^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2) Performance Counters for Linux
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3) ------------------------------
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 4)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 5) Performance counters are special hardware registers available on most modern
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 6) CPUs. These registers count the number of certain types of hw events: such
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 7) as instructions executed, cachemisses suffered, or branches mis-predicted -
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 8) without slowing down the kernel or applications. These registers can also
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 9) trigger interrupts when a threshold number of events have passed - and can
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 10) thus be used to profile the code that runs on that CPU.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 11)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 12) The Linux Performance Counter subsystem provides an abstraction of these
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 13) hardware capabilities. It provides per task and per CPU counters, counter
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 14) groups, and it provides event capabilities on top of those. It
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 15) provides "virtual" 64-bit counters, regardless of the width of the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 16) underlying hardware counters.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 17)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 18) Performance counters are accessed via special file descriptors.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 19) There's one file descriptor per virtual counter used.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 20)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 21) The special file descriptor is opened via the sys_perf_event_open()
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 22) system call:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 23)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 24) int sys_perf_event_open(struct perf_event_attr *hw_event_uptr,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 25) pid_t pid, int cpu, int group_fd,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 26) unsigned long flags);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 27)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 28) The syscall returns the new fd. The fd can be used via the normal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 29) VFS system calls: read() can be used to read the counter, fcntl()
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 30) can be used to set the blocking mode, etc.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 31)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 32) Multiple counters can be kept open at a time, and the counters
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 33) can be poll()ed.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 34)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 35) When creating a new counter fd, 'perf_event_attr' is:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 36)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 37) struct perf_event_attr {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 38) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 39) * The MSB of the config word signifies if the rest contains cpu
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 40) * specific (raw) counter configuration data, if unset, the next
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 41) * 7 bits are an event type and the rest of the bits are the event
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 42) * identifier.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 43) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 44) __u64 config;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 45)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 46) __u64 irq_period;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 47) __u32 record_type;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 48) __u32 read_format;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 49)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 50) __u64 disabled : 1, /* off by default */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 51) inherit : 1, /* children inherit it */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 52) pinned : 1, /* must always be on PMU */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 53) exclusive : 1, /* only group on PMU */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 54) exclude_user : 1, /* don't count user */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 55) exclude_kernel : 1, /* ditto kernel */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 56) exclude_hv : 1, /* ditto hypervisor */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 57) exclude_idle : 1, /* don't count when idle */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 58) mmap : 1, /* include mmap data */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 59) munmap : 1, /* include munmap data */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 60) comm : 1, /* include comm data */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 61)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 62) __reserved_1 : 52;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 63)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 64) __u32 extra_config_len;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 65) __u32 wakeup_events; /* wakeup every n events */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 66)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 67) __u64 __reserved_2;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 68) __u64 __reserved_3;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 69) };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 70)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 71) The 'config' field specifies what the counter should count. It
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 72) is divided into 3 bit-fields:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 73)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 74) raw_type: 1 bit (most significant bit) 0x8000_0000_0000_0000
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 75) type: 7 bits (next most significant) 0x7f00_0000_0000_0000
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 76) event_id: 56 bits (least significant) 0x00ff_ffff_ffff_ffff
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 77)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 78) If 'raw_type' is 1, then the counter will count a hardware event
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 79) specified by the remaining 63 bits of event_config. The encoding is
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 80) machine-specific.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 81)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 82) If 'raw_type' is 0, then the 'type' field says what kind of counter
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 83) this is, with the following encoding:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 84)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 85) enum perf_type_id {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 86) PERF_TYPE_HARDWARE = 0,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 87) PERF_TYPE_SOFTWARE = 1,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 88) PERF_TYPE_TRACEPOINT = 2,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 89) };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 90)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 91) A counter of PERF_TYPE_HARDWARE will count the hardware event
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 92) specified by 'event_id':
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 93)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 94) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 95) * Generalized performance counter event types, used by the hw_event.event_id
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 96) * parameter of the sys_perf_event_open() syscall:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 97) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 98) enum perf_hw_id {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 99) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 100) * Common hardware events, generalized by the kernel:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 101) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 102) PERF_COUNT_HW_CPU_CYCLES = 0,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 103) PERF_COUNT_HW_INSTRUCTIONS = 1,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 104) PERF_COUNT_HW_CACHE_REFERENCES = 2,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 105) PERF_COUNT_HW_CACHE_MISSES = 3,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 106) PERF_COUNT_HW_BRANCH_INSTRUCTIONS = 4,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 107) PERF_COUNT_HW_BRANCH_MISSES = 5,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 108) PERF_COUNT_HW_BUS_CYCLES = 6,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 109) };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 110)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 111) These are standardized types of events that work relatively uniformly
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 112) on all CPUs that implement Performance Counters support under Linux,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 113) although there may be variations (e.g., different CPUs might count
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 114) cache references and misses at different levels of the cache hierarchy).
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 115) If a CPU is not able to count the selected event, then the system call
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 116) will return -EINVAL.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 117)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 118) More hw_event_types are supported as well, but they are CPU-specific
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 119) and accessed as raw events. For example, to count "External bus
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 120) cycles while bus lock signal asserted" events on Intel Core CPUs, pass
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 121) in a 0x4064 event_id value and set hw_event.raw_type to 1.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 122)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 123) A counter of type PERF_TYPE_SOFTWARE will count one of the available
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 124) software events, selected by 'event_id':
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 125)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 126) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 127) * Special "software" counters provided by the kernel, even if the hardware
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 128) * does not support performance counters. These counters measure various
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 129) * physical and sw events of the kernel (and allow the profiling of them as
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 130) * well):
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 131) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 132) enum perf_sw_ids {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 133) PERF_COUNT_SW_CPU_CLOCK = 0,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 134) PERF_COUNT_SW_TASK_CLOCK = 1,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 135) PERF_COUNT_SW_PAGE_FAULTS = 2,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 136) PERF_COUNT_SW_CONTEXT_SWITCHES = 3,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 137) PERF_COUNT_SW_CPU_MIGRATIONS = 4,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 138) PERF_COUNT_SW_PAGE_FAULTS_MIN = 5,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 139) PERF_COUNT_SW_PAGE_FAULTS_MAJ = 6,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 140) PERF_COUNT_SW_ALIGNMENT_FAULTS = 7,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 141) PERF_COUNT_SW_EMULATION_FAULTS = 8,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 142) };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 143)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 144) Counters of the type PERF_TYPE_TRACEPOINT are available when the ftrace event
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 145) tracer is available, and event_id values can be obtained from
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 146) /debug/tracing/events/*/*/id
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 147)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 148)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 149) Counters come in two flavours: counting counters and sampling
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 150) counters. A "counting" counter is one that is used for counting the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 151) number of events that occur, and is characterised by having
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 152) irq_period = 0.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 153)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 154)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 155) A read() on a counter returns the current value of the counter and possible
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 156) additional values as specified by 'read_format', each value is a u64 (8 bytes)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 157) in size.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 158)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 159) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 160) * Bits that can be set in hw_event.read_format to request that
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 161) * reads on the counter should return the indicated quantities,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 162) * in increasing order of bit value, after the counter value.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 163) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 164) enum perf_event_read_format {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 165) PERF_FORMAT_TOTAL_TIME_ENABLED = 1,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 166) PERF_FORMAT_TOTAL_TIME_RUNNING = 2,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 167) };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 168)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 169) Using these additional values one can establish the overcommit ratio for a
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 170) particular counter allowing one to take the round-robin scheduling effect
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 171) into account.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 172)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 173)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 174) A "sampling" counter is one that is set up to generate an interrupt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 175) every N events, where N is given by 'irq_period'. A sampling counter
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 176) has irq_period > 0. The record_type controls what data is recorded on each
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 177) interrupt:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 178)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 179) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 180) * Bits that can be set in hw_event.record_type to request information
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 181) * in the overflow packets.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 182) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 183) enum perf_event_record_format {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 184) PERF_RECORD_IP = 1U << 0,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 185) PERF_RECORD_TID = 1U << 1,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 186) PERF_RECORD_TIME = 1U << 2,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 187) PERF_RECORD_ADDR = 1U << 3,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 188) PERF_RECORD_GROUP = 1U << 4,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 189) PERF_RECORD_CALLCHAIN = 1U << 5,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 190) };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 191)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 192) Such (and other) events will be recorded in a ring-buffer, which is
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 193) available to user-space using mmap() (see below).
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 194)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 195) The 'disabled' bit specifies whether the counter starts out disabled
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 196) or enabled. If it is initially disabled, it can be enabled by ioctl
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 197) or prctl (see below).
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 198)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 199) The 'inherit' bit, if set, specifies that this counter should count
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 200) events on descendant tasks as well as the task specified. This only
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 201) applies to new descendents, not to any existing descendents at the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 202) time the counter is created (nor to any new descendents of existing
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 203) descendents).
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 204)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 205) The 'pinned' bit, if set, specifies that the counter should always be
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 206) on the CPU if at all possible. It only applies to hardware counters
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 207) and only to group leaders. If a pinned counter cannot be put onto the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 208) CPU (e.g. because there are not enough hardware counters or because of
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 209) a conflict with some other event), then the counter goes into an
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 210) 'error' state, where reads return end-of-file (i.e. read() returns 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 211) until the counter is subsequently enabled or disabled.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 212)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 213) The 'exclusive' bit, if set, specifies that when this counter's group
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 214) is on the CPU, it should be the only group using the CPU's counters.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 215) In future, this will allow sophisticated monitoring programs to supply
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 216) extra configuration information via 'extra_config_len' to exploit
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 217) advanced features of the CPU's Performance Monitor Unit (PMU) that are
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 218) not otherwise accessible and that might disrupt other hardware
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 219) counters.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 220)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 221) The 'exclude_user', 'exclude_kernel' and 'exclude_hv' bits provide a
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 222) way to request that counting of events be restricted to times when the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 223) CPU is in user, kernel and/or hypervisor mode.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 224)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 225) Furthermore the 'exclude_host' and 'exclude_guest' bits provide a way
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 226) to request counting of events restricted to guest and host contexts when
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 227) using Linux as the hypervisor.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 228)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 229) The 'mmap' and 'munmap' bits allow recording of PROT_EXEC mmap/munmap
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 230) operations, these can be used to relate userspace IP addresses to actual
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 231) code, even after the mapping (or even the whole process) is gone,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 232) these events are recorded in the ring-buffer (see below).
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 233)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 234) The 'comm' bit allows tracking of process comm data on process creation.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 235) This too is recorded in the ring-buffer (see below).
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 236)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 237) The 'pid' parameter to the sys_perf_event_open() system call allows the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 238) counter to be specific to a task:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 239)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 240) pid == 0: if the pid parameter is zero, the counter is attached to the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 241) current task.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 242)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 243) pid > 0: the counter is attached to a specific task (if the current task
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 244) has sufficient privilege to do so)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 245)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 246) pid < 0: all tasks are counted (per cpu counters)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 247)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 248) The 'cpu' parameter allows a counter to be made specific to a CPU:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 249)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 250) cpu >= 0: the counter is restricted to a specific CPU
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 251) cpu == -1: the counter counts on all CPUs
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 252)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 253) (Note: the combination of 'pid == -1' and 'cpu == -1' is not valid.)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 254)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 255) A 'pid > 0' and 'cpu == -1' counter is a per task counter that counts
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 256) events of that task and 'follows' that task to whatever CPU the task
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 257) gets schedule to. Per task counters can be created by any user, for
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 258) their own tasks.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 259)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 260) A 'pid == -1' and 'cpu == x' counter is a per CPU counter that counts
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 261) all events on CPU-x. Per CPU counters need CAP_PERFMON or CAP_SYS_ADMIN
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 262) privilege.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 263)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 264) The 'flags' parameter is currently unused and must be zero.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 265)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 266) The 'group_fd' parameter allows counter "groups" to be set up. A
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 267) counter group has one counter which is the group "leader". The leader
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 268) is created first, with group_fd = -1 in the sys_perf_event_open call
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 269) that creates it. The rest of the group members are created
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 270) subsequently, with group_fd giving the fd of the group leader.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 271) (A single counter on its own is created with group_fd = -1 and is
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 272) considered to be a group with only 1 member.)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 273)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 274) A counter group is scheduled onto the CPU as a unit, that is, it will
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 275) only be put onto the CPU if all of the counters in the group can be
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 276) put onto the CPU. This means that the values of the member counters
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 277) can be meaningfully compared, added, divided (to get ratios), etc.,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 278) with each other, since they have counted events for the same set of
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 279) executed instructions.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 280)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 281)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 282) Like stated, asynchronous events, like counter overflow or PROT_EXEC mmap
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 283) tracking are logged into a ring-buffer. This ring-buffer is created and
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 284) accessed through mmap().
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 285)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 286) The mmap size should be 1+2^n pages, where the first page is a meta-data page
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 287) (struct perf_event_mmap_page) that contains various bits of information such
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 288) as where the ring-buffer head is.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 289)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 290) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 291) * Structure of the page that can be mapped via mmap
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 292) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 293) struct perf_event_mmap_page {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 294) __u32 version; /* version number of this structure */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 295) __u32 compat_version; /* lowest version this is compat with */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 296)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 297) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 298) * Bits needed to read the hw counters in user-space.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 299) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 300) * u32 seq;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 301) * s64 count;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 302) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 303) * do {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 304) * seq = pc->lock;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 305) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 306) * barrier()
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 307) * if (pc->index) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 308) * count = pmc_read(pc->index - 1);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 309) * count += pc->offset;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 310) * } else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 311) * goto regular_read;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 312) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 313) * barrier();
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 314) * } while (pc->lock != seq);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 315) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 316) * NOTE: for obvious reason this only works on self-monitoring
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 317) * processes.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 318) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 319) __u32 lock; /* seqlock for synchronization */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 320) __u32 index; /* hardware counter identifier */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 321) __s64 offset; /* add to hardware counter value */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 322)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 323) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 324) * Control data for the mmap() data buffer.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 325) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 326) * User-space reading this value should issue an rmb(), on SMP capable
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 327) * platforms, after reading this value -- see perf_event_wakeup().
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 328) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 329) __u32 data_head; /* head in the data section */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 330) };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 331)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 332) NOTE: the hw-counter userspace bits are arch specific and are currently only
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 333) implemented on powerpc.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 334)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 335) The following 2^n pages are the ring-buffer which contains events of the form:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 336)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 337) #define PERF_RECORD_MISC_KERNEL (1 << 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 338) #define PERF_RECORD_MISC_USER (1 << 1)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 339) #define PERF_RECORD_MISC_OVERFLOW (1 << 2)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 340)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 341) struct perf_event_header {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 342) __u32 type;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 343) __u16 misc;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 344) __u16 size;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 345) };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 346)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 347) enum perf_event_type {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 348)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 349) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 350) * The MMAP events record the PROT_EXEC mappings so that we can
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 351) * correlate userspace IPs to code. They have the following structure:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 352) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 353) * struct {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 354) * struct perf_event_header header;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 355) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 356) * u32 pid, tid;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 357) * u64 addr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 358) * u64 len;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 359) * u64 pgoff;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 360) * char filename[];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 361) * };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 362) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 363) PERF_RECORD_MMAP = 1,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 364) PERF_RECORD_MUNMAP = 2,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 365)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 366) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 367) * struct {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 368) * struct perf_event_header header;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 369) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 370) * u32 pid, tid;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 371) * char comm[];
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 372) * };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 373) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 374) PERF_RECORD_COMM = 3,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 375)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 376) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 377) * When header.misc & PERF_RECORD_MISC_OVERFLOW the event_type field
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 378) * will be PERF_RECORD_*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 379) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 380) * struct {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 381) * struct perf_event_header header;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 382) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 383) * { u64 ip; } && PERF_RECORD_IP
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 384) * { u32 pid, tid; } && PERF_RECORD_TID
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 385) * { u64 time; } && PERF_RECORD_TIME
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 386) * { u64 addr; } && PERF_RECORD_ADDR
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 387) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 388) * { u64 nr;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 389) * { u64 event, val; } cnt[nr]; } && PERF_RECORD_GROUP
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 390) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 391) * { u16 nr,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 392) * hv,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 393) * kernel,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 394) * user;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 395) * u64 ips[nr]; } && PERF_RECORD_CALLCHAIN
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 396) * };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 397) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 398) };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 399)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 400) NOTE: PERF_RECORD_CALLCHAIN is arch specific and currently only implemented
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 401) on x86.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 402)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 403) Notification of new events is possible through poll()/select()/epoll() and
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 404) fcntl() managing signals.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 405)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 406) Normally a notification is generated for every page filled, however one can
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 407) additionally set perf_event_attr.wakeup_events to generate one every
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 408) so many counter overflow events.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 409)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 410) Future work will include a splice() interface to the ring-buffer.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 411)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 412)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 413) Counters can be enabled and disabled in two ways: via ioctl and via
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 414) prctl. When a counter is disabled, it doesn't count or generate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 415) events but does continue to exist and maintain its count value.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 416)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 417) An individual counter can be enabled with
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 418)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 419) ioctl(fd, PERF_EVENT_IOC_ENABLE, 0);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 420)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 421) or disabled with
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 422)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 423) ioctl(fd, PERF_EVENT_IOC_DISABLE, 0);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 424)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 425) For a counter group, pass PERF_IOC_FLAG_GROUP as the third argument.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 426) Enabling or disabling the leader of a group enables or disables the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 427) whole group; that is, while the group leader is disabled, none of the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 428) counters in the group will count. Enabling or disabling a member of a
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 429) group other than the leader only affects that counter - disabling an
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 430) non-leader stops that counter from counting but doesn't affect any
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 431) other counter.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 432)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 433) Additionally, non-inherited overflow counters can use
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 434)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 435) ioctl(fd, PERF_EVENT_IOC_REFRESH, nr);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 436)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 437) to enable a counter for 'nr' events, after which it gets disabled again.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 438)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 439) A process can enable or disable all the counter groups that are
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 440) attached to it, using prctl:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 441)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 442) prctl(PR_TASK_PERF_EVENTS_ENABLE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 443)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 444) prctl(PR_TASK_PERF_EVENTS_DISABLE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 445)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 446) This applies to all counters on the current process, whether created
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 447) by this process or by another, and doesn't affect any counters that
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 448) this process has created on other processes. It only enables or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 449) disables the group leaders, not any other members in the groups.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 450)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 451)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 452) Arch requirements
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 453) -----------------
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 454)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 455) If your architecture does not have hardware performance metrics, you can
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 456) still use the generic software counters based on hrtimers for sampling.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 457)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 458) So to start with, in order to add HAVE_PERF_EVENTS to your Kconfig, you
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 459) will need at least this:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 460) - asm/perf_event.h - a basic stub will suffice at first
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 461) - support for atomic64 types (and associated helper functions)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 462)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 463) If your architecture does have hardware capabilities, you can override the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 464) weak stub hw_perf_event_init() to register hardware counters.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 465)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 466) Architectures that have d-cache aliassing issues, such as Sparc and ARM,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 467) should select PERF_USE_VMALLOC in order to avoid these for perf mmap().
Orange Pi5 kernel
Deprecated Linux kernel 5.10.110 for OrangePi 5/5B/5+ boards
3 Commits
0 Branches
0 Tags