Skip to contentSkip to conversation

per-event settings fixes

RFC In review Arnaldo Carvalho de Melo 5 participants 20 messages Jan 16, 2018 – Jan 17, 2018 perf events

lore
Share & tools
lore

Arnaldo Carvalho de Melo

author maintainer
#0 lore raw

cover per-event settings fixes

From: Arnaldo Carvalho de Melo <[email protected]>

Hi guys,

Jiri asked me to post this series, so here it is, please take a
look, I'd love to harvest your Reviewed-by/Tested-by/Acked-by,

- Arnaldo

Arnaldo Carvalho de Melo (5):
perf callchain: Fix attr.sample_max_stack setting
perf unwind: Do not look at globals
perf trace: Setup DWARF callchains for non-syscall events when --max-stack is used perf trace: Allow overriding global --max-stack per event
perf callchains: Ask for PERF_RECORD_MMAP for data mmaps for DWARF unwinding

 tools/perf/builtin-trace.c               | 19 ++++++++++++++++---
 tools/perf/util/evsel.c                  | 24 +++++++++++++++++-------
 tools/perf/util/unwind-libunwind-local.c |  9 ---------
 3 files changed, 33 insertions(+), 19 deletions(-)
Signature · 1 line
2.14.3
author maintainer
#1 lore raw

PATCH 1/5 perf callchain: Fix attr.sample_max_stack setting

From: Arnaldo Carvalho de Melo <[email protected]>

When setting the "dwarf" unwinder for a specific event and not specifying the max-stack, the attr.sample_max_stack ended up using an uninitialized callchain_param.max_stack, fix it by using designated initializers for that callchain_param variable, zeroing all non explicitely initialized struct members.

Here is what happened:

  # perf trace -vv --no-syscalls --max-stack 4 -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
  callchain: type DWARF
  callchain: stack dump size 8192
  perf_event_attr:
    type                             2
    size                             112
    config                           0x730
    { sample_period, sample_freq }   1
    sample_type                      IP|TID|TIME|ADDR|CALLCHAIN|CPU|PERIOD|RAW|REGS_USER|STACK_USER|DATA_SRC
    exclude_callchain_user           1
    { wakeup_events, wakeup_watermark } 1
    sample_regs_user                 0xff0fff
    sample_stack_user                8192
    sample_max_stack                 50656
  sys_perf_event_open failed, error -75
  Value too large for defined data type
  # perf trace -vv --no-syscalls --max-stack 4 -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
  callchain: type DWARF
  callchain: stack dump size 8192
  perf_event_attr:
    type                             2
    size                             112
    config                           0x730
    sample_type                      IP|TID|TIME|ADDR|CALLCHAIN|CPU|PERIOD|RAW|REGS_USER|STACK_USER|DATA_SRC
    exclude_callchain_user           1
    sample_regs_user                 0xff0fff
    sample_stack_user                8192
    sample_max_stack                 30448
  sys_perf_event_open failed, error -75
  Value too large for defined data type
  #

Now the attr.sample_max_stack is set to zero and the above works as expected:

  # perf trace --no-syscalls --max-stack 4 -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.072 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.072/0.072/0.072/0.000 ms

       0.000 probe_libc:inet_pton:(7feb7a998350))
                                         __inet_pton (inlined)
                                         gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)
                                         __GI_getaddrinfo (inlined)
                                         [0xffffaa39b6108f3f] (/usr/bin/ping)

#

Cc Adrian Hunter <[email protected]>
Cc David Ahern <[email protected]>
Cc Hendrick Brueckner <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Thomas Richter <[email protected]>
Cc Wang Nan <[email protected]>
Link https://lkml.kernel.org/n/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/util/evsel.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
tools/perf/util/evsel.c +4 -4
index efa2e629a669..8f971a2301d1 100644
@@ -731,14 +731,14 @@ static void apply_config_terms(struct perf_evsel *evsel,
 	struct perf_evsel_config_term *term;
 	struct list_head *config_terms = &evsel->config_terms;
 	struct perf_event_attr *attr = &evsel->attr;
-	struct callchain_param param;
+	/* callgraph default */
+	struct callchain_param param = {
+		.record_mode = callchain_param.record_mode,
+	};
 	u32 dump_size = 0;
 	int max_stack = 0;
 	const char *callgraph_buf = NULL;
 
-	/* callgraph default */
-	param.record_mode = callchain_param.record_mode;
-
 	list_for_each_entry(term, config_terms, list) {
 		switch (term->type) {
 		case PERF_EVSEL__CONFIG_TERM_PERIOD:
Signature · 1 line
2.14.3
author maintainer
#2 lore raw

PATCH 2/5 perf unwind: Do not look at globals

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

When setting up DWARF callchains on specific events, without using 'record' or 'trace' --call-graph, but instead doing it like:

perf trace -e cycles/call-graph=dwarf/

The unwind__prepare_access() call in thread__insert_map() when we process PERF_RECORD_MMAP(2) metadata events were not being performed, precluding us from using per-event DWARF callchains, handling them just when we asked for all events to be DWARF, using "--call-graph dwarf".

We do it in the PERF_RECORD_MMAP because we have to look at one of the executable maps to figure out the executable type (64-bit, 32-bit) of the DSO laid out in that mmap. Also to look at the architecture where the perf.data file was recorded.

All this probably should be deferred to when we process a sample for some thread that has callchains, so that we do this processing only for the threads with samples, not for all of them.

For now, fix using DWARF on specific events.

Before:

  # perf trace --no-syscalls -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.048 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.048/0.048/0.048/0.000 ms

     0.000 probe_libc:inet_pton:(7fe9597bb350))

Problem processing probe_libc:inet_pton callchain, skipping... #

After:

  # perf trace --no-syscalls -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.060 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.060/0.060/0.060/0.000 ms

       0.000 probe_libc:inet_pton:(7fd4aa930350))
                                         __inet_pton (inlined)
                                         gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)
                                         __GI_getaddrinfo (inlined)
                                         [0xffffaa804e51af3f] (/usr/bin/ping)
                                         __libc_start_main (/usr/lib64/libc-2.26.so)
                                         [0xffffaa804e51b379] (/usr/bin/ping)
  #
  # perf trace --call-graph=dwarf --no-syscalls -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.057 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.057/0.057/0.057/0.000 ms

       0.000 probe_libc:inet_pton:(7f9363b9e350))
                                         __inet_pton (inlined)
                                         gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)
                                         __GI_getaddrinfo (inlined)
                                         [0xffffa9e8a14e0f3f] (/usr/bin/ping)
                                         __libc_start_main (/usr/lib64/libc-2.26.so)
                                         [0xffffa9e8a14e1379] (/usr/bin/ping)
  #
  # perf trace --call-graph=fp --no-syscalls -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.077 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.077/0.077/0.077/0.000 ms

       0.000 probe_libc:inet_pton:(7f4947e1c350))
                                         __inet_pton (inlined)
                                         gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)
                                         __GI_getaddrinfo (inlined)
                                         [0xffffaa716d88ef3f] (/usr/bin/ping)
                                         __libc_start_main (/usr/lib64/libc-2.26.so)
                                         [0xffffaa716d88f379] (/usr/bin/ping)
  #
  # perf trace --no-syscalls -e probe_libc:inet_pton/call-graph=fp/ ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.078 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.078/0.078/0.078/0.000 ms

       0.000 probe_libc:inet_pton:(7fa157696350))
                                         __GI___inet_pton (/usr/lib64/libc-2.26.so)
                                         getaddrinfo (/usr/lib64/libc-2.26.so)
                                         [0xffffa9ba39c74f40] (/usr/bin/ping)

#

Cc Adrian Hunter <[email protected]>
Cc David Ahern <[email protected]>
Cc Hendrick Brueckner <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Thomas Richter <[email protected]>
Cc Wang Nan <[email protected]>
Link https://lkml.kernel.org/n/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/util/unwind-libunwind-local.c | 9 ---------
 1 file changed, 9 deletions(-)
tools/perf/util/unwind-libunwind-local.c +0 -9
index 7a42f703e858..02dc5a9d8f72 100644
@@ -631,9 +631,6 @@ static unw_accessors_t accessors = {
 
 static int _unwind__prepare_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
-		return 0;
-
 	thread->addr_space = unw_create_addr_space(&accessors, 0);
 	if (!thread->addr_space) {
 		pr_err("unwind: Can't create unwind address space.\n");

@@ -646,17 +643,11 @@ static int _unwind__prepare_access(struct thread *thread)
 
 static void _unwind__flush_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
-		return;
-
 	unw_flush_cache(thread->addr_space, 0, 0);
 }
 
 static void _unwind__finish_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
-		return;
-
 	unw_destroy_addr_space(thread->addr_space);
 }
Signature · 1 line
2.14.3
#3 lore raw

Re: [PATCH 2/5] perf unwind: Do not look at globals

On Tue, Jan 16, 2018 at 11:24:35AM -0300, Arnaldo Carvalho de Melo wrote:

SNIP

Quoted 18 lines
Cc Adrian Hunter <[email protected]>
Cc David Ahern <[email protected]>
Cc Hendrick Brueckner <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Thomas Richter <[email protected]>
Cc Wang Nan <[email protected]>
Link https://lkml.kernel.org/n/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/util/unwind-libunwind-local.c | 9 ---------
 1 file changed, 9 deletions(-)
tools/perf/util/unwind-libunwind-local.c +0 -0
index 7a42f703e858..02dc5a9d8f72 100644

tools/perf/util/unwind-libunwind-local.c static unw_accessors_t accessors = { · line 636

Quoted 6 lines
tools/perf/util/unwind-libunwind-local.c +0 -3
@@ -631,9 +631,6 @@ static unw_accessors_t accessors = {
 
 static int _unwind__prepare_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
-		return 0;
-

this would create thread->addr_space also for data without dwarf callchains data, so I think we need to keep it

it should get set in apply_config_terms which calls parse_callchain_record once it detects some 'call-graph' term setup.. something's probably wrong there?

jirka

Quoted 24 lines
tools/perf/util/unwind-libunwind-local.c +0 -6
 	thread->addr_space = unw_create_addr_space(&accessors, 0);
 	if (!thread->addr_space) {
 		pr_err("unwind: Can't create unwind address space.\n");

@@ -646,17 +643,11 @@ static int _unwind__prepare_access(struct thread *thread)
 
 static void _unwind__flush_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
-		return;
-
 	unw_flush_cache(thread->addr_space, 0, 0);
 }
 
 static void _unwind__finish_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
-		return;
-
 	unw_destroy_addr_space(thread->addr_space);
 }
Signature · 1 line
2.14.3
author maintainer
#4 lore raw

Re: [PATCH 2/5] perf unwind: Do not look at globals

Em Tue, Jan 16, 2018 at 04:19:15PM +0100, Jiri Olsa escreveu:

Quoted 32 lines · Arnaldo Carvalho de Melo

SNIP

Quoted 18 lines
Cc Adrian Hunter <[email protected]>
Cc David Ahern <[email protected]>
Cc Hendrick Brueckner <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Thomas Richter <[email protected]>
Cc Wang Nan <[email protected]>
Link https://lkml.kernel.org/n/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/util/unwind-libunwind-local.c | 9 ---------
 1 file changed, 9 deletions(-)
tools/perf/util/unwind-libunwind-local.c +0 -0
index 7a42f703e858..02dc5a9d8f72 100644

tools/perf/util/unwind-libunwind-local.c static unw_accessors_t accessors = { · line 636

Quoted 6 lines
tools/perf/util/unwind-libunwind-local.c +0 -3
@@ -631,9 +631,6 @@ static unw_accessors_t accessors = {
 
 static int _unwind__prepare_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
-		return 0;
-

this would create thread->addr_space also for data without dwarf callchains data, so I think we need to keep it

Quoted 1 line

it should get set in apply_config_terms which calls parse_callchain_record

No, it should not set the global parameter, as this is just for a specific event, i.e. we can have something like:

	perf record -e cycles/call-graph=dwarf/ \
		    -e instructions/call-graph=lbr/
		    -e cache-misses/call-graph=fp

And then get these samples for the same thread.

If you want to avoid creating thread->addr_space, and we do want that, sure, a followup patch should address that, we need to postpone allocating it till we get a DWARF callchain in a sample for that specific thread, when we then should allocate thread->addr_space.

But then we need to break that prepare routine, the part that needs the map needs to set something in the 'struct thread' to mark what kind of unwind ops should be used if we ever get a sample with a DWARF callchain for that thread later on the perf.data (or live session), and when that happens, look if the thread->addr_space is allocated and allocate it if not.

Quoted 2 lines

once it detects some 'call-graph' term setup.. something's probably wrong there?

See above. The fix in this patch is the quickest, i.e. we make sure that if we ever find DWARF callchains, what we need will be there. We could introduce a new global variable that would be touched by apply_config_terms() now, and touch that, not the global config, that may be used by other code, that would think that hey, DWARF is globally configured, when it is just by some of the events.

But if we do that, we will waste some space anyway since not all threads will have DWARF callchains, e.g. in a system wide session.

- Arnaldo

Quoted 26 lines

jirka

Quoted 24 lines
tools/perf/util/unwind-libunwind-local.c +0 -6
 	thread->addr_space = unw_create_addr_space(&accessors, 0);
 	if (!thread->addr_space) {
 		pr_err("unwind: Can't create unwind address space.\n");

@@ -646,17 +643,11 @@ static int _unwind__prepare_access(struct thread *thread)
 
 static void _unwind__flush_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
-		return;
-
 	unw_flush_cache(thread->addr_space, 0, 0);
 }
 
 static void _unwind__finish_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
-		return;
-
 	unw_destroy_addr_space(thread->addr_space);
 }
Signature · 1 line
2.14.3
author maintainer
#5 lore raw

Re: [PATCH 2/5] perf unwind: Do not look at globals

Em Tue, Jan 16, 2018 at 12:36:21PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 11 lines

Em Tue, Jan 16, 2018 at 04:19:15PM +0100, Jiri Olsa escreveu:

Quoted 1 line

it should get set in apply_config_terms which calls parse_callchain_record

No, it should not set the global parameter, as this is just for a specific event, i.e. we can have something like:

	perf record -e cycles/call-graph=dwarf/ \
		    -e instructions/call-graph=lbr/
		    -e cache-misses/call-graph=fp

And then get these samples for the same thread.

<SNIP>

Quoted 9 lines
Quoted 2 lines

once it detects some 'call-graph' term setup.. something's probably wrong there?

See above. The fix in this patch is the quickest, i.e. we make sure that if we ever find DWARF callchains, what we need will be there. We could introduce a new global variable that would be touched by apply_config_terms() now, and touch that, not the global config, that may be used by other code, that would think that hey, DWARF is globally configured, when it is just by some of the events.

So, look at the patch at the end of this message, better (for now), replacing this one (2/5).

I retested everything, same results.

Quoted 2 lines

But if we do that, we will waste some space anyway since not all threads will have DWARF callchains, e.g. in a system wide session.

- Arnaldo

commit 8a302012fabdb387ab45a0d0b283411a8fd05b32

Author: Arnaldo Carvalho de Melo <[email protected]>
Date:   Mon Jan 15 16:48:46 2018 -0300
    perf unwind: Do not look just at the global callchain_param.record_mode
    
    When setting up DWARF callchains on specific events, without using
    'record' or 'trace' --call-graph, but instead doing it like:
    
            perf trace -e cycles/call-graph=dwarf/
    
    The unwind__prepare_access() call in thread__insert_map() when we
    process PERF_RECORD_MMAP(2) metadata events were not being performed,
    precluding us from using per-event DWARF callchains, handling them just
    when we asked for all events to be DWARF, using "--call-graph dwarf".
    
    We do it in the PERF_RECORD_MMAP because we have to look at one of the
    executable maps to figure out the executable type (64-bit, 32-bit) of
    the DSO laid out in that mmap. Also to look at the architecture where
    the perf.data file was recorded.
    
    All this probably should be deferred to when we process a sample for
    some thread that has callchains, so that we do this processing only for
    the threads with samples, not for all of them.
    
    For now, fix using DWARF on specific events.
    
    Before:
    
      # perf trace --no-syscalls -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
      PING ::1(::1) 56 data bytes
      64 bytes from ::1: icmp_seq=1 ttl=64 time=0.048 ms
    
      --- ::1 ping statistics ---
      1 packets transmitted, 1 received, 0% packet loss, time 0ms
      rtt min/avg/max/mdev = 0.048/0.048/0.048/0.000 ms
         0.000 probe_libc:inet_pton:(7fe9597bb350))
      Problem processing probe_libc:inet_pton callchain, skipping...
      #
    
    After:
    
      # perf trace --no-syscalls -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
      PING ::1(::1) 56 data bytes
      64 bytes from ::1: icmp_seq=1 ttl=64 time=0.060 ms
    
      --- ::1 ping statistics ---
      1 packets transmitted, 1 received, 0% packet loss, time 0ms
      rtt min/avg/max/mdev = 0.060/0.060/0.060/0.000 ms
           0.000 probe_libc:inet_pton:(7fd4aa930350))
                                             __inet_pton (inlined)
                                             gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)
                                             __GI_getaddrinfo (inlined)
                                             [0xffffaa804e51af3f] (/usr/bin/ping)
                                             __libc_start_main (/usr/lib64/libc-2.26.so)
                                             [0xffffaa804e51b379] (/usr/bin/ping)
      #
      # perf trace --call-graph=dwarf --no-syscalls -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
      PING ::1(::1) 56 data bytes
      64 bytes from ::1: icmp_seq=1 ttl=64 time=0.057 ms
    
      --- ::1 ping statistics ---
      1 packets transmitted, 1 received, 0% packet loss, time 0ms
      rtt min/avg/max/mdev = 0.057/0.057/0.057/0.000 ms
           0.000 probe_libc:inet_pton:(7f9363b9e350))
                                             __inet_pton (inlined)
                                             gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)
                                             __GI_getaddrinfo (inlined)
                                             [0xffffa9e8a14e0f3f] (/usr/bin/ping)
                                             __libc_start_main (/usr/lib64/libc-2.26.so)
                                             [0xffffa9e8a14e1379] (/usr/bin/ping)
      #
      # perf trace --call-graph=fp --no-syscalls -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
      PING ::1(::1) 56 data bytes
      64 bytes from ::1: icmp_seq=1 ttl=64 time=0.077 ms
    
      --- ::1 ping statistics ---
      1 packets transmitted, 1 received, 0% packet loss, time 0ms
      rtt min/avg/max/mdev = 0.077/0.077/0.077/0.000 ms
           0.000 probe_libc:inet_pton:(7f4947e1c350))
                                             __inet_pton (inlined)
                                             gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)
                                             __GI_getaddrinfo (inlined)
                                             [0xffffaa716d88ef3f] (/usr/bin/ping)
                                             __libc_start_main (/usr/lib64/libc-2.26.so)
                                             [0xffffaa716d88f379] (/usr/bin/ping)
      #
      # perf trace --no-syscalls -e probe_libc:inet_pton/call-graph=fp/ ping -6 -c 1 ::1
      PING ::1(::1) 56 data bytes
      64 bytes from ::1: icmp_seq=1 ttl=64 time=0.078 ms
    
      --- ::1 ping statistics ---
      1 packets transmitted, 1 received, 0% packet loss, time 0ms
      rtt min/avg/max/mdev = 0.078/0.078/0.078/0.000 ms
           0.000 probe_libc:inet_pton:(7fa157696350))
                                             __GI___inet_pton (/usr/lib64/libc-2.26.so)
                                             getaddrinfo (/usr/lib64/libc-2.26.so)
                                             [0xffffa9ba39c74f40] (/usr/bin/ping)
      #
    
    Cc: Adrian Hunter <[email protected]>
    Cc: David Ahern <[email protected]>
    Cc: Hendrick Brueckner <[email protected]>
    Cc: Jiri Olsa <[email protected]>
    Cc: Namhyung Kim <[email protected]>
    Cc: Thomas Richter <[email protected]>
    Cc: Wang Nan <[email protected]>
    Link: https://lkml.kernel.org/n/[email protected]
    Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>
tools/perf/builtin-c2c.c +3 -2
index c0debc3f79b6..c0815a37fdb5 100644
@@ -2390,9 +2390,10 @@ static int setup_callchain(struct perf_evlist *evlist)
 	enum perf_call_graph_mode mode = CALLCHAIN_NONE;
 
 	if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-	    (sample_type & PERF_SAMPLE_STACK_USER))
+	    (sample_type & PERF_SAMPLE_STACK_USER)) {
 		mode = CALLCHAIN_DWARF;
-	else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+		dwarf_callchain_users = true;
+	} else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
 		mode = CALLCHAIN_LBR;
 	else if (sample_type & PERF_SAMPLE_CALLCHAIN)
 		mode = CALLCHAIN_FP;
tools/perf/builtin-report.c +3 -2
index dd4df9a5cd06..6593779224d5 100644
@@ -338,9 +338,10 @@ static int report__setup_sample_type(struct report *rep)
 
 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-		    (sample_type & PERF_SAMPLE_STACK_USER))
+		    (sample_type & PERF_SAMPLE_STACK_USER)) {
 			callchain_param.record_mode = CALLCHAIN_DWARF;
-		else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+			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;
tools/perf/builtin-script.c +3 -2
index c1cce474c0f1..08bc818f371b 100644
@@ -2919,9 +2919,10 @@ static void script__setup_sample_type(struct perf_script *script)
 
 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-		    (sample_type & PERF_SAMPLE_STACK_USER))
+		    (sample_type & PERF_SAMPLE_STACK_USER)) {
 			callchain_param.record_mode = CALLCHAIN_DWARF;
-		else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+			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;
tools/perf/tests/dwarf-unwind.c +1 -0
index ac40e05bcab4..260418969120 100644
@@ -173,6 +173,7 @@ int test__dwarf_unwind(struct test *test __maybe_unused, int subtest __maybe_unu
 	}
 
 	callchain_param.record_mode = CALLCHAIN_DWARF;
+	dwarf_callchain_users = true;
 
 	if (init_live_machine(machine)) {
 		pr_err("Could not init machine\n");
tools/perf/util/callchain.c +10 -0
index 082505d08d72..32ef7bdca1cf 100644
@@ -37,6 +37,15 @@ struct callchain_param callchain_param = {
 	CALLCHAIN_PARAM_DEFAULT
 };
 
+/*
+ * Are there any events usind DWARF callchains?
+ *
+ * I.e.
+ *
+ * -e cycles/call-graph=dwarf/
+ */
+bool dwarf_callchain_users;
+
 struct callchain_param callchain_param_default = {
 	CALLCHAIN_PARAM_DEFAULT
 };

@@ -265,6 +274,7 @@ int parse_callchain_record(const char *arg, struct callchain_param *param)
 			ret = 0;
 			param->record_mode = CALLCHAIN_DWARF;
 			param->dump_size = default_stack_dump_size;
+			dwarf_callchain_users = true;
 
 			tok = strtok_r(NULL, ",", &saveptr);
 			if (tok) {
tools/perf/util/callchain.h +2 -0
index b79ef2478a57..154560b1eb65 100644
@@ -89,6 +89,8 @@ enum chain_value {
 	CCVAL_COUNT,
 };
 
+extern bool dwarf_callchain_users;
+
 struct callchain_param {
 	bool			enabled;
 	enum perf_call_graph_mode record_mode;
tools/perf/util/unwind-libunwind-local.c +3 -6
index 7a42f703e858..af873044d33a 100644
@@ -631,9 +631,8 @@ static unw_accessors_t accessors = {
 
 static int _unwind__prepare_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
+	if (!dwarf_callchain_users)
 		return 0;
-
 	thread->addr_space = unw_create_addr_space(&accessors, 0);
 	if (!thread->addr_space) {
 		pr_err("unwind: Can't create unwind address space.\n");

@@ -646,17 +645,15 @@ static int _unwind__prepare_access(struct thread *thread)
 
 static void _unwind__flush_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
+	if (!dwarf_callchain_users)
 		return;
-
 	unw_flush_cache(thread->addr_space, 0, 0);
 }
 
 static void _unwind__finish_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
+	if (!dwarf_callchain_users)
 		return;
-
 	unw_destroy_addr_space(thread->addr_space);
 }
#6 lore raw

Re: [PATCH 2/5] perf unwind: Do not look at globals

On Tue, Jan 16, 2018 at 03:26:50PM -0300, Arnaldo Carvalho de Melo wrote:

SNIP

Quoted 69 lines
tools/perf/builtin-c2c.c +3 -2
index c0debc3f79b6..c0815a37fdb5 100644
@@ -2390,9 +2390,10 @@ static int setup_callchain(struct perf_evlist *evlist)
 	enum perf_call_graph_mode mode = CALLCHAIN_NONE;
 
 	if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-	    (sample_type & PERF_SAMPLE_STACK_USER))
+	    (sample_type & PERF_SAMPLE_STACK_USER)) {
 		mode = CALLCHAIN_DWARF;
-	else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+		dwarf_callchain_users = true;
+	} else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
 		mode = CALLCHAIN_LBR;
 	else if (sample_type & PERF_SAMPLE_CALLCHAIN)
 		mode = CALLCHAIN_FP;
tools/perf/builtin-report.c +3 -2
index dd4df9a5cd06..6593779224d5 100644
@@ -338,9 +338,10 @@ static int report__setup_sample_type(struct report *rep)
 
 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-		    (sample_type & PERF_SAMPLE_STACK_USER))
+		    (sample_type & PERF_SAMPLE_STACK_USER)) {
 			callchain_param.record_mode = CALLCHAIN_DWARF;
-		else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+			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;
tools/perf/builtin-script.c +3 -2
index c1cce474c0f1..08bc818f371b 100644
@@ -2919,9 +2919,10 @@ static void script__setup_sample_type(struct perf_script *script)
 
 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-		    (sample_type & PERF_SAMPLE_STACK_USER))
+		    (sample_type & PERF_SAMPLE_STACK_USER)) {
 			callchain_param.record_mode = CALLCHAIN_DWARF;
-		else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+			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;
tools/perf/tests/dwarf-unwind.c +1 -0
index ac40e05bcab4..260418969120 100644
@@ -173,6 +173,7 @@ int test__dwarf_unwind(struct test *test __maybe_unused, int subtest __maybe_unu
 	}
 
 	callchain_param.record_mode = CALLCHAIN_DWARF;
+	dwarf_callchain_users = true;
 
 	if (init_live_machine(machine)) {
 		pr_err("Could not init machine\n");
tools/perf/util/callchain.c +0 -0
index 082505d08d72..32ef7bdca1cf 100644
@@ -37,6 +37,15 @@ struct callchain_param callchain_param = {
 	CALLCHAIN_PARAM_DEFAULT

tools/perf/util/callchain.c struct callchain_param callchain_param = { · line 47

Quoted 10 lines
tools/perf/util/callchain.c +8 -0
 };
 
+/*
+ * Are there any events usind DWARF callchains?
+ *
+ * I.e.
+ *
+ * -e cycles/call-graph=dwarf/
+ */
+bool dwarf_callchain_users;

hum, I don't follow.. this bool seems to mirror the usage of 'param->record_mode = CALLCHAIN_DWARF', whats the difference?

also, the patch title says 'Do not look at globals', while inside you add new global dwarf_callchain_users and work with it.. what do I miss?

I'll check tomorrow with clean head ;-)

jirka

author maintainer
#7 lore raw

Re: [PATCH 2/5] perf unwind: Do not look at globals

Em Tue, Jan 16, 2018 at 08:49:09PM +0100, Jiri Olsa escreveu:

Quoted 87 lines · Arnaldo Carvalho de Melo

SNIP

Quoted 69 lines
tools/perf/builtin-c2c.c +3 -2
index c0debc3f79b6..c0815a37fdb5 100644
@@ -2390,9 +2390,10 @@ static int setup_callchain(struct perf_evlist *evlist)
 	enum perf_call_graph_mode mode = CALLCHAIN_NONE;
 
 	if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-	    (sample_type & PERF_SAMPLE_STACK_USER))
+	    (sample_type & PERF_SAMPLE_STACK_USER)) {
 		mode = CALLCHAIN_DWARF;
-	else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+		dwarf_callchain_users = true;
+	} else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
 		mode = CALLCHAIN_LBR;
 	else if (sample_type & PERF_SAMPLE_CALLCHAIN)
 		mode = CALLCHAIN_FP;
tools/perf/builtin-report.c +3 -2
index dd4df9a5cd06..6593779224d5 100644
@@ -338,9 +338,10 @@ static int report__setup_sample_type(struct report *rep)
 
 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-		    (sample_type & PERF_SAMPLE_STACK_USER))
+		    (sample_type & PERF_SAMPLE_STACK_USER)) {
 			callchain_param.record_mode = CALLCHAIN_DWARF;
-		else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+			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;
tools/perf/builtin-script.c +3 -2
index c1cce474c0f1..08bc818f371b 100644
@@ -2919,9 +2919,10 @@ static void script__setup_sample_type(struct perf_script *script)
 
 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-		    (sample_type & PERF_SAMPLE_STACK_USER))
+		    (sample_type & PERF_SAMPLE_STACK_USER)) {
 			callchain_param.record_mode = CALLCHAIN_DWARF;
-		else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+			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;
tools/perf/tests/dwarf-unwind.c +1 -0
index ac40e05bcab4..260418969120 100644
@@ -173,6 +173,7 @@ int test__dwarf_unwind(struct test *test __maybe_unused, int subtest __maybe_unu
 	}
 
 	callchain_param.record_mode = CALLCHAIN_DWARF;
+	dwarf_callchain_users = true;
 
 	if (init_live_machine(machine)) {
 		pr_err("Could not init machine\n");
tools/perf/util/callchain.c +0 -0
index 082505d08d72..32ef7bdca1cf 100644
@@ -37,6 +37,15 @@ struct callchain_param callchain_param = {
 	CALLCHAIN_PARAM_DEFAULT

tools/perf/util/callchain.c struct callchain_param callchain_param = { · line 47

Quoted 10 lines
tools/perf/util/callchain.c +8 -0
 };
 
+/*
+ * Are there any events usind DWARF callchains?
+ *
+ * I.e.
+ *
+ * -e cycles/call-graph=dwarf/
+ */
+bool dwarf_callchain_users;

hum, I don't follow.. this bool seems to mirror the usage of 'param->record_mode = CALLCHAIN_DWARF', whats the difference?

Quoted 1 line

also, the patch title says 'Do not look at globals', while inside you

The first version didn't look at globals, the second one doesn't look at an _specific_ global variable, the global config for --call-graph, which is a global variable, callchain_param, which _we_ can't touch at apply_config_terms(), since that is about _just_ that event, not all of them.

Quoted 3 lines

add new global dwarf_callchain_users and work with it.. what do I miss?

I'll check tomorrow with clean head ;-)

Look closely at apply_config_terms() it passes a _local_ variable to

                        perf_evsel__config_callchain(evsel, opts, &param);

It will not affect any globals that tools/perf/util/unwind-libunwind-local.c could possibly use... and that is the problem. :-)

The right fix, as I said, is more involved and may allow us to remove these two global variables, both callchain_param and dwarf_callchain_users.

We need to have per-evsel unwind ops, per thread addr_space continues to be used by the dwarf unwinder _for the events sampled in that thread_, etc.

The prepare_unwind is to be made to evsel and thread (for thread we need to look at one of its executable maps, to determine if it is 32-bit or 64-bit, etc, but not necessarily at that insert_map part, etc).

- Arnaldo

maintainer
#8 lore raw

Re: [PATCH 2/5] perf unwind: Do not look at globals

Hi Arnaldo,

Quoted 88 lines · Arnaldo Carvalho de Melo

Em Tue, Jan 16, 2018 at 08:49:09PM +0100, Jiri Olsa escreveu:

Quoted 87 lines · Arnaldo Carvalho de Melo

SNIP

Quoted 69 lines
tools/perf/builtin-c2c.c +3 -2
index c0debc3f79b6..c0815a37fdb5 100644
@@ -2390,9 +2390,10 @@ static int setup_callchain(struct perf_evlist *evlist)
 	enum perf_call_graph_mode mode = CALLCHAIN_NONE;
 
 	if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-	    (sample_type & PERF_SAMPLE_STACK_USER))
+	    (sample_type & PERF_SAMPLE_STACK_USER)) {
 		mode = CALLCHAIN_DWARF;
-	else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+		dwarf_callchain_users = true;
+	} else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
 		mode = CALLCHAIN_LBR;
 	else if (sample_type & PERF_SAMPLE_CALLCHAIN)
 		mode = CALLCHAIN_FP;
tools/perf/builtin-report.c +3 -2
index dd4df9a5cd06..6593779224d5 100644
@@ -338,9 +338,10 @@ static int report__setup_sample_type(struct report *rep)
 
 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-		    (sample_type & PERF_SAMPLE_STACK_USER))
+		    (sample_type & PERF_SAMPLE_STACK_USER)) {
 			callchain_param.record_mode = CALLCHAIN_DWARF;
-		else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+			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;
tools/perf/builtin-script.c +3 -2
index c1cce474c0f1..08bc818f371b 100644
@@ -2919,9 +2919,10 @@ static void script__setup_sample_type(struct perf_script *script)
 
 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-		    (sample_type & PERF_SAMPLE_STACK_USER))
+		    (sample_type & PERF_SAMPLE_STACK_USER)) {
 			callchain_param.record_mode = CALLCHAIN_DWARF;
-		else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+			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;
tools/perf/tests/dwarf-unwind.c +1 -0
index ac40e05bcab4..260418969120 100644
@@ -173,6 +173,7 @@ int test__dwarf_unwind(struct test *test __maybe_unused, int subtest __maybe_unu
 	}
 
 	callchain_param.record_mode = CALLCHAIN_DWARF;
+	dwarf_callchain_users = true;
 
 	if (init_live_machine(machine)) {
 		pr_err("Could not init machine\n");
tools/perf/util/callchain.c +0 -0
index 082505d08d72..32ef7bdca1cf 100644
@@ -37,6 +37,15 @@ struct callchain_param callchain_param = {
 	CALLCHAIN_PARAM_DEFAULT

tools/perf/util/callchain.c struct callchain_param callchain_param = { · line 47

Quoted 10 lines
tools/perf/util/callchain.c +8 -0
 };
 
+/*
+ * Are there any events usind DWARF callchains?
+ *
+ * I.e.
+ *
+ * -e cycles/call-graph=dwarf/
+ */
+bool dwarf_callchain_users;

hum, I don't follow.. this bool seems to mirror the usage of 'param->record_mode = CALLCHAIN_DWARF', whats the difference?

Quoted 7 lines
Quoted 1 line

also, the patch title says 'Do not look at globals', while inside you

The first version didn't look at globals, the second one doesn't look at an _specific_ global variable, the global config for --call-graph, which is a global variable, callchain_param, which _we_ can't touch at apply_config_terms(), since that is about _just_ that event, not all of them.

Right, we need to call the prepare routine when any of event requires DWARF unwind even though the global callchain_param is FP, for example.

Quoted 16 lines
Quoted 4 lines

add new global dwarf_callchain_users and work with it.. what do I miss?

I'll check tomorrow with clean head ;-)

Look closely at apply_config_terms() it passes a _local_ variable to

                        perf_evsel__config_callchain(evsel, opts, &param);

It will not affect any globals that tools/perf/util/unwind-libunwind-local.c could possibly use... and that is the problem. :-)

The right fix, as I said, is more involved and may allow us to remove these two global variables, both callchain_param and dwarf_callchain_users.

Quoted 7 lines

We need to have per-evsel unwind ops, per thread addr_space continues to be used by the dwarf unwinder _for the events sampled in that thread_, etc.

The prepare_unwind is to be made to evsel and thread (for thread we need to look at one of its executable maps, to determine if it is 32-bit or 64-bit, etc, but not necessarily at that insert_map part, etc).

Yep, but I'm ok with the proposed solution right now.

Thanks,
Namhyung

#9 lore raw

Re: [PATCH 2/5] perf unwind: Do not look at globals

In reply to Arnaldo Carvalho de Melo

On Tue, Jan 16, 2018 at 05:05:20PM -0300, Arnaldo Carvalho de Melo wrote:

SNIP

Quoted 23 lines
Quoted 13 lines

tools/perf/util/callchain.c struct callchain_param callchain_param = { · line 47

Quoted 9 lines
tools/perf/util/callchain.c +8 -0
+/*
+ * Are there any events usind DWARF callchains?
+ *
+ * I.e.
+ *
+ * -e cycles/call-graph=dwarf/
+ */
+bool dwarf_callchain_users;

hum, I don't follow.. this bool seems to mirror the usage of 'param->record_mode = CALLCHAIN_DWARF', whats the difference?

Quoted 1 line

also, the patch title says 'Do not look at globals', while inside you

The first version didn't look at globals, the second one doesn't look at an _specific_ global variable, the global config for --call-graph, which is a global variable, callchain_param, which _we_ can't touch at apply_config_terms(), since that is about _just_ that event, not all of them.

Quoted 2 lines

add new global dwarf_callchain_users and work with it.. what do I miss?

Quoted 8 lines
Quoted 1 line

I'll check tomorrow with clean head ;-)

Look closely at apply_config_terms() it passes a _local_ variable to

                        perf_evsel__config_callchain(evsel, opts, &param);

It will not affect any globals that tools/perf/util/unwind-libunwind-local.c could possibly use... and that is the problem. :-)

ah ok, that one is local.. just to set up the attrs

then adding that new variable is ok, though it could be aded inside struct callchain_param, to keep callchain config stuff together

thanks,
jirka

merge bot
#10 lore raw

[tip:perf/core] perf unwind: Do not look just at the global callchain_param.record_mode

In reply to Arnaldo Carvalho de Melo
perf unwind: Do not look just at the global callchain_param.record_mode
Commit-ID:  eabad8c6856f185f876b54c426c2cc69fe0f0a7d
Gitweb:     https://git.kernel.org/tip/eabad8c6856f185f876b54c426c2cc69fe0f0a7d
Author:     Arnaldo Carvalho de Melo <[email protected]>
AuthorDate: Mon, 15 Jan 2018 16:48:46 -0300
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Wed, 17 Jan 2018 10:23:32 -0300

perf unwind: Do not look just at the global callchain_param.record_mode

When setting up DWARF callchains on specific events, without using 'record' or 'trace' --call-graph, but instead doing it like:

perf trace -e cycles/call-graph=dwarf/

The unwind__prepare_access() call in thread__insert_map() when we process PERF_RECORD_MMAP(2) metadata events were not being performed, precluding us from using per-event DWARF callchains, handling them just when we asked for all events to be DWARF, using "--call-graph dwarf".

We do it in the PERF_RECORD_MMAP because we have to look at one of the executable maps to figure out the executable type (64-bit, 32-bit) of the DSO laid out in that mmap. Also to look at the architecture where the perf.data file was recorded.

All this probably should be deferred to when we process a sample for some thread that has callchains, so that we do this processing only for the threads with samples, not for all of them.

For now, fix using DWARF on specific events.

Before:

  # perf trace --no-syscalls -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.048 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.048/0.048/0.048/0.000 ms

     0.000 probe_libc:inet_pton:(7fe9597bb350))

Problem processing probe_libc:inet_pton callchain, skipping... #

After:

  # perf trace --no-syscalls -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.060 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.060/0.060/0.060/0.000 ms

       0.000 probe_libc:inet_pton:(7fd4aa930350))
                                         __inet_pton (inlined)
                                         gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)
                                         __GI_getaddrinfo (inlined)
                                         [0xffffaa804e51af3f] (/usr/bin/ping)
                                         __libc_start_main (/usr/lib64/libc-2.26.so)
                                         [0xffffaa804e51b379] (/usr/bin/ping)
  #
  # perf trace --call-graph=dwarf --no-syscalls -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.057 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.057/0.057/0.057/0.000 ms

       0.000 probe_libc:inet_pton:(7f9363b9e350))
                                         __inet_pton (inlined)
                                         gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)
                                         __GI_getaddrinfo (inlined)
                                         [0xffffa9e8a14e0f3f] (/usr/bin/ping)
                                         __libc_start_main (/usr/lib64/libc-2.26.so)
                                         [0xffffa9e8a14e1379] (/usr/bin/ping)
  #
  # perf trace --call-graph=fp --no-syscalls -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.077 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.077/0.077/0.077/0.000 ms

       0.000 probe_libc:inet_pton:(7f4947e1c350))
                                         __inet_pton (inlined)
                                         gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)
                                         __GI_getaddrinfo (inlined)
                                         [0xffffaa716d88ef3f] (/usr/bin/ping)
                                         __libc_start_main (/usr/lib64/libc-2.26.so)
                                         [0xffffaa716d88f379] (/usr/bin/ping)
  #
  # perf trace --no-syscalls -e probe_libc:inet_pton/call-graph=fp/ ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.078 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.078/0.078/0.078/0.000 ms

       0.000 probe_libc:inet_pton:(7fa157696350))
                                         __GI___inet_pton (/usr/lib64/libc-2.26.so)
                                         getaddrinfo (/usr/lib64/libc-2.26.so)
                                         [0xffffa9ba39c74f40] (/usr/bin/ping)

#

Acked-by Namhyung Kim <[email protected]>
Cc Adrian Hunter <[email protected]>
Cc David Ahern <[email protected]>
Cc Hendrick Brueckner <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Thomas Richter <[email protected]>
Cc Wang Nan <[email protected]>
Link https://lkml.kernel.org/r/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/builtin-c2c.c                 |  5 +++--
 tools/perf/builtin-report.c              |  5 +++--
 tools/perf/builtin-script.c              |  5 +++--
 tools/perf/tests/dwarf-unwind.c          |  1 +
 tools/perf/util/callchain.c              | 10 ++++++++++
 tools/perf/util/callchain.h              |  2 ++
 tools/perf/util/unwind-libunwind-local.c |  9 +++------
 7 files changed, 25 insertions(+), 12 deletions(-)
tools/perf/builtin-c2c.c +3 -2
index c0debc3..c0815a3 100644
@@ -2390,9 +2390,10 @@ static int setup_callchain(struct perf_evlist *evlist)
 	enum perf_call_graph_mode mode = CALLCHAIN_NONE;
 
 	if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-	    (sample_type & PERF_SAMPLE_STACK_USER))
+	    (sample_type & PERF_SAMPLE_STACK_USER)) {
 		mode = CALLCHAIN_DWARF;
-	else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+		dwarf_callchain_users = true;
+	} else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
 		mode = CALLCHAIN_LBR;
 	else if (sample_type & PERF_SAMPLE_CALLCHAIN)
 		mode = CALLCHAIN_FP;
tools/perf/builtin-report.c +3 -2
index dd4df9a..6593779 100644
@@ -338,9 +338,10 @@ static int report__setup_sample_type(struct report *rep)
 
 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-		    (sample_type & PERF_SAMPLE_STACK_USER))
+		    (sample_type & PERF_SAMPLE_STACK_USER)) {
 			callchain_param.record_mode = CALLCHAIN_DWARF;
-		else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+			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;
tools/perf/builtin-script.c +3 -2
index c1cce47..08bc818 100644
@@ -2919,9 +2919,10 @@ static void script__setup_sample_type(struct perf_script *script)
 
 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
-		    (sample_type & PERF_SAMPLE_STACK_USER))
+		    (sample_type & PERF_SAMPLE_STACK_USER)) {
 			callchain_param.record_mode = CALLCHAIN_DWARF;
-		else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+			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;
tools/perf/tests/dwarf-unwind.c +1 -0
index ac40e05..26041896 100644
@@ -173,6 +173,7 @@ int test__dwarf_unwind(struct test *test __maybe_unused, int subtest __maybe_unu
 	}
 
 	callchain_param.record_mode = CALLCHAIN_DWARF;
+	dwarf_callchain_users = true;
 
 	if (init_live_machine(machine)) {
 		pr_err("Could not init machine\n");
tools/perf/util/callchain.c +10 -0
index 082505d0..32ef7bd 100644
@@ -37,6 +37,15 @@ struct callchain_param callchain_param = {
 	CALLCHAIN_PARAM_DEFAULT
 };
 
+/*
+ * Are there any events usind DWARF callchains?
+ *
+ * I.e.
+ *
+ * -e cycles/call-graph=dwarf/
+ */
+bool dwarf_callchain_users;
+
 struct callchain_param callchain_param_default = {
 	CALLCHAIN_PARAM_DEFAULT
 };

@@ -265,6 +274,7 @@ int parse_callchain_record(const char *arg, struct callchain_param *param)
 			ret = 0;
 			param->record_mode = CALLCHAIN_DWARF;
 			param->dump_size = default_stack_dump_size;
+			dwarf_callchain_users = true;
 
 			tok = strtok_r(NULL, ",", &saveptr);
 			if (tok) {
tools/perf/util/callchain.h +2 -0
index b79ef24..154560b 100644
@@ -89,6 +89,8 @@ enum chain_value {
 	CCVAL_COUNT,
 };
 
+extern bool dwarf_callchain_users;
+
 struct callchain_param {
 	bool			enabled;
 	enum perf_call_graph_mode record_mode;
tools/perf/util/unwind-libunwind-local.c +3 -6
index 7a42f70..af87304 100644
@@ -631,9 +631,8 @@ static unw_accessors_t accessors = {
 
 static int _unwind__prepare_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
+	if (!dwarf_callchain_users)
 		return 0;
-
 	thread->addr_space = unw_create_addr_space(&accessors, 0);
 	if (!thread->addr_space) {
 		pr_err("unwind: Can't create unwind address space.\n");

@@ -646,17 +645,15 @@ static int _unwind__prepare_access(struct thread *thread)
 
 static void _unwind__flush_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
+	if (!dwarf_callchain_users)
 		return;
-
 	unw_flush_cache(thread->addr_space, 0, 0);
 }
 
 static void _unwind__finish_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
+	if (!dwarf_callchain_users)
 		return;
-
 	unw_destroy_addr_space(thread->addr_space);
 }
#11 lore raw

Re: [PATCH 2/5] perf unwind: Do not look at globals

In reply to Arnaldo Carvalho de Melo
Quoted 42 lines · Arnaldo Carvalho de Melo

Em Tue, Jan 16, 2018 at 04:19:15PM +0100, Jiri Olsa escreveu:

Quoted 32 lines · Arnaldo Carvalho de Melo

SNIP

Quoted 18 lines
Cc Adrian Hunter <[email protected]>
Cc David Ahern <[email protected]>
Cc Hendrick Brueckner <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Thomas Richter <[email protected]>
Cc Wang Nan <[email protected]>
Link https://lkml.kernel.org/n/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/util/unwind-libunwind-local.c | 9 ---------
 1 file changed, 9 deletions(-)
tools/perf/util/unwind-libunwind-local.c +0 -0
index 7a42f703e858..02dc5a9d8f72 100644

tools/perf/util/unwind-libunwind-local.c static unw_accessors_t accessors = { · line 636

Quoted 6 lines
tools/perf/util/unwind-libunwind-local.c +0 -3
@@ -631,9 +631,6 @@ static unw_accessors_t accessors = {
 
 static int _unwind__prepare_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
-		return 0;
-

this would create thread->addr_space also for data without dwarf callchains data, so I think we need to keep it

Quoted 1 line

it should get set in apply_config_terms which calls parse_callchain_record

No, it should not set the global parameter, as this is just for a specific event, i.e. we can have something like:

	perf record -e cycles/call-graph=dwarf/ \
		    -e instructions/call-graph=lbr/
		    -e cache-misses/call-graph=fp
Quoted 6 lines

And then get these samples for the same thread.

If you want to avoid creating thread->addr_space, and we do want that, sure, a followup patch should address that, we need to postpone allocating it till we get a DWARF callchain in a sample for that specific thread, when we then should allocate thread->addr_space.

yea that could be fixed.. however current code prevents to create that for data without DWARF data, and your change forces it

I guess that's addressed in your next email.. going to read it ;-)

jirka

author maintainer
#12 lore raw

Re: [PATCH 2/5] perf unwind: Do not look at globals

Em Tue, Jan 16, 2018 at 08:30:35PM +0100, Jiri Olsa escreveu:

Quoted 45 lines · Arnaldo Carvalho de Melo
Quoted 44 lines

Em Tue, Jan 16, 2018 at 04:19:15PM +0100, Jiri Olsa escreveu:

Quoted 32 lines · Arnaldo Carvalho de Melo

SNIP

Quoted 18 lines
Cc Adrian Hunter <[email protected]>
Cc David Ahern <[email protected]>
Cc Hendrick Brueckner <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Thomas Richter <[email protected]>
Cc Wang Nan <[email protected]>
Link https://lkml.kernel.org/n/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/util/unwind-libunwind-local.c | 9 ---------
 1 file changed, 9 deletions(-)
tools/perf/util/unwind-libunwind-local.c +0 -0
index 7a42f703e858..02dc5a9d8f72 100644

tools/perf/util/unwind-libunwind-local.c static unw_accessors_t accessors = { · line 636

Quoted 6 lines
tools/perf/util/unwind-libunwind-local.c +0 -3
@@ -631,9 +631,6 @@ static unw_accessors_t accessors = {
 
 static int _unwind__prepare_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
-		return 0;
-

this would create thread->addr_space also for data without dwarf callchains data, so I think we need to keep it

Quoted 1 line

it should get set in apply_config_terms which calls parse_callchain_record

No, it should not set the global parameter, as this is just for a specific event, i.e. we can have something like:

	perf record -e cycles/call-graph=dwarf/ \
		    -e instructions/call-graph=lbr/
		    -e cache-misses/call-graph=fp

And then get these samples for the same thread.

Quoted 8 lines
Quoted 4 lines

If you want to avoid creating thread->addr_space, and we do want that, sure, a followup patch should address that, we need to postpone allocating it till we get a DWARF callchain in a sample for that specific thread, when we then should allocate thread->addr_space.

yea that could be fixed.. however current code prevents to create that for data without DWARF data, and your change forces it

Humm? Which code?

[root@jouet ~]# perf record -e cycles/call-graph=dwarf/ -e instructions/call-graph=lbr/ -e cache-misses/call-graph=fp/ sleep 2
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.140 MB perf.data (43 samples) ]
[root@jouet ~]# perf evlist -v
cycles/call-graph=dwarf/: size: 112, { sample_period, sample_freq }: 4000, sample_type: IP|TID|TIME|ADDR|CALLCHAIN|PERIOD|REGS_USER|STACK_USER|IDENTIFIER|DATA_SRC, read_format: ID, disabled: 1, inherit: 1, mmap: 1, comm: 1, freq: 1, enable_on_exec: 1, task: 1, mmap_data: 1, sample_id_all: 1, exclude_guest: 1, exclude_callchain_user: 1, mmap2: 1, comm_exec: 1, sample_regs_user: 0xff0fff, sample_stack_user: 8192 instructions/call-graph=lbr/: size: 112, config: 0x1, { sample_period, sample_freq }: 4000, sample_type: IP|TID|TIME|CALLCHAIN|PERIOD|BRANCH_STACK|IDENTIFIER, read_format: ID, disabled: 1, inherit: 1, freq: 1, enable_on_exec: 1, sample_id_all: 1, exclude_guest: 1, branch_sample_type: USER|CALL_STACK|NO_FLAGS|NO_CYCLES cache-misses/call-graph=fp/: size: 112, config: 0x3, { sample_period, sample_freq }: 4000, sample_type: IP|TID|TIME|CALLCHAIN|PERIOD|IDENTIFIER, read_format: ID, disabled: 1, inherit: 1, freq: 1, enable_on_exec: 1, sample_id_all: 1, exclude_guest: 1
[root@jouet ~]# perf report --no-header | head -20
no symbols found in /usr/bin/sleep, maybe install a debug package?
# To display the perf.data header info, please use --header/--header-only options.
#
#
# Total Lost Samples: 0
#
# Samples: 15 of event 'cycles/call-graph=dwarf/'
# Event count (approx.): 2153016
#

# Children      Self  Command  Shared Object     Symbol                                 
# ........  ........  .......  ................  .......................................
#
    48.98%     0.00%  sleep    ld-2.26.so        [.] _start
            |
            ---_start
               |          
               |--35.15%--_dl_start
               |          _dl_sysdep_start
               |          dl_main
               |          |          
               |          |--18.44%--_dl_relocate_object

[root@jouet ~]#

- Arnaldo

#13 lore raw

Re: [PATCH 2/5] perf unwind: Do not look at globals

Quoted 51 lines · Arnaldo Carvalho de Melo

Em Tue, Jan 16, 2018 at 08:30:35PM +0100, Jiri Olsa escreveu:

Quoted 50 lines · Arnaldo Carvalho de Melo
Quoted 48 lines

Em Tue, Jan 16, 2018 at 04:19:15PM +0100, Jiri Olsa escreveu:

Quoted 32 lines · Arnaldo Carvalho de Melo

SNIP

Quoted 18 lines
Cc Adrian Hunter <[email protected]>
Cc David Ahern <[email protected]>
Cc Hendrick Brueckner <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Thomas Richter <[email protected]>
Cc Wang Nan <[email protected]>
Link https://lkml.kernel.org/n/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/util/unwind-libunwind-local.c | 9 ---------
 1 file changed, 9 deletions(-)
tools/perf/util/unwind-libunwind-local.c +0 -0
index 7a42f703e858..02dc5a9d8f72 100644

tools/perf/util/unwind-libunwind-local.c static unw_accessors_t accessors = { · line 636

Quoted 6 lines
tools/perf/util/unwind-libunwind-local.c +0 -3
@@ -631,9 +631,6 @@ static unw_accessors_t accessors = {
 
 static int _unwind__prepare_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
-		return 0;
-

this would create thread->addr_space also for data without dwarf callchains data, so I think we need to keep it

Quoted 1 line

it should get set in apply_config_terms which calls parse_callchain_record

No, it should not set the global parameter, as this is just for a specific event, i.e. we can have something like:

	perf record -e cycles/call-graph=dwarf/ \
		    -e instructions/call-graph=lbr/
		    -e cache-misses/call-graph=fp

And then get these samples for the same thread.

If you want to avoid creating thread->addr_space, and we do want that, sure, a followup patch should address that, we need to postpone allocating it till we get a DWARF callchain in a sample for that specific thread, when we then should allocate thread->addr_space.

Quoted 5 lines
Quoted 3 lines

yea that could be fixed.. however current code prevents to create that for data without DWARF data, and your change forces it

Humm? Which code?

these bits

jirka

---

tools/perf/util/unwind-libunwind-local.c +0 -9
index 7a42f703e858..02dc5a9d8f72 100644
@@ -631,9 +631,6 @@ static unw_accessors_t accessors = {
 
 static int _unwind__prepare_access(struct thread *thread)
 {
-       if (callchain_param.record_mode != CALLCHAIN_DWARF)
-               return 0;
-
        thread->addr_space = unw_create_addr_space(&accessors, 0);
        if (!thread->addr_space) {
                pr_err("unwind: Can't create unwind address space.\n");

@@ -646,17 +643,11 @@ static int _unwind__prepare_access(struct thread *thread)
 
 static void _unwind__flush_access(struct thread *thread)
 {
-       if (callchain_param.record_mode != CALLCHAIN_DWARF)
-               return;
-
        unw_flush_cache(thread->addr_space, 0, 0);
 }
 
 static void _unwind__finish_access(struct thread *thread)
 {
-       if (callchain_param.record_mode != CALLCHAIN_DWARF)
-               return;
-
        unw_destroy_addr_space(thread->addr_space);
 }
author maintainer
#14 lore raw

Re: [PATCH 2/5] perf unwind: Do not look at globals

Em Tue, Jan 16, 2018 at 08:55:55PM +0100, Jiri Olsa escreveu:

Quoted 52 lines · Arnaldo Carvalho de Melo
Quoted 51 lines

Em Tue, Jan 16, 2018 at 08:30:35PM +0100, Jiri Olsa escreveu:

Quoted 50 lines · Arnaldo Carvalho de Melo
Quoted 48 lines

Em Tue, Jan 16, 2018 at 04:19:15PM +0100, Jiri Olsa escreveu:

Quoted 32 lines · Arnaldo Carvalho de Melo

SNIP

Quoted 18 lines
Cc Adrian Hunter <[email protected]>
Cc David Ahern <[email protected]>
Cc Hendrick Brueckner <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Thomas Richter <[email protected]>
Cc Wang Nan <[email protected]>
Link https://lkml.kernel.org/n/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/util/unwind-libunwind-local.c | 9 ---------
 1 file changed, 9 deletions(-)
tools/perf/util/unwind-libunwind-local.c +0 -0
index 7a42f703e858..02dc5a9d8f72 100644

tools/perf/util/unwind-libunwind-local.c static unw_accessors_t accessors = { · line 636

Quoted 6 lines
tools/perf/util/unwind-libunwind-local.c +0 -3
@@ -631,9 +631,6 @@ static unw_accessors_t accessors = {
 
 static int _unwind__prepare_access(struct thread *thread)
 {
-	if (callchain_param.record_mode != CALLCHAIN_DWARF)
-		return 0;
-

this would create thread->addr_space also for data without dwarf callchains data, so I think we need to keep it

Quoted 1 line

it should get set in apply_config_terms which calls parse_callchain_record

No, it should not set the global parameter, as this is just for a specific event, i.e. we can have something like:

	perf record -e cycles/call-graph=dwarf/ \
		    -e instructions/call-graph=lbr/
		    -e cache-misses/call-graph=fp

And then get these samples for the same thread.

If you want to avoid creating thread->addr_space, and we do want that, sure, a followup patch should address that, we need to postpone allocating it till we get a DWARF callchain in a sample for that specific thread, when we then should allocate thread->addr_space.

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

yea that could be fixed.. however current code prevents to create that for data without DWARF data, and your change forces it

Humm? Which code?

these bits

Those are replaced already, based on your comments.

Quoted 35 lines

jirka

---

tools/perf/util/unwind-libunwind-local.c +0 -9
index 7a42f703e858..02dc5a9d8f72 100644
@@ -631,9 +631,6 @@ static unw_accessors_t accessors = {
 
 static int _unwind__prepare_access(struct thread *thread)
 {
-       if (callchain_param.record_mode != CALLCHAIN_DWARF)
-               return 0;
-
        thread->addr_space = unw_create_addr_space(&accessors, 0);
        if (!thread->addr_space) {
                pr_err("unwind: Can't create unwind address space.\n");

@@ -646,17 +643,11 @@ static int _unwind__prepare_access(struct thread *thread)
 
 static void _unwind__flush_access(struct thread *thread)
 {
-       if (callchain_param.record_mode != CALLCHAIN_DWARF)
-               return;
-
        unw_flush_cache(thread->addr_space, 0, 0);
 }
 
 static void _unwind__finish_access(struct thread *thread)
 {
-       if (callchain_param.record_mode != CALLCHAIN_DWARF)
-               return;
-
        unw_destroy_addr_space(thread->addr_space);
 }
author maintainer
#15 lore raw

PATCH 3/5 perf trace: Setup DWARF callchains for non-syscall events when --max-stack is used

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

If we use:

	perf trace --max-stack=4

then the syscall events will use DWARF callchains, when available (libunwind enabled in the build) and the printing will stop at 4 levels.

When we introduced support for tracepoint events this ended up not applying for them, fix it.

Before:

  # perf trace --call-graph=dwarf --no-syscalls -e probe_libc:inet_pton ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.058 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.058/0.058/0.058/0.000 ms

       0.000 probe_libc:inet_pton:(7fc6c2a16350))

#

After:

  # perf trace --call-graph=dwarf --no-syscalls -e probe_libc:inet_pton ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.087 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.087/0.087/0.087/0.000 ms

       0.000 probe_libc:inet_pton:(7fbf9a041350))
                                         __inet_pton (inlined)
                                         gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)
                                         __GI_getaddrinfo (inlined)
                                         [0xffffaa947cb67f3f] (/usr/bin/ping)
                                         __libc_start_main (/usr/lib64/libc-2.26.so)
                                         [0xffffaa947cb68379] (/usr/bin/ping)

#

Reported-by Thomas Richter <[email protected]>
Cc Adrian Hunter <[email protected]>
Cc David Ahern <[email protected]>
Cc Hendrick Brueckner <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Wang Nan <[email protected]>
Link https://lkml.kernel.org/n/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/builtin-trace.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
tools/perf/builtin-trace.c +3 -2
index 0362974854e9..ee85c29dbf70 100644
@@ -2350,7 +2350,7 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
 		goto out_delete_evlist;
 	}
 
-	perf_evlist__config(evlist, &trace->opts, NULL);
+	perf_evlist__config(evlist, &trace->opts, &callchain_param);
 
 	signal(SIGCHLD, sig_handler);
 	signal(SIGINT, sig_handler);

@@ -3065,8 +3065,9 @@ int cmd_trace(int argc, const char **argv)
 	}
 
 #ifdef HAVE_DWARF_UNWIND_SUPPORT
-	if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled && trace.trace_syscalls)
+	if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
 		record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
+	}
 #endif
 
 	if (callchain_param.enabled) {
Signature · 1 line
2.14.3
author maintainer
#16 lore raw

PATCH 4/5 perf trace: Allow overriding global --max-stack per event

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

The per-event max-stack setting wasn't overriding the global --max-stack setting:

  # perf trace --no-syscalls --max-stack 4 -e probe_libc:inet_pton/call-graph=dwarf,max-stack=2/ ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.072 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.072/0.072/0.072/0.000 ms

       0.000 probe_libc:inet_pton:(7feb7a998350))
                                         __inet_pton (inlined)
                                         gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)
                                         __GI_getaddrinfo (inlined)
                                         [0xffffaa39b6108f3f] (/usr/bin/ping)

#

Fix it:

  # perf trace --no-syscalls --max-stack 4 -e probe_libc:inet_pton/call-graph=dwarf,max-stack=2/ ping -6 -c 1 ::1
  PING ::1(::1) 56 data bytes
  64 bytes from ::1: icmp_seq=1 ttl=64 time=0.073 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.073/0.073/0.073/0.000 ms

       0.000 probe_libc:inet_pton:(7f1083221350))
                                         __inet_pton (inlined)
                                         gaih_inet.constprop.7 (/usr/lib64/libc-2.26.so)

#

Cc Adrian Hunter <[email protected]>
Cc David Ahern <[email protected]>
Cc Hendrick Brueckner <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Thomas Richter <[email protected]>
Cc Wang Nan <[email protected]>
Link https://lkml.kernel.org/n/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/builtin-trace.c | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)
tools/perf/builtin-trace.c +13 -1
index ee85c29dbf70..531d43bf57e1 100644
@@ -1644,7 +1644,7 @@ static int trace__resolve_callchain(struct trace *trace, struct perf_evsel *evse
 	struct addr_location al;
 
 	if (machine__resolve(trace->host, &al, sample) < 0 ||
-	    thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, trace->max_stack))
+	    thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, evsel->attr.sample_max_stack))
 		return -1;
 
 	return 0;

@@ -2423,6 +2423,18 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
 	trace->multiple_threads = thread_map__pid(evlist->threads, 0) == -1 ||
 				  evlist->threads->nr > 1 ||
 				  perf_evlist__first(evlist)->attr.inherit;
+
+	/*
+	 * Now that we already used evsel->attr to ask the kernel to setup the
+	 * events, lets reuse evsel->attr.sample_max_stack as the limit in
+	 * trace__resolve_callchain(), allowing per-event max-stack settings
+	 * to override an explicitely set --max-stack global setting.
+	 */
+	evlist__for_each_entry(evlist, evsel) {
+		if ((evsel->attr.sample_type & PERF_SAMPLE_CALLCHAIN) &&
+		    evsel->attr.sample_max_stack == 0)
+			evsel->attr.sample_max_stack = trace->max_stack;
+	}
 again:
 	before = trace->nr_events;
Signature · 1 line
2.14.3
author maintainer
#17 lore raw

PATCH 5/5 perf callchains: Ask for PERF_RECORD_MMAP for data mmaps for DWARF unwinding

In reply to Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

When we use a global DWARF setting as in:

	perf record --call-graph dwarf

According to 5c0cf22477ea ("perf record: Store data mmaps for dwarf unwind") we need to set up some extra perf_event_attr bits.

But when we instead do a per event dwarf setting:

perf record -e cycles/call-graph=dwarf/

This was not being done, make them equivalent.

This didn't produce any output changes in my tests while fixing up loose ends in the per-event settings, I found it just by comparing the perf_event_attr fields trying to find an explanation for those problems.

Cc Adrian Hunter <[email protected]>
Cc David Ahern <[email protected]>
Cc Hendrick Brueckner <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Noel Grandin <[email protected]>
Cc Thomas Richter <[email protected]>
Cc Wang Nan <[email protected]>
Link https://lkml.kernel.org/n/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/util/evsel.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)
tools/perf/util/evsel.c +13 -3
index 8f971a2301d1..85eb84dfdf91 100644
@@ -726,7 +726,7 @@ perf_evsel__reset_callgraph(struct perf_evsel *evsel,
 }
 
 static void apply_config_terms(struct perf_evsel *evsel,
-			       struct record_opts *opts)
+			       struct record_opts *opts, bool track)
 {
 	struct perf_evsel_config_term *term;
 	struct list_head *config_terms = &evsel->config_terms;

@@ -797,6 +797,8 @@ static void apply_config_terms(struct perf_evsel *evsel,
 
 	/* User explicitly set per-event callgraph, clear the old setting and reset. */
 	if ((callgraph_buf != NULL) || (dump_size > 0) || max_stack) {
+		bool sample_address = false;
+
 		if (max_stack) {
 			param.max_stack = max_stack;
 			if (callgraph_buf == NULL)

@@ -816,6 +818,8 @@ static void apply_config_terms(struct perf_evsel *evsel,
 					       evsel->name);
 					return;
 				}
+				if (param.record_mode == CALLCHAIN_DWARF)
+					sample_address = true;
 			}
 		}
 		if (dump_size > 0) {

@@ -828,8 +832,14 @@ static void apply_config_terms(struct perf_evsel *evsel,
 			perf_evsel__reset_callgraph(evsel, &callchain_param);
 
 		/* set perf-event callgraph */
-		if (param.enabled)
+		if (param.enabled) {
+			if (sample_address) {
+				perf_evsel__set_sample_bit(evsel, ADDR);
+				perf_evsel__set_sample_bit(evsel, DATA_SRC);
+				evsel->attr.mmap_data = track;
+			}
 			perf_evsel__config_callchain(evsel, opts, &param);
+		}
 	}
 }
 

@@ -1060,7 +1070,7 @@ void perf_evsel__config(struct perf_evsel *evsel, struct record_opts *opts,
 	 * Apply event specific term settings,
 	 * it overloads any global configuration.
 	 */
-	apply_config_terms(evsel, opts);
+	apply_config_terms(evsel, opts, track);
 
 	evsel->ignore_missing_thread = opts->ignore_missing_thread;
 }
Signature · 1 line
2.14.3
#18 lore raw
In reply to Arnaldo Carvalho de Melo
Quoted 16 lines · Arnaldo Carvalho de Melo

From: Arnaldo Carvalho de Melo <[email protected]>

Hi guys,

Jiri asked me to post this series, so here it is, please take a
look, I'd love to harvest your Reviewed-by/Tested-by/Acked-by,

- Arnaldo

Arnaldo Carvalho de Melo (5):
perf callchain: Fix attr.sample_max_stack setting
perf unwind: Do not look at globals
perf trace: Setup DWARF callchains for non-syscall events when --max-stack is used perf trace: Allow overriding global --max-stack per event
perf callchains: Ask for PERF_RECORD_MMAP for data mmaps for DWARF unwinding

Quoted 5 lines
 tools/perf/builtin-trace.c               | 19 ++++++++++++++++---
 tools/perf/util/evsel.c                  | 24 +++++++++++++++++-------
 tools/perf/util/unwind-libunwind-local.c |  9 ---------
 3 files changed, 33 insertions(+), 19 deletions(-)

I have tested your patches on my s390x. I applied
them on top of your perf/core branch.

Here is the output:
[root@s8360047 perf]# ./perf trace --no-syscalls

        -e probe_libc:inet_pton ping -6 -c 1 ::1

PING ::1(::1) 56 data bytes
64 bytes from ::1: icmp_seq=1 ttl=64 time=0.046 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.046/0.046/0.046/0.000 ms

     0.000 probe_libc:inet_pton:(3ffa4ec2060))

[root@s8360047 perf]#

Test ok !

[root@s8360047 perf]# ./perf trace --no-syscalls

        -e probe_libc:inet_pton/max-stack=5,call-graph=dwarf/ ping -6 -c 1 ::1

PING ::1(::1) 56 data bytes
64 bytes from ::1: icmp_seq=1 ttl=64 time=0.054 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.054/0.054/0.054/0.000 ms

     0.000 probe_libc:inet_pton:(3ff7e742060))
           __GI___inet_pton (/usr/lib64/libc-2.26.so)
           gaih_inet (inlined)
           __GI_getaddrinfo (inlined)
           main (/usr/bin/ping)
           __libc_start_main (/usr/lib64/libc-2.26.so)

[root@s8360047 perf]#

Test ok!

[root@s8360047 perf]# ./perf trace --no-syscalls

        -e probe_libc:inet_pton/call-graph=dwarf/ ping -6 -c 1 ::1

PING ::1(::1) 56 data bytes
64 bytes from ::1: icmp_seq=1 ttl=64 time=0.062 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.062/0.062/0.062/0.000 ms

     0.000 probe_libc:inet_pton:(3ff86642060))
           __GI___inet_pton (/usr/lib64/libc-2.26.so)
           gaih_inet (inlined)
           __GI_getaddrinfo (inlined)
           main (/usr/bin/ping)
           __libc_start_main (/usr/lib64/libc-2.26.so)
           _start (/usr/bin/ping)

[root@s8360047 perf]#

Test ok:

[root@s8360047 perf]# ./perf trace --no-syscalls --call-graph dwarf

        --max-stack 4  -e probe_libc:inet_pton ping -6 -c 1 ::1

PING ::1(::1) 56 data bytes
64 bytes from ::1: icmp_seq=1 ttl=64 time=0.031 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.031/0.031/0.031/0.000 ms

     0.000 probe_libc:inet_pton:(3ffa1bc2060))
           __GI___inet_pton (/usr/lib64/libc-2.26.so)
           gaih_inet (inlined)
           __GI_getaddrinfo (inlined)
           main (/usr/bin/ping)

[root@s8360047 perf]#

Test ok, also when --max-stack 4 is omitted the complete backchain is printed.

[root@s8360047 perf]# ./perf trace --no-syscalls --call-graph dwarf

        -e probe_libc:inet_pton/max-stack=3/ ping -6 -c 1 ::1

PING ::1(::1) 56 data bytes
64 bytes from ::1: icmp_seq=1 ttl=64 time=0.047 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.047/0.047/0.047/0.000 ms

     0.000 probe_libc:inet_pton:(3ffb4bc2060))

[root@s8360047 perf]#

Test fails, I would have expected some callchain output.

[root@s8360047 perf]# ./perf trace --no-syscalls --call-graph dwarf --max-stack 4

        -e probe_libc:inet_pton/max-stack=3/ ping -6 -c 1 ::1

PING ::1(::1) 56 data bytes
64 bytes from ::1: icmp_seq=1 ttl=64 time=0.055 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.055/0.055/0.055/0.000 ms

     0.000 probe_libc:inet_pton:(3ff7e6c2060))

[root@s8360047 perf]#

Test fails, I would have expected some callchain output.

It looks like /max-stack=3/ without ,call-graph=dwarf does not trigger the stack unwinding.

Signature · 5 lines
Thomas Richter, Dept 3303, IBM LTC Boeblingen Germany
--
Vorsitzende des Aufsichtsrats: Martina Koederitz 
Geschäftsführung: Dirk Wittkopp
Sitz der Gesellschaft: Böblingen / Registergericht: Amtsgericht Stuttgart, HRB 243294
author maintainer
#19 lore raw

Em Tue, Jan 16, 2018 at 04:27:52PM +0100, Thomas-Mich Richter escreveu:

Quoted 16 lines

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.047/0.047/0.047/0.000 ms

     0.000 probe_libc:inet_pton:(3ffb4bc2060))

[root@s8360047 perf]#

Test fails, I would have expected some callchain output.

[root@s8360047 perf]# ./perf trace --no-syscalls --call-graph dwarf --max-stack 4

        -e probe_libc:inet_pton/max-stack=3/ ping -6 -c 1 ::1

PING ::1(::1) 56 data bytes
64 bytes from ::1: icmp_seq=1 ttl=64 time=0.055 ms

--- ::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 0.055/0.055/0.055/0.000 ms

Quoted 7 lines
     0.000 probe_libc:inet_pton:(3ff7e6c2060))

[root@s8360047 perf]#

Test fails, I would have expected some callchain output.

It looks like /max-stack=3/ without ,call-graph=dwarf does not trigger the stack unwinding.

Ok, will work on those, then redo all the tests you did, grr, tricky stuff.

- Arnaldo