public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
From: Arnaldo Carvalho de Melo <acme@kernel.org>
To: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>, Namhyung Kim <namhyung@kernel.org>,
	Clark Williams <williams@redhat.com>,
	linux-kernel@vger.kernel.org, linux-perf-users@vger.kernel.org,
	Adrian Hunter <adrian.hunter@intel.com>,
	Arnaldo Carvalho de Melo <acme@redhat.com>
Subject: [PATCH 48/53] perf thread-stack: Represent jmps to the start of a different symbol
Date: Wed,  6 Feb 2019 15:48:58 -0300	[thread overview]
Message-ID: <20190206184903.24054-49-acme@kernel.org> (raw)
In-Reply-To: <20190206184903.24054-1-acme@kernel.org>

From: Adrian Hunter <adrian.hunter@intel.com>

The compiler might optimize a call/ret combination by making it a jmp.
However the thread-stack does not presently cater for that, so that such
control flow is not visible in the call graph. Make it visible by
recording on the stack a branch to the start of a different symbol.
Note, that means when a ret pops the stack, all jmps must be popped off
first.

Example:

  $ cat jmp-to-fn.c
  __attribute__((noinline)) int bar(void)
  {
          return -1;
  }

  __attribute__((noinline)) int foo(void)
  {
          return bar() + 1;
  }

  int main()
  {
          return foo();
  }
  $ gcc -ggdb3 -Wall -Wextra -O2 -o jmp-to-fn jmp-to-fn.c
  $ objdump -d jmp-to-fn
  <SNIP>
  0000000000001040 <main>:
      1040:       31 c0                   xor    %eax,%eax
      1042:       e9 09 01 00 00          jmpq   1150 <foo>
  <SNIP>
  0000000000001140 <bar>:
      1140:       b8 ff ff ff ff          mov    $0xffffffff,%eax
      1145:       c3                      retq
  <SNIP>
  0000000000001150 <foo>:
      1150:       31 c0                   xor    %eax,%eax
      1152:       e8 e9 ff ff ff          callq  1140 <bar>
      1157:       83 c0 01                add    $0x1,%eax
      115a:       c3                      retq
  <SNIP>
  $ perf record -o jmp-to-fn.perf.data -e intel_pt/cyc/u ./jmp-to-fn
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0,017 MB jmp-to-fn.perf.data ]
  $ perf script -i jmp-to-fn.perf.data --itrace=be -s ~/libexec/perf-core/scripts/python/export-to-sqlite.py jmp-to-fn.db branches calls
  2019-01-08 13:24:58.783069 Creating database...
  2019-01-08 13:24:58.794650 Writing records...
  2019-01-08 13:24:59.008050 Adding indexes
  2019-01-08 13:24:59.015802 Done
  $  ~/libexec/perf-core/scripts/python/exported-sql-viewer.py jmp-to-fn.db

Before:

    main
        -> bar

After:

    main
        -> foo
            -> bar

Committer testing:

Install the python2-pyside package, then select these menu options
on the GUI:

   "Reports"
      "Context sensitive callgraphs"

Then go on expanding the symbols, to get, full picture when doing this
on a fedora:29 with gcc version 8.2.1 20181215 (Red Hat 8.2.1-6) (GCC):

jmp-to-fn
  PID:TID
    _start                (ld-2.28.so)
      __libc_start_main
        main
          foo
            bar

To verify that indeed, this fixes the problem.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/20190109091835.5570-5-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 .../scripts/python/export-to-postgresql.py    |  2 +-
 tools/perf/scripts/python/export-to-sqlite.py |  2 +-
 tools/perf/util/thread-stack.c                | 30 +++++++++++++++++--
 tools/perf/util/thread-stack.h                |  3 ++
 4 files changed, 33 insertions(+), 4 deletions(-)

diff --git a/tools/perf/scripts/python/export-to-postgresql.py b/tools/perf/scripts/python/export-to-postgresql.py
index 0564dd7377f2..30130213da7e 100644
--- a/tools/perf/scripts/python/export-to-postgresql.py
+++ b/tools/perf/scripts/python/export-to-postgresql.py
@@ -478,7 +478,7 @@ if perf_db_export_calls:
 			'branch_count,'
 			'call_id,'
 			'return_id,'
-			'CASE WHEN flags=1 THEN \'no call\' WHEN flags=2 THEN \'no return\' WHEN flags=3 THEN \'no call/return\' ELSE \'\' END AS flags,'
+			'CASE WHEN flags=0 THEN \'\' WHEN flags=1 THEN \'no call\' WHEN flags=2 THEN \'no return\' WHEN flags=3 THEN \'no call/return\' WHEN flags=6 THEN \'jump\' ELSE flags END AS flags,'
 			'parent_call_path_id'
 		' FROM calls INNER JOIN call_paths ON call_paths.id = call_path_id')
 
diff --git a/tools/perf/scripts/python/export-to-sqlite.py b/tools/perf/scripts/python/export-to-sqlite.py
index 245caf2643ed..ed237f2ed03f 100644
--- a/tools/perf/scripts/python/export-to-sqlite.py
+++ b/tools/perf/scripts/python/export-to-sqlite.py
@@ -320,7 +320,7 @@ if perf_db_export_calls:
 			'branch_count,'
 			'call_id,'
 			'return_id,'
-			'CASE WHEN flags=1 THEN \'no call\' WHEN flags=2 THEN \'no return\' WHEN flags=3 THEN \'no call/return\' ELSE \'\' END AS flags,'
+			'CASE WHEN flags=0 THEN \'\' WHEN flags=1 THEN \'no call\' WHEN flags=2 THEN \'no return\' WHEN flags=3 THEN \'no call/return\' WHEN flags=6 THEN \'jump\' ELSE flags END AS flags,'
 			'parent_call_path_id'
 		' FROM calls INNER JOIN call_paths ON call_paths.id = call_path_id')
 
diff --git a/tools/perf/util/thread-stack.c b/tools/perf/util/thread-stack.c
index 7f8eff018c16..f52c0f90915d 100644
--- a/tools/perf/util/thread-stack.c
+++ b/tools/perf/util/thread-stack.c
@@ -38,6 +38,7 @@
  * @cp: call path
  * @no_call: a 'call' was not seen
  * @trace_end: a 'call' but trace ended
+ * @non_call: a branch but not a 'call' to the start of a different symbol
  */
 struct thread_stack_entry {
 	u64 ret_addr;
@@ -47,6 +48,7 @@ struct thread_stack_entry {
 	struct call_path *cp;
 	bool no_call;
 	bool trace_end;
+	bool non_call;
 };
 
 /**
@@ -268,6 +270,8 @@ static int thread_stack__call_return(struct thread *thread,
 		cr.flags |= CALL_RETURN_NO_CALL;
 	if (no_return)
 		cr.flags |= CALL_RETURN_NO_RETURN;
+	if (tse->non_call)
+		cr.flags |= CALL_RETURN_NON_CALL;
 
 	return crp->process(&cr, crp->data);
 }
@@ -510,6 +514,7 @@ static int thread_stack__push_cp(struct thread_stack *ts, u64 ret_addr,
 	tse->cp = cp;
 	tse->no_call = no_call;
 	tse->trace_end = trace_end;
+	tse->non_call = false;
 
 	return 0;
 }
@@ -531,14 +536,16 @@ static int thread_stack__pop_cp(struct thread *thread, struct thread_stack *ts,
 							 timestamp, ref, false);
 	}
 
-	if (ts->stack[ts->cnt - 1].ret_addr == ret_addr) {
+	if (ts->stack[ts->cnt - 1].ret_addr == ret_addr &&
+	    !ts->stack[ts->cnt - 1].non_call) {
 		return thread_stack__call_return(thread, ts, --ts->cnt,
 						 timestamp, ref, false);
 	} else {
 		size_t i = ts->cnt - 1;
 
 		while (i--) {
-			if (ts->stack[i].ret_addr != ret_addr)
+			if (ts->stack[i].ret_addr != ret_addr ||
+			    ts->stack[i].non_call)
 				continue;
 			i += 1;
 			while (ts->cnt > i) {
@@ -757,6 +764,25 @@ int thread_stack__process(struct thread *thread, struct comm *comm,
 		err = thread_stack__trace_begin(thread, ts, sample->time, ref);
 	} else if (sample->flags & PERF_IP_FLAG_TRACE_END) {
 		err = thread_stack__trace_end(ts, sample, ref);
+	} else if (sample->flags & PERF_IP_FLAG_BRANCH &&
+		   from_al->sym != to_al->sym && to_al->sym &&
+		   to_al->addr == to_al->sym->start) {
+		struct call_path_root *cpr = ts->crp->cpr;
+		struct call_path *cp;
+
+		/*
+		 * The compiler might optimize a call/ret combination by making
+		 * it a jmp. Make that visible by recording on the stack a
+		 * branch to the start of a different symbol. Note, that means
+		 * when a ret pops the stack, all jmps must be popped off first.
+		 */
+		cp = call_path__findnew(cpr, ts->stack[ts->cnt - 1].cp,
+					to_al->sym, sample->addr,
+					ts->kernel_start);
+		err = thread_stack__push_cp(ts, 0, sample->time, ref, cp, false,
+					    false);
+		if (!err)
+			ts->stack[ts->cnt - 1].non_call = true;
 	}
 
 	return err;
diff --git a/tools/perf/util/thread-stack.h b/tools/perf/util/thread-stack.h
index 1f626f4a1c40..b7c04e19ad41 100644
--- a/tools/perf/util/thread-stack.h
+++ b/tools/perf/util/thread-stack.h
@@ -35,10 +35,13 @@ struct call_path;
  *
  * CALL_RETURN_NO_CALL: 'return' but no matching 'call'
  * CALL_RETURN_NO_RETURN: 'call' but no matching 'return'
+ * CALL_RETURN_NON_CALL: a branch but not a 'call' to the start of a different
+ *                       symbol
  */
 enum {
 	CALL_RETURN_NO_CALL	= 1 << 0,
 	CALL_RETURN_NO_RETURN	= 1 << 1,
+	CALL_RETURN_NON_CALL	= 1 << 2,
 };
 
 /**
-- 
2.20.1


  parent reply	other threads:[~2019-02-06 18:51 UTC|newest]

Thread overview: 55+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-02-06 18:48 [GIT PULL 00/53] perf/core improvements and fixes Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 01/53] perf powerpc: Add missing headers to skip-callchain-idx.c Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 02/53] perf arm pmu: Add missing linux/string.h header Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 03/53] perf srccode: Move struct definition from map.h to srccode.h Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 04/53] perf callchain: Uninline callchain_cursor_reset() to remove map.h dependency Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 05/53] perf symbols: Introduce map_symbol.h Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 06/53] pref tools: Add missing map.h includes Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 07/53] perf map: Move structs and prototypes for map groups to a separate header Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 08/53] perf tests: Add missing headers so far obtained indirectly Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 09/53] perf hist: Remove symbol.h from hist.h, just fwd decls are needed Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 10/53] perf tools: Add missing include for symbols.h Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 11/53] perf evsel: No need to include symbol.h in evsel.h, symbol_conf.h is enough Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 12/53] perf thread: Don't include symbol.h, " Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 13/53] perf tools: Add missing include <callchain.h> in various places Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 14/53] perf hist: Remove the needless callchain.h include from hist.h Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 15/53] perf tests pmu: Add missing headers Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 16/53] perf pmu: Remove needless evsel.h include, only needs one fwd decl Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 17/53] perf kvm stat: Replace kvm-stat.h includes with forward declarations Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 18/53] perf powerpc kvm-stat: Add missing evlist.h header Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 19/53] perf bpf-loader: Remove unecessary includes from bpf-loader.h Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 20/53] perf/aux: Make perf_event accessible to setup_aux() Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 21/53] coresight: perf: Add "sinks" group to PMU directory Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 22/53] coresight: Use event attributes for sink selection Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 23/53] perf pmu: Move EVENT_SOURCE_DEVICE_PATH to PMU header file Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 24/53] perf arm cs-etm: Use event attributes to send sink information to kernel Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 25/53] perf coresight: Remove set_drv_config() API Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 26/53] perf pmu: Remove set_drv_config API Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 27/53] perf record: Allocate affinity masks Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 28/53] perf record: Bind the AIO user space buffers to nodes Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 29/53] perf tools: Add fallback versions for CPU_{OR,EQUAL}() Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 30/53] perf record: Apply affinity masks when reading mmap buffers Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 31/53] perf clang: Do not use 'return std::move(something)' Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 32/53] perf hists: Add argument to hists__resort_cb_t callback Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 33/53] perf evsel: Add output_resort_cb method Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 34/53] perf report: Move symbol annotation to the resort phase Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 35/53] perf tools: Add documentation for BPF event selection Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 36/53] perf cs-etm: Add last instruction information in packet Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 37/53] perf cs-etm: Set sample flags for instruction range packet Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 38/53] perf cs-etm: Set sample flags for trace discontinuity Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 39/53] perf cs-etm: Add exception number in exception packet Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 40/53] perf cs-etm: Change tuple from traceID-CPU# to traceID-metadata Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 41/53] perf cs-etm: Add traceID in packet Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 42/53] perf cs-etm: Set sample flags for exception packet Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 43/53] perf cs-etm: Set sample flags for exception return packet Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 44/53] perf vendor events intel: Fix Load_Miss_Real_Latency on CLX Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 45/53] perf tools: Fix split_kallsyms_for_kcore() for trampoline symbols Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 46/53] perf thread-stack: Tidy thread_stack__push_cp() usage Arnaldo Carvalho de Melo
2019-02-06 18:48 ` [PATCH 47/53] perf thread-stack: Tidy thread_stack__no_call_return() by adding more local variables Arnaldo Carvalho de Melo
2019-02-06 18:48 ` Arnaldo Carvalho de Melo [this message]
2019-02-06 18:48 ` [PATCH 49/53] perf auxtrace: Define auxtrace record alignment Arnaldo Carvalho de Melo
2019-02-06 18:49 ` [PATCH 50/53] perf intel-pt: Fix overlap calculation for padding Arnaldo Carvalho de Melo
2019-02-06 18:49 ` [PATCH 51/53] perf intel-pt: Fix CYC timestamp calculation after OVF Arnaldo Carvalho de Melo
2019-02-06 18:49 ` [PATCH 52/53] perf intel-pt: Packet splitting can happen only on 32-bit Arnaldo Carvalho de Melo
2019-02-06 18:49 ` [PATCH 53/53] perf auxtrace: Add timestamp to auxtrace errors Arnaldo Carvalho de Melo
2019-02-09 12:17 ` [GIT PULL 00/53] perf/core improvements and fixes Ingo Molnar

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20190206184903.24054-49-acme@kernel.org \
    --to=acme@kernel.org \
    --cc=acme@redhat.com \
    --cc=adrian.hunter@intel.com \
    --cc=jolsa@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-perf-users@vger.kernel.org \
    --cc=mingo@kernel.org \
    --cc=namhyung@kernel.org \
    --cc=williams@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox