Patches contributed by Eötvös Lorand University
commit 6c009ecef8cca28c7c09eb16d0802e37915a76e1
Merge: 98c2aaf8be5b d508afb437da
Author: Ingo Molnar <mingo@elte.hu>
Date: Tue Apr 7 12:05:21 2009 +0200
Merge branch 'linus' into perfcounters/core
Merge reason: need the upstream facility added by:
7f1e2ca: hrtimer: fix rq->lock inversion (again)
Signed-off-by: Ingo Molnar <mingo@elte.hu>
commit 98c2aaf8be5baf7193be37fb28bce8e7327158bc
Author: Ingo Molnar <mingo@elte.hu>
Date: Tue Apr 7 11:30:17 2009 +0200
x86, perfcounters: add atomic64_xchg()
Complete atomic64_t support on the 32-bit side by adding atomic64_xch().
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20090406094518.445450972@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
diff --git a/arch/x86/include/asm/atomic_32.h b/arch/x86/include/asm/atomic_32.h
index 977250ed8b89..aff9f1fcdcd7 100644
--- a/arch/x86/include/asm/atomic_32.h
+++ b/arch/x86/include/asm/atomic_32.h
@@ -291,19 +291,37 @@ atomic64_cmpxchg(atomic64_t *ptr, unsigned long long old_val,
}
/**
- * atomic64_set - set atomic64 variable
+ * atomic64_xchg - xchg atomic64 variable
* @ptr: pointer to type atomic64_t
* @new_val: value to assign
+ * @old_val: old value that was there
*
- * Atomically sets the value of @ptr to @new_val.
+ * Atomically xchgs the value of @ptr to @new_val and returns
+ * the old value.
*/
-static inline void atomic64_set(atomic64_t *ptr, unsigned long long new_val)
+
+static inline unsigned long long
+atomic64_xchg(atomic64_t *ptr, unsigned long long new_val)
{
unsigned long long old_val;
do {
old_val = atomic_read(ptr);
} while (atomic64_cmpxchg(ptr, old_val, new_val) != old_val);
+
+ return old_val;
+}
+
+/**
+ * atomic64_set - set atomic64 variable
+ * @ptr: pointer to type atomic64_t
+ * @new_val: value to assign
+ *
+ * Atomically sets the value of @ptr to @new_val.
+ */
+static inline void atomic64_set(atomic64_t *ptr, unsigned long long new_val)
+{
+ atomic64_xchg(ptr, new_val);
}
/**
commit 5e34437840d33554f69380584311743b39e8fbeb
Merge: 77d05632baee d508afb437da
Author: Ingo Molnar <mingo@elte.hu>
Date: Tue Apr 7 11:15:40 2009 +0200
Merge branch 'linus' into core/softlockup
Conflicts:
kernel/sysctl.c
commit 77d05632baee21b1cef8730d7c06aa69601e4dca
Author: Ingo Molnar <mingo@elte.hu>
Date: Tue Apr 7 08:55:36 2009 +0200
softlockup: make DETECT_HUNG_TASK default depend on DETECT_SOFTLOCKUP
Don't offer a default-y option when the user has turned off
CONFIG_DETECT_SOFTLOCKUP already.
Do offer it as 'y' only if DETECT_SOFTLOCKUP is on already.
This makes it match previous behavior - where the hung-task check was
embedded i CONFIG_DETECT_SOFTLOCKUP code.
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
index 4934eaa21e1e..898e07c33c3c 100644
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -189,7 +189,7 @@ config BOOTPARAM_SOFTLOCKUP_PANIC_VALUE
config DETECT_HUNG_TASK
bool "Detect Hung Tasks"
depends on DEBUG_KERNEL
- default y
+ default DETECT_SOFTLOCKUP
help
Say Y here to enable the kernel to detect "hung tasks",
which are bugs that cause the task to be stuck in
commit 81cdbe0509542324ad7d3282ab67c2b6716df663
Author: Ingo Molnar <mingo@elte.hu>
Date: Mon Mar 23 22:29:50 2009 +0100
perf_counter tools: increase cpu-cycles again
Commit b7368fdd7d decreased the CPU cycles interval 100-fold, but
this is causig kerneltop failures on my Nehalem box:
aldebaran:/home/mingo/linux/linux/Documentation/perf_counter>
./kerneltop
KernelTop refresh period: 2 seconds
ERROR: failed to keep up with mmap data
10,000 cycles is way too short.
What we should do instead on mostly-idle systems is some sort of
read/poll timeout, so that we display something every 2 seconds
for sure.
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
diff --git a/Documentation/perf_counter/kerneltop.c b/Documentation/perf_counter/kerneltop.c
index 8f9a303f28d8..2ab29b5e32e8 100644
--- a/Documentation/perf_counter/kerneltop.c
+++ b/Documentation/perf_counter/kerneltop.c
@@ -212,7 +212,7 @@ struct source_line {
const unsigned int default_count[] = {
- 10000,
+ 1000000,
1000000,
10000,
10000,
commit 193e8df1b465f206f8a286202680702e1c153367
Author: Ingo Molnar <mingo@elte.hu>
Date: Mon Mar 23 22:23:16 2009 +0100
perf_counter tools: fix build warning in kerneltop.c
Fix:
kerneltop.c: In function ‘record_ip’:
kerneltop.c:1005: warning: format ‘%016llx’ expects type ‘long long unsigned int’, but argument 2 has type ‘uint64_t’
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Paul Mackerras <paulus@samba.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Orig-LKML-Reference: <20090323172417.677932499@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
diff --git a/Documentation/perf_counter/kerneltop.c b/Documentation/perf_counter/kerneltop.c
index 25e80bc44556..8f9a303f28d8 100644
--- a/Documentation/perf_counter/kerneltop.c
+++ b/Documentation/perf_counter/kerneltop.c
@@ -1002,7 +1002,7 @@ static void record_ip(uint64_t ip, int counter)
assert(left <= middle && middle <= right);
if (!(left <= ip && ip <= right)) {
printf(" left: %016lx\n", left);
- printf(" ip: %016llx\n", ip);
+ printf(" ip: %016lx\n", (unsigned long)ip);
printf("right: %016lx\n", right);
}
assert(left <= ip && ip <= right);
commit 383c5f8cd7d253be0017d8d5a39cbb78aaf70206
Author: Ingo Molnar <mingo@elte.hu>
Date: Mon Mar 23 21:49:25 2009 +0100
perf_counter tools: tidy up in-kernel dependencies
Remove now unified perfstat.c and perf_counter.h, and link to the
in-kernel perf_counter.h.
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Paul Mackerras <paulus@samba.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Orig-LKML-Reference: <20090323172417.677932499@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
diff --git a/Documentation/perf_counter/Makefile b/Documentation/perf_counter/Makefile
index b45749753fcb..666da95a7877 100644
--- a/Documentation/perf_counter/Makefile
+++ b/Documentation/perf_counter/Makefile
@@ -2,7 +2,7 @@ BINS = kerneltop perfstat
all: $(BINS)
-kerneltop: kerneltop.c perfcounters.h
+kerneltop: kerneltop.c ../../include/linux/perf_counter.h
cc -O6 -Wall -lrt `pkg-config --cflags --libs glib-2.0` -o $@ $<
perfstat: kerneltop
diff --git a/Documentation/perf_counter/kerneltop.c b/Documentation/perf_counter/kerneltop.c
index 80b790553ec8..25e80bc44556 100644
--- a/Documentation/perf_counter/kerneltop.c
+++ b/Documentation/perf_counter/kerneltop.c
@@ -88,7 +88,7 @@
#include <linux/unistd.h>
-#include "include/linux/perf_counter.h"
+#include "../../include/linux/perf_counter.h"
/*
diff --git a/Documentation/perf_counter/perfcounters.h b/Documentation/perf_counter/perfcounters.h
deleted file mode 100644
index 32e24b9154ab..000000000000
--- a/Documentation/perf_counter/perfcounters.h
+++ /dev/null
@@ -1,142 +0,0 @@
-/*
- * Ioctls that can be done on a perf counter fd:
- */
-#define PERF_COUNTER_IOC_ENABLE _IO('$', 0)
-#define PERF_COUNTER_IOC_DISABLE _IO('$', 1)
-
-/*
- * prctl(PR_TASK_PERF_COUNTERS_DISABLE) will (cheaply) disable all
- * counters in the current task.
- */
-#define PR_TASK_PERF_COUNTERS_DISABLE 31
-#define PR_TASK_PERF_COUNTERS_ENABLE 32
-
-#define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
-
-#define rdclock() \
-({ \
- struct timespec ts; \
- \
- clock_gettime(CLOCK_MONOTONIC, &ts); \
- ts.tv_sec * 1000000000ULL + ts.tv_nsec; \
-})
-
-/*
- * Pick up some kernel type conventions:
- */
-#define __user
-#define asmlinkage
-
-typedef unsigned int __u32;
-typedef unsigned long long __u64;
-typedef long long __s64;
-
-/*
- * User-space ABI bits:
- */
-
-/*
- * Generalized performance counter event types, used by the hw_event.type
- * parameter of the sys_perf_counter_open() syscall:
- */
-enum hw_event_types {
- /*
- * Common hardware events, generalized by the kernel:
- */
- PERF_COUNT_CPU_CYCLES = 0,
- PERF_COUNT_INSTRUCTIONS = 1,
- PERF_COUNT_CACHE_REFERENCES = 2,
- PERF_COUNT_CACHE_MISSES = 3,
- PERF_COUNT_BRANCH_INSTRUCTIONS = 4,
- PERF_COUNT_BRANCH_MISSES = 5,
- PERF_COUNT_BUS_CYCLES = 6,
-
- PERF_HW_EVENTS_MAX = 7,
-
- /*
- * Special "software" counters provided by the kernel, even if
- * the hardware does not support performance counters. These
- * counters measure various physical and sw events of the
- * kernel (and allow the profiling of them as well):
- */
- PERF_COUNT_CPU_CLOCK = -1,
- PERF_COUNT_TASK_CLOCK = -2,
- PERF_COUNT_PAGE_FAULTS = -3,
- PERF_COUNT_CONTEXT_SWITCHES = -4,
- PERF_COUNT_CPU_MIGRATIONS = -5,
-
- PERF_SW_EVENTS_MIN = -6,
-};
-
-/*
- * IRQ-notification data record type:
- */
-enum perf_counter_record_type {
- PERF_RECORD_SIMPLE = 0,
- PERF_RECORD_IRQ = 1,
- PERF_RECORD_GROUP = 2,
-};
-
-/*
- * Hardware event to monitor via a performance monitoring counter:
- */
-struct perf_counter_hw_event {
- __s64 type;
-
- __u64 irq_period;
- __u64 record_type;
- __u64 read_format;
-
- __u64 disabled : 1, /* off by default */
- nmi : 1, /* NMI sampling */
- raw : 1, /* raw event type */
- inherit : 1, /* children inherit it */
- pinned : 1, /* must always be on PMU */
- exclusive : 1, /* only group on PMU */
- exclude_user : 1, /* don't count user */
- exclude_kernel : 1, /* ditto kernel */
- exclude_hv : 1, /* ditto hypervisor */
- exclude_idle : 1, /* don't count when idle */
-
- __reserved_1 : 54;
-
- __u32 extra_config_len;
- __u32 __reserved_4;
-
- __u64 __reserved_2;
- __u64 __reserved_3;
-};
-
-
-#ifdef __x86_64__
-# define __NR_perf_counter_open 295
-#endif
-
-#ifdef __i386__
-# define __NR_perf_counter_open 333
-#endif
-
-#ifdef __powerpc__
-#define __NR_perf_counter_open 319
-#endif
-
-asmlinkage int sys_perf_counter_open(
-
- struct perf_counter_hw_event *hw_event_uptr __user,
- pid_t pid,
- int cpu,
- int group_fd,
- unsigned long flags)
-{
- int ret;
-
- ret = syscall(
- __NR_perf_counter_open, hw_event_uptr, pid, cpu, group_fd, flags);
-#if defined(__x86_64__) || defined(__i386__)
- if (ret < 0 && ret > -4096) {
- errno = -ret;
- ret = -1;
- }
-#endif
- return ret;
-}
diff --git a/Documentation/perf_counter/perfstat.c b/Documentation/perf_counter/perfstat.c
deleted file mode 100644
index fd594468e655..000000000000
--- a/Documentation/perf_counter/perfstat.c
+++ /dev/null
@@ -1,251 +0,0 @@
-/*
- * perfstat: /usr/bin/time -alike performance counter statistics utility
- *
- * It summarizes the counter events of all tasks (and child tasks),
- * covering all CPUs that the command (or workload) executes on.
- * It only counts the per-task events of the workload started,
- * independent of how many other tasks run on those CPUs.
- *
- * Build with: cc -O2 -g -lrt -Wall -W -o perfstat perfstat.c
- *
- * Sample output:
- *
-
- $ ./perfstat -e 1 -e 3 -e 5 ls -lR /usr/include/ >/dev/null
-
- Performance counter stats for 'ls':
-
- 163516953 instructions
- 2295 cache-misses
- 2855182 branch-misses
-
- *
- * Copyright (C) 2008, Red Hat Inc, Ingo Molnar <mingo@redhat.com>
- *
- * Released under the GPLv2 (not later).
- *
- * Percpu counter support by: Yanmin Zhang <yanmin_zhang@linux.intel.com>
- * Symbolic event options by: Wu Fengguang <fengguang.wu@intel.com>
- */
-#define _GNU_SOURCE
-
-#include <assert.h>
-#include <getopt.h>
-#include <stdint.h>
-#include <stdlib.h>
-#include <string.h>
-#include <unistd.h>
-#include <ctype.h>
-#include <errno.h>
-#include <fcntl.h>
-#include <stdio.h>
-#include <time.h>
-
-#include <sys/syscall.h>
-#include <sys/ioctl.h>
-#include <sys/prctl.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <sys/time.h>
-#include <sys/wait.h>
-#include <sys/uio.h>
-
-#include <linux/unistd.h>
-
-#include "perfcounters.h"
-
-static int nr_cpus = 0;
-
-static int system_wide = 0;
-
-static void display_help(void)
-{
- unsigned int i;
- int e;
-
- printf(
- "Usage: perfstat [<events...>] <cmd...>\n\n"
- "PerfStat Options (up to %d event types can be specified):\n\n",
- MAX_COUNTERS);
- printf(
- " -e EVENT --event=EVENT # symbolic-name abbreviations");
-
- for (i = 0, e = PERF_HW_EVENTS_MAX; i < ARRAY_SIZE(event_symbols); i++) {
- if (e != event_symbols[i].event) {
- e = event_symbols[i].event;
- printf(
- "\n %2d: %-20s", e, event_symbols[i].symbol);
- } else
- printf(" %s", event_symbols[i].symbol);
- }
-
- printf("\n"
- " rNNN: raw event type\n\n"
- " -s # system-wide collection\n\n"
- " -c <cmd..> --command=<cmd..> # command+arguments to be timed.\n"
- "\n");
- exit(0);
-}
-
-static void process_options(int argc, char *argv[])
-{
- for (;;) {
- int option_index = 0;
- /** Options for getopt */
- static struct option long_options[] = {
- {"event", required_argument, NULL, 'e'},
- {"help", no_argument, NULL, 'h'},
- {"command", no_argument, NULL, 'c'},
- {NULL, 0, NULL, 0 }
- };
- int c = getopt_long(argc, argv, "+:e:c:s",
- long_options, &option_index);
- if (c == -1)
- break;
-
- switch (c) {
- case 'c':
- break;
- case 's':
- system_wide = 1;
- break;
- case 'e':
- parse_events(optarg);
- break;
- default:
- break;
- }
- }
- if (optind == argc)
- goto err;
-
- if (!nr_counters)
- nr_counters = 8;
- return;
-
-err:
- display_help();
-}
-
-char fault_here[1000000];
-
-static int fd[MAX_NR_CPUS][MAX_COUNTERS];
-
-static void create_counter(int counter)
-{
- struct perf_counter_hw_event hw_event;
-
- memset(&hw_event, 0, sizeof(hw_event));
- hw_event.type = event_id[counter];
- hw_event.raw = event_raw[counter];
- hw_event.record_type = PERF_RECORD_SIMPLE;
- hw_event.nmi = 0;
-
- if (system_wide) {
- int cpu;
- for (cpu = 0; cpu < nr_cpus; cpu ++) {
- fd[cpu][counter] = sys_perf_counter_open(&hw_event, -1, cpu, -1, 0);
- if (fd[cpu][counter] < 0) {
- printf("perfstat error: syscall returned with %d (%s)\n",
- fd[cpu][counter], strerror(errno));
- exit(-1);
- }
-
- }
- } else {
- hw_event.inherit = 1;
- hw_event.disabled = 1;
-
- fd[0][counter] = sys_perf_counter_open(&hw_event, 0, -1, -1, 0);
- if (fd[0][counter] < 0) {
- printf("perfstat error: syscall returned with %d (%s)\n",
- fd[0][counter], strerror(errno));
- exit(-1);
- }
- }
-}
-
-
-int main(int argc, char *argv[])
-{
- unsigned long long t0, t1;
- int counter;
- ssize_t res;
- int status;
- int pid;
-
- process_options(argc, argv);
-
- if (system_wide) {
- nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
- assert(nr_cpus <= MAX_NR_CPUS);
- assert(nr_cpus >= 0);
- } else
- nr_cpus = 1;
-
- for (counter = 0; counter < nr_counters; counter++)
- create_counter(counter);
-
- argc -= optind;
- argv += optind;
-
- /*
- * Enable counters and exec the command:
- */
- t0 = rdclock();
- prctl(PR_TASK_PERF_COUNTERS_ENABLE);
-
- if ((pid = fork()) < 0)
- perror("failed to fork");
- if (!pid) {
- if (execvp(argv[0], argv)) {
- perror(argv[0]);
- exit(-1);
- }
- }
- while (wait(&status) >= 0)
- ;
- prctl(PR_TASK_PERF_COUNTERS_DISABLE);
- t1 = rdclock();
-
- fflush(stdout);
-
- fprintf(stderr, "\n");
- fprintf(stderr, " Performance counter stats for \'%s\':\n",
- argv[0]);
- fprintf(stderr, "\n");
-
- for (counter = 0; counter < nr_counters; counter++) {
- int cpu;
- __u64 count, single_count;
-
- count = 0;
- for (cpu = 0; cpu < nr_cpus; cpu ++) {
- res = read(fd[cpu][counter],
- (char *) &single_count, sizeof(single_count));
- assert(res == sizeof(single_count));
- count += single_count;
- }
-
- if (!event_raw[counter] &&
- (event_id[counter] == PERF_COUNT_CPU_CLOCK ||
- event_id[counter] == PERF_COUNT_TASK_CLOCK)) {
-
- double msecs = (double)count / 1000000;
-
- fprintf(stderr, " %14.6f %-20s (msecs)\n",
- msecs, event_name(counter));
- } else {
- fprintf(stderr, " %14Ld %-20s (events)\n",
- count, event_name(counter));
- }
- if (!counter)
- fprintf(stderr, "\n");
- }
- fprintf(stderr, "\n");
- fprintf(stderr, " Wall-clock time elapsed: %12.6f msecs\n",
- (double)(t1-t0)/1e6);
- fprintf(stderr, "\n");
-
- return 0;
-}
commit e0143bad9dbf2a8fad4c5430562bceba196b66ea
Author: Ingo Molnar <mingo@elte.hu>
Date: Mon Mar 23 21:29:59 2009 +0100
perf_counter: add sample user-space to Documentation/perf_counter/
Initial version of kerneltop.c and perfstat.c.
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
diff --git a/Documentation/perf_counter/Makefile b/Documentation/perf_counter/Makefile
new file mode 100644
index 000000000000..b45749753fcb
--- /dev/null
+++ b/Documentation/perf_counter/Makefile
@@ -0,0 +1,12 @@
+BINS = kerneltop perfstat
+
+all: $(BINS)
+
+kerneltop: kerneltop.c perfcounters.h
+ cc -O6 -Wall -lrt `pkg-config --cflags --libs glib-2.0` -o $@ $<
+
+perfstat: kerneltop
+ ln -sf kerneltop perfstat
+
+clean:
+ rm $(BINS)
diff --git a/Documentation/perf_counter/kerneltop.c b/Documentation/perf_counter/kerneltop.c
new file mode 100644
index 000000000000..cf0e30bab5d5
--- /dev/null
+++ b/Documentation/perf_counter/kerneltop.c
@@ -0,0 +1,956 @@
+/*
+ * kerneltop.c: show top kernel functions - performance counters showcase
+
+ Build with:
+
+ cc -O6 -Wall `pkg-config --cflags --libs glib-2.0` -o kerneltop kerneltop.c
+
+ Sample output:
+
+------------------------------------------------------------------------------
+ KernelTop: 2669 irqs/sec [NMI, cache-misses/cache-refs], (all, cpu: 2)
+------------------------------------------------------------------------------
+
+ weight RIP kernel function
+ ______ ________________ _______________
+
+ 35.20 - ffffffff804ce74b : skb_copy_and_csum_dev
+ 33.00 - ffffffff804cb740 : sock_alloc_send_skb
+ 31.26 - ffffffff804ce808 : skb_push
+ 22.43 - ffffffff80510004 : tcp_established_options
+ 19.00 - ffffffff8027d250 : find_get_page
+ 15.76 - ffffffff804e4fc9 : eth_type_trans
+ 15.20 - ffffffff804d8baa : dst_release
+ 14.86 - ffffffff804cf5d8 : skb_release_head_state
+ 14.00 - ffffffff802217d5 : read_hpet
+ 12.00 - ffffffff804ffb7f : __ip_local_out
+ 11.97 - ffffffff804fc0c8 : ip_local_deliver_finish
+ 8.54 - ffffffff805001a3 : ip_queue_xmit
+
+ Started by Ingo Molnar <mingo@redhat.com>
+
+ Improvements and fixes by:
+
+ Arjan van de Ven <arjan@linux.intel.com>
+ Yanmin Zhang <yanmin.zhang@intel.com>
+ Mike Galbraith <efault@gmx.de>
+
+ Released under the GPL v2. (and only v2, not any later version)
+
+ */
+#define _GNU_SOURCE
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/time.h>
+#include <unistd.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include <getopt.h>
+#include <assert.h>
+#include <fcntl.h>
+#include <stdio.h>
+#include <errno.h>
+#include <ctype.h>
+#include <time.h>
+
+#include <glib.h>
+
+#include <sys/syscall.h>
+#include <sys/ioctl.h>
+#include <sys/poll.h>
+#include <sys/prctl.h>
+#include <sys/wait.h>
+#include <sys/uio.h>
+
+#include <linux/unistd.h>
+
+#ifdef __x86_64__
+# define __NR_perf_counter_open 295
+#endif
+
+#ifdef __i386__
+# define __NR_perf_counter_open 333
+#endif
+
+/*
+ * Pick up some kernel type conventions:
+ */
+#define __user
+#define asmlinkage
+
+typedef unsigned int __u32;
+typedef unsigned long long __u64;
+typedef long long __s64;
+
+/*
+ * User-space ABI bits:
+ */
+
+/*
+ * Generalized performance counter event types, used by the hw_event.type
+ * parameter of the sys_perf_counter_open() syscall:
+ */
+enum hw_event_types {
+ /*
+ * Common hardware events, generalized by the kernel:
+ */
+ PERF_COUNT_CPU_CYCLES = 0,
+ PERF_COUNT_INSTRUCTIONS = 1,
+ PERF_COUNT_CACHE_REFERENCES = 2,
+ PERF_COUNT_CACHE_MISSES = 3,
+ PERF_COUNT_BRANCH_INSTRUCTIONS = 4,
+ PERF_COUNT_BRANCH_MISSES = 5,
+ PERF_COUNT_BUS_CYCLES = 6,
+
+ PERF_HW_EVENTS_MAX = 7,
+
+ /*
+ * Special "software" counters provided by the kernel, even if
+ * the hardware does not support performance counters. These
+ * counters measure various physical and sw events of the
+ * kernel (and allow the profiling of them as well):
+ */
+ PERF_COUNT_CPU_CLOCK = -1,
+ PERF_COUNT_TASK_CLOCK = -2,
+ PERF_COUNT_PAGE_FAULTS = -3,
+ PERF_COUNT_CONTEXT_SWITCHES = -4,
+ PERF_COUNT_CPU_MIGRATIONS = -5,
+
+ PERF_SW_EVENTS_MIN = -6,
+};
+
+/*
+ * IRQ-notification data record type:
+ */
+enum perf_counter_record_type {
+ PERF_RECORD_SIMPLE = 0,
+ PERF_RECORD_IRQ = 1,
+ PERF_RECORD_GROUP = 2,
+};
+
+/*
+ * Hardware event to monitor via a performance monitoring counter:
+ */
+struct perf_counter_hw_event {
+ __s64 type;
+
+ __u64 irq_period;
+ __u64 record_type;
+ __u64 read_format;
+
+ __u64 disabled : 1, /* off by default */
+ nmi : 1, /* NMI sampling */
+ raw : 1, /* raw event type */
+ inherit : 1, /* children inherit it */
+ pinned : 1, /* must always be on PMU */
+ exclusive : 1, /* only group on PMU */
+ exclude_user : 1, /* don't count user */
+ exclude_kernel : 1, /* ditto kernel */
+ exclude_hv : 1, /* ditto hypervisor */
+ exclude_idle : 1, /* don't count when idle */
+
+ __reserved_1 : 54;
+
+ __u32 extra_config_len;
+ __u32 __reserved_4;
+
+ __u64 __reserved_2;
+ __u64 __reserved_3;
+};
+
+/*
+ * Ioctls that can be done on a perf counter fd:
+ */
+#define PERF_COUNTER_IOC_ENABLE _IO('$', 0)
+#define PERF_COUNTER_IOC_DISABLE _IO('$', 1)
+
+asmlinkage int sys_perf_counter_open(
+
+ struct perf_counter_hw_event *hw_event_uptr __user,
+ pid_t pid,
+ int cpu,
+ int group_fd,
+ unsigned long flags)
+{
+ int ret;
+
+ ret = syscall(
+ __NR_perf_counter_open, hw_event_uptr, pid, cpu, group_fd, flags);
+#if defined(__x86_64__) || defined(__i386__)
+ if (ret < 0 && ret > -4096) {
+ errno = -ret;
+ ret = -1;
+ }
+#endif
+ return ret;
+}
+
+const char *event_types [] = {
+ "CPU cycles",
+ "instructions",
+ "cache-refs",
+ "cache-misses",
+ "branches",
+ "branch-misses",
+ "bus cycles"
+};
+
+const unsigned int default_count[] = {
+ 1000000,
+ 1000000,
+ 10000,
+ 10000,
+ 1000000,
+ 10000,
+};
+
+/*
+ * prctl(PR_TASK_PERF_COUNTERS_DISABLE) will (cheaply) disable all
+ * counters in the current task.
+ */
+#define PR_TASK_PERF_COUNTERS_DISABLE 31
+#define PR_TASK_PERF_COUNTERS_ENABLE 32
+
+#define MAX_COUNTERS 8
+
+static int nr_counters = -1;
+
+static __u64 count_filter = 100;
+
+#define MAX_NR_CPUS 256
+
+static int event_count[MAX_COUNTERS];
+static unsigned long event_id[MAX_COUNTERS];
+static int event_raw[MAX_COUNTERS];
+
+static int tid = -1;
+static int profile_cpu = -1;
+static int nr_cpus = 0;
+static int nmi = 1;
+static int group = 0;
+
+static char *vmlinux;
+
+static char *sym_filter;
+static unsigned long filter_start;
+static unsigned long filter_end;
+
+static int delay_secs = 2;
+static int zero;
+static int dump_symtab;
+
+struct source_line {
+ uint64_t EIP;
+ unsigned long count;
+ char *line;
+};
+
+static GList *lines;
+
+static void display_help(void)
+{
+ printf(
+ "Usage: kerneltop [<options>]\n\n"
+ "KernelTop Options (up to %d event types can be specified at once):\n\n",
+ MAX_COUNTERS);
+ printf(
+ " -e EID --event_id=EID # event type ID [default: 0]\n"
+ " 0: CPU cycles\n"
+ " 1: instructions\n"
+ " 2: cache accesses\n"
+ " 3: cache misses\n"
+ " 4: branch instructions\n"
+ " 5: branch prediction misses\n"
+ " 6: bus cycles\n\n"
+ " rNNN: raw PMU events (eventsel+umask)\n\n"
+ " -c CNT --count=CNT # event period to sample\n\n"
+ " -C CPU --cpu=CPU # CPU (-1 for all) [default: -1]\n"
+ " -p PID --pid=PID # PID of sampled task (-1 for all) [default: -1]\n\n"
+ " -d delay --delay=<seconds> # sampling/display delay [default: 2]\n"
+ " -f CNT --filter=CNT # min-event-count filter [default: 100]\n\n"
+ " -s symbol --symbol=<symbol> # function to be showed annotated one-shot\n"
+ " -x path --vmlinux=<path> # the vmlinux binary, required for -s use:\n"
+ " -z --zero # zero counts after display\n"
+ " -D --dump_symtab # dump symbol table to stderr on startup\n"
+ "\n");
+
+ exit(0);
+}
+
+static void process_options(int argc, char *argv[])
+{
+ int error = 0, counter;
+
+ for (;;) {
+ int option_index = 0;
+ /** Options for getopt */
+ static struct option long_options[] = {
+ {"count", required_argument, NULL, 'c'},
+ {"cpu", required_argument, NULL, 'C'},
+ {"delay", required_argument, NULL, 'd'},
+ {"dump_symtab", no_argument, NULL, 'D'},
+ {"event_id", required_argument, NULL, 'e'},
+ {"filter", required_argument, NULL, 'f'},
+ {"group", required_argument, NULL, 'g'},
+ {"help", no_argument, NULL, 'h'},
+ {"nmi", required_argument, NULL, 'n'},
+ {"pid", required_argument, NULL, 'p'},
+ {"vmlinux", required_argument, NULL, 'x'},
+ {"symbol", required_argument, NULL, 's'},
+ {"zero", no_argument, NULL, 'z'},
+ {NULL, 0, NULL, 0 }
+ };
+ int c = getopt_long(argc, argv, "c:C:d:De:f:g:hn:p:s:x:z",
+ long_options, &option_index);
+ if (c == -1)
+ break;
+
+ switch (c) {
+ case 'c':
+ if (nr_counters == -1)
+ nr_counters = 0;
+ event_count[nr_counters] = atoi(optarg); break;
+ case 'C':
+ /* CPU and PID are mutually exclusive */
+ if (tid != -1) {
+ printf("WARNING: CPU switch overriding PID\n");
+ sleep(1);
+ tid = -1;
+ }
+ profile_cpu = atoi(optarg); break;
+ case 'd': delay_secs = atoi(optarg); break;
+ case 'D': dump_symtab = 1; break;
+
+ case 'e':
+ nr_counters++;
+ if (nr_counters == MAX_COUNTERS) {
+ error = 1;
+ break;
+ }
+ if (*optarg == 'r') {
+ event_raw[nr_counters] = 1;
+ ++optarg;
+ }
+ event_id[nr_counters] = strtol(optarg, NULL, 16);
+ break;
+
+ case 'f': count_filter = atoi(optarg); break;
+ case 'g': group = atoi(optarg); break;
+ case 'h': display_help(); break;
+ case 'n': nmi = atoi(optarg); break;
+ case 'p':
+ /* CPU and PID are mutually exclusive */
+ if (profile_cpu != -1) {
+ printf("WARNING: PID switch overriding CPU\n");
+ sleep(1);
+ profile_cpu = -1;
+ }
+ tid = atoi(optarg); break;
+ case 's': sym_filter = strdup(optarg); break;
+ case 'x': vmlinux = strdup(optarg); break;
+ case 'z': zero = 1; break;
+ default: error = 1; break;
+ }
+ }
+ if (error)
+ display_help();
+
+ nr_counters++;
+ if (nr_counters < 1)
+ nr_counters = 1;
+
+ for (counter = 0; counter < nr_counters; counter++) {
+ if (event_count[counter])
+ continue;
+
+ if (event_id[counter] < PERF_HW_EVENTS_MAX)
+ event_count[counter] = default_count[event_id[counter]];
+ else
+ event_count[counter] = 100000;
+ }
+}
+
+static uint64_t min_ip;
+static uint64_t max_ip = -1ll;
+
+struct sym_entry {
+ unsigned long long addr;
+ char *sym;
+ unsigned long count[MAX_COUNTERS];
+ int skip;
+ GList *source;
+};
+
+#define MAX_SYMS 100000
+
+static int sym_table_count;
+
+struct sym_entry *sym_filter_entry;
+
+static struct sym_entry sym_table[MAX_SYMS];
+
+static void show_details(struct sym_entry *sym);
+
+/*
+ * Ordering weight: count-1 * count-1 * ... / count-n
+ */
+static double sym_weight(const struct sym_entry *sym)
+{
+ double weight;
+ int counter;
+
+ weight = sym->count[0];
+
+ for (counter = 1; counter < nr_counters-1; counter++)
+ weight *= sym->count[counter];
+
+ weight /= (sym->count[counter] + 1);
+
+ return weight;
+}
+
+static int compare(const void *__sym1, const void *__sym2)
+{
+ const struct sym_entry *sym1 = __sym1, *sym2 = __sym2;
+
+ return sym_weight(sym1) < sym_weight(sym2);
+}
+
+static time_t last_refresh;
+static long events;
+static long userspace_events;
+static const char CONSOLE_CLEAR[] = "[H[2J";
+
+static struct sym_entry tmp[MAX_SYMS];
+
+static void print_sym_table(void)
+{
+ int i, printed;
+ int counter;
+ float events_per_sec = events/delay_secs;
+ float kevents_per_sec = (events-userspace_events)/delay_secs;
+
+ memcpy(tmp, sym_table, sizeof(sym_table[0])*sym_table_count);
+ qsort(tmp, sym_table_count, sizeof(tmp[0]), compare);
+
+ write(1, CONSOLE_CLEAR, strlen(CONSOLE_CLEAR));
+
+ printf(
+"------------------------------------------------------------------------------\n");
+ printf( " KernelTop:%8.0f irqs/sec kernel:%3.1f%% [%s, ",
+ events_per_sec,
+ 100.0 - (100.0*((events_per_sec-kevents_per_sec)/events_per_sec)),
+ nmi ? "NMI" : "IRQ");
+
+ if (nr_counters == 1)
+ printf("%d ", event_count[0]);
+
+ for (counter = 0; counter < nr_counters; counter++) {
+ if (counter)
+ printf("/");
+
+ if (event_id[counter] < PERF_HW_EVENTS_MAX)
+ printf( "%s", event_types[event_id[counter]]);
+ else
+ printf( "raw:%04lx", event_id[counter]);
+ }
+
+ printf( "], ");
+
+ if (tid != -1)
+ printf(" (tid: %d", tid);
+ else
+ printf(" (all");
+
+ if (profile_cpu != -1)
+ printf(", cpu: %d)\n", profile_cpu);
+ else {
+ if (tid != -1)
+ printf(")\n");
+ else
+ printf(", %d CPUs)\n", nr_cpus);
+ }
+
+ printf("------------------------------------------------------------------------------\n\n");
+
+ if (nr_counters == 1)
+ printf(" events");
+ else
+ printf(" weight events");
+
+ printf(" RIP kernel function\n"
+ " ______ ______ ________________ _______________\n\n"
+ );
+
+ printed = 0;
+ for (i = 0; i < sym_table_count; i++) {
+ int count;
+
+ if (nr_counters == 1) {
+ if (printed <= 18 &&
+ tmp[i].count[0] >= count_filter) {
+ printf("%19.2f - %016llx : %s\n",
+ sym_weight(tmp + i), tmp[i].addr, tmp[i].sym);
+ printed++;
+ }
+ } else {
+ if (printed <= 18 &&
+ tmp[i].count[0] >= count_filter) {
+ printf("%8.1f %10ld - %016llx : %s\n",
+ sym_weight(tmp + i),
+ tmp[i].count[0],
+ tmp[i].addr, tmp[i].sym);
+ printed++;
+ }
+ }
+ /*
+ * Add decay to the counts:
+ */
+ for (count = 0; count < nr_counters; count++)
+ sym_table[i].count[count] = zero ? 0 : sym_table[i].count[count] * 7 / 8;
+ }
+
+ if (sym_filter_entry)
+ show_details(sym_filter_entry);
+
+ last_refresh = time(NULL);
+
+ {
+ struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
+
+ if (poll(&stdin_poll, 1, 0) == 1) {
+ printf("key pressed - exiting.\n");
+ exit(0);
+ }
+ }
+}
+
+static int read_symbol(FILE *in, struct sym_entry *s)
+{
+ static int filter_match = 0;
+ char *sym, stype;
+ char str[500];
+ int rc, pos;
+
+ rc = fscanf(in, "%llx %c %499s", &s->addr, &stype, str);
+ if (rc == EOF)
+ return -1;
+
+ assert(rc == 3);
+
+ /* skip until end of line: */
+ pos = strlen(str);
+ do {
+ rc = fgetc(in);
+ if (rc == '\n' || rc == EOF || pos >= 499)
+ break;
+ str[pos] = rc;
+ pos++;
+ } while (1);
+ str[pos] = 0;
+
+ sym = str;
+
+ /* Filter out known duplicates and non-text symbols. */
+ if (!strcmp(sym, "_text"))
+ return 1;
+ if (!min_ip && !strcmp(sym, "_stext"))
+ return 1;
+ if (!strcmp(sym, "_etext") || !strcmp(sym, "_sinittext"))
+ return 1;
+ if (stype != 'T' && stype != 't')
+ return 1;
+ if (!strncmp("init_module", sym, 11) || !strncmp("cleanup_module", sym, 14))
+ return 1;
+ if (strstr(sym, "_text_start") || strstr(sym, "_text_end"))
+ return 1;
+
+ s->sym = malloc(strlen(str));
+ assert(s->sym);
+
+ strcpy((char *)s->sym, str);
+ s->skip = 0;
+
+ /* Tag events to be skipped. */
+ if (!strcmp("default_idle", s->sym) || !strcmp("cpu_idle", s->sym))
+ s->skip = 1;
+ if (!strcmp("enter_idle", s->sym) || !strcmp("exit_idle", s->sym))
+ s->skip = 1;
+
+ if (filter_match == 1) {
+ filter_end = s->addr;
+ filter_match = -1;
+ if (filter_end - filter_start > 10000) {
+ printf("hm, too large filter symbol <%s> - skipping.\n",
+ sym_filter);
+ printf("symbol filter start: %016lx\n", filter_start);
+ printf(" end: %016lx\n", filter_end);
+ filter_end = filter_start = 0;
+ sym_filter = NULL;
+ sleep(1);
+ }
+ }
+ if (filter_match == 0 && sym_filter && !strcmp(s->sym, sym_filter)) {
+ filter_match = 1;
+ filter_start = s->addr;
+ }
+
+ return 0;
+}
+
+int compare_addr(const void *__sym1, const void *__sym2)
+{
+ const struct sym_entry *sym1 = __sym1, *sym2 = __sym2;
+
+ return sym1->addr > sym2->addr;
+}
+
+static void sort_symbol_table(void)
+{
+ int i, dups;
+
+ do {
+ qsort(sym_table, sym_table_count, sizeof(sym_table[0]), compare_addr);
+ for (i = 0, dups = 0; i < sym_table_count; i++) {
+ if (sym_table[i].addr == sym_table[i+1].addr) {
+ sym_table[i+1].addr = -1ll;
+ dups++;
+ }
+ }
+ sym_table_count -= dups;
+ } while(dups);
+}
+
+static void parse_symbols(void)
+{
+ struct sym_entry *last;
+
+ FILE *kallsyms = fopen("/proc/kallsyms", "r");
+
+ if (!kallsyms) {
+ printf("Could not open /proc/kallsyms - no CONFIG_KALLSYMS_ALL=y?\n");
+ exit(-1);
+ }
+
+ while (!feof(kallsyms)) {
+ if (read_symbol(kallsyms, &sym_table[sym_table_count]) == 0) {
+ sym_table_count++;
+ assert(sym_table_count <= MAX_SYMS);
+ }
+ }
+
+ sort_symbol_table();
+ min_ip = sym_table[0].addr;
+ max_ip = sym_table[sym_table_count-1].addr;
+ last = sym_table + sym_table_count++;
+
+ last->addr = -1ll;
+ last->sym = "<end>";
+
+ if (filter_end) {
+ int count;
+ for (count=0; count < sym_table_count; count ++) {
+ if (!strcmp(sym_table[count].sym, sym_filter)) {
+ sym_filter_entry = &sym_table[count];
+ break;
+ }
+ }
+ }
+ if (dump_symtab) {
+ int i;
+
+ for (i = 0; i < sym_table_count; i++)
+ fprintf(stderr, "%llx %s\n",
+ sym_table[i].addr, sym_table[i].sym);
+ }
+}
+
+
+static void parse_vmlinux(char *filename)
+{
+ FILE *file;
+ char command[PATH_MAX*2];
+ if (!filename)
+ return;
+
+ sprintf(command, "objdump --start-address=0x%016lx --stop-address=0x%016lx -dS %s", filter_start, filter_end, filename);
+
+ file = popen(command, "r");
+ if (!file)
+ return;
+
+ while (!feof(file)) {
+ struct source_line *src;
+ size_t dummy = 0;
+ char *c;
+
+ src = malloc(sizeof(struct source_line));
+ assert(src != NULL);
+ memset(src, 0, sizeof(struct source_line));
+
+ if (getline(&src->line, &dummy, file) < 0)
+ break;
+ if (!src->line)
+ break;
+
+ c = strchr(src->line, '\n');
+ if (c)
+ *c = 0;
+
+ lines = g_list_prepend(lines, src);
+
+ if (strlen(src->line)>8 && src->line[8] == ':')
+ src->EIP = strtoull(src->line, NULL, 16);
+ if (strlen(src->line)>8 && src->line[16] == ':')
+ src->EIP = strtoull(src->line, NULL, 16);
+ }
+ pclose(file);
+ lines = g_list_reverse(lines);
+}
+
+static void record_precise_ip(uint64_t ip)
+{
+ struct source_line *line;
+ GList *item;
+
+ item = g_list_first(lines);
+ while (item) {
+ line = item->data;
+ if (line->EIP == ip)
+ line->count++;
+ if (line->EIP > ip)
+ break;
+ item = g_list_next(item);
+ }
+}
+
+static void lookup_sym_in_vmlinux(struct sym_entry *sym)
+{
+ struct source_line *line;
+ GList *item;
+ char pattern[PATH_MAX];
+ sprintf(pattern, "<%s>:", sym->sym);
+
+ item = g_list_first(lines);
+ while (item) {
+ line = item->data;
+ if (strstr(line->line, pattern)) {
+ sym->source = item;
+ break;
+ }
+ item = g_list_next(item);
+ }
+}
+
+void show_lines(GList *item_queue, int item_queue_count)
+{
+ int i;
+ struct source_line *line;
+
+ for (i = 0; i < item_queue_count; i++) {
+ line = item_queue->data;
+ printf("%8li\t%s\n", line->count, line->line);
+ item_queue = g_list_next(item_queue);
+ }
+}
+
+#define TRACE_COUNT 3
+
+static void show_details(struct sym_entry *sym)
+{
+ struct source_line *line;
+ GList *item;
+ int displayed = 0;
+ GList *item_queue = NULL;
+ int item_queue_count = 0;
+
+ if (!sym->source)
+ lookup_sym_in_vmlinux(sym);
+ if (!sym->source)
+ return;
+
+ printf("Showing details for %s\n", sym->sym);
+
+ item = sym->source;
+ while (item) {
+ line = item->data;
+ if (displayed && strstr(line->line, ">:"))
+ break;
+
+ if (!item_queue_count)
+ item_queue = item;
+ item_queue_count ++;
+
+ if (line->count >= count_filter) {
+ show_lines(item_queue, item_queue_count);
+ item_queue_count = 0;
+ item_queue = NULL;
+ } else if (item_queue_count > TRACE_COUNT) {
+ item_queue = g_list_next(item_queue);
+ item_queue_count --;
+ }
+
+ line->count = 0;
+ displayed++;
+ if (displayed > 300)
+ break;
+ item = g_list_next(item);
+ }
+}
+
+/*
+ * Binary search in the histogram table and record the hit:
+ */
+static void record_ip(uint64_t ip, int counter)
+{
+ int left_idx, middle_idx, right_idx, idx;
+ unsigned long left, middle, right;
+
+ record_precise_ip(ip);
+
+ left_idx = 0;
+ right_idx = sym_table_count-1;
+ assert(ip <= max_ip && ip >= min_ip);
+
+ while (left_idx + 1 < right_idx) {
+ middle_idx = (left_idx + right_idx) / 2;
+
+ left = sym_table[ left_idx].addr;
+ middle = sym_table[middle_idx].addr;
+ right = sym_table[ right_idx].addr;
+
+ if (!(left <= middle && middle <= right)) {
+ printf("%016lx...\n%016lx...\n%016lx\n", left, middle, right);
+ printf("%d %d %d\n", left_idx, middle_idx, right_idx);
+ }
+ assert(left <= middle && middle <= right);
+ if (!(left <= ip && ip <= right)) {
+ printf(" left: %016lx\n", left);
+ printf(" ip: %016lx\n", ip);
+ printf("right: %016lx\n", right);
+ }
+ assert(left <= ip && ip <= right);
+ /*
+ * [ left .... target .... middle .... right ]
+ * => right := middle
+ */
+ if (ip < middle) {
+ right_idx = middle_idx;
+ continue;
+ }
+ /*
+ * [ left .... middle ... target ... right ]
+ * => left := middle
+ */
+ left_idx = middle_idx;
+ }
+
+ idx = left_idx;
+
+ if (!sym_table[idx].skip)
+ sym_table[idx].count[counter]++;
+ else events--;
+}
+
+static void process_event(uint64_t ip, int counter)
+{
+ events++;
+
+ if (ip < min_ip || ip > max_ip) {
+ userspace_events++;
+ return;
+ }
+
+ record_ip(ip, counter);
+}
+
+int main(int argc, char *argv[])
+{
+ struct pollfd event_array[MAX_NR_CPUS][MAX_COUNTERS];
+ struct perf_counter_hw_event hw_event;
+ int fd[MAX_NR_CPUS][MAX_COUNTERS];
+ int i, counter, group_fd;
+ unsigned int cpu;
+ uint64_t ip;
+ ssize_t res;
+ int ret;
+
+ process_options(argc, argv);
+
+ nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
+ if (tid != -1 || profile_cpu != -1)
+ nr_cpus = 1;
+
+ assert(nr_cpus <= MAX_NR_CPUS);
+
+ for (i = 0; i < nr_cpus; i++) {
+ group_fd = -1;
+ for (counter = 0; counter < nr_counters; counter++) {
+
+ cpu = profile_cpu;
+ if (tid == -1 && profile_cpu == -1)
+ cpu = i;
+
+ memset(&hw_event, 0, sizeof(hw_event));
+ hw_event.type = event_id[counter];
+ hw_event.raw = event_raw[counter];
+ hw_event.irq_period = event_count[counter];
+ hw_event.record_type = PERF_RECORD_IRQ;
+ hw_event.nmi = nmi;
+
+ fd[i][counter] = sys_perf_counter_open(&hw_event, tid, cpu, group_fd, 0);
+ fcntl(fd[i][counter], F_SETFL, O_NONBLOCK);
+ if (fd[i][counter] < 0) {
+ printf("kerneltop error: syscall returned with %d (%s)\n",
+ fd[i][counter], strerror(-fd[i][counter]));
+ if (fd[i][counter] == -1)
+ printf("Are you root?\n");
+ exit(-1);
+ }
+ assert(fd[i][counter] >= 0);
+
+ /*
+ * First counter acts as the group leader:
+ */
+ if (group && group_fd == -1)
+ group_fd = fd[i][counter];
+
+ event_array[i][counter].fd = fd[i][counter];
+ event_array[i][counter].events = POLLIN;
+ }
+ }
+
+ parse_symbols();
+ if (vmlinux && sym_filter_entry)
+ parse_vmlinux(vmlinux);
+
+ printf("KernelTop refresh period: %d seconds\n", delay_secs);
+ last_refresh = time(NULL);
+
+ while (1) {
+ int hits = events;
+
+ for (i = 0; i < nr_cpus; i++) {
+ for (counter = 0; counter < nr_counters; counter++) {
+ res = read(fd[i][counter], (char *) &ip, sizeof(ip));
+ if (res > 0) {
+ assert(res == sizeof(ip));
+
+ process_event(ip, counter);
+ }
+ }
+ }
+
+ if (time(NULL) >= last_refresh + delay_secs) {
+ print_sym_table();
+ events = userspace_events = 0;
+ }
+
+ if (hits == events)
+ ret = poll(event_array[0], nr_cpus, 1000);
+ hits = events;
+ }
+
+ return 0;
+}
diff --git a/Documentation/perf_counter/perfstat.c b/Documentation/perf_counter/perfstat.c
new file mode 100644
index 000000000000..9a5808fbcf90
--- /dev/null
+++ b/Documentation/perf_counter/perfstat.c
@@ -0,0 +1,521 @@
+/*
+ * perfstat: /usr/bin/time -alike performance counter statistics utility
+ *
+ * It summarizes the counter events of all tasks (and child tasks),
+ * covering all CPUs that the command (or workload) executes on.
+ * It only counts the per-task events of the workload started,
+ * independent of how many other tasks run on those CPUs.
+ *
+ * Build with: cc -O2 -g -lrt -Wall -W -o perfstat perfstat.c
+ *
+ * Sample output:
+ *
+
+ $ ./perfstat -e 1 -e 3 -e 5 ls -lR /usr/include/ >/dev/null
+
+ Performance counter stats for 'ls':
+
+ 163516953 instructions
+ 2295 cache-misses
+ 2855182 branch-misses
+
+ *
+ * Copyright (C) 2008, Red Hat Inc, Ingo Molnar <mingo@redhat.com>
+ *
+ * Released under the GPLv2 (not later).
+ *
+ * Percpu counter support by: Yanmin Zhang <yanmin_zhang@linux.intel.com>
+ * Symbolic event options by: Wu Fengguang <fengguang.wu@intel.com>
+ */
+#define _GNU_SOURCE
+
+#include <assert.h>
+#include <getopt.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+#include <ctype.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <stdio.h>
+#include <time.h>
+
+#include <sys/syscall.h>
+#include <sys/ioctl.h>
+#include <sys/prctl.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/time.h>
+#include <sys/wait.h>
+#include <sys/uio.h>
+
+#include <linux/unistd.h>
+
+#ifdef __x86_64__
+# define __NR_perf_counter_open 295
+#endif
+
+#ifdef __i386__
+# define __NR_perf_counter_open 333
+#endif
+
+#ifdef __powerpc__
+#define __NR_perf_counter_open 319
+#endif
+
+/*
+ * Pick up some kernel type conventions:
+ */
+#define __user
+#define asmlinkage
+
+typedef unsigned int __u32;
+typedef unsigned long long __u64;
+typedef long long __s64;
+
+#define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
+
+/*
+ * User-space ABI bits:
+ */
+
+/*
+ * Generalized performance counter event types, used by the hw_event.type
+ * parameter of the sys_perf_counter_open() syscall:
+ */
+enum hw_event_types {
+ /*
+ * Common hardware events, generalized by the kernel:
+ */
+ PERF_COUNT_CPU_CYCLES = 0,
+ PERF_COUNT_INSTRUCTIONS = 1,
+ PERF_COUNT_CACHE_REFERENCES = 2,
+ PERF_COUNT_CACHE_MISSES = 3,
+ PERF_COUNT_BRANCH_INSTRUCTIONS = 4,
+ PERF_COUNT_BRANCH_MISSES = 5,
+ PERF_COUNT_BUS_CYCLES = 6,
+
+ PERF_HW_EVENTS_MAX = 7,
+
+ /*
+ * Special "software" counters provided by the kernel, even if
+ * the hardware does not support performance counters. These
+ * counters measure various physical and sw events of the
+ * kernel (and allow the profiling of them as well):
+ */
+ PERF_COUNT_CPU_CLOCK = -1,
+ PERF_COUNT_TASK_CLOCK = -2,
+ PERF_COUNT_PAGE_FAULTS = -3,
+ PERF_COUNT_CONTEXT_SWITCHES = -4,
+ PERF_COUNT_CPU_MIGRATIONS = -5,
+
+ PERF_SW_EVENTS_MIN = -6,
+};
+
+/*
+ * IRQ-notification data record type:
+ */
+enum perf_counter_record_type {
+ PERF_RECORD_SIMPLE = 0,
+ PERF_RECORD_IRQ = 1,
+ PERF_RECORD_GROUP = 2,
+};
+
+/*
+ * Hardware event to monitor via a performance monitoring counter:
+ */
+struct perf_counter_hw_event {
+ __s64 type;
+
+ __u64 irq_period;
+ __u64 record_type;
+ __u64 read_format;
+
+ __u64 disabled : 1, /* off by default */
+ nmi : 1, /* NMI sampling */
+ raw : 1, /* raw event type */
+ inherit : 1, /* children inherit it */
+ pinned : 1, /* must always be on PMU */
+ exclusive : 1, /* only group on PMU */
+ exclude_user : 1, /* don't count user */
+ exclude_kernel : 1, /* ditto kernel */
+ exclude_hv : 1, /* ditto hypervisor */
+ exclude_idle : 1, /* don't count when idle */
+
+ __reserved_1 : 54;
+
+ __u32 extra_config_len;
+ __u32 __reserved_4;
+
+ __u64 __reserved_2;
+ __u64 __reserved_3;
+};
+
+/*
+ * Ioctls that can be done on a perf counter fd:
+ */
+#define PERF_COUNTER_IOC_ENABLE _IO('$', 0)
+#define PERF_COUNTER_IOC_DISABLE _IO('$', 1)
+
+asmlinkage int sys_perf_counter_open(
+
+ struct perf_counter_hw_event *hw_event_uptr __user,
+ pid_t pid,
+ int cpu,
+ int group_fd,
+ unsigned long flags)
+{
+ int ret;
+
+ ret = syscall(
+ __NR_perf_counter_open, hw_event_uptr, pid, cpu, group_fd, flags);
+#if defined(__x86_64__) || defined(__i386__)
+ if (ret < 0 && ret > -4096) {
+ errno = -ret;
+ ret = -1;
+ }
+#endif
+ return ret;
+}
+
+
+static char *hw_event_names [] = {
+ "CPU cycles",
+ "instructions",
+ "cache references",
+ "cache misses",
+ "branches",
+ "branch misses",
+ "bus cycles",
+};
+
+static char *sw_event_names [] = {
+ "cpu clock ticks",
+ "task clock ticks",
+ "pagefaults",
+ "context switches",
+ "CPU migrations",
+};
+
+struct event_symbol {
+ int event;
+ char *symbol;
+};
+
+static struct event_symbol event_symbols [] = {
+ {PERF_COUNT_CPU_CYCLES, "cpu-cycles", },
+ {PERF_COUNT_CPU_CYCLES, "cycles", },
+ {PERF_COUNT_INSTRUCTIONS, "instructions", },
+ {PERF_COUNT_CACHE_REFERENCES, "cache-references", },
+ {PERF_COUNT_CACHE_MISSES, "cache-misses", },
+ {PERF_COUNT_BRANCH_INSTRUCTIONS, "branch-instructions", },
+ {PERF_COUNT_BRANCH_INSTRUCTIONS, "branches", },
+ {PERF_COUNT_BRANCH_MISSES, "branch-misses", },
+ {PERF_COUNT_BUS_CYCLES, "bus-cycles", },
+ {PERF_COUNT_CPU_CLOCK, "cpu-ticks", },
+ {PERF_COUNT_CPU_CLOCK, "ticks", },
+ {PERF_COUNT_TASK_CLOCK, "task-ticks", },
+ {PERF_COUNT_PAGE_FAULTS, "page-faults", },
+ {PERF_COUNT_PAGE_FAULTS, "faults", },
+ {PERF_COUNT_CONTEXT_SWITCHES, "context-switches", },
+ {PERF_COUNT_CONTEXT_SWITCHES, "cs", },
+ {PERF_COUNT_CPU_MIGRATIONS, "cpu-migrations", },
+ {PERF_COUNT_CPU_MIGRATIONS, "migrations", },
+};
+
+#define MAX_COUNTERS 64
+#define MAX_NR_CPUS 256
+
+static int nr_counters = 0;
+static int nr_cpus = 0;
+
+static int event_id[MAX_COUNTERS] =
+ { -2, -5, -4, -3, 0, 1, 2, 3};
+
+static int event_raw[MAX_COUNTERS];
+
+static int system_wide = 0;
+
+static void display_help(void)
+{
+ unsigned int i;
+ int e;
+
+ printf(
+ "Usage: perfstat [<events...>] <cmd...>\n\n"
+ "PerfStat Options (up to %d event types can be specified):\n\n",
+ MAX_COUNTERS);
+ printf(
+ " -e EVENT --event=EVENT # symbolic-name abbreviations");
+
+ for (i = 0, e = PERF_HW_EVENTS_MAX; i < ARRAY_SIZE(event_symbols); i++) {
+ if (e != event_symbols[i].event) {
+ e = event_symbols[i].event;
+ printf(
+ "\n %2d: %-20s", e, event_symbols[i].symbol);
+ } else
+ printf(" %s", event_symbols[i].symbol);
+ }
+
+ printf("\n"
+ " rNNN: raw event type\n\n"
+ " -s # system-wide collection\n\n"
+ " -c <cmd..> --command=<cmd..> # command+arguments to be timed.\n"
+ "\n");
+ exit(0);
+}
+
+static int type_valid(int type)
+{
+ if (type >= PERF_HW_EVENTS_MAX)
+ return 0;
+ if (type <= PERF_SW_EVENTS_MIN)
+ return 0;
+
+ return 1;
+}
+
+static char *event_name(int ctr)
+{
+ int type = event_id[ctr];
+ static char buf[32];
+
+ if (event_raw[ctr]) {
+ sprintf(buf, "raw 0x%x", type);
+ return buf;
+ }
+ if (!type_valid(type))
+ return "unknown";
+
+ if (type >= 0)
+ return hw_event_names[type];
+
+ return sw_event_names[-type-1];
+}
+
+/*
+ * Each event can have multiple symbolic names.
+ * Symbolic names are (almost) exactly matched.
+ */
+static int match_event_symbols(char *str)
+{
+ unsigned int i;
+
+ if (isdigit(str[0]) || str[0] == '-')
+ return atoi(str);
+
+ for (i = 0; i < ARRAY_SIZE(event_symbols); i++) {
+ if (!strncmp(str, event_symbols[i].symbol,
+ strlen(event_symbols[i].symbol)))
+ return event_symbols[i].event;
+ }
+
+ return PERF_HW_EVENTS_MAX;
+}
+
+static void parse_events(char *str)
+{
+ int type, raw;
+
+again:
+ nr_counters++;
+ if (nr_counters == MAX_COUNTERS)
+ display_help();
+
+ raw = 0;
+ if (*str == 'r') {
+ raw = 1;
+ ++str;
+ type = strtol(str, NULL, 16);
+ } else {
+ type = match_event_symbols(str);
+ if (!type_valid(type))
+ display_help();
+ }
+
+ event_id[nr_counters] = type;
+ event_raw[nr_counters] = raw;
+
+ str = strstr(str, ",");
+ if (str) {
+ str++;
+ goto again;
+ }
+}
+
+static void process_options(int argc, char *argv[])
+{
+ for (;;) {
+ int option_index = 0;
+ /** Options for getopt */
+ static struct option long_options[] = {
+ {"event", required_argument, NULL, 'e'},
+ {"help", no_argument, NULL, 'h'},
+ {"command", no_argument, NULL, 'c'},
+ {NULL, 0, NULL, 0 }
+ };
+ int c = getopt_long(argc, argv, "+:e:c:s",
+ long_options, &option_index);
+ if (c == -1)
+ break;
+
+ switch (c) {
+ case 'c':
+ break;
+ case 's':
+ system_wide = 1;
+ break;
+ case 'e':
+ parse_events(optarg);
+ break;
+ default:
+ break;
+ }
+ }
+ if (optind == argc)
+ goto err;
+
+ if (!nr_counters)
+ nr_counters = 8;
+ else
+ nr_counters++;
+ return;
+
+err:
+ display_help();
+}
+
+char fault_here[1000000];
+
+#define PR_TASK_PERF_COUNTERS_DISABLE 31
+#define PR_TASK_PERF_COUNTERS_ENABLE 32
+
+static int fd[MAX_NR_CPUS][MAX_COUNTERS];
+
+static void create_counter(int counter)
+{
+ struct perf_counter_hw_event hw_event;
+
+ memset(&hw_event, 0, sizeof(hw_event));
+ hw_event.type = event_id[counter];
+ hw_event.raw = event_raw[counter];
+ hw_event.record_type = PERF_RECORD_SIMPLE;
+ hw_event.nmi = 0;
+
+ if (system_wide) {
+ int cpu;
+ for (cpu = 0; cpu < nr_cpus; cpu ++) {
+ fd[cpu][counter] = sys_perf_counter_open(&hw_event, -1, cpu, -1, 0);
+ if (fd[cpu][counter] < 0) {
+ printf("perfstat error: syscall returned with %d (%s)\n",
+ fd[cpu][counter], strerror(errno));
+ exit(-1);
+ }
+
+ }
+ } else {
+ hw_event.inherit = 1;
+ hw_event.disabled = 1;
+
+ fd[0][counter] = sys_perf_counter_open(&hw_event, 0, -1, -1, 0);
+ if (fd[0][counter] < 0) {
+ printf("perfstat error: syscall returned with %d (%s)\n",
+ fd[0][counter], strerror(errno));
+ exit(-1);
+ }
+ }
+}
+
+
+#define rdclock() \
+({ \
+ struct timespec ts; \
+ \
+ clock_gettime(CLOCK_MONOTONIC, &ts); \
+ ts.tv_sec * 1000000000ULL + ts.tv_nsec; \
+})
+
+int main(int argc, char *argv[])
+{
+ unsigned long long t0, t1;
+ int counter;
+ ssize_t res;
+ int status;
+ int pid;
+
+ process_options(argc, argv);
+
+ if (system_wide) {
+ nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
+ assert(nr_cpus <= MAX_NR_CPUS);
+ assert(nr_cpus >= 0);
+ } else
+ nr_cpus = 1;
+
+ for (counter = 0; counter < nr_counters; counter++)
+ create_counter(counter);
+
+ argc -= optind;
+ argv += optind;
+
+ /*
+ * Enable counters and exec the command:
+ */
+ t0 = rdclock();
+ prctl(PR_TASK_PERF_COUNTERS_ENABLE);
+
+ if ((pid = fork()) < 0)
+ perror("failed to fork");
+ if (!pid) {
+ if (execvp(argv[0], argv)) {
+ perror(argv[0]);
+ exit(-1);
+ }
+ }
+ while (wait(&status) >= 0)
+ ;
+ prctl(PR_TASK_PERF_COUNTERS_DISABLE);
+ t1 = rdclock();
+
+ fflush(stdout);
+
+ fprintf(stderr, "\n");
+ fprintf(stderr, " Performance counter stats for \'%s\':\n",
+ argv[0]);
+ fprintf(stderr, "\n");
+
+ for (counter = 0; counter < nr_counters; counter++) {
+ int cpu;
+ __u64 count, single_count;
+
+ count = 0;
+ for (cpu = 0; cpu < nr_cpus; cpu ++) {
+ res = read(fd[cpu][counter],
+ (char *) &single_count, sizeof(single_count));
+ assert(res == sizeof(single_count));
+ count += single_count;
+ }
+
+ if (!event_raw[counter] &&
+ (event_id[counter] == PERF_COUNT_CPU_CLOCK ||
+ event_id[counter] == PERF_COUNT_TASK_CLOCK)) {
+
+ double msecs = (double)count / 1000000;
+
+ fprintf(stderr, " %14.6f %-20s (msecs)\n",
+ msecs, event_name(counter));
+ } else {
+ fprintf(stderr, " %14Ld %-20s (events)\n",
+ count, event_name(counter));
+ }
+ if (!counter)
+ fprintf(stderr, "\n");
+ }
+ fprintf(stderr, "\n");
+ fprintf(stderr, " Wall-clock time elapsed: %12.6f msecs\n",
+ (double)(t1-t0)/1e6);
+ fprintf(stderr, "\n");
+
+ return 0;
+}
commit 6f9f791eb53b56097cd311a1a5517a8e89bdaf35
Author: Ingo Molnar <mingo@elte.hu>
Date: Mon Mar 23 21:26:52 2009 +0100
perf_counter: create Documentation/perf_counter/ and move perfcounters.txt there
We'll have more files in that directory, prepare for that.
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
diff --git a/Documentation/perf-counters.txt b/Documentation/perf_counter/design.txt
similarity index 100%
rename from Documentation/perf-counters.txt
rename to Documentation/perf_counter/design.txt
commit 7bb497bd885eedd0f56dfe3cc1b5ff20710d33b9
Author: Ingo Molnar <mingo@elte.hu>
Date: Wed Mar 18 08:59:21 2009 +0100
perf_counter: fix crash on perfmon v1 systems
Impact: fix boot crash on Intel Perfmon Version 1 systems
Intel Perfmon v1 does not support the global MSRs, nor does
it offer the generalized MSR ranges. So support v2 and later
CPUs only.
Also mark pmc_ops as read-mostly - to avoid false cacheline
sharing.
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c
index 22dab06c08a4..6cba9d47b711 100644
--- a/arch/x86/kernel/cpu/perf_counter.c
+++ b/arch/x86/kernel/cpu/perf_counter.c
@@ -57,12 +57,14 @@ struct pmc_x86_ops {
int max_events;
};
-static struct pmc_x86_ops *pmc_ops;
+static struct pmc_x86_ops *pmc_ops __read_mostly;
static DEFINE_PER_CPU(struct cpu_hw_counters, cpu_hw_counters) = {
.enabled = 1,
};
+static __read_mostly int intel_perfmon_version;
+
/*
* Intel PerfMon v3. Used on Core2 and later.
*/
@@ -613,7 +615,7 @@ void perf_counter_print_debug(void)
cpu = smp_processor_id();
cpuc = &per_cpu(cpu_hw_counters, cpu);
- if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
+ if (intel_perfmon_version >= 2) {
rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl);
rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
rdmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, overflow);
@@ -930,10 +932,10 @@ static struct pmc_x86_ops pmc_amd_ops = {
static struct pmc_x86_ops *pmc_intel_init(void)
{
+ union cpuid10_edx edx;
union cpuid10_eax eax;
- unsigned int ebx;
unsigned int unused;
- union cpuid10_edx edx;
+ unsigned int ebx;
/*
* Check whether the Architectural PerfMon supports
@@ -943,8 +945,12 @@ static struct pmc_x86_ops *pmc_intel_init(void)
if (eax.split.mask_length <= ARCH_PERFMON_BRANCH_MISSES_RETIRED)
return NULL;
+ intel_perfmon_version = eax.split.version_id;
+ if (intel_perfmon_version < 2)
+ return NULL;
+
pr_info("Intel Performance Monitoring support detected.\n");
- pr_info("... version: %d\n", eax.split.version_id);
+ pr_info("... version: %d\n", intel_perfmon_version);
pr_info("... bit width: %d\n", eax.split.bit_width);
pr_info("... mask length: %d\n", eax.split.mask_length);