perf/urgent fixes
Arnaldo Carvalho de Melo
maintainerPATCH 1/5 perf stat: Fix per-pkg event reporting bug
In reply to Arnaldo Carvalho de MeloFrom: Stephane Eranian <[email protected]>
Per-pkg events need to be captured once per processor socket. The code in check_per_pkg() ensures only one value per processor package is used. However there is a problem with this function in case the first CPU of the package does not measure anything for the per-pkg event, but other CPUs do.
Consider the following:
$ create cgroup FOO; echo $$ >FOO/tasks; taskset -c 1 noploop &
$ perf stat -a -I 1000 -e intel_cqm/llc_occupancy/ -G FOO sleep 100
1.00000 <not counted> Bytes intel_cqm/llc_occupancy/ FOOThe reason for this is that CPU0 in the cgroup has nothing running on it. Yet check_per_plg() will mark socket0 as processed and no other event value will be considered for the socket.
This patch fixes the problem by having check_per_pkg() only consider events which actually ran.
--- tools/perf/util/stat.c | 16 ++++++++++++++-- 1 file changed, 14 insertions(+), 2 deletions(-)
tools/perf/util/stat.c +14 -2
@@ -196,7 +196,8 @@ static void zero_per_pkg(struct perf_evsel *counter)memset (counter->per_pkg_mask,0 ,MAX_NR_CPUS ); } -static int check_per_pkg (struct perf_evsel *counter,int cpu,bool *skip) +static int check_per_pkg (struct perf_evsel *counter, +struct perf_counts_values *vals,int cpu,bool *skip) {unsigned long *mask = counter->per_pkg_mask;struct cpu_map *cpus =perf_evsel__cpus (counter);
@@ -218,6 +219,17 @@ static int check_per_pkg(struct perf_evsel *counter, int cpu, bool *skip) counter->per_pkg_mask = mask; } +/* +* we do not consider an event that has not run as a good +* instance to mark a package as used (skip=1). Otherwise +* we may run into a situation where the first CPU in a package +* is not running anything, yet the second is, and this function +* would mark the package as used after the first CPU and would +* not read the values from the second CPU. +*/ +if (!(vals->run && vals->ena)) +return 0 ; + s =cpu_map__get_socket (cpus, cpu);if (s <0 )return -1 ;
@@ -235,7 +247,7 @@ process_counter_values(struct perf_stat_config *config, struct perf_evsel *evselstatic struct perf_counts_values zero;bool skip = false; -if (check_per_pkg (evsel, cpu, &skip)) { +if (check_per_pkg (evsel, count, cpu, &skip)) {pr_err ("failed to read per-pkg counter\n" );return -1 ; }
Signature · 1 line
2.1.0
Arnaldo Carvalho de Melo
maintainerPATCH 3/5 tools build: Add test for presence of numa_num_possible_cpus() in libnuma
In reply to Arnaldo Carvalho de MeloFrom: Arnaldo Carvalho de Melo <[email protected]>
The existing numa test checks only if numa.h and numa_available() are present, but that can be satisfied with an old libnuma that is not enough for the 'perf bench numa' entry, so add a test to check for that:
[acme@rhel5 linux]$ make NO_AUXTRACE=1 NO_LIBPERL=1 -C tools/perf O=/tmp/build/perf install-bin
make: Entering directory `/home/acme/git/linux/tools/perf'
BUILD: Doing 'make -j2' parallel build
Auto-detecting system features:
... libelf: [ on ]
... libnuma: [ on ]
... numa_num_possible_cpus: [ OFF ]
... libperl: [ on ]
<SNIP>
config/Makefile:577: Old numa library found, disables 'perf bench numa mem' benchmark, please install numactl-devel/libnuma-devel/libnuma-dev >= 2.0.8
INSTALL binaries
<SNIP>
This fixes the build on old systems such as RHEL/CentOS 5.11.
--- tools/build/Makefile.feature | 2 ++ tools/build/feature/Makefile | 4 ++++ tools/build/feature/test-all.c | 5 +++++ tools/build/feature/test-numa_num_possible_cpus.c | 6 ++++++ tools/perf/config/Makefile | 11 ++++++++--- 5 files changed, 25 insertions(+), 3 deletions(-) create mode 100644 tools/build/feature/test-numa_num_possible_cpus.c
tools/build/Makefile.feature +2 -0
@@ -41,6 +41,7 @@ FEATURE_TESTS ?= \ libelf-getphdrnum \ libelf-mmap \ libnuma \ + numa_num_possible_cpus \ libperl \ libpython \ libpython-version \
@@ -61,6 +62,7 @@ FEATURE_DISPLAY ?= \ libbfd \ libelf \ libnuma \ + numa_num_possible_cpus \ libperl \ libpython \ libslang \
tools/build/feature/Makefile +4 -0
@@ -19,6 +19,7 @@ FILES= \ test-libelf-getphdrnum.bin \ test-libelf-mmap.bin \ test-libnuma.bin \ + test-numa_num_possible_cpus.bin \ test-libperl.bin \ test-libpython.bin \ test-libpython-version.bin \
@@ -87,6 +88,9 @@ test-libelf-getphdrnum.bin: test-libnuma.bin: $(BUILD) -lnuma +test-numa_num_possible_cpus.bin: + $(BUILD) -lnuma + test-libunwind.bin: $(BUILD) -lelf
tools/build/feature/test-all.c +5 -0
@@ -77,6 +77,10 @@# include "test-libnuma.c" #undef main +#define main main_test_numa_num_possible_cpus +# include "test-numa_num_possible_cpus.c" +#undef main +#define main main_test_timerfd# include "test-timerfd.c" #undef main
@@ -136,6 +140,7 @@ int main(int argc, char *argv[])main_test_libbfd ();main_test_backtrace ();main_test_libnuma (); +main_test_numa_num_possible_cpus ();main_test_timerfd ();main_test_stackprotector_all ();main_test_libdw_dwarf_unwind ();
tools/build/feature/test-numa_num_possible_cpus.c new file +6 -0
@@ -0,0 +1,6 @@ +#include <numa.h> + +int main (void ) +{ +return numa_num_possible_cpus (); +}
tools/perf/config/Makefile +8 -3
@@ -573,9 +573,14 @@ ifndef NO_LIBNUMA msg := $(warning No numa.h found, disables 'perf bench numa mem' benchmark, please install numactl-devel/libnuma-devel/libnuma-dev); NO_LIBNUMA := 1 else - CFLAGS += -DHAVE_LIBNUMA_SUPPORT - EXTLIBS += -lnuma - $(call detected,CONFIG_NUMA) + ifeq ($(feature-numa_num_possible_cpus), 0) + msg := $(warning Old numa library found, disables 'perf bench numa mem' benchmark, please install numactl-devel/libnuma-devel/libnuma-dev >= 2.0.8); + NO_LIBNUMA := 1 + else + CFLAGS += -DHAVE_LIBNUMA_SUPPORT + EXTLIBS += -lnuma + $(call detected,CONFIG_NUMA) + endif endif endif
Signature · 1 line
2.1.0
Arnaldo Carvalho de Melo
maintainerPATCH 5/5 perf tools: Bool functions shouldn't return -1
In reply to Arnaldo Carvalho de MeloFrom: Peter Senna Tschudin <[email protected]>
Returning a negative value for a boolean function seem to have the undesired effect of returning true. Replace -1 by false in a bool-returning function.
The diff of the .s file before and after the change (for x86_64):
3907c3907
< movl $1, %ebx
---
Quoted 1 line
xorl %ebx, %ebx
while if -1 is replaced by true, the diff is empty.
This issue was found by the following Coccinelle semantic patch:
<smpl>
@@
identifier f;
constant C;
typedef bool;
@@
bool f (...){
<+...
* return -C;
...+>
}
</smpl>--- tools/perf/util/util.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/util.c +1 -1
@@ -709,7 +709,7 @@ bool find_process(const char *name) dir =opendir (procfs__mountpoint ());if (!dir) -return -1 ; +return false;/* Walk through the directory. */ while (ret && (d =readdir (dir)) !=NULL ) {
Signature · 1 line
2.1.0
Arnaldo Carvalho de Melo
maintainerPATCH 2/5 Revert "perf symbols: Fix mismatched declarations for elf_getphdrnum"
In reply to Arnaldo Carvalho de MeloFrom: Arnaldo Carvalho de Melo <[email protected]>
This reverts commit f785f2357673d520a0b7b468973cdd197f336494.
We have a test to check if elf_getphdrnum() is present, so, if it fails, we'll get:
[acme@rhel5 linux]$ cat /tmp/build/perf/feature/test-libelf-getphdrnum.make.output
cc1: warnings being treated as errors
test-libelf-getphdrnum.c: In function ‘main’:
test-libelf-getphdrnum.c:7: warning: implicit declaration of function ‘elf_getphdrnum’ [acme@rhel5 linux]$
And this block will not be compiled:
#ifndef HAVE_ELF_GETPHDRNUM_SUPPORT static int elf_getphdrnum(Elf *elf, size_t *dst) ... #endif
So, if elf_getphdrnum() is being defined somewhere, there is a problem with the test that is not detecting that function, go fix it.
--- tools/perf/util/symbol-elf.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/symbol-elf.c +1 -1
@@ -38,7 +38,7 @@ static inline char *bfd_demangle(void __maybe_unused *v,#endif #ifndef HAVE_ELF_GETPHDRNUM_SUPPORT -int elf_getphdrnum (Elf *elf,size_t *dst) +static int elf_getphdrnum (Elf *elf,size_t *dst) {GElf_Ehdr gehdr;GElf_Ehdr *ehdr;
Signature · 1 line
2.1.0
Arnaldo Carvalho de Melo
maintainerPATCH 4/5 tools build: Add test for presence of __get_cpuid() gcc builtin
In reply to Arnaldo Carvalho de MeloFrom: Arnaldo Carvalho de Melo <[email protected]>
The auxtrace code needed by Intel PT uses the __get_cpuid() gcc builtin, that is not present in old systems, breaking the build.
Add a test to check for that builtin and disable AUXTRACE in those systems.
[acme@rhel5 linux]$ make NO_LIBPERL=1 -C tools/perf O=/tmp/build/perf install-bin
make: Entering directory `/home/acme/git/linux/tools/perf'
BUILD: Doing 'make -j2' parallel build
Auto-detecting system features:
<SNIP>
... lzma: [ on ]
... get_cpuid: [ OFF ]
<SNIP>
config/Makefile:630: Your gcc lacks the __get_cpuid() builtin, disables support for auxtrace/Intel PT, please install a newer gcc
MKDIR /tmp/build/perf/util/
<SNIP>
This fixes the build on old systems such as RHEL/CentOS 5.11.
--- tools/build/Makefile.feature | 6 ++++-- tools/build/feature/Makefile | 6 +++++- tools/build/feature/test-all.c | 5 +++++ tools/build/feature/test-get_cpuid.c | 7 +++++++ tools/perf/config/Makefile | 9 +++++++-- 5 files changed, 28 insertions(+), 5 deletions(-) create mode 100644 tools/build/feature/test-get_cpuid.c
tools/build/Makefile.feature +4 -2
@@ -52,7 +52,8 @@ FEATURE_TESTS ?= \ timerfd \ libdw-dwarf-unwind \ zlib \ - lzma + lzma \ + get_cpuid FEATURE_DISPLAY ?= \ dwarf \
@@ -69,7 +70,8 @@ FEATURE_DISPLAY ?= \ libunwind \ libdw-dwarf-unwind \ zlib \ - lzma + lzma \ + get_cpuid # Set FEATURE_CHECK_(C|LD)FLAGS-all for all FEATURE_TESTS features. # If in the future we need per-feature checks/flags for features not
tools/build/feature/Makefile +5 -1
@@ -35,7 +35,8 @@ FILES= \ test-compile-x32.bin \ test-zlib.bin \ test-lzma.bin \ - test-bpf.bin + test-bpf.bin \ + test-get_cpuid.bin CC := $(CROSS_COMPILE)gcc -MD PKG_CONFIG := $(CROSS_COMPILE)pkg-config
@@ -166,6 +167,9 @@ test-zlib.bin: test-lzma.bin: $(BUILD) -llzma +test-get_cpuid.bin: + $(BUILD) + test-bpf.bin: $(BUILD)
tools/build/feature/test-all.c +5 -0
@@ -121,6 +121,10 @@# include "test-lzma.c" #undef main +#define main main_test_get_cpuid +# include "test-get_cpuid.c" +#undef main +int main (int argc,char *argv[]) {main_test_libpython ();
@@ -148,6 +152,7 @@ int main(int argc, char *argv[])main_test_zlib ();main_test_pthread_attr_setaffinity_np ();main_test_lzma (); +main_test_get_cpuid ();return 0 ; }
tools/build/feature/test-get_cpuid.c new file +7 -0
@@ -0,0 +1,7 @@ +#include <cpuid.h> + +int main (void ) +{ +unsigned int eax =0 , ebx =0 , ecx =0 , edx =0 ; +return __get_cpuid (0x15 , &eax, &ebx, &ecx, &edx); +}
tools/perf/config/Makefile +7 -2
@@ -626,8 +626,13 @@ ifdef LIBBABELTRACE endif ifndef NO_AUXTRACE - $(call detected,CONFIG_AUXTRACE) - CFLAGS += -DHAVE_AUXTRACE_SUPPORT + ifeq ($(feature-get_cpuid), 0) + msg := $(warning Your gcc lacks the __get_cpuid() builtin, disables support for auxtrace/Intel PT, please install a newer gcc); + NO_AUXTRACE := 1 + else + $(call detected,CONFIG_AUXTRACE) + CFLAGS += -DHAVE_AUXTRACE_SUPPORT + endif endif # Among the variables below, these:
Signature · 1 line
2.1.0
Arnaldo Carvalho de Melo
maintainerHi Ingo,
Please consider pulling,
- Arnaldo
The following changes since commit f6cf87f748ff9480f97ff9c5caf6d6faacf52aa1:
Merge tag 'perf-urgent-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/urgent (2015-09-16 09:06:54 +0200)
are available in the git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git tags/perf-urgent-for-mingo
for you to fetch changes up to bf6445631c6f00882b25516a174d5073ce0c6f81:
perf tools: Bool functions shouldn't return -1 (2015-09-17 15:31:52 -0300)
----------------------------------------------------------------
perf/urgent fixes:
User visible:
- When handling perf_event_open() returning EBUSY and not being able to opendir the procfs mount point we would tell the user that the oprofile daemon was found by returning -1 on as the return for a bool function, oops, fix it, found with Coccinelle (Peter Senna Tschudin).
- Fix per-pkg event reporting bug in 'perf stat' (Stephane Eranian)
Developer visible:
- Fix missing prototype for function provided when it isn't present in the libelf present, fixing the build on RHEL/CentOS 5.1 systems, for instance (Arnaldo Carvalho de Melo)
- Detect if the gcc and libnuma have the features needed to avoid requiring the use of NO_LIBNUMA and/or NO_AUXTRACE to build on older systems (Arnaldo Carvalho de Melo)
----------------------------------------------------------------
Arnaldo Carvalho de Melo (3):
Revert "perf symbols: Fix mismatched declarations for elf_getphdrnum" tools build: Add test for presence of numa_num_possible_cpus() in libnuma tools build: Add test for presence of __get_cpuid() gcc builtin
Peter Senna Tschudin (1):
perf tools: Bool functions shouldn't return -1
Stephane Eranian (1):
perf stat: Fix per-pkg event reporting bug
tools/build/Makefile.feature | 8 ++++++-- tools/build/feature/Makefile | 10 +++++++++- tools/build/feature/test-all.c | 10 ++++++++++ tools/build/feature/test-get_cpuid.c | 7 +++++++ tools/build/feature/test-numa_num_possible_cpus.c | 6 ++++++ tools/perf/config/Makefile | 20 +++++++++++++++----- tools/perf/util/stat.c | 16 ++++++++++++++-- tools/perf/util/symbol-elf.c | 2 +- tools/perf/util/util.c | 2 +- 9 files changed, 69 insertions(+), 12 deletions(-) create mode 100644 tools/build/feature/test-get_cpuid.c create mode 100644 tools/build/feature/test-numa_num_possible_cpus.c
Quoted 54 lines · Arnaldo Carvalho de Melo
Hi Ingo,
Please consider pulling,
- Arnaldo
The following changes since commit
f6cf87f748ff9480f97ff9c5caf6d6faacf52aa1:Merge tag 'perf-urgent-for-mingo' of
git://git.kernel.org/pub/scm/linux/kernel/git/acme/linuxinto perf/urgent (2015-09-16 09:06:54 +0200)are available in the git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.gittags/perf-urgent-for-mingofor you to fetch changes up to
bf6445631c6f00882b25516a174d5073ce0c6f81:perf tools: Bool functions shouldn't return -1 (2015-09-17 15:31:52 -0300)
----------------------------------------------------------------
perf/urgent fixes:User visible:
- When handling
perf_event_open()returning EBUSY and not being able to opendir the procfs mount point we would tell the user that the oprofile daemon was found by returning -1 on as the return for a bool function, oops, fix it, found with Coccinelle (Peter Senna Tschudin).- Fix per-pkg event reporting bug in 'perf stat' (Stephane Eranian)
Developer visible:
- Fix missing prototype for function provided when it isn't present in the libelf present, fixing the build on RHEL/CentOS 5.1 systems, for instance (Arnaldo Carvalho de Melo)
- Detect if the gcc and libnuma have the features needed to avoid requiring the use of
NO_LIBNUMAand/orNO_AUXTRACEto build on older systems (Arnaldo Carvalho de Melo)Signed-off-by Arnaldo Carvalho de Melo <[email protected]>----------------------------------------------------------------
Arnaldo Carvalho de Melo (3):
Revert "perf symbols: Fix mismatched declarations forelf_getphdrnum" tools build: Add test for presence ofnuma_num_possible_cpus()in libnuma tools build: Add test for presence of__get_cpuid()gcc builtinPeter Senna Tschudin (1):
perf tools: Bool functions shouldn't return -1Stephane Eranian (1):
perf stat: Fix per-pkg event reporting bug
Quoted 12 lines
tools/build/Makefile.feature | 8 ++++++-- tools/build/feature/Makefile | 10 +++++++++- tools/build/feature/test-all.c | 10 ++++++++++ tools/build/feature/test-get_cpuid.c | 7 +++++++ tools/build/feature/test-numa_num_possible_cpus.c | 6 ++++++ tools/perf/config/Makefile | 20 +++++++++++++++----- tools/perf/util/stat.c | 16 ++++++++++++++-- tools/perf/util/symbol-elf.c | 2 +- tools/perf/util/util.c | 2 +- 9 files changed, 69 insertions(+), 12 deletions(-) create mode 100644 tools/build/feature/test-get_cpuid.c create mode 100644 tools/build/feature/test-numa_num_possible_cpus.c
Pulled, thanks Arnaldo!
Ingo