- 25 Oct, 2023 6 commits
-
-
Namhyung Kim authored
Currently lock contention timestamp is maintained in a hash map keyed by pid. That means it needs to get and release a map element (which is proctected by spinlock!) on each contention begin and end pair. This can impact on performance if there are a lot of contention (usually from spinlocks). It used to go with task local storage but it had an issue on memory allocation in some critical paths. Although it's addressed in recent kernels IIUC, the tool should support old kernels too. So it cannot simply switch to the task local storage at least for now. As spinlocks create lots of contention and they disabled preemption during the spinning, it can use per-cpu array to keep the timestamp to avoid overhead in hashmap update and delete. In contention_begin, it's easy to check the lock types since it can see the flags. But contention_end cannot see it. So let's try to per-cpu array first (unconditionally) if it has an active element (lock != 0). Then it should be used and per-task tstamp map should not be used until the per-cpu array element is cleared which means nested spinlock contention (if any) was finished and it nows see (the outer) lock. Signed-off-by:
Namhyung Kim <namhyung@kernel.org> Acked-by:
Ian Rogers <irogers@google.com> Cc: Hao Luo <haoluo@google.com> Cc: Song Liu <song@kernel.org> Cc: bpf@vger.kernel.org Link: https://lore.kernel.org/r/20231020204741.1869520-3-namhyung@kernel.org
-
Namhyung Kim authored
When pelem is NULL, it'd create a new entry with zero data. But it might be preempted by IRQ/NMI just before calling bpf_map_update_elem() then there's a chance to call it twice for the same pid. So it'd be better to use BPF_NOEXIST flag and check the return value to prevent the race. Signed-off-by:
-
Namhyung Kim authored
It checks the current lock to calculated the delta of contention time. The address is saved in the tstamp map which is allocated at begining of contention and released at end of contention. But it's possible for bpf_map_delete_elem() to fail. In that case, the element in the tstamp map kept for the current lock and it makes the next contention for the same lock tracked incorrectly. Specificially the next contention begin will see the existing element for the task and it'd just return. Then the next contention end will see the element and calculate the time using the timestamp for the previous begin. This can result in a large value for two small contentions happened from time to time. Let's clear the lock address so that it can be updated next time even if the bpf_map_delete_elem() failed. Signed-off-by:
-
Yang Jihong authored
evsel__increase_rlimit() helper does nothing with evsel, and description of the functionality is inaccurate, rename it and move to util/rlimit.c. By the way, fix a checkppatch warning about misplaced license tag: WARNING: Misplaced SPDX-License-Identifier tag - use line 1 instead #160: FILE: tools/perf/util/rlimit.h:3: /* SPDX-License-Identifier: LGPL-2.1 */ No functional change. Signed-off-by:
Yang Jihong <yangjihong1@huawei.com> Link: https://lore.kernel.org/r/20231023033144.1011896-1-yangjihong1@huawei.comSigned-off-by:
-
Namhyung Kim authored
The -G/--cgroups option is to put sender and receiver in different cgroups in order to measure cgroup context switch overheads. Users need to make sure the cgroups exist and accessible. The following example should the effect of this change. Please don't forget taskset before the perf bench to measure cgroup switches properly. Otherwise each task would run on a different CPU and generate cgroup switches regardless of this change. # perf stat -e context-switches,cgroup-switches \ > taskset -c 0 perf bench sched pipe -l 10000 > /dev/null Performance counter stats for 'taskset -c 0 perf bench sched pipe -l 10000': 20,001 context-switches 2 cgroup-switches 0.053449651 seconds time elapsed 0.011286000 seconds user 0.041869000 seconds sys # perf stat -e context-switches,cgroup-switches \ > taskset -c 0 perf bench sched pipe -l 10000 -G AAA,BBB > /dev/null Performance counter stats for 'taskset -c 0 perf bench sched pipe -l 10000 -G AAA,BBB': 20,001 context-switches 20,001 cgroup-switches 0.052768627 seconds time elapsed 0.006284000 seconds user 0.046266000 seconds sys Signed-off-by:Arnaldo Carvalho de Melo <acme@redhat.com> Link: https://lore.kernel.org/r/20231017202342.1353124-1-namhyung@kernel.org
-
Michael Petlan authored
CoreSight might be not available, in such case, skip the tests. Signed-off-by:
Michael Petlan <mpetlan@redhat.com> Reviewed-by:
Leo Yan <leo.yan@linaro.org> Reviewed-by:
Carsten Haitzler <carsten.haitzler@arm.com> Cc: vmolnaro@redhat.com Link: https://lore.kernel.org/r/20231019091137.22525-1-mpetlan@redhat.comSigned-off-by:
-
- 20 Oct, 2023 3 commits
-
-
Yang Jihong authored
If using parallel threads to collect data, perf record needs at least 6 fds per CPU. (one for sys_perf_event_open, four for pipe msg and ack of the pipe, see record__thread_data_open_pipes(), and one for open perf.data.XXX) For an environment with more than 100 cores, if perf record uses both `-a` and `--threads` options, it is easy to exceed the upper limit of the file descriptor number, when we run out of them try to increase the limits. Before: $ ulimit -n 1024 $ lscpu | grep 'On-line CPU(s)' On-line CPU(s) list: 0-159 $ perf record --threads -a sleep 1 Failed to create data directory: Too many open files After: $ ulimit -n 1024 $ lscpu | grep 'On-line CPU(s)' On-line CPU(s) list: 0-159 $ perf record --threads -a sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.394 MB perf.data (1576 samples) ] Signed-off-by:
-
Kajol Jain authored
The CPI_STALL_RATIO metric group can be used to present the high level CPI stall breakdown metrics in powerpc, which will show: - DISPATCH_STALL_CPI ( Dispatch stall cycles per insn ) - ISSUE_STALL_CPI ( Issue stall cycles per insn ) - EXECUTION_STALL_CPI ( Execution stall cycles per insn ) - COMPLETION_STALL_CPI ( Completion stall cycles per insn ) Commit cf26e043 ("perf vendor events power10: Add JSON metric events to present CPI stall cycles in powerpc)" which added the CPI_STALL_RATIO metric group, also modified the PMC value used in PM_RUN_INST_CMPL event from PMC4 to PMC5, to avoid multiplexing of events. But that got revert in recent changes. Fix this issue by changing back the PMC value used in PM_RUN_INST_CMPL to PMC5. Result with the fix: ./perf stat --metric-no-group -M CPI_STALL_RATIO <workload> Performance counter stats for 'workload': 68,745,426 PM_CMPL_STALL # 0.21 COMPLETION_STALL_CPI 7,692,827 PM_ISSUE_STALL # 0.02 ISSUE_STALL_CPI 322,638,223 PM_RUN_INST_CMPL # 0.05 DISPATCH_STALL_CPI # 0.48 EXECUTION_STALL_CPI 16,858,553 PM_DISP_STALL_CYC 153,880,133 PM_EXEC_STALL 0.089774592 seconds time elapsed "--metric-no-group" is used for forcing PM_RUN_INST_CMPL to be scheduled in all group for more accuracy. Fixes: 7d473f47 ("perf vendor events: Move JSON/events to appropriate files for power10 platform") Reported-by:
Disha Goel <disgoel@linux.vnet.ibm.com> Signed-off-by:
Kajol Jain <kjain@linux.ibm.com> Reviewed-by:
Athira Rajeev <atrajeev@linux.vnet.ibm.com> Tested-by: Disha Goel<disgoel@linux.ibm.com> Cc: maddy@linux.ibm.com Link: https://lore.kernel.org/r/20231016143110.244255-1-kjain@linux.ibm.comSigned-off-by:
-
Thomas Richter authored
Perf test case 111 Check open filename arg using perf trace + vfs_getname fails on s390. This is caused by a failing function bpf_probe_read() in file util/bpf_skel/augmented_raw_syscalls.bpf.c. The root cause is the lookup by address. Function bpf_probe_read() is used. This function works only for architectures with ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE. On s390 is not possible to determine from the address to which address space the address belongs to (user or kernel space). Replace bpf_probe_read() by bpf_probe_read_kernel() and bpf_probe_read_str() by bpf_probe_read_user_str() to explicity specify the address space the address refers to. Output before: # ./perf trace -eopen,openat -- touch /tmp/111 libbpf: prog 'sys_enter': BPF program load failed: Invalid argument libbpf: prog 'sys_enter': -- BEGIN PROG LOAD LOG -- reg type unsupported for arg#0 function sys_enter#75 0: R1=ctx(off=0,imm=0) R10=fp0 ; int sys_enter(struct syscall_enter_args *args) 0: (bf) r6 = r1 ; R1=ctx(off=0,imm=0) R6_w=ctx(off=0,imm=0) ; return bpf_get_current_pid_tgid(); 1: (85) call bpf_get_current_pid_tgid#14 ; R0_w=scalar() 2: (63) *(u32 *)(r10 -8) = r0 ; R0_w=scalar() R10=fp0 fp-8=????mmmm 3: (bf) r2 = r10 ; R2_w=fp0 R10=fp0 ; ..... lines deleted here ..... 23: (bf) r3 = r6 ; R3_w=ctx(off=0,imm=0) R6=ctx(off=0,imm=0) 24: (85) call bpf_probe_read#4 unknown func bpf_probe_read#4 processed 23 insns (limit 1000000) max_states_per_insn 0 \ total_states 2 peak_states 2 mark_read 2 -- END PROG LOAD LOG -- libbpf: prog 'sys_enter': failed to load: -22 libbpf: failed to load object 'augmented_raw_syscalls_bpf' libbpf: failed to load BPF skeleton 'augmented_raw_syscalls_bpf': -22 .... Output after: # ./perf test -Fv 111 111: Check open filename arg using perf trace + vfs_getname : --- start --- 1.085 ( 0.011 ms): touch/320753 openat(dfd: CWD, filename: \ "/tmp/temporary_file.SWH85", \ flags: CREAT|NOCTTY|NONBLOCK|WRONLY, mode: IRUGO|IWUGO) = 3 ---- end ---- Check open filename arg using perf trace + vfs_getname: Ok # Test with the sleep command shows: Output before: # ./perf trace -e *sleep sleep 1.234567890 0.000 (1234.681 ms): sleep/63114 clock_nanosleep(rqtp: \ { .tv_sec: 0, .tv_nsec: 0 }, rmtp: 0x3ffe0979720) = 0 # Output after: # ./perf trace -e *sleep sleep 1.234567890 0.000 (1234.686 ms): sleep/64277 clock_nanosleep(rqtp: \ { .tv_sec: 1, .tv_nsec: 234567890 }, rmtp: 0x3fff3df9ea0) = 0 # Fixes: 14e4b9f4 ("perf trace: Raw augmented syscalls fix libbpf 1.0+ compatibility") Signed-off-by:Thomas Richter <tmricht@linux.ibm.com> Co-developed-by:
Ilya Leoshkevich <iii@linux.ibm.com> Tested-by:
-
- 18 Oct, 2023 3 commits
-
-
Namhyung Kim authored
The recent change made it possible to generate vmlinux.h from BTF and to ignore the file. But we also have a minimal vmlinux.h that will be used by default. It should not be ignored by GIT. Fixes: b7a2d774 ("perf build: Add ability to build with a generated vmlinux.h") Reported-by:
kernel test robot <lkp@intel.com> Reviewed-by:
-
Jiri Olsa authored
As Dmitry described in [1] changelog the current way of detecting -s option is broken for new make. Changing the tools/build -s option detection the same way as it was fixed for root Makefile in [1]. [1] 4bf73588 ("kbuild: Port silent mode detection to future gnu make.") Cc: Dmitry Goncharov <dgoncharov@users.sf.net> Signed-off-by:
Jiri Olsa <jolsa@kernel.org> Tested-by:
-
Jiri Olsa authored
As Dmitry described in [1] changelog the current way of detecting -s option is broken for new make. Changing the tools/build -s option detection the same way as it was fixed for root Makefile in [1]. [1] 4bf73588 ("kbuild: Port silent mode detection to future gnu make.") Cc: Dmitry Goncharov <dgoncharov@users.sf.net> Signed-off-by:
-
- 17 Oct, 2023 19 commits
-
-
Changbin Du authored
A comma is missed at the end of line. Signed-off-by:
Changbin Du <changbin.du@huawei.com> Link: https://lore.kernel.org/r/20231017015524.797065-1-changbin.du@huawei.comSigned-off-by:
-
Athira Rajeev authored
Running shellcheck on stat_all_metricgroups.sh reports below warning: In ./tests/shell/stat_all_metricgroups.sh line 7: function ParanoidAndNotRoot() ^-- SC2112: 'function' keyword is non-standard. Delete it. As per the format, "function" is a non-standard keyword that can be used to declare functions. Fix this by removing the "function" keyword from ParanoidAndNotRoot function Signed-off-by:
-
Athira Rajeev authored
Running shellcheck on record_sideband.sh throws below warning: In tests/shell/record_sideband.sh line 25: if ! perf record -o ${perfdata} -BN --no-bpf-event -C $1 true 2>&1 >/dev/null ^--^ SC2069: To redirect stdout+stderr, 2>&1 must be last (or use '{ cmd > file; } 2>&1' to clarify). This shows shellcheck warning SC2069 where the redirection order needs to be fixed. Use "cmd > /dev/null 2>&1" to fix the redirection of perf record output Fixes: 23b97c7e ("perf test: Add test case for record sideband events") Signed-off-by: -
Athira Rajeev authored
Running shellcheck on lock_contention.sh generates below warning In tests/shell/lock_contention.sh line 36: if [ `nproc` -lt 4 ]; then ^-----^ SC2046: Quote this to prevent word splitting. Here since nproc will generate a single word output and there is no possibility of word splitting, this warning can be ignored. Use exception for this with "disable" option in shellcheck. This warning is observed after commit: "commit 29441ab3 ("perf test lock_contention.sh: Skip test if not enough CPUs")" Fixes: 29441ab3 ("perf test lock_contention.sh: Skip test if not enough CPUs") Signed-off-by:
-
Athira Rajeev authored
Simple expression parser test fails in powerpc as below: 4: Simple expression parser test child forked, pid 170385 Using CPUID 004e2102 division by zero syntax error syntax error FAILED tests/expr.c:65 parse test failed test child finished with -1 Simple expression parser: FAILED! This is observed after commit: 'commit 9d5da30e ("perf jevents: Add a new expression builtin strcmp_cpuid_str()")' With this commit, a new expression builtin strcmp_cpuid_str got added. This function takes an 'ID' type value, which is a string. So expression parse for strcmp_cpuid_str expects const char * as cpuid value type. In case of powerpc, CPU IDs are numbers. Hence it doesn't get interpreted correctly by bison parser. Example in case of power9, cpuid string returns as: 004e2102 cpuid of string type is expected in two cases: 1. char *get_cpuid_str(struct perf_pmu *pmu __maybe_unused); Testcase "tests/expr.c" uses "perf_pmu__getcpuid" which calls get_cpuid_str to get the cpuid string. 2. cpuid field in :struct pmu_events_map struct pmu_events_map { const char *arch; const char *cpuid; Here cpuid field is used in "perf_pmu__find_events_table" function as "strcmp_cpuid_str(map->cpuid, cpuid)". The value for cpuid field is picked from mapfile.csv. Fix the mapfile.csv and get_cpuid_str function to prefix cpuid with 0x so that it gets correctly interpreted by the bison parser Signed-off-by: -
Leo Yan authored
When users pass the option '--timestamp' or '-T' in the record command, all events will set the PERF_SAMPLE_TIME bit in the attribution. In this case, the AUX event will record the kernel timestamp, but it doesn't mean Arm CoreSight enables timestamp packets in its hardware tracing. If the option '--timestamp' or '-T' is set, this patch always enables Arm CoreSight timestamp, as a result, the bit 28 in event's config is to be set. Before: # perf record -e cs_etm// --per-thread --timestamp -- ls # perf script --header-only ... # event : name = cs_etm//, , id = { 69 }, type = 12, size = 136, config = 0, { sample_period, sample_freq } = 1, sample_type = IP|TID|TIME|CPU|IDENTIFIER, read_format = ID|LOST, disabled = 1, enable_on_exec = 1, sample_id_all = 1, exclude_guest = 1 ... After: # perf record -e cs_etm// --per-thread --timestamp -- ls # perf script --header-only ... # event : name = cs_etm//, , id = { 49 }, type = 12, size = 136, config = 0x10000000, { sample_period, sample_freq } = 1, sample_type = IP|TID|TIME|CPU|IDENTIFIER, read_format = ID|LOST, disabled = 1, enable_on_exec = 1, sample_id_all = 1, exclude_guest = 1 ... Signed-off-by:James Clark <james.clark@arm.com> Acked-by:
Suzuki K Poulose <suzuki.poulose@arm.com> Cc: Will Deacon <will@kernel.org> Cc: Mike Leach <mike.leach@linaro.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: John Garry <john.g.garry@oracle.com> Cc: linux-arm-kernel@lists.infradead.org Cc: coresight@lists.linaro.org Link: https://lore.kernel.org/r/20231014074159.1667880-3-leo.yan@linaro.orgSigned-off-by:
-
Leo Yan authored
So far, it's impossible to validate timestamp trace in Arm CoreSight when the perf is in the per-thread mode. E.g. for the command: perf record -e cs_etm/timestamp/ --per-thread -- ls The command enables config 'timestamp' for 'cs_etm' event in the per-thread mode. In this case, the function cs_etm_validate_config() directly bails out and skips validation. Given profiled process can be scheduled on any CPUs in the per-thread mode, this patch validates timestamp tracing for all CPUs when detect the CPU map is empty. Signed-off-by:
-
Ian Rogers authored
The default config is computed during creation of the PMU and may do things like scanning sysfs, when the PMU may just be used as part of scanning. Change default_config to perf_event_attr_init_default, a callback that is used when a default config needs initializing. This avoids holding onto the memory for a perf_event_attr and copying. On a tigerlake laptop running the pmu-scan benchmark: Before: Running 'internals/pmu-scan' benchmark: Computing performance of sysfs PMU event scan for 100 times Average core PMU scanning took: 28.780 usec (+- 0.503 usec) Average PMU scanning took: 283.480 usec (+- 18.471 usec) Number of openat syscalls: 30,227 After: Running 'internals/pmu-scan' benchmark: Computing performance of sysfs PMU event scan for 100 times Average core PMU scanning took: 27.880 usec (+- 0.169 usec) Average PMU scanning took: 245.260 usec (+- 15.758 usec) Number of openat syscalls: 28,914 Over 3 runs it is a nearly 12% reduction in execution time and a 4.3% of openat calls. Signed-off-by:
Adrian Hunter <adrian.hunter@intel.com> Cc: Ravi Bangoria <ravi.bangoria@amd.com> Cc: James Clark <james.clark@arm.com> Cc: Suzuki K Poulose <suzuki.poulose@arm.com> Cc: Yang Jihong <yangjihong1@huawei.com> Cc: Will Deacon <will@kernel.org> Cc: Leo Yan <leo.yan@linaro.org> Cc: Mike Leach <mike.leach@linaro.org> Cc: Jing Zhang <renyu.zj@linux.alibaba.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Thomas Richter <tmricht@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: John Garry <john.g.garry@oracle.com> Cc: linux-arm-kernel@lists.infradead.org Cc: coresight@lists.linaro.org Link: https://lore.kernel.org/r/20231012175645.1849503-8-irogers@google.comSigned-off-by:
-
Ian Rogers authored
strcmp_cpuid_str performs regular expression comparisons and so per CPUID linear searches over the perf_events_map are expensive. Add a helper function called map_for_pmu that does the search but also caches the map specific to a PMU. As the PMU may differ, also cache the CPUID string so that PMUs with the same CPUID string don't require the linear search and regular expression comparisons. This speeds loading PMUs as the search is done once per PMU to find the appropriate tables. Signed-off-by:
-
Ian Rogers authored
Add const to related APIs, this is so they can be used to default initialize a perf_event_attr from a const pmu. Signed-off-by:
-
Ian Rogers authored
File APIs don't alter the struct pmu so allow const ones to be passed. Signed-off-by:
-
Ian Rogers authored
Avoid setting PMU values in arm_spe_pmu_default_config, move to perf_pmu__arch_init. Signed-off-by:
-
Ian Rogers authored
Avoid setting PMU values in intel_pt_pmu_default_config, move to perf_pmu__arch_init. Signed-off-by:
-
Ian Rogers authored
Assign default_config as part of the init. perf_pmu__get_default_config was doing more than just getting the default config and so this is intended to better align with the code. Signed-off-by:
-
Adrian Hunter authored
Use get_unaligned_le64() instead of memcpy_le64(..., 8) because it produces simpler code. Signed-off-by:
-
Adrian Hunter authored
Avoid unaligned access by using get_unaligned_le16(), get_unaligned_le32() and get_unaligned_le64(). Signed-off-by:
-
Adrian Hunter authored
Use definitions from tools/include/linux/kernel.h Signed-off-by:
-
Adrian Hunter authored
Simplify and remove unnecessary constant expressions. Signed-off-by:
-
Adrian Hunter authored
Add get_unaligned_le16(), get_unaligned_le32 and get_unaligned_le64, same as include/asm-generic/unaligned.h. And add include/asm-generic/unaligned.h to check-headers.sh bringing tools/include/asm-generic/unaligned.h up to date so that the kernel and tools versions match. Use diagnostic pragmas to ignore -Wpacked used by perf build. Signed-off-by:
-
- 12 Oct, 2023 9 commits
-
-
Besar Wicaksono authored
The decoder creation for raw trace uses metadata from the first CPU. On per-cpu mode, this metadata is incorrectly used for every decoder. On per-process/per-thread traces, the first CPU is CPU0. If CPU0 trace is not enabled, its metadata will be marked unused and the decoder is not created. Perf report dump skips the decoding part because the decoder is missing. To fix this, use metadata of the CPU associated with sample object. Signed-off-by:
Besar Wicaksono <bwicaksono@nvidia.com> Reviewed-by:
-
Ian Rogers authored
Memory leaks were detected by clang-tidy. Signed-off-by:
-
Ian Rogers authored
Memory leaks were detected by clang-tidy. Signed-off-by:
-
Ian Rogers authored
If opendir failed then closedir was passed NULL which is erroneous. Caught by clang-tidy. Signed-off-by:
-
Ian Rogers authored
io__getline will free the line on error but it doesn't clear the out argument. This may lead to the line being freed twice, like in tools/perf/util/srcline.c as detected by clang-tidy. Signed-off-by:
-
Ian Rogers authored
Add missing free on an error path as detected by clang-tidy. Signed-off-by:
-
Ian Rogers authored
If a memory allocation fails then the strdup-ed string needs freeing. Detected by clang-tidy. Signed-off-by:
-
Ian Rogers authored
On success path the sib_core and sib_thr values weren't being freed. Detected by clang-tidy. Signed-off-by:
-
Ian Rogers authored
On other code paths browser->he_selection is NULL checked, add a missing case reported by clang-tidy. Signed-off-by:
-
