B. see instead perf_env__arch() or
+ * machine__normalize_is() if a normalized arch is needed.
  */
 bool machine__is(struct machine *machine, const char *arch)
 {
 	return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
 }
 
+bool machine__normalize_is(struct machine *machine, const char *arch)
+{
+	return machine && !strcmp(perf_env__arch(machine->env), arch);
+}
+
 int machine__nr_cpus_avail(struct machine *machine)
 {
 	return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
diff --git a/tools/perf/util/machine.h b/tools/perf/util/machine.h
index a143087eeb47..665535153411 100644
--- a/tools/perf/util/machine.h
+++ b/tools/perf/util/machine.h
@@ -208,6 +208,7 @@ static inline bool machine__is_host(struct machine *machine)
 }
 
 bool machine__is(struct machine *machine, const char *arch);
+bool machine__normalize_is(struct machine *machine, const char *arch);
 int machine__nr_cpus_avail(struct machine *machine);
 
 struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid);
-- 
2.25.1


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

^ permalink raw reply related	[flat|nested] 22+ messages in thread

* Re: [PATCH v4 6/6] perf tools: determine if LR is the return address
  2021-12-15 15:11   ` German Gomez
@ 2021-12-15 16:33     ` Mark Rutland
  -1 siblings, 0 replies; 22+ messages in thread
From: Mark Rutland @ 2021-12-15 16:33 UTC (permalink / raw)
  To: German Gomez
  Cc: linux-kernel, linux-perf-users, acme, Alexandre Truong,
	John Garry, Will Deacon, Mathieu Poirier, Leo Yan,
	Alexander Shishkin, Jiri Olsa, Namhyung Kim, linux-arm-kernel

Hi,

On Wed, Dec 15, 2021 at 03:11:38PM +0000, German Gomez wrote:
> From: Alexandre Truong <alexandre.truong@arm.com>
> 
> On arm64 and frame pointer mode (e.g: perf record --callgraph fp),
> use dwarf unwind info to check if the link register is the return
> address in order to inject it to the frame pointer stack.

This series looks good overall, but as a general note the commit messages are a
bit hard to read because they jump into implementation details of the patch
(i.e. the change the patch makes) before explaining the problem (i.e. what the
patch is trying to solve).

It would be nice to have a short introduction, e.g.

  When unwinding using frame pointers on arm64, the return address of the
  current leaf function may be missed. The return address of a leaf function
  may live in the LR and/or a frame record (and the location can change within
  a function), so it is necessary to use DWARF to identify where to look for
  the return address at any given point during a function.

  For example:

  unsigned long foo(unsigned long i)
  {
          i += 2;
	  i += 5;
  }

  ... could be compiled as:

  foo:
  	// return addr in LR
  	add	x0, x0, #2
	// return addr in LR
	stp	x29, x30, [SP, #-16]!
	// return addr in LR
	mov	x29, sp
	// return addr in LR *and* frame record
	add	x0, x0, #5
	// return addr in LR *and* frame record
	ldp	x29, x30, [sp], #16
	// return addr in LR
	ret

> Write the following application:
> 
> 	int a = 10;
> 
> 	void f2(void)
> 	{
> 		for (int i = 0; i < 1000000; i++)
> 			a *= a;
> 	}
> 
> 	void f1()
> 	{
> 		for (int i = 0; i < 10; i++)
> 			f2();
> 	}
> 
> 	int main(void)
> 	{
> 		f1();
> 		return 0;
> 	}
> 
> with the following compilation flags:
>         gcc -fno-omit-frame-pointer -fno-inline -O2
> 
> The compiler omits the frame pointer for f2 on arm. This is a problem
> with any leaf call, for example an application with many different
> calls to malloc() would always omit the calling frame, even if it
> can be determined.

I think the wording here is slightly misleading. For f2, the compiler *doesn't
create a frame record*, but the frame pointer (to the caller's frame record)
remains and is not omitted.

Also, I think it's woth noting (as per the example I gave above) this applies
to *any* function which is the current leaf function, regardless of whether
that function creates a frame record at some point. For example, if `f1` is
interrupted before it creates its own frame record (or after it destroys the
frame record), the FP will point at the record created by `main` (containing
the caller of main), and `main` itself will be missing from the unwind as it
will only exist in the LR.

> 	./perf record --call-graph fp ./a.out
> 	./perf report
> 
> currently gives the following stack:
> 
> 0xffffea52f361
> _start
> __libc_start_main
> main
> f2
> 
> After this change, perf report correctly shows f1() calling f2(),
> even though it was missing from the frame pointer unwind:
> 
> 	./perf report
> 
> 0xffffea52f361
> _start
> __libc_start_main
> main
> f1
> f2
> 
> Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
> Signed-off-by: German Gomez <german.gomez@arm.com>
> ---
>  tools/perf/util/Build                         |  1 +
>  .../util/arm64-frame-pointer-unwind-support.c | 63 +++++++++++++++++++
>  .../util/arm64-frame-pointer-unwind-support.h | 10 +++
>  tools/perf/util/machine.c                     | 19 ++++--
>  tools/perf/util/machine.h                     |  1 +
>  5 files changed, 89 insertions(+), 5 deletions(-)
>  create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.c
>  create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.h
> 
> diff --git a/tools/perf/util/Build b/tools/perf/util/Build
> index 2e5bfbb69960..03d4c647bd86 100644
> --- a/tools/perf/util/Build
> +++ b/tools/perf/util/Build
> @@ -1,3 +1,4 @@
> +perf-y += arm64-frame-pointer-unwind-support.o
>  perf-y += annotate.o
>  perf-y += block-info.o
>  perf-y += block-range.o
> diff --git a/tools/perf/util/arm64-frame-pointer-unwind-support.c b/tools/perf/util/arm64-frame-pointer-unwind-support.c
> new file mode 100644
> index 000000000000..4f5ecf51ed38
> --- /dev/null
> +++ b/tools/perf/util/arm64-frame-pointer-unwind-support.c
> @@ -0,0 +1,63 @@
> +// SPDX-License-Identifier: GPL-2.0
> +#include "arm64-frame-pointer-unwind-support.h"
> +#include "callchain.h"
> +#include "event.h"
> +#include "perf_regs.h" // SMPL_REG_MASK
> +#include "unwind.h"
> +
> +#define perf_event_arm_regs perf_event_arm64_regs
> +#include "../arch/arm64/include/uapi/asm/perf_regs.h"
> +#undef perf_event_arm_regs
> +
> +struct entries {
> +	u64 stack[2];
> +	size_t length;
> +};
> +
> +static bool get_leaf_frame_caller_enabled(struct perf_sample *sample)
> +{
> +	return callchain_param.record_mode == CALLCHAIN_FP && sample->user_regs.regs
> +		&& sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_LR);
> +}
> +
> +static int add_entry(struct unwind_entry *entry, void *arg)
> +{
> +	struct entries *entries = arg;
> +
> +	entries->stack[entries->length++] = entry->ip;
> +	return 0;
> +}
> +
> +u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thread, int usr_idx)
> +{
> +	int ret;
> +	struct entries entries = {};
> +	struct regs_dump old_regs = sample->user_regs;
> +
> +	if (!get_leaf_frame_caller_enabled(sample))
> +		return 0;
> +
> +	/*
> +	 * If PC and SP are not recorded, get the value of PC from the stack
> +	 * and set its mask. SP is not used when doing the unwinding but it
> +	 * still needs to be set to prevent failures.
> +	 */

To prevent failures where? Is this something libunwind requires?

Thanks,
Mark.

> +	if (!(sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_PC))) {
> +		sample->user_regs.cache_mask |= SMPL_REG_MASK(PERF_REG_ARM64_PC);
> +		sample->user_regs.cache_regs[PERF_REG_ARM64_PC] = sample->callchain->ips[usr_idx+1];
> +	}
> +
> +	if (!(sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_SP))) {
> +		sample->user_regs.cache_mask |= SMPL_REG_MASK(PERF_REG_ARM64_SP);
> +		sample->user_regs.cache_regs[PERF_REG_ARM64_SP] = 0;
> +	}
> +
> +	ret = unwind__get_entries(add_entry, &entries, thread, sample, 2);
> +	sample->user_regs = old_regs;
> +
> +	if (ret || entries.length != 2)
> +		return ret;
> +
> +	return callchain_param.order == ORDER_CALLER ? entries.stack[0] : entries.stack[1];
> +}
> diff --git a/tools/perf/util/arm64-frame-pointer-unwind-support.h b/tools/perf/util/arm64-frame-pointer-unwind-support.h
> new file mode 100644
> index 000000000000..32af9ce94398
> --- /dev/null
> +++ b/tools/perf/util/arm64-frame-pointer-unwind-support.h
> @@ -0,0 +1,10 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +#ifndef __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H
> +#define __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H
> +
> +#include "event.h"
> +#include "thread.h"
> +
> +u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thread, int user_idx);
> +
> +#endif /* __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H */
> diff --git a/tools/perf/util/machine.c b/tools/perf/util/machine.c
> index 3eddad009f78..a00fd6796b35 100644
> --- a/tools/perf/util/machine.c
> +++ b/tools/perf/util/machine.c
> @@ -34,6 +34,7 @@
>  #include "bpf-event.h"
>  #include <internal/lib.h> // page_size
>  #include "cgroup.h"
> +#include "arm64-frame-pointer-unwind-support.h"
>  
>  #include <linux/ctype.h>
>  #include <symbol/kallsyms.h>
> @@ -2710,10 +2711,13 @@ static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
>  	return err;
>  }
>  
> -static u64 get_leaf_frame_caller(struct perf_sample *sample __maybe_unused,
> -		struct thread *thread __maybe_unused, int usr_idx __maybe_unused)
> +static u64 get_leaf_frame_caller(struct perf_sample *sample,
> +		struct thread *thread, int usr_idx)
>  {
> -	return 0;
> +	if (machine__normalize_is(thread->maps->machine, "arm64"))
> +		return get_leaf_frame_caller_aarch64(sample, thread, usr_idx);
> +	else
> +		return 0;
>  }
>  
>  static int thread__resolve_callchain_sample(struct thread *thread,
> @@ -3114,14 +3118,19 @@ int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
>  }
>  
>  /*
> - * Compares the raw arch string. N.B. see instead perf_env__arch() if a
> - * normalized arch is needed.
> + * Compares the raw arch string. N.B. see instead perf_env__arch() or
> + * machine__normalize_is() if a normalized arch is needed.
>   */
>  bool machine__is(struct machine *machine, const char *arch)
>  {
>  	return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
>  }
>  
> +bool machine__normalize_is(struct machine *machine, const char *arch)
> +{
> +	return machine && !strcmp(perf_env__arch(machine->env), arch);
> +}
> +
>  int machine__nr_cpus_avail(struct machine *machine)
>  {
>  	return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
> diff --git a/tools/perf/util/machine.h b/tools/perf/util/machine.h
> index a143087eeb47..665535153411 100644
> --- a/tools/perf/util/machine.h
> +++ b/tools/perf/util/machine.h
> @@ -208,6 +208,7 @@ static inline bool machine__is_host(struct machine *machine)
>  }
>  
>  bool machine__is(struct machine *machine, const char *arch);
> +bool machine__normalize_is(struct machine *machine, const char *arch);
>  int machine__nr_cpus_avail(struct machine *machine);
>  
>  struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid);
> -- 
> 2.25.1
> 

^ permalink raw reply	[flat|nested] 22+ messages in thread

* Re: [PATCH v4 6/6] perf tools: determine if LR is the return address
@ 2021-12-15 16:33     ` Mark Rutland
  0 siblings, 0 replies; 22+ messages in thread
From: Mark Rutland @ 2021-12-15 16:33 UTC (permalink / raw)
  To: German Gomez
  Cc: linux-kernel, linux-perf-users, acme, Alexandre Truong,
	John Garry, Will Deacon, Mathieu Poirier, Leo Yan,
	Alexander Shishkin, Jiri Olsa, Namhyung Kim, linux-arm-kernel

Hi,

On Wed, Dec 15, 2021 at 03:11:38PM +0000, German Gomez wrote:
> From: Alexandre Truong <alexandre.truong@arm.com>
> 
> On arm64 and frame pointer mode (e.g: perf record --callgraph fp),
> use dwarf unwind info to check if the link register is the return
> address in order to inject it to the frame pointer stack.

This series looks good overall, but as a general note the commit messages are a
bit hard to read because they jump into implementation details of the patch
(i.e. the change the patch makes) before explaining the problem (i.e. what the
patch is trying to solve).

It would be nice to have a short introduction, e.g.

  When unwinding using frame pointers on arm64, the return address of the
  current leaf function may be missed. The return address of a leaf function
  may live in the LR and/or a frame record (and the location can change within
  a function), so it is necessary to use DWARF to identify where to look for
  the return address at any given point during a function.

  For example:

  unsigned long foo(unsigned long i)
  {
          i += 2;
	  i += 5;
  }

  ... could be compiled as:

  foo:
  	// return addr in LR
  	add	x0, x0, #2
	// return addr in LR
	stp	x29, x30, [SP, #-16]!
	// return addr in LR
	mov	x29, sp
	// return addr in LR *and* frame record
	add	x0, x0, #5
	// return addr in LR *and* frame record
	ldp	x29, x30, [sp], #16
	// return addr in LR
	ret

> Write the following application:
> 
> 	int a = 10;
> 
> 	void f2(void)
> 	{
> 		for (int i = 0; i < 1000000; i++)
> 			a *= a;
> 	}
> 
> 	void f1()
> 	{
> 		for (int i = 0; i < 10; i++)
> 			f2();
> 	}
> 
> 	int main(void)
> 	{
> 		f1();
> 		return 0;
> 	}
> 
> with the following compilation flags:
>         gcc -fno-omit-frame-pointer -fno-inline -O2
> 
> The compiler omits the frame pointer for f2 on arm. This is a problem
> with any leaf call, for example an application with many different
> calls to malloc() would always omit the calling frame, even if it
> can be determined.

I think the wording here is slightly misleading. For f2, the compiler *doesn't
create a frame record*, but the frame pointer (to the caller's frame record)
remains and is not omitted.

Also, I think it's woth noting (as per the example I gave above) this applies
to *any* function which is the current leaf function, regardless of whether
that function creates a frame record at some point. For example, if `f1` is
interrupted before it creates its own frame record (or after it destroys the
frame record), the FP will point at the record created by `main` (containing
the caller of main), and `main` itself will be missing from the unwind as it
will only exist in the LR.

> 	./perf record --call-graph fp ./a.out
> 	./perf report
> 
> currently gives the following stack:
> 
> 0xffffea52f361
> _start
> __libc_start_main
> main
> f2
> 
> After this change, perf report correctly shows f1() calling f2(),
> even though it was missing from the frame pointer unwind:
> 
> 	./perf report
> 
> 0xffffea52f361
> _start
> __libc_start_main
> main
> f1
> f2
> 
> Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
> Signed-off-by: German Gomez <german.gomez@arm.com>
> ---
>  tools/perf/util/Build                         |  1 +
>  .../util/arm64-frame-pointer-unwind-support.c | 63 +++++++++++++++++++
>  .../util/arm64-frame-pointer-unwind-support.h | 10 +++
>  tools/perf/util/machine.c                     | 19 ++++--
>  tools/perf/util/machine.h                     |  1 +
>  5 files changed, 89 insertions(+), 5 deletions(-)
>  create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.c
>  create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.h
> 
> diff --git a/tools/perf/util/Build b/tools/perf/util/Build
> index 2e5bfbb69960..03d4c647bd86 100644
> --- a/tools/perf/util/Build
> +++ b/tools/perf/util/Build
> @@ -1,3 +1,4 @@
> +perf-y += arm64-frame-pointer-unwind-support.o
>  perf-y += annotate.o
>  perf-y += block-info.o
>  perf-y += block-range.o
> diff --git a/tools/perf/util/arm64-frame-pointer-unwind-support.c b/tools/perf/util/arm64-frame-pointer-unwind-support.c
> new file mode 100644
> index 000000000000..4f5ecf51ed38
> --- /dev/null
> +++ b/tools/perf/util/arm64-frame-pointer-unwind-support.c
> @@ -0,0 +1,63 @@
> +// SPDX-License-Identifier: GPL-2.0
> +#include "arm64-frame-pointer-unwind-support.h"
> +#include "callchain.h"
> +#include "event.h"
> +#include "perf_regs.h" // SMPL_REG_MASK
> +#include "unwind.h"
> +
> +#define perf_event_arm_regs perf_event_arm64_regs
> +#include "../arch/arm64/include/uapi/asm/perf_regs.h"
> +#undef perf_event_arm_regs
> +
> +struct entries {
> +	u64 stack[2];
> +	size_t length;
> +};
> +
> +static bool get_leaf_frame_caller_enabled(struct perf_sample *sample)
> +{
> +	return callchain_param.record_mode == CALLCHAIN_FP && sample->user_regs.regs
> +		&& sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_LR);
> +}
> +
> +static int add_entry(struct unwind_entry *entry, void *arg)
> +{
> +	struct entries *entries = arg;
> +
> +	entries->stack[entries->length++] = entry->ip;
> +	return 0;
> +}
> +
> +u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thread, int usr_idx)
> +{
> +	int ret;
> +	struct entries entries = {};
> +	struct regs_dump old_regs = sample->user_regs;
> +
> +	if (!get_leaf_frame_caller_enabled(sample))
> +		return 0;
> +
> +	/*
> +	 * If PC and SP are not recorded, get the value of PC from the stack
> +	 * and set its mask. SP is not used when doing the unwinding but it
> +	 * still needs to be set to prevent failures.
> +	 */

To prevent failures where? Is this something libunwind requires?

Thanks,
Mark.

> +	if (!(sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_PC))) {
> +		sample->user_regs.cache_mask |= SMPL_REG_MASK(PERF_REG_ARM64_PC);
> +		sample->user_regs.cache_regs[PERF_REG_ARM64_PC] = sample->callchain->ips[usr_idx+1];
> +	}
> +
> +	if (!(sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_SP))) {
> +		sample->user_regs.cache_mask |= SMPL_REG_MASK(PERF_REG_ARM64_SP);
> +		sample->user_regs.cache_regs[PERF_REG_ARM64_SP] = 0;
> +	}
> +
> +	ret = unwind__get_entries(add_entry, &entries, thread, sample, 2);
> +	sample->user_regs = old_regs;
> +
> +	if (ret || entries.length != 2)
> +		return ret;
> +
> +	return callchain_param.order == ORDER_CALLER ? entries.stack[0] : entries.stack[1];
> +}
> diff --git a/tools/perf/util/arm64-frame-pointer-unwind-support.h b/tools/perf/util/arm64-frame-pointer-unwind-support.h
> new file mode 100644
> index 000000000000..32af9ce94398
> --- /dev/null
> +++ b/tools/perf/util/arm64-frame-pointer-unwind-support.h
> @@ -0,0 +1,10 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +#ifndef __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H
> +#define __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H
> +
> +#include "event.h"
> +#include "thread.h"
> +
> +u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thread, int user_idx);
> +
> +#endif /* __PERF_ARM_FRAME_POINTER_UNWIND_SUPPORT_H */
> diff --git a/tools/perf/util/machine.c b/tools/perf/util/machine.c
> index 3eddad009f78..a00fd6796b35 100644
> --- a/tools/perf/util/machine.c
> +++ b/tools/perf/util/machine.c
> @@ -34,6 +34,7 @@
>  #include "bpf-event.h"
>  #include <internal/lib.h> // page_size
>  #include "cgroup.h"
> +#include "arm64-frame-pointer-unwind-support.h"
>  
>  #include <linux/ctype.h>
>  #include <symbol/kallsyms.h>
> @@ -2710,10 +2711,13 @@ static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
>  	return err;
>  }
>  
> -static u64 get_leaf_frame_caller(struct perf_sample *sample __maybe_unused,
> -		struct thread *thread __maybe_unused, int usr_idx __maybe_unused)
> +static u64 get_leaf_frame_caller(struct perf_sample *sample,
> +		struct thread *thread, int usr_idx)
>  {
> -	return 0;
> +	if (machine__normalize_is(thread->maps->machine, "arm64"))
> +		return get_leaf_frame_caller_aarch64(sample, thread, usr_idx);
> +	else
> +		return 0;
>  }
>  
>  static int thread__resolve_callchain_sample(struct thread *thread,
> @@ -3114,14 +3118,19 @@ int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
>  }
>  
>  /*
> - * Compares the raw arch string. N.B. see instead perf_env__arch() if a
> - * normalized arch is needed.
> + * Compares the raw arch string. N.B. see instead perf_env__arch() or
> + * machine__normalize_is() if a normalized arch is needed.
>   */
>  bool machine__is(struct machine *machine, const char *arch)
>  {
>  	return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
>  }
>  
> +bool machine__normalize_is(struct machine *machine, const char *arch)
> +{
> +	return machine && !strcmp(perf_env__arch(machine->env), arch);
> +}
> +
>  int machine__nr_cpus_avail(struct machine *machine)
>  {
>  	return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
> diff --git a/tools/perf/util/machine.h b/tools/perf/util/machine.h
> index a143087eeb47..665535153411 100644
> --- a/tools/perf/util/machine.h
> +++ b/tools/perf/util/machine.h
> @@ -208,6 +208,7 @@ static inline bool machine__is_host(struct machine *machine)
>  }
>  
>  bool machine__is(struct machine *machine, const char *arch);
> +bool machine__normalize_is(struct machine *machine, const char *arch);
>  int machine__nr_cpus_avail(struct machine *machine);
>  
>  struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid);
> -- 
> 2.25.1
> 

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

^ permalink raw reply	[flat|nested] 22+ messages in thread

* Re: [PATCH v4 4/6] perf tools: enable dwarf_callchain_users on arm64
  2021-12-15 15:11   ` German Gomez
@ 2021-12-15 16:37     ` Mark Rutland
  -1 siblings, 0 replies; 22+ messages in thread
From: Mark Rutland @ 2021-12-15 16:37 UTC (permalink / raw)
  To: German Gomez
  Cc: linux-kernel, linux-perf-users, acme, Alexandre Truong,
	John Garry, Will Deacon, Mathieu Poirier, Leo Yan,
	Alexander Shishkin, Jiri Olsa, Namhyung Kim, linux-arm-kernel

On Wed, Dec 15, 2021 at 03:11:36PM +0000, German Gomez wrote:
> From: Alexandre Truong <alexandre.truong@arm.com>
> 
> On arm64, enable dwarf_callchain_users which will be needed
> to do a dwarf unwind in order to get the caller of the leaf frame.
> 
> Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
> Signed-off-by: German Gomez <german.gomez@arm.com>
> ---
>  tools/perf/builtin-report.c | 4 ++--
>  tools/perf/builtin-script.c | 4 ++--
>  tools/perf/util/callchain.c | 9 ++++++++-
>  tools/perf/util/callchain.h | 2 +-
>  4 files changed, 13 insertions(+), 6 deletions(-)
> 
> diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
> index 8167ebfe776a..a31ad60ba66e 100644
> --- a/tools/perf/builtin-report.c
> +++ b/tools/perf/builtin-report.c
> @@ -410,7 +410,7 @@ static int report__setup_sample_type(struct report *rep)
>  		}
>  	}
>  
> -	callchain_param_setup(sample_type);
> +	callchain_param_setup(sample_type, perf_env__arch(&rep->session->header.env));
>  
>  	if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
>  		ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
> @@ -1124,7 +1124,7 @@ static int process_attr(struct perf_tool *tool __maybe_unused,
>  	 * on events sample_type.
>  	 */
>  	sample_type = evlist__combined_sample_type(*pevlist);
> -	callchain_param_setup(sample_type);
> +	callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env));
>  	return 0;
>  }
>  
> diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
> index ab7d575f97f2..d308adfd1176 100644
> --- a/tools/perf/builtin-script.c
> +++ b/tools/perf/builtin-script.c
> @@ -2318,7 +2318,7 @@ static int process_attr(struct perf_tool *tool, union perf_event *event,
>  	 * on events sample_type.
>  	 */
>  	sample_type = evlist__combined_sample_type(evlist);
> -	callchain_param_setup(sample_type);
> +	callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env));
>  
>  	/* Enable fields for callchain entries */
>  	if (symbol_conf.use_callchain &&
> @@ -3468,7 +3468,7 @@ static void script__setup_sample_type(struct perf_script *script)
>  	struct perf_session *session = script->session;
>  	u64 sample_type = evlist__combined_sample_type(session->evlist);
>  
> -	callchain_param_setup(sample_type);
> +	callchain_param_setup(sample_type, perf_env__arch(session->machines.host.env));
>  
>  	if (script->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
>  		pr_warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
> diff --git a/tools/perf/util/callchain.c b/tools/perf/util/callchain.c
> index 8e2777133bd9..aaab9a674807 100644
> --- a/tools/perf/util/callchain.c
> +++ b/tools/perf/util/callchain.c
> @@ -1600,7 +1600,7 @@ void callchain_cursor_reset(struct callchain_cursor *cursor)
>  		map__zput(node->ms.map);
>  }
>  
> -void callchain_param_setup(u64 sample_type)
> +void callchain_param_setup(u64 sample_type, const char *arch)
>  {
>  	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
>  		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
> @@ -1612,6 +1612,13 @@ void callchain_param_setup(u64 sample_type)
>  		else
>  			callchain_param.record_mode = CALLCHAIN_FP;
>  	}
> +
> +	/*
> +	 * It's possible to determine the caller of leaf frames with omitted
> +	 * frame pointers on aarch64 using libunwind, so enable it.
> +	 */

I reckon it's worth mentioning *why* we need to do this; how about:

	/*
	 * It's necessary to use libunwind to reliably determine the caller of
	 * a leaf function on aarch64, as otherwise we cannot know whether to
	 * start from the LR or FP.
	 *
	 * Always starting from the LR can result in duplicate or entirely
	 * erroneous entries. Always skipping the LR and starting from the FP
	 * can result in missing entries.
	 */

Other than that, this looks fine to me!

Thanks,
Mark.

> +	if (callchain_param.record_mode == CALLCHAIN_FP && !strcmp(arch, "arm64"))
> +		dwarf_callchain_users = true;
>  }
>  
>  static bool chain_match(struct callchain_list *base_chain,
> diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h
> index 77fba053c677..d95615daed73 100644
> --- a/tools/perf/util/callchain.h
> +++ b/tools/perf/util/callchain.h
> @@ -300,7 +300,7 @@ int callchain_branch_counts(struct callchain_root *root,
>  			    u64 *branch_count, u64 *predicted_count,
>  			    u64 *abort_count, u64 *cycles_count);
>  
> -void callchain_param_setup(u64 sample_type);
> +void callchain_param_setup(u64 sample_type, const char *arch);
>  
>  bool callchain_cnode_matched(struct callchain_node *base_cnode,
>  			     struct callchain_node *pair_cnode);
> -- 
> 2.25.1
> 

^ permalink raw reply	[flat|nested] 22+ messages in thread

* Re: [PATCH v4 4/6] perf tools: enable dwarf_callchain_users on arm64
@ 2021-12-15 16:37     ` Mark Rutland
  0 siblings, 0 replies; 22+ messages in thread
From: Mark Rutland @ 2021-12-15 16:37 UTC (permalink / raw)
  To: German Gomez
  Cc: linux-kernel, linux-perf-users, acme, Alexandre Truong,
	John Garry, Will Deacon, Mathieu Poirier, Leo Yan,
	Alexander Shishkin, Jiri Olsa, Namhyung Kim, linux-arm-kernel

On Wed, Dec 15, 2021 at 03:11:36PM +0000, German Gomez wrote:
> From: Alexandre Truong <alexandre.truong@arm.com>
> 
> On arm64, enable dwarf_callchain_users which will be needed
> to do a dwarf unwind in order to get the caller of the leaf frame.
> 
> Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
> Signed-off-by: German Gomez <german.gomez@arm.com>
> ---
>  tools/perf/builtin-report.c | 4 ++--
>  tools/perf/builtin-script.c | 4 ++--
>  tools/perf/util/callchain.c | 9 ++++++++-
>  tools/perf/util/callchain.h | 2 +-
>  4 files changed, 13 insertions(+), 6 deletions(-)
> 
> diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
> index 8167ebfe776a..a31ad60ba66e 100644
> --- a/tools/perf/builtin-report.c
> +++ b/tools/perf/builtin-report.c
> @@ -410,7 +410,7 @@ static int report__setup_sample_type(struct report *rep)
>  		}
>  	}
>  
> -	callchain_param_setup(sample_type);
> +	callchain_param_setup(sample_type, perf_env__arch(&rep->session->header.env));
>  
>  	if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
>  		ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
> @@ -1124,7 +1124,7 @@ static int process_attr(struct perf_tool *tool __maybe_unused,
>  	 * on events sample_type.
>  	 */
>  	sample_type = evlist__combined_sample_type(*pevlist);
> -	callchain_param_setup(sample_type);
> +	callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env));
>  	return 0;
>  }
>  
> diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
> index ab7d575f97f2..d308adfd1176 100644
> --- a/tools/perf/builtin-script.c
> +++ b/tools/perf/builtin-script.c
> @@ -2318,7 +2318,7 @@ static int process_attr(struct perf_tool *tool, union perf_event *event,
>  	 * on events sample_type.
>  	 */
>  	sample_type = evlist__combined_sample_type(evlist);
> -	callchain_param_setup(sample_type);
> +	callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env));
>  
>  	/* Enable fields for callchain entries */
>  	if (symbol_conf.use_callchain &&
> @@ -3468,7 +3468,7 @@ static void script__setup_sample_type(struct perf_script *script)
>  	struct perf_session *session = script->session;
>  	u64 sample_type = evlist__combined_sample_type(session->evlist);
>  
> -	callchain_param_setup(sample_type);
> +	callchain_param_setup(sample_type, perf_env__arch(session->machines.host.env));
>  
>  	if (script->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
>  		pr_warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
> diff --git a/tools/perf/util/callchain.c b/tools/perf/util/callchain.c
> index 8e2777133bd9..aaab9a674807 100644
> --- a/tools/perf/util/callchain.c
> +++ b/tools/perf/util/callchain.c
> @@ -1600,7 +1600,7 @@ void callchain_cursor_reset(struct callchain_cursor *cursor)
>  		map__zput(node->ms.map);
>  }
>  
> -void callchain_param_setup(u64 sample_type)
> +void callchain_param_setup(u64 sample_type, const char *arch)
>  {
>  	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
>  		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
> @@ -1612,6 +1612,13 @@ void callchain_param_setup(u64 sample_type)
>  		else
>  			callchain_param.record_mode = CALLCHAIN_FP;
>  	}
> +
> +	/*
> +	 * It's possible to determine the caller of leaf frames with omitted
> +	 * frame pointers on aarch64 using libunwind, so enable it.
> +	 */

I reckon it's worth mentioning *why* we need to do this; how about:

	/*
	 * It's necessary to use libunwind to reliably determine the caller of
	 * a leaf function on aarch64, as otherwise we cannot know whether to
	 * start from the LR or FP.
	 *
	 * Always starting from the LR can result in duplicate or entirely
	 * erroneous entries. Always skipping the LR and starting from the FP
	 * can result in missing entries.
	 */

Other than that, this looks fine to me!

Thanks,
Mark.

> +	if (callchain_param.record_mode == CALLCHAIN_FP && !strcmp(arch, "arm64"))
> +		dwarf_callchain_users = true;
>  }
>  
>  static bool chain_match(struct callchain_list *base_chain,
> diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h
> index 77fba053c677..d95615daed73 100644
> --- a/tools/perf/util/callchain.h
> +++ b/tools/perf/util/callchain.h
> @@ -300,7 +300,7 @@ int callchain_branch_counts(struct callchain_root *root,
>  			    u64 *branch_count, u64 *predicted_count,
>  			    u64 *abort_count, u64 *cycles_count);
>  
> -void callchain_param_setup(u64 sample_type);
> +void callchain_param_setup(u64 sample_type, const char *arch);
>  
>  bool callchain_cnode_matched(struct callchain_node *base_cnode,
>  			     struct callchain_node *pair_cnode);
> -- 
> 2.25.1
> 

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

^ permalink raw reply	[flat|nested] 22+ messages in thread

* Re: [PATCH v4 6/6] perf tools: determine if LR is the return address
  2021-12-15 16:33     ` Mark Rutland
@ 2021-12-17 11:57       ` German Gomez
  -1 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-17 11:57 UTC (permalink / raw)
  To: Mark Rutland
  Cc: linux-kernel, linux-perf-users, acme, Alexandre Truong,
	John Garry, Will Deacon, Mathieu Poirier, Leo Yan,
	Alexander Shishkin, Jiri Olsa, Namhyung Kim, linux-arm-kernel

Hi Mark,

Thanks for your review comments

On 15/12/2021 16:33, Mark Rutland wrote:
> Hi,
>
> On Wed, Dec 15, 2021 at 03:11:38PM +0000, German Gomez wrote:
>> From: Alexandre Truong <alexandre.truong@arm.com>
>>
>> On arm64 and frame pointer mode (e.g: perf record --callgraph fp),
>> use dwarf unwind info to check if the link register is the return
>> address in order to inject it to the frame pointer stack.
> This series looks good overall, but as a general note the commit messages are a
> bit hard to read because they jump into implementation details of the patch
> (i.e. the change the patch makes) before explaining the problem (i.e. what the
> patch is trying to solve).
>
> It would be nice to have a short introduction, e.g.

Thanks for the suggestion! I'll run through the logs to see if I can
improve them.

>
>   When unwinding using frame pointers on arm64, the return address of the
>   current leaf function may be missed. The return address of a leaf function
>   may live in the LR and/or a frame record (and the location can change within
>   a function), so it is necessary to use DWARF to identify where to look for
>   the return address at any given point during a function.
>
>   For example:
>
>   unsigned long foo(unsigned long i)
>   {
>           i += 2;
> 	  i += 5;
>   }
>
>   ... could be compiled as:
>
>   foo:
>   	// return addr in LR
>   	add	x0, x0, #2
> 	// return addr in LR
> 	stp	x29, x30, [SP, #-16]!
> 	// return addr in LR
> 	mov	x29, sp
> 	// return addr in LR *and* frame record
> 	add	x0, x0, #5
> 	// return addr in LR *and* frame record
> 	ldp	x29, x30, [sp], #16
> 	// return addr in LR
> 	ret
>
>> Write the following application:
>>
>> 	int a = 10;
>>
>> 	void f2(void)
>> 	{
>> 		for (int i = 0; i < 1000000; i++)
>> 			a *= a;
>> 	}
>>
>> 	void f1()
>> 	{
>> 		for (int i = 0; i < 10; i++)
>> 			f2();
>> 	}
>>
>> 	int main(void)
>> 	{
>> 		f1();
>> 		return 0;
>> 	}
>>
>> with the following compilation flags:
>>         gcc -fno-omit-frame-pointer -fno-inline -O2
>>
>> The compiler omits the frame pointer for f2 on arm. This is a problem
>> with any leaf call, for example an application with many different
>> calls to malloc() would always omit the calling frame, even if it
>> can be determined.
> I think the wording here is slightly misleading. For f2, the compiler *doesn't
> create a frame record*, but the frame pointer (to the caller's frame record)
> remains and is not omitted.
>
> Also, I think it's woth noting (as per the example I gave above) this applies
> to *any* function which is the current leaf function, regardless of whether
> that function creates a frame record at some point. For example, if `f1` is
> interrupted before it creates its own frame record (or after it destroys the
> frame record), the FP will point at the record created by `main` (containing
> the caller of main), and `main` itself will be missing from the unwind as it
> will only exist in the LR.

I see! I hadn't considered this. I guess it's not as likely to happen
but it's worth noting indeed.

>
>> 	./perf record --call-graph fp ./a.out
>> 	./perf report
>>
>> currently gives the following stack:
>>
>> 0xffffea52f361
>> _start
>> __libc_start_main
>> main
>> f2
>>
>> After this change, perf report correctly shows f1() calling f2(),
>> even though it was missing from the frame pointer unwind:
>>
>> 	./perf report
>>
>> 0xffffea52f361
>> _start
>> __libc_start_main
>> main
>> f1
>> f2
>>
>> Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
>> Signed-off-by: German Gomez <german.gomez@arm.com>
>> ---
>>  tools/perf/util/Build                         |  1 +
>>  .../util/arm64-frame-pointer-unwind-support.c | 63 +++++++++++++++++++
>>  .../util/arm64-frame-pointer-unwind-support.h | 10 +++
>>  tools/perf/util/machine.c                     | 19 ++++--
>>  tools/perf/util/machine.h                     |  1 +
>>  5 files changed, 89 insertions(+), 5 deletions(-)
>>  create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.c
>>  create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.h
>>
>> diff --git a/tools/perf/util/Build b/tools/perf/util/Build
>> index 2e5bfbb69960..03d4c647bd86 100644
>> --- a/tools/perf/util/Build
>> +++ b/tools/perf/util/Build
>> @@ -1,3 +1,4 @@
>> +perf-y += arm64-frame-pointer-unwind-support.o
>>  perf-y += annotate.o
>>  perf-y += block-info.o
>>  perf-y += block-range.o
>> diff --git a/tools/perf/util/arm64-frame-pointer-unwind-support.c b/tools/perf/util/arm64-frame-pointer-unwind-support.c
>> new file mode 100644
>> index 000000000000..4f5ecf51ed38
>> --- /dev/null
>> +++ b/tools/perf/util/arm64-frame-pointer-unwind-support.c
>> @@ -0,0 +1,63 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +#include "arm64-frame-pointer-unwind-support.h"
>> +#include "callchain.h"
>> +#include "event.h"
>> +#include "perf_regs.h" // SMPL_REG_MASK
>> +#include "unwind.h"
>> +
>> +#define perf_event_arm_regs perf_event_arm64_regs
>> +#include "../arch/arm64/include/uapi/asm/perf_regs.h"
>> +#undef perf_event_arm_regs
>> +
>> +struct entries {
>> +	u64 stack[2];
>> +	size_t length;
>> +};
>> +
>> +static bool get_leaf_frame_caller_enabled(struct perf_sample *sample)
>> +{
>> +	return callchain_param.record_mode == CALLCHAIN_FP && sample->user_regs.regs
>> +		&& sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_LR);
>> +}
>> +
>> +static int add_entry(struct unwind_entry *entry, void *arg)
>> +{
>> +	struct entries *entries = arg;
>> +
>> +	entries->stack[entries->length++] = entry->ip;
>> +	return 0;
>> +}
>> +
>> +u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thread, int usr_idx)
>> +{
>> +	int ret;
>> +	struct entries entries = {};
>> +	struct regs_dump old_regs = sample->user_regs;
>> +
>> +	if (!get_leaf_frame_caller_enabled(sample))
>> +		return 0;
>> +
>> +	/*
>> +	 * If PC and SP are not recorded, get the value of PC from the stack
>> +	 * and set its mask. SP is not used when doing the unwinding but it
>> +	 * still needs to be set to prevent failures.
>> +	 */
> To prevent failures where? Is this something libunwind requires?

Admittedly I haven't look very deep into libunwind, but SP seems to go
ignored when getting the last 2 entries only, so here we set it to any
value.

Thanks,
German

^ permalink raw reply	[flat|nested] 22+ messages in thread

* Re: [PATCH v4 6/6] perf tools: determine if LR is the return address
@ 2021-12-17 11:57       ` German Gomez
  0 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-17 11:57 UTC (permalink / raw)
  To: Mark Rutland
  Cc: linux-kernel, linux-perf-users, acme, Alexandre Truong,
	John Garry, Will Deacon, Mathieu Poirier, Leo Yan,
	Alexander Shishkin, Jiri Olsa, Namhyung Kim, linux-arm-kernel

Hi Mark,

Thanks for your review comments

On 15/12/2021 16:33, Mark Rutland wrote:
> Hi,
>
> On Wed, Dec 15, 2021 at 03:11:38PM +0000, German Gomez wrote:
>> From: Alexandre Truong <alexandre.truong@arm.com>
>>
>> On arm64 and frame pointer mode (e.g: perf record --callgraph fp),
>> use dwarf unwind info to check if the link register is the return
>> address in order to inject it to the frame pointer stack.
> This series looks good overall, but as a general note the commit messages are a
> bit hard to read because they jump into implementation details of the patch
> (i.e. the change the patch makes) before explaining the problem (i.e. what the
> patch is trying to solve).
>
> It would be nice to have a short introduction, e.g.

Thanks for the suggestion! I'll run through the logs to see if I can
improve them.

>
>   When unwinding using frame pointers on arm64, the return address of the
>   current leaf function may be missed. The return address of a leaf function
>   may live in the LR and/or a frame record (and the location can change within
>   a function), so it is necessary to use DWARF to identify where to look for
>   the return address at any given point during a function.
>
>   For example:
>
>   unsigned long foo(unsigned long i)
>   {
>           i += 2;
> 	  i += 5;
>   }
>
>   ... could be compiled as:
>
>   foo:
>   	// return addr in LR
>   	add	x0, x0, #2
> 	// return addr in LR
> 	stp	x29, x30, [SP, #-16]!
> 	// return addr in LR
> 	mov	x29, sp
> 	// return addr in LR *and* frame record
> 	add	x0, x0, #5
> 	// return addr in LR *and* frame record
> 	ldp	x29, x30, [sp], #16
> 	// return addr in LR
> 	ret
>
>> Write the following application:
>>
>> 	int a = 10;
>>
>> 	void f2(void)
>> 	{
>> 		for (int i = 0; i < 1000000; i++)
>> 			a *= a;
>> 	}
>>
>> 	void f1()
>> 	{
>> 		for (int i = 0; i < 10; i++)
>> 			f2();
>> 	}
>>
>> 	int main(void)
>> 	{
>> 		f1();
>> 		return 0;
>> 	}
>>
>> with the following compilation flags:
>>         gcc -fno-omit-frame-pointer -fno-inline -O2
>>
>> The compiler omits the frame pointer for f2 on arm. This is a problem
>> with any leaf call, for example an application with many different
>> calls to malloc() would always omit the calling frame, even if it
>> can be determined.
> I think the wording here is slightly misleading. For f2, the compiler *doesn't
> create a frame record*, but the frame pointer (to the caller's frame record)
> remains and is not omitted.
>
> Also, I think it's woth noting (as per the example I gave above) this applies
> to *any* function which is the current leaf function, regardless of whether
> that function creates a frame record at some point. For example, if `f1` is
> interrupted before it creates its own frame record (or after it destroys the
> frame record), the FP will point at the record created by `main` (containing
> the caller of main), and `main` itself will be missing from the unwind as it
> will only exist in the LR.

I see! I hadn't considered this. I guess it's not as likely to happen
but it's worth noting indeed.

>
>> 	./perf record --call-graph fp ./a.out
>> 	./perf report
>>
>> currently gives the following stack:
>>
>> 0xffffea52f361
>> _start
>> __libc_start_main
>> main
>> f2
>>
>> After this change, perf report correctly shows f1() calling f2(),
>> even though it was missing from the frame pointer unwind:
>>
>> 	./perf report
>>
>> 0xffffea52f361
>> _start
>> __libc_start_main
>> main
>> f1
>> f2
>>
>> Signed-off-by: Alexandre Truong <alexandre.truong@arm.com>
>> Signed-off-by: German Gomez <german.gomez@arm.com>
>> ---
>>  tools/perf/util/Build                         |  1 +
>>  .../util/arm64-frame-pointer-unwind-support.c | 63 +++++++++++++++++++
>>  .../util/arm64-frame-pointer-unwind-support.h | 10 +++
>>  tools/perf/util/machine.c                     | 19 ++++--
>>  tools/perf/util/machine.h                     |  1 +
>>  5 files changed, 89 insertions(+), 5 deletions(-)
>>  create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.c
>>  create mode 100644 tools/perf/util/arm64-frame-pointer-unwind-support.h
>>
>> diff --git a/tools/perf/util/Build b/tools/perf/util/Build
>> index 2e5bfbb69960..03d4c647bd86 100644
>> --- a/tools/perf/util/Build
>> +++ b/tools/perf/util/Build
>> @@ -1,3 +1,4 @@
>> +perf-y += arm64-frame-pointer-unwind-support.o
>>  perf-y += annotate.o
>>  perf-y += block-info.o
>>  perf-y += block-range.o
>> diff --git a/tools/perf/util/arm64-frame-pointer-unwind-support.c b/tools/perf/util/arm64-frame-pointer-unwind-support.c
>> new file mode 100644
>> index 000000000000..4f5ecf51ed38
>> --- /dev/null
>> +++ b/tools/perf/util/arm64-frame-pointer-unwind-support.c
>> @@ -0,0 +1,63 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +#include "arm64-frame-pointer-unwind-support.h"
>> +#include "callchain.h"
>> +#include "event.h"
>> +#include "perf_regs.h" // SMPL_REG_MASK
>> +#include "unwind.h"
>> +
>> +#define perf_event_arm_regs perf_event_arm64_regs
>> +#include "../arch/arm64/include/uapi/asm/perf_regs.h"
>> +#undef perf_event_arm_regs
>> +
>> +struct entries {
>> +	u64 stack[2];
>> +	size_t length;
>> +};
>> +
>> +static bool get_leaf_frame_caller_enabled(struct perf_sample *sample)
>> +{
>> +	return callchain_param.record_mode == CALLCHAIN_FP && sample->user_regs.regs
>> +		&& sample->user_regs.mask & SMPL_REG_MASK(PERF_REG_ARM64_LR);
>> +}
>> +
>> +static int add_entry(struct unwind_entry *entry, void *arg)
>> +{
>> +	struct entries *entries = arg;
>> +
>> +	entries->stack[entries->length++] = entry->ip;
>> +	return 0;
>> +}
>> +
>> +u64 get_leaf_frame_caller_aarch64(struct perf_sample *sample, struct thread *thread, int usr_idx)
>> +{
>> +	int ret;
>> +	struct entries entries = {};
>> +	struct regs_dump old_regs = sample->user_regs;
>> +
>> +	if (!get_leaf_frame_caller_enabled(sample))
>> +		return 0;
>> +
>> +	/*
>> +	 * If PC and SP are not recorded, get the value of PC from the stack
>> +	 * and set its mask. SP is not used when doing the unwinding but it
>> +	 * still needs to be set to prevent failures.
>> +	 */
> To prevent failures where? Is this something libunwind requires?

Admittedly I haven't look very deep into libunwind, but SP seems to go
ignored when getting the last 2 entries only, so here we set it to any
value.

Thanks,
German

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

^ permalink raw reply	[flat|nested] 22+ messages in thread

* Re: [PATCH v4 4/6] perf tools: enable dwarf_callchain_users on arm64
  2021-12-15 16:37     ` Mark Rutland
@ 2021-12-17 12:08       ` German Gomez
  -1 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-17 12:08 UTC (permalink / raw)
  To: Mark Rutland
  Cc: linux-kernel, linux-perf-users, acme, Alexandre Truong,
	John Garry, Will Deacon, Mathieu Poirier, Leo Yan,
	Alexander Shishkin, Jiri Olsa, Namhyung Kim, linux-arm-kernel


On 15/12/2021 16:37, Mark Rutland wrote:
> On Wed, Dec 15, 2021 at 03:11:36PM +0000, German Gomez wrote:
>> [...]
>>
>> +
>> +	/*
>> +	 * It's possible to determine the caller of leaf frames with omitted
>> +	 * frame pointers on aarch64 using libunwind, so enable it.
>> +	 */
> I reckon it's worth mentioning *why* we need to do this; how about:
>
> 	/*
> 	 * It's necessary to use libunwind to reliably determine the caller of
> 	 * a leaf function on aarch64, as otherwise we cannot know whether to
> 	 * start from the LR or FP.
> 	 *
> 	 * Always starting from the LR can result in duplicate or entirely
> 	 * erroneous entries. Always skipping the LR and starting from the FP
> 	 * can result in missing entries.
> 	 */
>
> Other than that, this looks fine to me!
>
> Thanks,
> Mark.


Ack, I will update this

Thanks,
German

>
>> +	if (callchain_param.record_mode == CALLCHAIN_FP && !strcmp(arch, "arm64"))
>> +		dwarf_callchain_users = true;
>>  }
>>  
>>  static bool chain_match(struct callchain_list *base_chain,
>> diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h
>> index 77fba053c677..d95615daed73 100644
>> --- a/tools/perf/util/callchain.h
>> +++ b/tools/perf/util/callchain.h
>> @@ -300,7 +300,7 @@ int callchain_branch_counts(struct callchain_root *root,
>>  			    u64 *branch_count, u64 *predicted_count,
>>  			    u64 *abort_count, u64 *cycles_count);
>>  
>> -void callchain_param_setup(u64 sample_type);
>> +void callchain_param_setup(u64 sample_type, const char *arch);
>>  
>>  bool callchain_cnode_matched(struct callchain_node *base_cnode,
>>  			     struct callchain_node *pair_cnode);
>> -- 
>> 2.25.1
>>

^ permalink raw reply	[flat|nested] 22+ messages in thread

* Re: [PATCH v4 4/6] perf tools: enable dwarf_callchain_users on arm64
@ 2021-12-17 12:08       ` German Gomez
  0 siblings, 0 replies; 22+ messages in thread
From: German Gomez @ 2021-12-17 12:08 UTC (permalink / raw)
  To: Mark Rutland
  Cc: linux-kernel, linux-perf-users, acme, Alexandre Truong,
	John Garry, Will Deacon, Mathieu Poirier, Leo Yan,
	Alexander Shishkin, Jiri Olsa, Namhyung Kim, linux-arm-kernel


On 15/12/2021 16:37, Mark Rutland wrote:
> On Wed, Dec 15, 2021 at 03:11:36PM +0000, German Gomez wrote:
>> [...]
>>
>> +
>> +	/*
>> +	 * It's possible to determine the caller of leaf frames with omitted
>> +	 * frame pointers on aarch64 using libunwind, so enable it.
>> +	 */
> I reckon it's worth mentioning *why* we need to do this; how about:
>
> 	/*
> 	 * It's necessary to use libunwind to reliably determine the caller of
> 	 * a leaf function on aarch64, as otherwise we cannot know whether to
> 	 * start from the LR or FP.
> 	 *
> 	 * Always starting from the LR can result in duplicate or entirely
> 	 * erroneous entries. Always skipping the LR and starting from the FP
> 	 * can result in missing entries.
> 	 */
>
> Other than that, this looks fine to me!
>
> Thanks,
> Mark.


Ack, I will update this

Thanks,
German

>
>> +	if (callchain_param.record_mode == CALLCHAIN_FP && !strcmp(arch, "arm64"))
>> +		dwarf_callchain_users = true;
>>  }
>>  
>>  static bool chain_match(struct callchain_list *base_chain,
>> diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h
>> index 77fba053c677..d95615daed73 100644
>> --- a/tools/perf/util/callchain.h
>> +++ b/tools/perf/util/callchain.h
>> @@ -300,7 +300,7 @@ int callchain_branch_counts(struct callchain_root *root,
>>  			    u64 *branch_count, u64 *predicted_count,
>>  			    u64 *abort_count, u64 *cycles_count);
>>  
>> -void callchain_param_setup(u64 sample_type);
>> +void callchain_param_setup(u64 sample_type, const char *arch);
>>  
>>  bool callchain_cnode_matched(struct callchain_node *base_cnode,
>>  			     struct callchain_node *pair_cnode);
>> -- 
>> 2.25.1
>>

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

^ permalink raw reply	[flat|nested] 22+ messages in thread

end of thread, other threads:[~2021-12-17 12:10 UTC | newest]

Thread overview: 22+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-12-15 15:11 [PATCH v4 0/6] Fix missing leaf-function callers when recording German Gomez
2021-12-15 15:11 ` German Gomez
2021-12-15 15:11 ` [PATCH v4 1/6] perf tools: record ARM64 LR register automatically German Gomez
2021-12-15 15:11   ` German Gomez
2021-12-15 15:11 ` [PATCH v4 2/6] perf tools: add a mechanism to inject stack frames German Gomez
2021-12-15 15:11   ` German Gomez
2021-12-15 15:11 ` [PATCH v4 3/6] perf tools: Refactor script__setup_sample_type() German Gomez
2021-12-15 15:11   ` German Gomez
2021-12-15 15:11 ` [PATCH v4 4/6] perf tools: enable dwarf_callchain_users on arm64 German Gomez
2021-12-15 15:11   ` German Gomez
2021-12-15 16:37   ` Mark Rutland
2021-12-15 16:37     ` Mark Rutland
2021-12-17 12:08     ` German Gomez
2021-12-17 12:08       ` German Gomez
2021-12-15 15:11 ` [PATCH v4 5/6] perf tools: Refactor SMPL_REG macro in perf_regs.h German Gomez
2021-12-15 15:11   ` German Gomez
2021-12-15 15:11 ` [PATCH v4 6/6] perf tools: determine if LR is the return address German Gomez
2021-12-15 15:11   ` German Gomez
2021-12-15 16:33   ` Mark Rutland
2021-12-15 16:33     ` Mark Rutland
2021-12-17 11:57     ` German Gomez
2021-12-17 11:57       ` German Gomez

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.