diff options
| author | Ingo Molnar <mingo@elte.hu> | 2010-10-16 14:17:25 -0400 |
|---|---|---|
| committer | Ingo Molnar <mingo@elte.hu> | 2010-10-16 14:17:25 -0400 |
| commit | f92f6e6ee35d2779aa62e70f78ad8e1cd417eb52 (patch) | |
| tree | 1e8e2ee34678a43d416c4bab58f9ca91673d4444 /arch | |
| parent | 66af86e2c630908b21cec018cb612576cf5f516e (diff) | |
| parent | cd254f295248c98b62ea824f361e10d80a081fe7 (diff) | |
Merge branch 'core' of git://git.kernel.org/pub/scm/linux/kernel/git/rric/oprofile into perf/core
Diffstat (limited to 'arch')
| -rw-r--r-- | arch/arm/kernel/perf_event.c | 6 | ||||
| -rw-r--r-- | arch/arm/oprofile/Makefile | 4 | ||||
| -rw-r--r-- | arch/arm/oprofile/common.c | 311 | ||||
| -rw-r--r-- | arch/sh/Kconfig | 13 | ||||
| -rw-r--r-- | arch/sh/kernel/perf_event.c | 18 | ||||
| -rw-r--r-- | arch/sh/oprofile/Makefile | 4 | ||||
| -rw-r--r-- | arch/sh/oprofile/common.c | 115 | ||||
| -rw-r--r-- | arch/sh/oprofile/op_impl.h | 33 | ||||
| -rw-r--r-- | arch/x86/oprofile/backtrace.c | 70 | ||||
| -rw-r--r-- | arch/x86/oprofile/nmi_int.c | 9 |
10 files changed, 134 insertions, 449 deletions
diff --git a/arch/arm/kernel/perf_event.c b/arch/arm/kernel/perf_event.c index 55addc85eaa7..6cc6521881aa 100644 --- a/arch/arm/kernel/perf_event.c +++ b/arch/arm/kernel/perf_event.c | |||
| @@ -123,6 +123,12 @@ armpmu_get_max_events(void) | |||
| 123 | } | 123 | } |
| 124 | EXPORT_SYMBOL_GPL(armpmu_get_max_events); | 124 | EXPORT_SYMBOL_GPL(armpmu_get_max_events); |
| 125 | 125 | ||
| 126 | int perf_num_counters(void) | ||
| 127 | { | ||
| 128 | return armpmu_get_max_events(); | ||
| 129 | } | ||
| 130 | EXPORT_SYMBOL_GPL(perf_num_counters); | ||
| 131 | |||
| 126 | #define HW_OP_UNSUPPORTED 0xFFFF | 132 | #define HW_OP_UNSUPPORTED 0xFFFF |
| 127 | 133 | ||
| 128 | #define C(_x) \ | 134 | #define C(_x) \ |
diff --git a/arch/arm/oprofile/Makefile b/arch/arm/oprofile/Makefile index e666eafed152..b2215c61cdf0 100644 --- a/arch/arm/oprofile/Makefile +++ b/arch/arm/oprofile/Makefile | |||
| @@ -6,4 +6,8 @@ DRIVER_OBJS = $(addprefix ../../../drivers/oprofile/, \ | |||
| 6 | oprofilefs.o oprofile_stats.o \ | 6 | oprofilefs.o oprofile_stats.o \ |
| 7 | timer_int.o ) | 7 | timer_int.o ) |
| 8 | 8 | ||
| 9 | ifeq ($(CONFIG_HW_PERF_EVENTS),y) | ||
| 10 | DRIVER_OBJS += $(addprefix ../../../drivers/oprofile/, oprofile_perf.o) | ||
| 11 | endif | ||
| 12 | |||
| 9 | oprofile-y := $(DRIVER_OBJS) common.o | 13 | oprofile-y := $(DRIVER_OBJS) common.o |
diff --git a/arch/arm/oprofile/common.c b/arch/arm/oprofile/common.c index d1fb5b2f5218..8aa974491dfc 100644 --- a/arch/arm/oprofile/common.c +++ b/arch/arm/oprofile/common.c | |||
| @@ -25,139 +25,10 @@ | |||
| 25 | #include <asm/ptrace.h> | 25 | #include <asm/ptrace.h> |
| 26 | 26 | ||
| 27 | #ifdef CONFIG_HW_PERF_EVENTS | 27 | #ifdef CONFIG_HW_PERF_EVENTS |
| 28 | /* | 28 | char *op_name_from_perf_id(void) |
| 29 | * Per performance monitor configuration as set via oprofilefs. | ||
| 30 | */ | ||
| 31 | struct op_counter_config { | ||
| 32 | unsigned long count; | ||
| 33 | unsigned long enabled; | ||
| 34 | unsigned long event; | ||
| 35 | unsigned long unit_mask; | ||
| 36 | unsigned long kernel; | ||
| 37 | unsigned long user; | ||
| 38 | struct perf_event_attr attr; | ||
| 39 | }; | ||
| 40 | |||
| 41 | static int op_arm_enabled; | ||
| 42 | static DEFINE_MUTEX(op_arm_mutex); | ||
| 43 | |||
| 44 | static struct op_counter_config *counter_config; | ||
| 45 | static struct perf_event **perf_events[nr_cpumask_bits]; | ||
| 46 | static int perf_num_counters; | ||
| 47 | |||
| 48 | /* | ||
| 49 | * Overflow callback for oprofile. | ||
| 50 | */ | ||
| 51 | static void op_overflow_handler(struct perf_event *event, int unused, | ||
| 52 | struct perf_sample_data *data, struct pt_regs *regs) | ||
| 53 | { | 29 | { |
| 54 | int id; | 30 | enum arm_perf_pmu_ids id = armpmu_get_pmu_id(); |
| 55 | u32 cpu = smp_processor_id(); | ||
| 56 | |||
| 57 | for (id = 0; id < perf_num_counters; ++id) | ||
| 58 | if (perf_events[cpu][id] == event) | ||
| 59 | break; | ||
| 60 | |||
| 61 | if (id != perf_num_counters) | ||
| 62 | oprofile_add_sample(regs, id); | ||
| 63 | else | ||
| 64 | pr_warning("oprofile: ignoring spurious overflow " | ||
| 65 | "on cpu %u\n", cpu); | ||
| 66 | } | ||
| 67 | |||
| 68 | /* | ||
| 69 | * Called by op_arm_setup to create perf attributes to mirror the oprofile | ||
| 70 | * settings in counter_config. Attributes are created as `pinned' events and | ||
| 71 | * so are permanently scheduled on the PMU. | ||
| 72 | */ | ||
| 73 | static void op_perf_setup(void) | ||
| 74 | { | ||
| 75 | int i; | ||
| 76 | u32 size = sizeof(struct perf_event_attr); | ||
| 77 | struct perf_event_attr *attr; | ||
| 78 | |||
| 79 | for (i = 0; i < perf_num_counters; ++i) { | ||
| 80 | attr = &counter_config[i].attr; | ||
| 81 | memset(attr, 0, size); | ||
| 82 | attr->type = PERF_TYPE_RAW; | ||
| 83 | attr->size = size; | ||
| 84 | attr->config = counter_config[i].event; | ||
| 85 | attr->sample_period = counter_config[i].count; | ||
| 86 | attr->pinned = 1; | ||
| 87 | } | ||
| 88 | } | ||
| 89 | |||
| 90 | static int op_create_counter(int cpu, int event) | ||
| 91 | { | ||
| 92 | int ret = 0; | ||
| 93 | struct perf_event *pevent; | ||
| 94 | |||
| 95 | if (!counter_config[event].enabled || (perf_events[cpu][event] != NULL)) | ||
| 96 | return ret; | ||
| 97 | |||
| 98 | pevent = perf_event_create_kernel_counter(&counter_config[event].attr, | ||
| 99 | cpu, NULL, | ||
| 100 | op_overflow_handler); | ||
| 101 | |||
| 102 | if (IS_ERR(pevent)) { | ||
| 103 | ret = PTR_ERR(pevent); | ||
| 104 | } else if (pevent->state != PERF_EVENT_STATE_ACTIVE) { | ||
| 105 | perf_event_release_kernel(pevent); | ||
| 106 | pr_warning("oprofile: failed to enable event %d " | ||
| 107 | "on CPU %d\n", event, cpu); | ||
| 108 | ret = -EBUSY; | ||
| 109 | } else { | ||
| 110 | perf_events[cpu][event] = pevent; | ||
| 111 | } | ||
| 112 | |||
| 113 | return ret; | ||
| 114 | } | ||
| 115 | 31 | ||
| 116 | static void op_destroy_counter(int cpu, int event) | ||
| 117 | { | ||
| 118 | struct perf_event *pevent = perf_events[cpu][event]; | ||
| 119 | |||
| 120 | if (pevent) { | ||
| 121 | perf_event_release_kernel(pevent); | ||
| 122 | perf_events[cpu][event] = NULL; | ||
| 123 | } | ||
| 124 | } | ||
| 125 | |||
| 126 | /* | ||
| 127 | * Called by op_arm_start to create active perf events based on the | ||
| 128 | * perviously configured attributes. | ||
| 129 | */ | ||
| 130 | static int op_perf_start(void) | ||
| 131 | { | ||
| 132 | int cpu, event, ret = 0; | ||
| 133 | |||
| 134 | for_each_online_cpu(cpu) { | ||
| 135 | for (event = 0; event < perf_num_counters; ++event) { | ||
| 136 | ret = op_create_counter(cpu, event); | ||
| 137 | if (ret) | ||
| 138 | goto out; | ||
| 139 | } | ||
| 140 | } | ||
| 141 | |||
| 142 | out: | ||
| 143 | return ret; | ||
| 144 | } | ||
| 145 | |||
| 146 | /* | ||
| 147 | * Called by op_arm_stop at the end of a profiling run. | ||
| 148 | */ | ||
| 149 | static void op_perf_stop(void) | ||
| 150 | { | ||
| 151 | int cpu, event; | ||
| 152 | |||
| 153 | for_each_online_cpu(cpu) | ||
| 154 | for (event = 0; event < perf_num_counters; ++event) | ||
| 155 | op_destroy_counter(cpu, event); | ||
| 156 | } | ||
| 157 | |||
| 158 | |||
| 159 | static char *op_name_from_perf_id(enum arm_perf_pmu_ids id) | ||
| 160 | { | ||
| 161 | switch (id) { | 32 | switch (id) { |
| 162 | case ARM_PERF_PMU_ID_XSCALE1: | 33 | case ARM_PERF_PMU_ID_XSCALE1: |
| 163 | return "arm/xscale1"; | 34 | return "arm/xscale1"; |
| @@ -176,116 +47,6 @@ static char *op_name_from_perf_id(enum arm_perf_pmu_ids id) | |||
| 176 | } | 47 | } |
| 177 | } | 48 | } |
| 178 | 49 | ||
| 179 | static int op_arm_create_files(struct super_block *sb, struct dentry *root) | ||
| 180 | { | ||
| 181 | unsigned int i; | ||
| 182 | |||
| 183 | for (i = 0; i < perf_num_counters; i++) { | ||
| 184 | struct dentry *dir; | ||
| 185 | char buf[4]; | ||
| 186 | |||
| 187 | snprintf(buf, sizeof buf, "%d", i); | ||
| 188 | dir = oprofilefs_mkdir(sb, root, buf); | ||
| 189 | oprofilefs_create_ulong(sb, dir, "enabled", &counter_config[i].enabled); | ||
| 190 | oprofilefs_create_ulong(sb, dir, "event", &counter_config[i].event); | ||
| 191 | oprofilefs_create_ulong(sb, dir, "count", &counter_config[i].count); | ||
| 192 | oprofilefs_create_ulong(sb, dir, "unit_mask", &counter_config[i].unit_mask); | ||
| 193 | oprofilefs_create_ulong(sb, dir, "kernel", &counter_config[i].kernel); | ||
| 194 | oprofilefs_create_ulong(sb, dir, "user", &counter_config[i].user); | ||
| 195 | } | ||
| 196 | |||
| 197 | return 0; | ||
| 198 | } | ||
| 199 | |||
| 200 | static int op_arm_setup(void) | ||
| 201 | { | ||
| 202 | spin_lock(&oprofilefs_lock); | ||
| 203 | op_perf_setup(); | ||
| 204 | spin_unlock(&oprofilefs_lock); | ||
| 205 | return 0; | ||
| 206 | } | ||
| 207 | |||
| 208 | static int op_arm_start(void) | ||
| 209 | { | ||
| 210 | int ret = -EBUSY; | ||
| 211 | |||
| 212 | mutex_lock(&op_arm_mutex); | ||
| 213 | if (!op_arm_enabled) { | ||
| 214 | ret = 0; | ||
| 215 | op_perf_start(); | ||
| 216 | op_arm_enabled = 1; | ||
| 217 | } | ||
| 218 | mutex_unlock(&op_arm_mutex); | ||
| 219 | return ret; | ||
| 220 | } | ||
| 221 | |||
| 222 | static void op_arm_stop(void) | ||
| 223 | { | ||
| 224 | mutex_lock(&op_arm_mutex); | ||
| 225 | if (op_arm_enabled) | ||
| 226 | op_perf_stop(); | ||
| 227 | op_arm_enabled = 0; | ||
| 228 | mutex_unlock(&op_arm_mutex); | ||
| 229 | } | ||
| 230 | |||
| 231 | #ifdef CONFIG_PM | ||
| 232 | static int op_arm_suspend(struct platform_device *dev, pm_message_t state) | ||
| 233 | { | ||
| 234 | mutex_lock(&op_arm_mutex); | ||
| 235 | if (op_arm_enabled) | ||
| 236 | op_perf_stop(); | ||
| 237 | mutex_unlock(&op_arm_mutex); | ||
| 238 | return 0; | ||
| 239 | } | ||
| 240 | |||
| 241 | static int op_arm_resume(struct platform_device *dev) | ||
| 242 | { | ||
| 243 | mutex_lock(&op_arm_mutex); | ||
| 244 | if (op_arm_enabled && op_perf_start()) | ||
| 245 | op_arm_enabled = 0; | ||
| 246 | mutex_unlock(&op_arm_mutex); | ||
| 247 | return 0; | ||
| 248 | } | ||
| 249 | |||
| 250 | static struct platform_driver oprofile_driver = { | ||
| 251 | .driver = { | ||
| 252 | .name = "arm-oprofile", | ||
| 253 | }, | ||
| 254 | .resume = op_arm_resume, | ||
| 255 | .suspend = op_arm_suspend, | ||
| 256 | }; | ||
| 257 | |||
| 258 | static struct platform_device *oprofile_pdev; | ||
| 259 | |||
| 260 | static int __init init_driverfs(void) | ||
| 261 | { | ||
| 262 | int ret; | ||
| 263 | |||
| 264 | ret = platform_driver_register(&oprofile_driver); | ||
| 265 | if (ret) | ||
| 266 | goto out; | ||
| 267 | |||
| 268 | oprofile_pdev = platform_device_register_simple( | ||
| 269 | oprofile_driver.driver.name, 0, NULL, 0); | ||
| 270 | if (IS_ERR(oprofile_pdev)) { | ||
| 271 | ret = PTR_ERR(oprofile_pdev); | ||
| 272 | platform_driver_unregister(&oprofile_driver); | ||
| 273 | } | ||
| 274 | |||
| 275 | out: | ||
| 276 | return ret; | ||
| 277 | } | ||
| 278 | |||
| 279 | static void exit_driverfs(void) | ||
| 280 | { | ||
| 281 | platform_device_unregister(oprofile_pdev); | ||
| 282 | platform_driver_unregister(&oprofile_driver); | ||
| 283 | } | ||
| 284 | #else | ||
| 285 | static int __init init_driverfs(void) { return 0; } | ||
| 286 | #define exit_driverfs() do { } while (0) | ||
| 287 | #endif /* CONFIG_PM */ | ||
| 288 | |||
| 289 | static int report_trace(struct stackframe *frame, void *d) | 50 | static int report_trace(struct stackframe *frame, void *d) |
| 290 | { | 51 | { |
| 291 | unsigned int *depth = d; | 52 | unsigned int *depth = d; |
| @@ -350,74 +111,14 @@ static void arm_backtrace(struct pt_regs * const regs, unsigned int depth) | |||
| 350 | 111 | ||
| 351 | int __init oprofile_arch_init(struct oprofile_operations *ops) | 112 | int __init oprofile_arch_init(struct oprofile_operations *ops) |
| 352 | { | 113 | { |
| 353 | int cpu, ret = 0; | ||
| 354 | |||
| 355 | perf_num_counters = armpmu_get_max_events(); | ||
| 356 | |||
| 357 | counter_config = kcalloc(perf_num_counters, | ||
| 358 | sizeof(struct op_counter_config), GFP_KERNEL); | ||
| 359 | |||
| 360 | if (!counter_config) { | ||
| 361 | pr_info("oprofile: failed to allocate %d " | ||
| 362 | "counters\n", perf_num_counters); | ||
| 363 | return -ENOMEM; | ||
| 364 | } | ||
| 365 | |||
| 366 | ret = init_driverfs(); | ||
| 367 | if (ret) { | ||
| 368 | kfree(counter_config); | ||
| 369 | counter_config = NULL; | ||
| 370 | return ret; | ||
| 371 | } | ||
| 372 | |||
| 373 | for_each_possible_cpu(cpu) { | ||
| 374 | perf_events[cpu] = kcalloc(perf_num_counters, | ||
| 375 | sizeof(struct perf_event *), GFP_KERNEL); | ||
| 376 | if (!perf_events[cpu]) { | ||
| 377 | pr_info("oprofile: failed to allocate %d perf events " | ||
| 378 | "for cpu %d\n", perf_num_counters, cpu); | ||
| 379 | while (--cpu >= 0) | ||
| 380 | kfree(perf_events[cpu]); | ||
| 381 | return -ENOMEM; | ||
| 382 | } | ||
| 383 | } | ||
| 384 | |||
| 385 | ops->backtrace = arm_backtrace; | 114 | ops->backtrace = arm_backtrace; |
| 386 | ops->create_files = op_arm_create_files; | ||
| 387 | ops->setup = op_arm_setup; | ||
| 388 | ops->start = op_arm_start; | ||
| 389 | ops->stop = op_arm_stop; | ||
| 390 | ops->shutdown = op_arm_stop; | ||
| 391 | ops->cpu_type = op_name_from_perf_id(armpmu_get_pmu_id()); | ||
| 392 | |||
| 393 | if (!ops->cpu_type) | ||
| 394 | ret = -ENODEV; | ||
| 395 | else | ||
| 396 | pr_info("oprofile: using %s\n", ops->cpu_type); | ||
| 397 | 115 | ||
| 398 | return ret; | 116 | return oprofile_perf_init(ops); |
| 399 | } | 117 | } |
| 400 | 118 | ||
| 401 | void oprofile_arch_exit(void) | 119 | void __exit oprofile_arch_exit(void) |
| 402 | { | 120 | { |
| 403 | int cpu, id; | 121 | oprofile_perf_exit(); |
| 404 | struct perf_event *event; | ||
| 405 | |||
| 406 | if (*perf_events) { | ||
| 407 | for_each_possible_cpu(cpu) { | ||
| 408 | for (id = 0; id < perf_num_counters; ++id) { | ||
| 409 | event = perf_events[cpu][id]; | ||
| 410 | if (event != NULL) | ||
| 411 | perf_event_release_kernel(event); | ||
| 412 | } | ||
| 413 | kfree(perf_events[cpu]); | ||
| 414 | } | ||
| 415 | } | ||
| 416 | |||
| 417 | if (counter_config) { | ||
| 418 | kfree(counter_config); | ||
| 419 | exit_driverfs(); | ||
| 420 | } | ||
| 421 | } | 122 | } |
| 422 | #else | 123 | #else |
| 423 | int __init oprofile_arch_init(struct oprofile_operations *ops) | 124 | int __init oprofile_arch_init(struct oprofile_operations *ops) |
| @@ -425,5 +126,5 @@ int __init oprofile_arch_init(struct oprofile_operations *ops) | |||
| 425 | pr_info("oprofile: hardware counters not available\n"); | 126 | pr_info("oprofile: hardware counters not available\n"); |
| 426 | return -ENODEV; | 127 | return -ENODEV; |
| 427 | } | 128 | } |
| 428 | void oprofile_arch_exit(void) {} | 129 | void __exit oprofile_arch_exit(void) {} |
| 429 | #endif /* CONFIG_HW_PERF_EVENTS */ | 130 | #endif /* CONFIG_HW_PERF_EVENTS */ |
diff --git a/arch/sh/Kconfig b/arch/sh/Kconfig index 33990fa95af0..35b6c3f85173 100644 --- a/arch/sh/Kconfig +++ b/arch/sh/Kconfig | |||
| @@ -249,6 +249,11 @@ config ARCH_SHMOBILE | |||
| 249 | select PM | 249 | select PM |
| 250 | select PM_RUNTIME | 250 | select PM_RUNTIME |
| 251 | 251 | ||
| 252 | config CPU_HAS_PMU | ||
| 253 | depends on CPU_SH4 || CPU_SH4A | ||
| 254 | default y | ||
| 255 | bool | ||
| 256 | |||
| 252 | if SUPERH32 | 257 | if SUPERH32 |
| 253 | 258 | ||
| 254 | choice | 259 | choice |
| @@ -738,6 +743,14 @@ config GUSA_RB | |||
| 738 | LLSC, this should be more efficient than the other alternative of | 743 | LLSC, this should be more efficient than the other alternative of |
| 739 | disabling interrupts around the atomic sequence. | 744 | disabling interrupts around the atomic sequence. |
| 740 | 745 | ||
| 746 | config HW_PERF_EVENTS | ||
| 747 | bool "Enable hardware performance counter support for perf events" | ||
| 748 | depends on PERF_EVENTS && CPU_HAS_PMU | ||
| 749 | default y | ||
| 750 | help | ||
| 751 | Enable hardware performance counter support for perf events. If | ||
| 752 | disabled, perf events will use software events only. | ||
| 753 | |||
| 741 | source "drivers/sh/Kconfig" | 754 | source "drivers/sh/Kconfig" |
| 742 | 755 | ||
| 743 | endmenu | 756 | endmenu |
diff --git a/arch/sh/kernel/perf_event.c b/arch/sh/kernel/perf_event.c index 036f7a9296fa..5a4b33435650 100644 --- a/arch/sh/kernel/perf_event.c +++ b/arch/sh/kernel/perf_event.c | |||
| @@ -59,6 +59,24 @@ static inline int sh_pmu_initialized(void) | |||
| 59 | return !!sh_pmu; | 59 | return !!sh_pmu; |
| 60 | } | 60 | } |
| 61 | 61 | ||
| 62 | const char *perf_pmu_name(void) | ||
| 63 | { | ||
| 64 | if (!sh_pmu) | ||
| 65 | return NULL; | ||
| 66 | |||
| 67 | return sh_pmu->name; | ||
| 68 | } | ||
| 69 | EXPORT_SYMBOL_GPL(perf_pmu_name); | ||
| 70 | |||
| 71 | int perf_num_counters(void) | ||
| 72 | { | ||
| 73 | if (!sh_pmu) | ||
| 74 | return 0; | ||
| 75 | |||
| 76 | return sh_pmu->num_events; | ||
| 77 | } | ||
| 78 | EXPORT_SYMBOL_GPL(perf_num_counters); | ||
| 79 | |||
| 62 | /* | 80 | /* |
| 63 | * Release the PMU if this is the last perf_event. | 81 | * Release the PMU if this is the last perf_event. |
| 64 | */ | 82 | */ |
diff --git a/arch/sh/oprofile/Makefile b/arch/sh/oprofile/Makefile index 4886c5c1786c..e85aae73e3dc 100644 --- a/arch/sh/oprofile/Makefile +++ b/arch/sh/oprofile/Makefile | |||
| @@ -6,4 +6,8 @@ DRIVER_OBJS = $(addprefix ../../../drivers/oprofile/, \ | |||
| 6 | oprofilefs.o oprofile_stats.o \ | 6 | oprofilefs.o oprofile_stats.o \ |
| 7 | timer_int.o ) | 7 | timer_int.o ) |
| 8 | 8 | ||
| 9 | ifeq ($(CONFIG_HW_PERF_EVENTS),y) | ||
| 10 | DRIVER_OBJS += $(addprefix ../../../drivers/oprofile/, oprofile_perf.o) | ||
| 11 | endif | ||
| 12 | |||
| 9 | oprofile-y := $(DRIVER_OBJS) common.o backtrace.o | 13 | oprofile-y := $(DRIVER_OBJS) common.o backtrace.o |
diff --git a/arch/sh/oprofile/common.c b/arch/sh/oprofile/common.c index ac604937f3ee..e10d89376f9b 100644 --- a/arch/sh/oprofile/common.c +++ b/arch/sh/oprofile/common.c | |||
| @@ -17,114 +17,45 @@ | |||
| 17 | #include <linux/init.h> | 17 | #include <linux/init.h> |
| 18 | #include <linux/errno.h> | 18 | #include <linux/errno.h> |
| 19 | #include <linux/smp.h> | 19 | #include <linux/smp.h> |
| 20 | #include <linux/perf_event.h> | ||
| 20 | #include <asm/processor.h> | 21 | #include <asm/processor.h> |
| 21 | #include "op_impl.h" | ||
| 22 | |||
| 23 | static struct op_sh_model *model; | ||
| 24 | |||
| 25 | static struct op_counter_config ctr[20]; | ||
| 26 | 22 | ||
| 23 | #ifdef CONFIG_HW_PERF_EVENTS | ||
| 27 | extern void sh_backtrace(struct pt_regs * const regs, unsigned int depth); | 24 | extern void sh_backtrace(struct pt_regs * const regs, unsigned int depth); |
| 28 | 25 | ||
| 29 | static int op_sh_setup(void) | 26 | char *op_name_from_perf_id(void) |
| 30 | { | ||
| 31 | /* Pre-compute the values to stuff in the hardware registers. */ | ||
| 32 | model->reg_setup(ctr); | ||
| 33 | |||
| 34 | /* Configure the registers on all cpus. */ | ||
| 35 | on_each_cpu(model->cpu_setup, NULL, 1); | ||
| 36 | |||
| 37 | return 0; | ||
| 38 | } | ||
| 39 | |||
| 40 | static int op_sh_create_files(struct super_block *sb, struct dentry *root) | ||
| 41 | { | 27 | { |
| 42 | int i, ret = 0; | 28 | const char *pmu; |
| 29 | char buf[20]; | ||
| 30 | int size; | ||
| 43 | 31 | ||
| 44 | for (i = 0; i < model->num_counters; i++) { | 32 | pmu = perf_pmu_name(); |
| 45 | struct dentry *dir; | 33 | if (!pmu) |
| 46 | char buf[4]; | 34 | return NULL; |
| 47 | 35 | ||
| 48 | snprintf(buf, sizeof(buf), "%d", i); | 36 | size = snprintf(buf, sizeof(buf), "sh/%s", pmu); |
| 49 | dir = oprofilefs_mkdir(sb, root, buf); | 37 | if (size > -1 && size < sizeof(buf)) |
| 38 | return buf; | ||
| 50 | 39 | ||
| 51 | ret |= oprofilefs_create_ulong(sb, dir, "enabled", &ctr[i].enabled); | 40 | return NULL; |
| 52 | ret |= oprofilefs_create_ulong(sb, dir, "event", &ctr[i].event); | ||
| 53 | ret |= oprofilefs_create_ulong(sb, dir, "kernel", &ctr[i].kernel); | ||
| 54 | ret |= oprofilefs_create_ulong(sb, dir, "user", &ctr[i].user); | ||
| 55 | |||
| 56 | if (model->create_files) | ||
| 57 | ret |= model->create_files(sb, dir); | ||
| 58 | else | ||
| 59 | ret |= oprofilefs_create_ulong(sb, dir, "count", &ctr[i].count); | ||
| 60 | |||
| 61 | /* Dummy entries */ | ||
| 62 | ret |= oprofilefs_create_ulong(sb, dir, "unit_mask", &ctr[i].unit_mask); | ||
| 63 | } | ||
| 64 | |||
| 65 | return ret; | ||
| 66 | } | 41 | } |
| 67 | 42 | ||
| 68 | static int op_sh_start(void) | 43 | int __init oprofile_arch_init(struct oprofile_operations *ops) |
| 69 | { | 44 | { |
| 70 | /* Enable performance monitoring for all counters. */ | 45 | ops->backtrace = sh_backtrace; |
| 71 | on_each_cpu(model->cpu_start, NULL, 1); | ||
| 72 | 46 | ||
| 73 | return 0; | 47 | return oprofile_perf_init(ops); |
| 74 | } | 48 | } |
| 75 | 49 | ||
| 76 | static void op_sh_stop(void) | 50 | void __exit oprofile_arch_exit(void) |
| 77 | { | 51 | { |
| 78 | /* Disable performance monitoring for all counters. */ | 52 | oprofile_perf_exit(); |
| 79 | on_each_cpu(model->cpu_stop, NULL, 1); | ||
| 80 | } | 53 | } |
| 81 | 54 | #else | |
| 82 | int __init oprofile_arch_init(struct oprofile_operations *ops) | 55 | int __init oprofile_arch_init(struct oprofile_operations *ops) |
| 83 | { | 56 | { |
| 84 | struct op_sh_model *lmodel = NULL; | 57 | pr_info("oprofile: hardware counters not available\n"); |
| 85 | int ret; | 58 | return -ENODEV; |
| 86 | |||
| 87 | /* | ||
| 88 | * Always assign the backtrace op. If the counter initialization | ||
| 89 | * fails, we fall back to the timer which will still make use of | ||
| 90 | * this. | ||
| 91 | */ | ||
| 92 | ops->backtrace = sh_backtrace; | ||
| 93 | |||
| 94 | /* | ||
| 95 | * XXX | ||
| 96 | * | ||
| 97 | * All of the SH7750/SH-4A counters have been converted to perf, | ||
| 98 | * this infrastructure hook is left for other users until they've | ||
| 99 | * had a chance to convert over, at which point all of this | ||
| 100 | * will be deleted. | ||
| 101 | */ | ||
| 102 | |||
| 103 | if (!lmodel) | ||
| 104 | return -ENODEV; | ||
| 105 | if (!(current_cpu_data.flags & CPU_HAS_PERF_COUNTER)) | ||
| 106 | return -ENODEV; | ||
| 107 | |||
| 108 | ret = lmodel->init(); | ||
| 109 | if (unlikely(ret != 0)) | ||
| 110 | return ret; | ||
| 111 | |||
| 112 | model = lmodel; | ||
| 113 | |||
| 114 | ops->setup = op_sh_setup; | ||
| 115 | ops->create_files = op_sh_create_files; | ||
| 116 | ops->start = op_sh_start; | ||
| 117 | ops->stop = op_sh_stop; | ||
| 118 | ops->cpu_type = lmodel->cpu_type; | ||
| 119 | |||
| 120 | printk(KERN_INFO "oprofile: using %s performance monitoring.\n", | ||
| 121 | lmodel->cpu_type); | ||
| 122 | |||
| 123 | return 0; | ||
| 124 | } | ||
| 125 | |||
| 126 | void oprofile_arch_exit(void) | ||
| 127 | { | ||
| 128 | if (model && model->exit) | ||
| 129 | model->exit(); | ||
| 130 | } | 59 | } |
| 60 | void __exit oprofile_arch_exit(void) {} | ||
| 61 | #endif /* CONFIG_HW_PERF_EVENTS */ | ||
diff --git a/arch/sh/oprofile/op_impl.h b/arch/sh/oprofile/op_impl.h deleted file mode 100644 index 1244479ceb29..000000000000 --- a/arch/sh/oprofile/op_impl.h +++ /dev/null | |||
| @@ -1,33 +0,0 @@ | |||
| 1 | #ifndef __OP_IMPL_H | ||
| 2 | #define __OP_IMPL_H | ||
| 3 | |||
| 4 | /* Per-counter configuration as set via oprofilefs. */ | ||
| 5 | struct op_counter_config { | ||
| 6 | unsigned long enabled; | ||
| 7 | unsigned long event; | ||
| 8 | |||
| 9 | unsigned long count; | ||
| 10 | |||
| 11 | /* Dummy values for userspace tool compliance */ | ||
| 12 | unsigned long kernel; | ||
| 13 | unsigned long user; | ||
| 14 | unsigned long unit_mask; | ||
| 15 | }; | ||
| 16 | |||
| 17 | /* Per-architecture configury and hooks. */ | ||
| 18 | struct op_sh_model { | ||
| 19 | void (*reg_setup)(struct op_counter_config *); | ||
| 20 | int (*create_files)(struct super_block *sb, struct dentry *dir); | ||
| 21 | void (*cpu_setup)(void *dummy); | ||
| 22 | int (*init)(void); | ||
| 23 | void (*exit)(void); | ||
| 24 | void (*cpu_start)(void *args); | ||
| 25 | void (*cpu_stop)(void *args); | ||
| 26 | char *cpu_type; | ||
| 27 | unsigned char num_counters; | ||
| 28 | }; | ||
| 29 | |||
| 30 | /* arch/sh/oprofile/common.c */ | ||
| 31 | extern void sh_backtrace(struct pt_regs * const regs, unsigned int depth); | ||
| 32 | |||
| 33 | #endif /* __OP_IMPL_H */ | ||
diff --git a/arch/x86/oprofile/backtrace.c b/arch/x86/oprofile/backtrace.c index 3855096c59b8..2d49d4e19a36 100644 --- a/arch/x86/oprofile/backtrace.c +++ b/arch/x86/oprofile/backtrace.c | |||
| @@ -14,6 +14,7 @@ | |||
| 14 | #include <asm/ptrace.h> | 14 | #include <asm/ptrace.h> |
| 15 | #include <asm/uaccess.h> | 15 | #include <asm/uaccess.h> |
| 16 | #include <asm/stacktrace.h> | 16 | #include <asm/stacktrace.h> |
| 17 | #include <linux/compat.h> | ||
| 17 | 18 | ||
| 18 | static void backtrace_warning_symbol(void *data, char *msg, | 19 | static void backtrace_warning_symbol(void *data, char *msg, |
| 19 | unsigned long symbol) | 20 | unsigned long symbol) |
| @@ -48,14 +49,12 @@ static struct stacktrace_ops backtrace_ops = { | |||
| 48 | .walk_stack = print_context_stack, | 49 | .walk_stack = print_context_stack, |
| 49 | }; | 50 | }; |
| 50 | 51 | ||
| 51 | struct frame_head { | 52 | #ifdef CONFIG_COMPAT |
| 52 | struct frame_head *bp; | 53 | static struct stack_frame_ia32 * |
| 53 | unsigned long ret; | 54 | dump_user_backtrace_32(struct stack_frame_ia32 *head) |
| 54 | } __attribute__((packed)); | ||
| 55 | |||
| 56 | static struct frame_head *dump_user_backtrace(struct frame_head *head) | ||
| 57 | { | 55 | { |
| 58 | struct frame_head bufhead[2]; | 56 | struct stack_frame_ia32 bufhead[2]; |
| 57 | struct stack_frame_ia32 *fp; | ||
| 59 | 58 | ||
| 60 | /* Also check accessibility of one struct frame_head beyond */ | 59 | /* Also check accessibility of one struct frame_head beyond */ |
| 61 | if (!access_ok(VERIFY_READ, head, sizeof(bufhead))) | 60 | if (!access_ok(VERIFY_READ, head, sizeof(bufhead))) |
| @@ -63,20 +62,66 @@ static struct frame_head *dump_user_backtrace(struct frame_head *head) | |||
| 63 | if (__copy_from_user_inatomic(bufhead, head, sizeof(bufhead))) | 62 | if (__copy_from_user_inatomic(bufhead, head, sizeof(bufhead))) |
| 64 | return NULL; | 63 | return NULL; |
| 65 | 64 | ||
| 66 | oprofile_add_trace(bufhead[0].ret); | 65 | fp = (struct stack_frame_ia32 *) compat_ptr(bufhead[0].next_frame); |
| 66 | |||
| 67 | oprofile_add_trace(bufhead[0].return_address); | ||
| 68 | |||
| 69 | /* frame pointers should strictly progress back up the stack | ||
| 70 | * (towards higher addresses) */ | ||
| 71 | if (head >= fp) | ||
| 72 | return NULL; | ||
| 73 | |||
| 74 | return fp; | ||
| 75 | } | ||
| 76 | |||
| 77 | static inline int | ||
| 78 | x86_backtrace_32(struct pt_regs * const regs, unsigned int depth) | ||
| 79 | { | ||
| 80 | struct stack_frame_ia32 *head; | ||
| 81 | |||
| 82 | /* User process is 32-bit */ | ||
| 83 | if (!current || !test_thread_flag(TIF_IA32)) | ||
| 84 | return 0; | ||
| 85 | |||
| 86 | head = (struct stack_frame_ia32 *) regs->bp; | ||
| 87 | while (depth-- && head) | ||
| 88 | head = dump_user_backtrace_32(head); | ||
| 89 | |||
| 90 | return 1; | ||
| 91 | } | ||
| 92 | |||
| 93 | #else | ||
| 94 | static inline int | ||
| 95 | x86_backtrace_32(struct pt_regs * const regs, unsigned int depth) | ||
| 96 | { | ||
| 97 | return 0; | ||
| 98 | } | ||
| 99 | #endif /* CONFIG_COMPAT */ | ||
| 100 | |||
| 101 | static struct stack_frame *dump_user_backtrace(struct stack_frame *head) | ||
| 102 | { | ||
| 103 | struct stack_frame bufhead[2]; | ||
| 104 | |||
| 105 | /* Also check accessibility of one struct stack_frame beyond */ | ||
| 106 | if (!access_ok(VERIFY_READ, head, sizeof(bufhead))) | ||
| 107 | return NULL; | ||
| 108 | if (__copy_from_user_inatomic(bufhead, head, sizeof(bufhead))) | ||
| 109 | return NULL; | ||
| 110 | |||
| 111 | oprofile_add_trace(bufhead[0].return_address); | ||
| 67 | 112 | ||
| 68 | /* frame pointers should strictly progress back up the stack | 113 | /* frame pointers should strictly progress back up the stack |
| 69 | * (towards higher addresses) */ | 114 | * (towards higher addresses) */ |
| 70 | if (head >= bufhead[0].bp) | 115 | if (head >= bufhead[0].next_frame) |
| 71 | return NULL; | 116 | return NULL; |
| 72 | 117 | ||
| 73 | return bufhead[0].bp; | 118 | return bufhead[0].next_frame; |
| 74 | } | 119 | } |
| 75 | 120 | ||
| 76 | void | 121 | void |
| 77 | x86_backtrace(struct pt_regs * const regs, unsigned int depth) | 122 | x86_backtrace(struct pt_regs * const regs, unsigned int depth) |
| 78 | { | 123 | { |
| 79 | struct frame_head *head = (struct frame_head *)frame_pointer(regs); | 124 | struct stack_frame *head = (struct stack_frame *)frame_pointer(regs); |
| 80 | 125 | ||
| 81 | if (!user_mode_vm(regs)) { | 126 | if (!user_mode_vm(regs)) { |
| 82 | unsigned long stack = kernel_stack_pointer(regs); | 127 | unsigned long stack = kernel_stack_pointer(regs); |
| @@ -86,6 +131,9 @@ x86_backtrace(struct pt_regs * const regs, unsigned int depth) | |||
| 86 | return; | 131 | return; |
| 87 | } | 132 | } |
| 88 | 133 | ||
| 134 | if (x86_backtrace_32(regs, depth)) | ||
| 135 | return; | ||
| 136 | |||
| 89 | while (depth-- && head) | 137 | while (depth-- && head) |
| 90 | head = dump_user_backtrace(head); | 138 | head = dump_user_backtrace(head); |
| 91 | } | 139 | } |
diff --git a/arch/x86/oprofile/nmi_int.c b/arch/x86/oprofile/nmi_int.c index f1575c9a2572..bd1489c3ce09 100644 --- a/arch/x86/oprofile/nmi_int.c +++ b/arch/x86/oprofile/nmi_int.c | |||
| @@ -695,9 +695,6 @@ static int __init ppro_init(char **cpu_type) | |||
| 695 | return 1; | 695 | return 1; |
| 696 | } | 696 | } |
| 697 | 697 | ||
| 698 | /* in order to get sysfs right */ | ||
| 699 | static int using_nmi; | ||
| 700 | |||
| 701 | int __init op_nmi_init(struct oprofile_operations *ops) | 698 | int __init op_nmi_init(struct oprofile_operations *ops) |
| 702 | { | 699 | { |
| 703 | __u8 vendor = boot_cpu_data.x86_vendor; | 700 | __u8 vendor = boot_cpu_data.x86_vendor; |
| @@ -705,8 +702,6 @@ int __init op_nmi_init(struct oprofile_operations *ops) | |||
| 705 | char *cpu_type = NULL; | 702 | char *cpu_type = NULL; |
| 706 | int ret = 0; | 703 | int ret = 0; |
| 707 | 704 | ||
| 708 | using_nmi = 0; | ||
| 709 | |||
| 710 | if (!cpu_has_apic) | 705 | if (!cpu_has_apic) |
| 711 | return -ENODEV; | 706 | return -ENODEV; |
| 712 | 707 | ||
| @@ -790,13 +785,11 @@ int __init op_nmi_init(struct oprofile_operations *ops) | |||
| 790 | if (ret) | 785 | if (ret) |
| 791 | return ret; | 786 | return ret; |
| 792 | 787 | ||
| 793 | using_nmi = 1; | ||
| 794 | printk(KERN_INFO "oprofile: using NMI interrupt.\n"); | 788 | printk(KERN_INFO "oprofile: using NMI interrupt.\n"); |
| 795 | return 0; | 789 | return 0; |
| 796 | } | 790 | } |
| 797 | 791 | ||
| 798 | void op_nmi_exit(void) | 792 | void op_nmi_exit(void) |
| 799 | { | 793 | { |
| 800 | if (using_nmi) | 794 | exit_sysfs(); |
| 801 | exit_sysfs(); | ||
| 802 | } | 795 | } |
