Skip to contentSkip to conversation

perf machine: Add machine->has_data_mmap field

PATCH In review Namhyung Kim 3 participants 9 messages Jun 20, 2023 – Jul 12, 2023 perf events

lore
Share & tools
lore

Namhyung Kim

author maintainer
#0 lore raw

PATCH 1/3 perf machine: Add machine->has_data_mmap field

So that it can indicate the it needs to collect data mappings. This is needed especially for kernel maps.

At first, I just wanted to add it to struct machines only and to use the machine->machines to check the value. But it turned out that some of machine didn't belong to any machines (eg. some test codes), so I just copied the value to individual struct machine.

---
 tools/perf/builtin-inject.c         |  2 +-
 tools/perf/builtin-record.c         |  2 +-
 tools/perf/tests/vmlinux-kallsyms.c |  4 ++--
 tools/perf/util/machine.c           | 12 +++++++-----
 tools/perf/util/machine.h           | 10 ++++++++--
 tools/perf/util/session.c           |  4 ++--
 tools/perf/util/session.h           |  4 ++--
 7 files changed, 23 insertions(+), 15 deletions(-)
tools/perf/builtin-inject.c +1 -1
index c8cf2fdd9cff..481adaa97a68 100644
@@ -2325,7 +2325,7 @@ int cmd_inject(int argc, const char **argv)
 	}
 
 	inject.session = __perf_session__new(&data, repipe,
-					     output_fd(&inject),
+					     output_fd(&inject), false,
 					     &inject.tool);
 	if (IS_ERR(inject.session)) {
 		ret = PTR_ERR(inject.session);
tools/perf/builtin-record.c +1 -1
index aec18db7ff23..b4d0154dcb18 100644
@@ -2373,7 +2373,7 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
 		signal(SIGUSR2, SIG_IGN);
 	}
 
-	session = perf_session__new(data, tool);
+	session = __perf_session__new(data, false, -1, rec->opts.sample_address, tool);
 	if (IS_ERR(session)) {
 		pr_err("Perf session creation failed.\n");
 		return PTR_ERR(session);
tools/perf/tests/vmlinux-kallsyms.c +2 -2
index 1078a93b01aa..1f1ba3ae88aa 100644
@@ -131,8 +131,8 @@ static int test__vmlinux_matches_kallsyms(struct test_suite *test __maybe_unused
 	 * Init the machines that will hold kernel, modules obtained from
 	 * both vmlinux + .ko files and from /proc/kallsyms split by modules.
 	 */
-	machine__init(&kallsyms, "", HOST_KERNEL_ID);
-	machine__init(&vmlinux, "", HOST_KERNEL_ID);
+	machine__init(&kallsyms, "", HOST_KERNEL_ID, false);
+	machine__init(&vmlinux, "", HOST_KERNEL_ID, false);
 
 	maps = machine__kernel_maps(&vmlinux);
 
tools/perf/util/machine.c +7 -5
index 4e62843d51b7..ddc0a2130caf 100644
@@ -110,7 +110,7 @@ static void thread__set_guest_comm(struct thread *thread, pid_t pid)
 	thread__set_comm(thread, comm, 0);
 }
 
-int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
+int machine__init(struct machine *machine, const char *root_dir, pid_t pid, bool has_data_mmap)
 {
 	int err = -ENOMEM;
 
@@ -128,6 +128,7 @@ int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
 	machine->env = NULL;
 
 	machine->pid = pid;
+	machine->has_data_mmap = has_data_mmap;
 
 	machine->id_hdr_size = 0;
 	machine->kptr_restrict_warned = false;
@@ -170,7 +171,7 @@ struct machine *machine__new_host(void)
 	struct machine *machine = malloc(sizeof(*machine));
 
 	if (machine != NULL) {
-		machine__init(machine, "", HOST_KERNEL_ID);
+		machine__init(machine, "", HOST_KERNEL_ID, false);
 
 		if (machine__create_kernel_maps(machine) < 0)
 			goto out_delete;
@@ -272,10 +273,11 @@ void machine__delete(struct machine *machine)
 	}
 }
 
-void machines__init(struct machines *machines)
+void __machines__init(struct machines *machines, bool have_data_mmap)
 {
-	machine__init(&machines->host, "", HOST_KERNEL_ID);
+	machine__init(&machines->host, "", HOST_KERNEL_ID, have_data_mmap);
 	machines->guests = RB_ROOT_CACHED;
+	machines->have_data_mmaps = have_data_mmap;
 }
 
 void machines__exit(struct machines *machines)
@@ -295,7 +297,7 @@ struct machine *machines__add(struct machines *machines, pid_t pid,
 	if (machine == NULL)
 		return NULL;
 
-	if (machine__init(machine, root_dir, pid) != 0) {
+	if (machine__init(machine, root_dir, pid, machines->have_data_mmaps) != 0) {
 		free(machine);
 		return NULL;
 	}
tools/perf/util/machine.h +8 -2
index d034ecaf89c1..f54e5c888a99 100644
@@ -46,6 +46,7 @@ struct machine {
 	bool		  comm_exec;
 	bool		  kptr_restrict_warned;
 	bool		  single_address_space;
+	bool		  has_data_mmap;
 	char		  *root_dir;
 	char		  *mmap_name;
 	char		  *kallsyms_filename;
@@ -160,9 +161,14 @@ typedef void (*machine__process_t)(struct machine *machine, void *data);
 struct machines {
 	struct machine host;
 	struct rb_root_cached guests;
+	bool have_data_mmaps;
 };
 
-void machines__init(struct machines *machines);
+void __machines__init(struct machines *machines, bool data_mmap);
+static inline void machines__init(struct machines *machines)
+{
+	__machines__init(machines, false);
+}
 void machines__exit(struct machines *machines);
 
 void machines__process_guests(struct machines *machines,
@@ -181,7 +187,7 @@ void machines__set_comm_exec(struct machines *machines, bool comm_exec);
 
 struct machine *machine__new_host(void);
 struct machine *machine__new_kallsyms(void);
-int machine__init(struct machine *machine, const char *root_dir, pid_t pid);
+int machine__init(struct machine *machine, const char *root_dir, pid_t pid, bool has_data_mmap);
 void machine__exit(struct machine *machine);
 void machine__delete_threads(struct machine *machine);
 void machine__delete(struct machine *machine);
tools/perf/util/session.c +2 -2
index 00d18c74c090..e09a02ec8f28 100644
@@ -191,7 +191,7 @@ static int ordered_events__deliver_event(struct ordered_events *oe,
 }
 
 struct perf_session *__perf_session__new(struct perf_data *data,
-					 bool repipe, int repipe_fd,
+					 bool repipe, int repipe_fd, bool data_mmap,
 					 struct perf_tool *tool)
 {
 	int ret = -ENOMEM;
@@ -205,7 +205,7 @@ struct perf_session *__perf_session__new(struct perf_data *data,
 	session->decomp_data.zstd_decomp = &session->zstd_data;
 	session->active_decomp = &session->decomp_data;
 	INIT_LIST_HEAD(&session->auxtrace_index);
-	machines__init(&session->machines);
+	__machines__init(&session->machines, data_mmap);
 	ordered_events__init(&session->ordered_events,
 			     ordered_events__deliver_event, NULL);
 
tools/perf/util/session.h +2 -2
index ee3715e8563b..465610e81e95 100644
@@ -64,13 +64,13 @@ struct decomp {
 struct perf_tool;
 
 struct perf_session *__perf_session__new(struct perf_data *data,
-					 bool repipe, int repipe_fd,
+					 bool repipe, int repipe_fd, bool data_mmap,
 					 struct perf_tool *tool);
 
 static inline struct perf_session *perf_session__new(struct perf_data *data,
 						     struct perf_tool *tool)
 {
-	return __perf_session__new(data, false, -1, tool);
+	return __perf_session__new(data, false, -1, false, tool);
 }
 
 void perf_session__delete(struct perf_session *session);
Signature · 1 line
2.41.0.185.g7c58973941-goog
author maintainer
#1 lore raw

PATCH 2/3 perf tools: Add kallsyms__get_symbol_start()

The kallsyms__get_symbol_start() to get any symbol address from kallsyms. The existing kallsyms__get_function_start() only allows text symbols so create this to allow data symbols too.

---
 tools/perf/util/event.c | 30 +++++++++++++++++++++++++++---
 tools/perf/util/event.h |  2 ++
 2 files changed, 29 insertions(+), 3 deletions(-)
tools/perf/util/event.c +27 -3
index 3860b0c74829..6fdda0eb3854 100644
@@ -93,8 +93,8 @@ struct process_symbol_args {
 	u64	   start;
 };
 
-static int find_symbol_cb(void *arg, const char *name, char type,
-			  u64 start)
+static int find_func_symbol_cb(void *arg, const char *name, char type,
+			       u64 start)
 {
 	struct process_symbol_args *args = arg;
 
@@ -110,12 +110,36 @@ static int find_symbol_cb(void *arg, const char *name, char type,
 	return 1;
 }
 
+static int find_any_symbol_cb(void *arg, const char *name,
+			      char type __maybe_unused, u64 start)
+{
+	struct process_symbol_args *args = arg;
+
+	if (strcmp(name, args->name))
+		return 0;
+
+	args->start = start;
+	return 1;
+}
+
 int kallsyms__get_function_start(const char *kallsyms_filename,
 				 const char *symbol_name, u64 *addr)
 {
 	struct process_symbol_args args = { .name = symbol_name, };
 
-	if (kallsyms__parse(kallsyms_filename, &args, find_symbol_cb) <= 0)
+	if (kallsyms__parse(kallsyms_filename, &args, find_func_symbol_cb) <= 0)
+		return -1;
+
+	*addr = args.start;
+	return 0;
+}
+
+int kallsyms__get_symbol_start(const char *kallsyms_filename,
+			       const char *symbol_name, u64 *addr)
+{
+	struct process_symbol_args args = { .name = symbol_name, };
+
+	if (kallsyms__parse(kallsyms_filename, &args, find_any_symbol_cb) <= 0)
 		return -1;
 
 	*addr = args.start;
tools/perf/util/event.h +2 -0
index de20e01c9d72..d8bcee2e9b93 100644
@@ -360,6 +360,8 @@ size_t perf_event__fprintf(union perf_event *event, struct machine *machine, FIL
 
 int kallsyms__get_function_start(const char *kallsyms_filename,
 				 const char *symbol_name, u64 *addr);
+int kallsyms__get_symbol_start(const char *kallsyms_filename,
+			       const char *symbol_name, u64 *addr);
 
 void event_attr_init(struct perf_event_attr *attr);
 
Signature · 1 line
2.41.0.185.g7c58973941-goog
reviewer Acked-by on v1 2/3
#2 lore raw

Re: [PATCH 2/3] perf tools: Add kallsyms__get_symbol_start()

Quoted 6 lines · Namhyung Kim

The kallsyms__get_symbol_start() to get any symbol address from kallsyms. The existing kallsyms__get_function_start() only allows text symbols so create this to allow data symbols too.

Thanks,
Ian

Quoted 74 lines
---
 tools/perf/util/event.c | 30 +++++++++++++++++++++++++++---
 tools/perf/util/event.h |  2 ++
 2 files changed, 29 insertions(+), 3 deletions(-)
tools/perf/util/event.c +27 -3
index 3860b0c74829..6fdda0eb3854 100644
@@ -93,8 +93,8 @@ struct process_symbol_args {
        u64        start;
 };
 
-static int find_symbol_cb(void *arg, const char *name, char type,
-                         u64 start)
+static int find_func_symbol_cb(void *arg, const char *name, char type,
+                              u64 start)
 {
        struct process_symbol_args *args = arg;
 
@@ -110,12 +110,36 @@ static int find_symbol_cb(void *arg, const char *name, char type,
        return 1;
 }
 
+static int find_any_symbol_cb(void *arg, const char *name,
+                             char type __maybe_unused, u64 start)
+{
+       struct process_symbol_args *args = arg;
+
+       if (strcmp(name, args->name))
+               return 0;
+
+       args->start = start;
+       return 1;
+}
+
 int kallsyms__get_function_start(const char *kallsyms_filename,
                                 const char *symbol_name, u64 *addr)
 {
        struct process_symbol_args args = { .name = symbol_name, };
 
-       if (kallsyms__parse(kallsyms_filename, &args, find_symbol_cb) <= 0)
+       if (kallsyms__parse(kallsyms_filename, &args, find_func_symbol_cb) <= 0)
+               return -1;
+
+       *addr = args.start;
+       return 0;
+}
+
+int kallsyms__get_symbol_start(const char *kallsyms_filename,
+                              const char *symbol_name, u64 *addr)
+{
+       struct process_symbol_args args = { .name = symbol_name, };
+
+       if (kallsyms__parse(kallsyms_filename, &args, find_any_symbol_cb) <= 0)
                return -1;
 
        *addr = args.start;
tools/perf/util/event.h +2 -1
index de20e01c9d72..d8bcee2e9b93 100644
@@ -360,6 +360,8 @@ size_t perf_event__fprintf(union perf_event *event, struct machine *machine, FIL
 
 int kallsyms__get_function_start(const char *kallsyms_filename,
                                 const char *symbol_name, u64 *addr);
+int kallsyms__get_symbol_start(const char *kallsyms_filename,
+                              const char *symbol_name, u64 *addr);
 
 void event_attr_init(struct perf_event_attr *attr);
 
--

2.41.0.185.g7c58973941-goog

author maintainer
#3 lore raw

PATCH 3/3 perf machine: Include data symbols in the kernel map

In reply to Namhyung Kim

When perf record -d is used, it needs data mmaps to symbolize global data. But it missed to collect kernel data maps so it cannot symbolize them. Instead of having a separate map, just increase the kernel map size to include the data section.

Probably we can have a separate kernel map for data, but the current code assumes a single kernel map. So it'd require more changes in other places and looks error-prone. I decided not to go that way for now.

Also it seems the kernel module size already includes the data section.

For example, my system has the following.

  $ grep -e _stext -e _etext -e _edata /proc/kallsyms
  ffffffff99800000 T _stext
  ffffffff9a601ac8 T _etext
  ffffffff9b446a00 D _edata

Size of the text section is (0x9a601ac8 - 0x99800000 = 0xe01ac8) and size of the data section is (0x9b446a00 - 0x99800000 = 0x1c46a00).

Before:

  $ perf record -d true
  $ perf report -D | grep MMAP | head -1
  0 0 0x460 [0x60]: PERF_RECORD_MMAP -1/0: [0xffffffff99800000(0xe01ac8) @ 0xffffffff99800000]: x [kernel.kallsyms]_text
                                                               ^^^^^^^^
                                                                 here
After:
  $ perf report -D | grep MMAP | head -1
  0 0 0x460 [0x60]: PERF_RECORD_MMAP -1/0: [0xffffffff99800000(0x1c46a00) @ 0xffffffff99800000]: x [kernel.kallsyms]_text
                                                               ^^^^^^^^^
---
 tools/perf/util/machine.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
tools/perf/util/machine.c +4 -1
index ddc0a2130caf..e93a66f6e0b3 100644
@@ -1218,7 +1218,10 @@ static int machine__get_running_kernel_start(struct machine *machine,
 
 	*start = addr;
 
-	err = kallsyms__get_function_start(filename, "_etext", &addr);
+	if (machine->has_data_mmap)
+		err = kallsyms__get_symbol_start(filename, "_edata", &addr);
+	else
+		err = kallsyms__get_function_start(filename, "_etext", &addr);
 	if (!err)
 		*end = addr;
 
Signature · 1 line
2.41.0.185.g7c58973941-goog
reviewer Acked-by on v1 3/3
#4 lore raw

Re: [PATCH 3/3] perf machine: Include data symbols in the kernel map

Quoted 22 lines · Namhyung Kim

When perf record -d is used, it needs data mmaps to symbolize global data. But it missed to collect kernel data maps so it cannot symbolize them. Instead of having a separate map, just increase the kernel map size to include the data section.

Probably we can have a separate kernel map for data, but the current code assumes a single kernel map. So it'd require more changes in other places and looks error-prone. I decided not to go that way for now.

Also it seems the kernel module size already includes the data section.

For example, my system has the following.

  $ grep -e _stext -e _etext -e _edata /proc/kallsyms
  ffffffff99800000 T _stext
  ffffffff9a601ac8 T _etext
  ffffffff9b446a00 D _edata

Size of the text section is (0x9a601ac8 - 0x99800000 = 0xe01ac8) and size of the data section is (0x9b446a00 - 0x99800000 = 0x1c46a00).

Quoted 7 lines

Before:

  $ perf record -d true
  $ perf report -D | grep MMAP | head -1
  0 0 0x460 [0x60]: PERF_RECORD_MMAP -1/0: [0xffffffff99800000(0xe01ac8) @ 0xffffffff99800000]: x [kernel.kallsyms]_text
                                                               ^^^^^^^^
                                                                 here

nit: should the ^^^ be under 0xe01ac8?

Quoted 4 lines

After:

  $ perf report -D | grep MMAP | head -1
  0 0 0x460 [0x60]: PERF_RECORD_MMAP -1/0: [0xffffffff99800000(0x1c46a00) @ 0xffffffff99800000]: x [kernel.kallsyms]_text
                                                               ^^^^^^^^^

nit: and here under 0x1c46a00 ?

Quoted 2 lines

Thanks,
Ian

Quoted 23 lines
---
 tools/perf/util/machine.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
tools/perf/util/machine.c +4 -2
index ddc0a2130caf..e93a66f6e0b3 100644
@@ -1218,7 +1218,10 @@ static int machine__get_running_kernel_start(struct machine *machine,
 
        *start = addr;
 
-       err = kallsyms__get_function_start(filename, "_etext", &addr);
+       if (machine->has_data_mmap)
+               err = kallsyms__get_symbol_start(filename, "_edata", &addr);
+       else
+               err = kallsyms__get_function_start(filename, "_etext", &addr);
        if (!err)
                *end = addr;
 
--

2.41.0.185.g7c58973941-goog

reviewer
#5 lore raw

Re: [PATCH 3/3] perf machine: Include data symbols in the kernel map

In reply to Namhyung Kim
Quoted 43 lines · Namhyung Kim

When perf record -d is used, it needs data mmaps to symbolize global data. But it missed to collect kernel data maps so it cannot symbolize them. Instead of having a separate map, just increase the kernel map size to include the data section.

Probably we can have a separate kernel map for data, but the current code assumes a single kernel map. So it'd require more changes in other places and looks error-prone. I decided not to go that way for now.

Also it seems the kernel module size already includes the data section.

For example, my system has the following.

  $ grep -e _stext -e _etext -e _edata /proc/kallsyms
  ffffffff99800000 T _stext
  ffffffff9a601ac8 T _etext
  ffffffff9b446a00 D _edata

Size of the text section is (0x9a601ac8 - 0x99800000 = 0xe01ac8) and size of the data section is (0x9b446a00 - 0x99800000 = 0x1c46a00).

Before:

  $ perf record -d true
  $ perf report -D | grep MMAP | head -1
  0 0 0x460 [0x60]: PERF_RECORD_MMAP -1/0: [0xffffffff99800000(0xe01ac8) @ 0xffffffff99800000]: x [kernel.kallsyms]_text
                                                               ^^^^^^^^
                                                                 here
After:
  $ perf report -D | grep MMAP | head -1
  0 0 0x460 [0x60]: PERF_RECORD_MMAP -1/0: [0xffffffff99800000(0x1c46a00) @ 0xffffffff99800000]: x [kernel.kallsyms]_text
                                                               ^^^^^^^^^
---
 tools/perf/util/machine.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
tools/perf/util/machine.c +0 -0
index ddc0a2130caf..e93a66f6e0b3 100644

tools/perf/util/machine.c static int machine__get_running_kernel_start(struct machine *machine, · line 1224

Quoted 8 lines
tools/perf/util/machine.c +4 -1
@@ -1218,7 +1218,10 @@ static int machine__get_running_kernel_start(struct machine *machine,
 
 	*start = addr;
 
-	err = kallsyms__get_function_start(filename, "_etext", &addr);
+	if (machine->has_data_mmap)
+		err = kallsyms__get_symbol_start(filename, "_edata", &addr);
+	else
+		err = kallsyms__get_function_start(filename, "_etext", &addr);

What is the downside of just extending it unconditionally?

Quoted 3 lines
tools/perf/util/machine.c +0 -0
 	if (!err)
 		*end = addr;
 
author maintainer
#6 lore raw

Re: [PATCH 3/3] perf machine: Include data symbols in the kernel map

Hi Adrian,

Quoted 48 lines · Adrian Hunter
Quoted 46 lines · Namhyung Kim

When perf record -d is used, it needs data mmaps to symbolize global data. But it missed to collect kernel data maps so it cannot symbolize them. Instead of having a separate map, just increase the kernel map size to include the data section.

Probably we can have a separate kernel map for data, but the current code assumes a single kernel map. So it'd require more changes in other places and looks error-prone. I decided not to go that way for now.

Also it seems the kernel module size already includes the data section.

For example, my system has the following.

  $ grep -e _stext -e _etext -e _edata /proc/kallsyms
  ffffffff99800000 T _stext
  ffffffff9a601ac8 T _etext
  ffffffff9b446a00 D _edata

Size of the text section is (0x9a601ac8 - 0x99800000 = 0xe01ac8) and size of the data section is (0x9b446a00 - 0x99800000 = 0x1c46a00).

Before:

  $ perf record -d true
  $ perf report -D | grep MMAP | head -1
  0 0 0x460 [0x60]: PERF_RECORD_MMAP -1/0: [0xffffffff99800000(0xe01ac8) @ 0xffffffff99800000]: x [kernel.kallsyms]_text
                                                               ^^^^^^^^
                                                                 here
After:
  $ perf report -D | grep MMAP | head -1
  0 0 0x460 [0x60]: PERF_RECORD_MMAP -1/0: [0xffffffff99800000(0x1c46a00) @ 0xffffffff99800000]: x [kernel.kallsyms]_text
                                                               ^^^^^^^^^
---
 tools/perf/util/machine.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
tools/perf/util/machine.c +0 -0
index ddc0a2130caf..e93a66f6e0b3 100644
@@ -1218,7 +1218,10 @@ static int machine__get_running_kernel_start(struct machine *machine,
 
      *start = addr;
 
Quoted 7 lines

tools/perf/util/machine.c static int machine__get_running_kernel_start(struct machine *machine, · line 1224

Quoted 5 lines
tools/perf/util/machine.c +4 -1
-     err = kallsyms__get_function_start(filename, "_etext", &addr);
+     if (machine->has_data_mmap)
+             err = kallsyms__get_symbol_start(filename, "_edata", &addr);
+     else
+             err = kallsyms__get_function_start(filename, "_etext", &addr);

What is the downside of just extending it unconditionally?

I don't know.. maybe some people would argue it needs the proper protection bits other than 'x' but this patch also breaks it. But as I said, I'm not sure if we really want to change that now.

That said, we can make it unconditional. :)

Thanks,
Namhyung

Quoted 5 lines
Quoted 4 lines
tools/perf/util/machine.c +0 -0
      if (!err)
              *end = addr;
 
reviewer
#7 lore raw

Re: [PATCH 3/3] perf machine: Include data symbols in the kernel map

Quoted 59 lines · Namhyung Kim

Hi Adrian,

Quoted 48 lines · Adrian Hunter
Quoted 46 lines · Namhyung Kim

When perf record -d is used, it needs data mmaps to symbolize global data. But it missed to collect kernel data maps so it cannot symbolize them. Instead of having a separate map, just increase the kernel map size to include the data section.

Probably we can have a separate kernel map for data, but the current code assumes a single kernel map. So it'd require more changes in other places and looks error-prone. I decided not to go that way for now.

Also it seems the kernel module size already includes the data section.

For example, my system has the following.

  $ grep -e _stext -e _etext -e _edata /proc/kallsyms
  ffffffff99800000 T _stext
  ffffffff9a601ac8 T _etext
  ffffffff9b446a00 D _edata

Size of the text section is (0x9a601ac8 - 0x99800000 = 0xe01ac8) and size of the data section is (0x9b446a00 - 0x99800000 = 0x1c46a00).

Before:

  $ perf record -d true
  $ perf report -D | grep MMAP | head -1
  0 0 0x460 [0x60]: PERF_RECORD_MMAP -1/0: [0xffffffff99800000(0xe01ac8) @ 0xffffffff99800000]: x [kernel.kallsyms]_text
                                                               ^^^^^^^^
                                                                 here
After:
  $ perf report -D | grep MMAP | head -1
  0 0 0x460 [0x60]: PERF_RECORD_MMAP -1/0: [0xffffffff99800000(0x1c46a00) @ 0xffffffff99800000]: x [kernel.kallsyms]_text
                                                               ^^^^^^^^^
---
 tools/perf/util/machine.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
tools/perf/util/machine.c +0 -0
index ddc0a2130caf..e93a66f6e0b3 100644
@@ -1218,7 +1218,10 @@ static int machine__get_running_kernel_start(struct machine *machine,
 
      *start = addr;
 
Quoted 7 lines

tools/perf/util/machine.c static int machine__get_running_kernel_start(struct machine *machine, · line 1224

Quoted 5 lines
tools/perf/util/machine.c +4 -1
-     err = kallsyms__get_function_start(filename, "_etext", &addr);
+     if (machine->has_data_mmap)
+             err = kallsyms__get_symbol_start(filename, "_edata", &addr);
+     else
+             err = kallsyms__get_function_start(filename, "_etext", &addr);

What is the downside of just extending it unconditionally?

Quoted 5 lines

I don't know.. maybe some people would argue it needs the proper protection bits other than 'x' but this patch also breaks it. But as I said, I'm not sure if we really want to change that now.

That said, we can make it unconditional. :)

Might as well to start with. Will need a big comment.

Also do we know if all arch's do it like that? Perhaps need to fallback to _etext if _edata is not found?

Quoted 9 lines

Thanks,
Namhyung

Quoted 5 lines
Quoted 4 lines
tools/perf/util/machine.c +0 -0
      if (!err)
              *end = addr;
 
reviewer Acked-by on v1 1/3
#8 lore raw
In reply to Namhyung Kim
Quoted 10 lines · Namhyung Kim

So that it can indicate the it needs to collect data mappings. This is needed especially for kernel maps.

At first, I just wanted to add it to struct machines only and to use the machine->machines to check the value. But it turned out that some of machine didn't belong to any machines (eg. some test codes), so I just copied the value to individual struct machine.

Quoted 73 lines
---
 tools/perf/builtin-inject.c         |  2 +-
 tools/perf/builtin-record.c         |  2 +-
 tools/perf/tests/vmlinux-kallsyms.c |  4 ++--
 tools/perf/util/machine.c           | 12 +++++++-----
 tools/perf/util/machine.h           | 10 ++++++++--
 tools/perf/util/session.c           |  4 ++--
 tools/perf/util/session.h           |  4 ++--
 7 files changed, 23 insertions(+), 15 deletions(-)
tools/perf/builtin-inject.c +1 -1
index c8cf2fdd9cff..481adaa97a68 100644
@@ -2325,7 +2325,7 @@ int cmd_inject(int argc, const char **argv)
        }
 
        inject.session = __perf_session__new(&data, repipe,
-                                            output_fd(&inject),
+                                            output_fd(&inject), false,
                                             &inject.tool);
        if (IS_ERR(inject.session)) {
                ret = PTR_ERR(inject.session);
tools/perf/builtin-record.c +1 -1
index aec18db7ff23..b4d0154dcb18 100644
@@ -2373,7 +2373,7 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
                signal(SIGUSR2, SIG_IGN);
        }
 
-       session = perf_session__new(data, tool);
+       session = __perf_session__new(data, false, -1, rec->opts.sample_address, tool);
        if (IS_ERR(session)) {
                pr_err("Perf session creation failed.\n");
                return PTR_ERR(session);
tools/perf/tests/vmlinux-kallsyms.c +2 -2
index 1078a93b01aa..1f1ba3ae88aa 100644
@@ -131,8 +131,8 @@ static int test__vmlinux_matches_kallsyms(struct test_suite *test __maybe_unused
         * Init the machines that will hold kernel, modules obtained from
         * both vmlinux + .ko files and from /proc/kallsyms split by modules.
         */
-       machine__init(&kallsyms, "", HOST_KERNEL_ID);
-       machine__init(&vmlinux, "", HOST_KERNEL_ID);
+       machine__init(&kallsyms, "", HOST_KERNEL_ID, false);
+       machine__init(&vmlinux, "", HOST_KERNEL_ID, false);
 
        maps = machine__kernel_maps(&vmlinux);
 
tools/perf/util/machine.c +2 -1
index 4e62843d51b7..ddc0a2130caf 100644
@@ -110,7 +110,7 @@ static void thread__set_guest_comm(struct thread *thread, pid_t pid)
        thread__set_comm(thread, comm, 0);
 }
 
-int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
+int machine__init(struct machine *machine, const char *root_dir, pid_t pid, bool has_data_mmap)
 {
        int err = -ENOMEM;
 
@@ -128,6 +128,7 @@ int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
        machine->env = NULL;
 
        machine->pid = pid;
+       machine->has_data_mmap = has_data_mmap;
 
        machine->id_hdr_size = 0;
        machine->kptr_restrict_warned = false;

tools/perf/util/machine.c struct machine *machine__new_host(void) · line 174

Quoted 5 lines
tools/perf/util/machine.c +1 -1
@@ -170,7 +171,7 @@ struct machine *machine__new_host(void)
        struct machine *machine = malloc(sizeof(*machine));
 
        if (machine != NULL) {
-               machine__init(machine, "", HOST_KERNEL_ID);
+               machine__init(machine, "", HOST_KERNEL_ID, false);

nit: when passing constants it can be nice to name the parameter, so here "/*data_mmap=*/false". There are some checkers based on this like:
https://clang.llvm.org/extra/clang-tidy/checks/bugprone/argument-comment.html

Thanks,
Ian

Quoted 108 lines
tools/perf/util/machine.c +4 -3
                if (machine__create_kernel_maps(machine) < 0)
                        goto out_delete;
@@ -272,10 +273,11 @@ void machine__delete(struct machine *machine)
        }
 }
 
-void machines__init(struct machines *machines)
+void __machines__init(struct machines *machines, bool have_data_mmap)
 {
-       machine__init(&machines->host, "", HOST_KERNEL_ID);
+       machine__init(&machines->host, "", HOST_KERNEL_ID, have_data_mmap);
        machines->guests = RB_ROOT_CACHED;
+       machines->have_data_mmaps = have_data_mmap;
 }
 
 void machines__exit(struct machines *machines)
@@ -295,7 +297,7 @@ struct machine *machines__add(struct machines *machines, pid_t pid,
        if (machine == NULL)
                return NULL;
 
-       if (machine__init(machine, root_dir, pid) != 0) {
+       if (machine__init(machine, root_dir, pid, machines->have_data_mmaps) != 0) {
                free(machine);
                return NULL;
        }
tools/perf/util/machine.h +8 -2
index d034ecaf89c1..f54e5c888a99 100644
@@ -46,6 +46,7 @@ struct machine {
        bool              comm_exec;
        bool              kptr_restrict_warned;
        bool              single_address_space;
+       bool              has_data_mmap;
        char              *root_dir;
        char              *mmap_name;
        char              *kallsyms_filename;
@@ -160,9 +161,14 @@ typedef void (*machine__process_t)(struct machine *machine, void *data);
 struct machines {
        struct machine host;
        struct rb_root_cached guests;
+       bool have_data_mmaps;
 };
 
-void machines__init(struct machines *machines);
+void __machines__init(struct machines *machines, bool data_mmap);
+static inline void machines__init(struct machines *machines)
+{
+       __machines__init(machines, false);
+}
 void machines__exit(struct machines *machines);
 
 void machines__process_guests(struct machines *machines,
@@ -181,7 +187,7 @@ void machines__set_comm_exec(struct machines *machines, bool comm_exec);
 
 struct machine *machine__new_host(void);
 struct machine *machine__new_kallsyms(void);
-int machine__init(struct machine *machine, const char *root_dir, pid_t pid);
+int machine__init(struct machine *machine, const char *root_dir, pid_t pid, bool has_data_mmap);
 void machine__exit(struct machine *machine);
 void machine__delete_threads(struct machine *machine);
 void machine__delete(struct machine *machine);
tools/perf/util/session.c +2 -2
index 00d18c74c090..e09a02ec8f28 100644
@@ -191,7 +191,7 @@ static int ordered_events__deliver_event(struct ordered_events *oe,
 }
 
 struct perf_session *__perf_session__new(struct perf_data *data,
-                                        bool repipe, int repipe_fd,
+                                        bool repipe, int repipe_fd, bool data_mmap,
                                         struct perf_tool *tool)
 {
        int ret = -ENOMEM;
@@ -205,7 +205,7 @@ struct perf_session *__perf_session__new(struct perf_data *data,
        session->decomp_data.zstd_decomp = &session->zstd_data;
        session->active_decomp = &session->decomp_data;
        INIT_LIST_HEAD(&session->auxtrace_index);
-       machines__init(&session->machines);
+       __machines__init(&session->machines, data_mmap);
        ordered_events__init(&session->ordered_events,
                             ordered_events__deliver_event, NULL);
 
tools/perf/util/session.h +2 -3
index ee3715e8563b..465610e81e95 100644
@@ -64,13 +64,13 @@ struct decomp {
 struct perf_tool;
 
 struct perf_session *__perf_session__new(struct perf_data *data,
-                                        bool repipe, int repipe_fd,
+                                        bool repipe, int repipe_fd, bool data_mmap,
                                         struct perf_tool *tool);
 
 static inline struct perf_session *perf_session__new(struct perf_data *data,
                                                     struct perf_tool *tool)
 {
-       return __perf_session__new(data, false, -1, tool);
+       return __perf_session__new(data, false, -1, false, tool);
 }
 
 void perf_session__delete(struct perf_session *session);
--

2.41.0.185.g7c58973941-goog