Skip to contentSkip to conversation

perf fixes

GIT PULL Ingo Molnar 3 participants 8 messages Sep 28, 2013 – Oct 1, 2013

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Linus,

Please pull the latest perf-urgent-for-linus git tree from:

git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git perf-urgent-for-linus

HEAD: 8a3da6c7d0031fcb6a0d17f9c7a68b0e01f52855 perf/x86: Fix PMU detection printout when no PMU is detected

A couple of tooling fixlets and a PMU detection printout fix.

Thanks,

Ingo

------------------>
Adrian Hunter (1):
perf machine: Fix path unpopulated in machine__create_modules()

Andi Kleen (1):
perf symbols: Demangle cloned functions

David Ahern (2):
perf trace: Add mmap2 handler
perf tools: Explicitly add libdl dependency

Ingo Molnar (1):
perf/x86: Fix PMU detection printout when no PMU is detected

Jiri Olsa (1):
perf kmem: Make it work again on non NUMA machines

Masami Hiramatsu (1):
perf probe: Fix probing symbols with optimization suffix

 arch/x86/kernel/cpu/perf_event.c |  2 +-
 tools/perf/builtin-kmem.c        |  2 +-
 tools/perf/builtin-trace.c       |  1 +
 tools/perf/config/Makefile       |  2 +-
 tools/perf/util/machine.c        |  2 +-
 tools/perf/util/probe-finder.c   | 77 +++++++++++++++++++++-------------------
 tools/perf/util/probe-finder.h   |  3 ++
 tools/perf/util/symbol-elf.c     | 27 +++++++++++++-
 8 files changed, 75 insertions(+), 41 deletions(-)
arch/x86/kernel/cpu/perf_event.c +1 -1
index a9c606b..897783b 100644
@@ -1506,7 +1506,7 @@ static int __init init_hw_perf_events(void)
 		err = amd_pmu_init();
 		break;
 	default:
-		return 0;
+		err = -ENOTSUPP;
 	}
 	if (err != 0) {
 		pr_cont("no PMU driver, software events only.\n");
tools/perf/builtin-kmem.c +1 -1
index c2dff9c..9b5f077 100644
@@ -101,7 +101,7 @@ static int setup_cpunode_map(void)
 
 	dir1 = opendir(PATH_SYS_NODE);
 	if (!dir1)
-		return -1;
+		return 0;
 
 	while ((dent1 = readdir(dir1)) != NULL) {
 		if (dent1->d_type != DT_DIR ||
tools/perf/builtin-trace.c +1 -0
index fd48534..71aa3e3 100644
@@ -1055,6 +1055,7 @@ static int trace__replay(struct trace *trace)
 
 	trace->tool.sample	  = trace__process_sample;
 	trace->tool.mmap	  = perf_event__process_mmap;
+	trace->tool.mmap2	  = perf_event__process_mmap2;
 	trace->tool.comm	  = perf_event__process_comm;
 	trace->tool.exit	  = perf_event__process_exit;
 	trace->tool.fork	  = perf_event__process_fork;
tools/perf/config/Makefile +1 -1
index 346ee92..5f6f9b3 100644
@@ -87,7 +87,7 @@ CFLAGS += -Wall
 CFLAGS += -Wextra
 CFLAGS += -std=gnu99
 
-EXTLIBS = -lelf -lpthread -lrt -lm
+EXTLIBS = -lelf -lpthread -lrt -lm -ldl
 
 ifeq ($(call try-cc,$(SOURCE_HELLO),$(CFLAGS) -Werror -fstack-protector-all,-fstack-protector-all),y)
   CFLAGS += -fstack-protector-all
tools/perf/util/machine.c +1 -1
index 933d14f..6188d28 100644
@@ -792,7 +792,7 @@ static int machine__create_modules(struct machine *machine)
 		modules = path;
 	}
 
-	if (symbol__restricted_filename(path, "/proc/modules"))
+	if (symbol__restricted_filename(modules, "/proc/modules"))
 		return -1;
 
 	file = fopen(modules, "r");
tools/perf/util/probe-finder.c +41 -36

123 lines not on this page · raw message

tools/perf/util/probe-finder.h +3 -0
index 17e94d0..3b7d630 100644
@@ -23,6 +23,7 @@ static inline int is_c_varname(const char *name)
 /* debug information structure */
 struct debuginfo {
 	Dwarf		*dbg;
+	Dwfl_Module	*mod;
 	Dwfl		*dwfl;
 	Dwarf_Addr	bias;
 };
@@ -77,6 +78,7 @@ struct probe_finder {
 
 struct trace_event_finder {
 	struct probe_finder	pf;
+	Dwfl_Module		*mod;		/* For solving symbols */
 	struct probe_trace_event *tevs;		/* Found trace events */
 	int			ntevs;		/* Number of trace events */
 	int			max_tevs;	/* Max number of trace events */
@@ -84,6 +86,7 @@ struct trace_event_finder {
 
 struct available_var_finder {
 	struct probe_finder	pf;
+	Dwfl_Module		*mod;		/* For solving symbols */
 	struct variable_list	*vls;		/* Found variable lists */
 	int			nvls;		/* Number of variable lists */
 	int			max_vls;	/* Max no. of variable lists */
tools/perf/util/symbol-elf.c +26 -1
index a9c829b..d2a888e 100644
@@ -928,8 +928,33 @@ int dso__load_sym(struct dso *dso, struct map *map,
 		 * to it...
 		 */
 		if (symbol_conf.demangle) {
-			demangled = bfd_demangle(NULL, elf_name,
+			/*
+			 * The demangler doesn't deal with cloned functions.
+			 * XXXX.clone.NUM or similar
+			 * Strip the dot part and readd it later.
+			 */
+			char *p = (char *)elf_name, *dot;
+			dot = strchr(elf_name, '.');
+			if (dot) {
+				p = strdup(elf_name);
+				if (!p)
+					goto new_symbol;
+				dot = strchr(p, '.');
+				*dot = 0;
+			}
+
+			demangled = bfd_demangle(NULL, p,
 						 DMGL_PARAMS | DMGL_ANSI);
+			if (dot)
+				*dot = '.';
+			if (demangled && dot) {
+				demangled = realloc(demangled, strlen(demangled) + strlen(dot) + 1);
+				if (!demangled)
+					goto new_symbol;
+				strcpy(demangled + (dot - p), dot);
+			}
+			if (p != elf_name)
+				free(p);
 			if (demangled != NULL)
 				elf_name = demangled;
 		}
#1 lore raw
Quoted 2 lines · Ingo Molnar

Andi Kleen (1):
perf symbols: Demangle cloned functions

The commit above e95ab53645:

commit de95ab53645a2f0015e0f68ee723f18dce2b8b51
Author: Andi Kleen <[email protected]>
Date: Thu Sep 12 08:16:38 2013 -0700

perf symbols: Demangle cloned functions

breaks "perf top" on my machine. I just see a gray screen with no text at all. Sometimes the following error messages are printed:
*** Error in perf': invalid fastbin entry (free): 0x00000000029b18c0 *** *** Error in perf': malloc(): memory corruption (fast): 0x0000000000ee0b10 ***

Reverting the commit "fixes" the issue.

Signature · 1 line
Markus
#2 lore raw

[GIT PULL] perf fix

Quoted 10 lines · Markus Trippelsdorf
Quoted 2 lines · Ingo Molnar

Andi Kleen (1):
perf symbols: Demangle cloned functions

The commit above e95ab53645:

commit de95ab53645a2f0015e0f68ee723f18dce2b8b51
Author: Andi Kleen <[email protected]>
Date: Thu Sep 12 08:16:38 2013 -0700

Quoted 8 lines

perf symbols: Demangle cloned functions

breaks "perf top" on my machine. I just see a gray screen with no text at all. Sometimes the following error messages are printed:
*** Error in perf': invalid fastbin entry (free): 0x00000000029b18c0 *** *** Error in perf': malloc(): memory corruption (fast): 0x0000000000ee0b10 ***

Reverting the commit "fixes" the issue.

Sorry about that.

Looking at the commit it fails on several quality levels:

- sloppy structure: the code is unreadable, it should have been a separate helper function

- sloppy flow: why the heck does it do strchr() twice? Not to
mention the fragile way of how the new flow tries to merge with the old instead of cleanly separating.

- sloppy types: it casts a const char * over into char *

- sloppy style: variable block not separated from statements,
meaningless temporary variable names, etc. etc.

- and it's not even a regression fix to begin with!

So the right resolution is a revert. Andi, please send a fixed patch with all of these issues and the crash fixed.

Linus,

Please pull the latest perf-urgent-for-linus git tree from:

git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git perf-urgent-for-linus

HEAD: 14951f22f1cc8375ead345d2ca08455e91f2152b Revert "perf symbols: Demangle cloned functions"

It fixes the 'perf top' regression Markus Trippelsdorf reported.

Thanks,

Ingo

------------------>
Ingo Molnar (1):
Revert "perf symbols: Demangle cloned functions"

 tools/perf/util/symbol-elf.c | 27 +--------------------------
 1 file changed, 1 insertion(+), 26 deletions(-)
tools/perf/util/symbol-elf.c +1 -26
index d2a888e..a9c829b 100644
@@ -928,33 +928,8 @@ int dso__load_sym(struct dso *dso, struct map *map,
 		 * to it...
 		 */
 		if (symbol_conf.demangle) {
-			/*
-			 * The demangler doesn't deal with cloned functions.
-			 * XXXX.clone.NUM or similar
-			 * Strip the dot part and readd it later.
-			 */
-			char *p = (char *)elf_name, *dot;
-			dot = strchr(elf_name, '.');
-			if (dot) {
-				p = strdup(elf_name);
-				if (!p)
-					goto new_symbol;
-				dot = strchr(p, '.');
-				*dot = 0;
-			}
-
-			demangled = bfd_demangle(NULL, p,
+			demangled = bfd_demangle(NULL, elf_name,
 						 DMGL_PARAMS | DMGL_ANSI);
-			if (dot)
-				*dot = '.';
-			if (demangled && dot) {
-				demangled = realloc(demangled, strlen(demangled) + strlen(dot) + 1);
-				if (!demangled)
-					goto new_symbol;
-				strcpy(demangled + (dot - p), dot);
-			}
-			if (p != elf_name)
-				free(p);
 			if (demangled != NULL)
 				elf_name = demangled;
 		}
#3 lore raw
In reply to Markus Trippelsdorf
Quoted 4 lines

breaks "perf top" on my machine. I just see a gray screen with no text at all. Sometimes the following error messages are printed:
*** Error in perf': invalid fastbin entry (free): 0x00000000029b18c0 *** *** Error in perf': malloc(): memory corruption (fast): 0x0000000000ee0b10 ***

Hmm, works for me. In fact I wrote it to fix a perf top issue.

What does valgrind say?

-Andi

Signature · 1 line
[email protected] -- Speaking for myself only
#4 lore raw
Quoted 8 lines · Andi Kleen
Quoted 4 lines

breaks "perf top" on my machine. I just see a gray screen with no text at all. Sometimes the following error messages are printed:
*** Error in perf': invalid fastbin entry (free): 0x00000000029b18c0 *** *** Error in perf': malloc(): memory corruption (fast): 0x0000000000ee0b10 ***

Hmm, works for me. In fact I wrote it to fix a perf top issue.

What does valgrind say?

Here's the -fsanitize=address output instead:

markus@x4 perf % perf top --stdio 2>&1 | asan_symbolize.py
=================================================================
==24617== ERROR: AddressSanitizer: heap-buffer-overflow on address 0x600c0000b638 at pc 0x7f2683fad79d bp 0x7fffbb8ec3b0 sp 0x7fffbb8ebb70 WRITE of size 9 at 0x600c0000b638 thread T0

    #0 0x7f2683fad79c in __interceptor_strcpy ??:?
    #1 0x4e422c in strcpy /usr/include/bits/string3.h:104
    #2 0x4ddeff in dso__load /usr/src/linux/tools/perf/util/symbol.c:1085
    #3 0x509058 in map__load /usr/src/linux/tools/perf/util/map.c:159
    #4 0x49fa6b in perf_event__preprocess_sample /usr/src/linux/tools/perf/util/event.c:762
    #5 0x455fa0 in perf_event__process_sample /usr/src/linux/tools/perf/builtin-top.c:719
    #6 0x45be8e in perf_top__mmap_read /usr/src/linux/tools/perf/builtin-top.c:870
    #7 0x41f88d in run_builtin /usr/src/linux/tools/perf/perf.c:319
    #8 0x41bc05 in run_argv /usr/src/linux/tools/perf/perf.c:376
    #9 0x7f2682d41a74 in __libc_start_main /home/markus/glibc/csu/libc-start.c:269
    #10 0x41e948 in _start /home/markus/glibc/csu/../sysdeps/x86_64/start.S:122
0x600c0000b63d is located 0 bytes to the right of 61-byte region [0x600c0000b600,0x600c0000b63d)
allocated by thread T0 here:
    #0 0x7f2683fb6d6f in __interceptor_realloc ??:?
    #1 0x4e420f in dso__load_sym /usr/src/linux/tools/perf/util/symbol-elf.c:951
    #2 0x4ddeff in dso__load /usr/src/linux/tools/perf/util/symbol.c:1085
    #3 0x509058 in map__load /usr/src/linux/tools/perf/util/map.c:159
    #4 0x49fa6b in perf_event__preprocess_sample /usr/src/linux/tools/perf/util/event.c:762
    #5 0x455fa0 in perf_event__process_sample /usr/src/linux/tools/perf/builtin-top.c:719
    #6 0x45be8e in perf_top__mmap_read /usr/src/linux/tools/perf/builtin-top.c:870
    #7 0x41f88d in run_builtin /usr/src/linux/tools/perf/perf.c:319
    #8 0x41bc05 in run_argv /usr/src/linux/tools/perf/perf.c:376
    #9 0x7f2682d41a74 in __libc_start_main /home/markus/glibc/csu/libc-start.c:269

Shadow bytes around the buggy address:
0x0c01ffff9670: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa 0x0c01ffff9680: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa 0x0c01ffff9690: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa 0x0c01ffff96a0: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa 0x0c01ffff96b0: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
=>0x0c01ffff96c0: 00 00 00 00 00 00 00[05]fa fa fa fa fd fd fd fd
0x0c01ffff96d0: fd fd fd fd fa fa fa fa fd fd fd fd fd fd fd fd 0x0c01ffff96e0: fa fa fa fa fd fd fd fd fd fd fd fa fa fa fa fa 0x0c01ffff96f0: fd fd fd fd fd fd fd fd fa fa fa fa fd fd fd fd 0x0c01ffff9700: fd fd fd fd fa fa fa fa fd fd fd fd fd fd fd fd 0x0c01ffff9710: fa fa fa fa fd fd fd fd fd fd fd fd fa fa fa fa
Shadow byte legend (one shadow byte represents 8 application bytes):
Addressable: 00
Partially addressable: 01 02 03 04 05 06 07

  Heap left redzone:     fa
  Heap righ redzone:     fb
  Freed Heap region:     fd
  Stack left redzone:    f1
  Stack mid redzone:     f2
  Stack right redzone:   f3
  Stack partial redzone: f4
  Stack after return:    f5
  Stack use after scope: f8
  Global redzone:        f9
  Global init order:     f6
  Poisoned by user:      f7
  ASan internal:         fe

==24617== ABORTING

And please note that cloned functions are demangled just fine on my machine even without your patch, e.g.:

10.61% libQtGui.so.4.8.5 [.] QFontEngineFT::stringToCMap(QChar const*, int, QGlyphLayout*, int*, QFlags<QTextEngine::ShaperFlag>) const [clone .part.20]

Signature · 1 line
Markus
#5 lore raw
In reply to Andi Kleen
Quoted 6 lines · Andi Kleen
Quoted 4 lines

breaks "perf top" on my machine. I just see a gray screen with no text at all. Sometimes the following error messages are printed:
*** Error in perf': invalid fastbin entry (free): 0x00000000029b18c0 *** *** Error in perf': malloc(): memory corruption (fast): 0x0000000000ee0b10 ***

Hmm, works for me. In fact I wrote it to fix a perf top issue.

What 'issue' and why wasn't it described in the changelog more accurately?

The blurb you wrote in the changelog:

| The libbfd C++ demangler doesn't seem to deal with cloned functions, 
| like symbol.clone.NUM.
|
| Just strip the dot part before demangling and add it back later.

Only alludes to a C++ symbol resolution annoyance and thus isn't very informative.

Ingo

#6 lore raw
Quoted 9 lines

The blurb you wrote in the changelog:

| The libbfd C++ demangler doesn't seem to deal with cloned functions, 
| like symbol.clone.NUM.
|
| Just strip the dot part before demangling and add it back later.

Only alludes to a C++ symbol resolution annoyance and thus isn't very informative.

The issue was that I saw a long unreadable line noise string in report/top, instead of a demangled function. After I stripped the postfix I got the demangled symbol as expected.

-Andi

Signature · 1 line
[email protected] -- Speaking for myself only
#7 lore raw
Quoted 12 lines · Andi Kleen
Quoted 12 lines
Quoted 1 line

Hmm, works for me. In fact I wrote it to fix a perf top issue.

What 'issue' and why wasn't it described in the changelog more accurately?

The blurb you wrote in the changelog:

| The libbfd C++ demangler doesn't seem to deal with cloned functions, 
| like symbol.clone.NUM.
|
| Just strip the dot part before demangling and add it back later.
Quoted 6 lines
Quoted 2 lines

Only alludes to a C++ symbol resolution annoyance and thus isn't very informative.

The issue was that I saw a long unreadable line noise string in report/top, instead of a demangled function. After I stripped the postfix I got the demangled symbol as expected.

We obviously don't want to trade output annoyances for crashes, nor does it justify the code uglies you introduced with the patch - which made the code obviously unrobust.

My guess would be that maybe it's the cast of a const pointer that possibly broke GCC's aliasing detection on Markus's system or so. But maybe there's a more direct bug in the code as well, the code you added is too ugly to be reviewed efficiently, please send a cleaned up version.

Thanks,

Ingo