Thomas Gleixner | b886d83c | 2019-06-01 10:08:55 +0200 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 2 | /* |
| 3 | * latencytop.c: Latency display infrastructure |
| 4 | * |
| 5 | * (C) Copyright 2008 Intel Corporation |
| 6 | * Author: Arjan van de Ven <arjan@linux.intel.com> |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 7 | */ |
Arjan van de Ven | ad0b0fd55 | 2009-02-10 11:42:26 -0800 | [diff] [blame] | 8 | |
| 9 | /* |
| 10 | * CONFIG_LATENCYTOP enables a kernel latency tracking infrastructure that is |
| 11 | * used by the "latencytop" userspace tool. The latency that is tracked is not |
| 12 | * the 'traditional' interrupt latency (which is primarily caused by something |
| 13 | * else consuming CPU), but instead, it is the latency an application encounters |
| 14 | * because the kernel sleeps on its behalf for various reasons. |
| 15 | * |
| 16 | * This code tracks 2 levels of statistics: |
| 17 | * 1) System level latency |
| 18 | * 2) Per process latency |
| 19 | * |
| 20 | * The latency is stored in fixed sized data structures in an accumulated form; |
| 21 | * if the "same" latency cause is hit twice, this will be tracked as one entry |
| 22 | * in the data structure. Both the count, total accumulated latency and maximum |
| 23 | * latency are tracked in this data structure. When the fixed size structure is |
| 24 | * full, no new causes are tracked until the buffer is flushed by writing to |
| 25 | * the /proc file; the userspace tool does this on a regular basis. |
| 26 | * |
| 27 | * A latency cause is identified by a stringified backtrace at the point that |
| 28 | * the scheduler gets invoked. The userland tool will use this string to |
| 29 | * identify the cause of the latency in human readable form. |
| 30 | * |
| 31 | * The information is exported via /proc/latency_stats and /proc/<pid>/latency. |
| 32 | * These files look like this: |
| 33 | * |
| 34 | * Latency Top version : v0.1 |
| 35 | * 70 59433 4897 i915_irq_wait drm_ioctl vfs_ioctl do_vfs_ioctl sys_ioctl |
| 36 | * | | | | |
| 37 | * | | | +----> the stringified backtrace |
| 38 | * | | +---------> The maximum latency for this entry in microseconds |
| 39 | * | +--------------> The accumulated latency for this entry (microseconds) |
| 40 | * +-------------------> The number of times this entry is hit |
| 41 | * |
| 42 | * (note: the average latency is the accumulated latency divided by the number |
| 43 | * of times) |
| 44 | */ |
| 45 | |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 46 | #include <linux/kallsyms.h> |
| 47 | #include <linux/seq_file.h> |
| 48 | #include <linux/notifier.h> |
| 49 | #include <linux/spinlock.h> |
| 50 | #include <linux/proc_fs.h> |
Mel Gorman | cb25176 | 2016-02-05 09:08:36 +0000 | [diff] [blame] | 51 | #include <linux/latencytop.h> |
Paul Gortmaker | 9984de1 | 2011-05-23 14:51:41 -0400 | [diff] [blame] | 52 | #include <linux/export.h> |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 53 | #include <linux/sched.h> |
Ingo Molnar | b17b015 | 2017-02-08 18:51:35 +0100 | [diff] [blame] | 54 | #include <linux/sched/debug.h> |
Ingo Molnar | 3905f9a | 2017-02-05 12:07:04 +0100 | [diff] [blame] | 55 | #include <linux/sched/stat.h> |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 56 | #include <linux/list.h> |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 57 | #include <linux/stacktrace.h> |
liaohua | 988f11e | 2022-04-07 15:29:48 +0800 | [diff] [blame] | 58 | #include <linux/sysctl.h> |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 59 | |
Thomas Gleixner | 757455d | 2009-09-17 17:31:09 +0200 | [diff] [blame] | 60 | static DEFINE_RAW_SPINLOCK(latency_lock); |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 61 | |
| 62 | #define MAXLR 128 |
| 63 | static struct latency_record latency_record[MAXLR]; |
| 64 | |
| 65 | int latencytop_enabled; |
| 66 | |
liaohua | 988f11e | 2022-04-07 15:29:48 +0800 | [diff] [blame] | 67 | #ifdef CONFIG_SYSCTL |
| 68 | static int sysctl_latencytop(struct ctl_table *table, int write, void *buffer, |
| 69 | size_t *lenp, loff_t *ppos) |
| 70 | { |
| 71 | int err; |
| 72 | |
| 73 | err = proc_dointvec(table, write, buffer, lenp, ppos); |
| 74 | if (latencytop_enabled) |
| 75 | force_schedstat_enabled(); |
| 76 | |
| 77 | return err; |
| 78 | } |
| 79 | |
| 80 | static struct ctl_table latencytop_sysctl[] = { |
| 81 | { |
| 82 | .procname = "latencytop", |
| 83 | .data = &latencytop_enabled, |
| 84 | .maxlen = sizeof(int), |
| 85 | .mode = 0644, |
| 86 | .proc_handler = sysctl_latencytop, |
| 87 | }, |
| 88 | {} |
| 89 | }; |
| 90 | #endif |
| 91 | |
Lin Feng | e02c9b0 | 2019-05-14 15:42:34 -0700 | [diff] [blame] | 92 | void clear_tsk_latency_tracing(struct task_struct *p) |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 93 | { |
| 94 | unsigned long flags; |
| 95 | |
Thomas Gleixner | 757455d | 2009-09-17 17:31:09 +0200 | [diff] [blame] | 96 | raw_spin_lock_irqsave(&latency_lock, flags); |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 97 | memset(&p->latency_record, 0, sizeof(p->latency_record)); |
| 98 | p->latency_record_count = 0; |
Thomas Gleixner | 757455d | 2009-09-17 17:31:09 +0200 | [diff] [blame] | 99 | raw_spin_unlock_irqrestore(&latency_lock, flags); |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 100 | } |
| 101 | |
| 102 | static void clear_global_latency_tracing(void) |
| 103 | { |
| 104 | unsigned long flags; |
| 105 | |
Thomas Gleixner | 757455d | 2009-09-17 17:31:09 +0200 | [diff] [blame] | 106 | raw_spin_lock_irqsave(&latency_lock, flags); |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 107 | memset(&latency_record, 0, sizeof(latency_record)); |
Thomas Gleixner | 757455d | 2009-09-17 17:31:09 +0200 | [diff] [blame] | 108 | raw_spin_unlock_irqrestore(&latency_lock, flags); |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 109 | } |
| 110 | |
| 111 | static void __sched |
Fabian Frederick | eaa1809 | 2014-06-04 16:11:21 -0700 | [diff] [blame] | 112 | account_global_scheduler_latency(struct task_struct *tsk, |
| 113 | struct latency_record *lat) |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 114 | { |
wuchi | b814751 | 2022-09-03 21:52:33 +0800 | [diff] [blame] | 115 | int firstnonnull = MAXLR; |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 116 | int i; |
| 117 | |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 118 | /* skip kernel threads for now */ |
| 119 | if (!tsk->mm) |
| 120 | return; |
| 121 | |
| 122 | for (i = 0; i < MAXLR; i++) { |
Dmitry Adamushko | 19fb518 | 2008-02-17 22:34:07 +0100 | [diff] [blame] | 123 | int q, same = 1; |
| 124 | |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 125 | /* Nothing stored: */ |
| 126 | if (!latency_record[i].backtrace[0]) { |
| 127 | if (firstnonnull > i) |
| 128 | firstnonnull = i; |
| 129 | continue; |
| 130 | } |
Arjan van de Ven | ad0b0fd55 | 2009-02-10 11:42:26 -0800 | [diff] [blame] | 131 | for (q = 0; q < LT_BACKTRACEDEPTH; q++) { |
Dmitry Adamushko | 19fb518 | 2008-02-17 22:34:07 +0100 | [diff] [blame] | 132 | unsigned long record = lat->backtrace[q]; |
| 133 | |
| 134 | if (latency_record[i].backtrace[q] != record) { |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 135 | same = 0; |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 136 | break; |
Dmitry Adamushko | 19fb518 | 2008-02-17 22:34:07 +0100 | [diff] [blame] | 137 | } |
| 138 | |
Thomas Gleixner | accddc4 | 2019-04-10 12:28:08 +0200 | [diff] [blame] | 139 | /* 0 entry marks end of backtrace: */ |
| 140 | if (!record) |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 141 | break; |
| 142 | } |
| 143 | if (same) { |
| 144 | latency_record[i].count++; |
| 145 | latency_record[i].time += lat->time; |
| 146 | if (lat->time > latency_record[i].max) |
| 147 | latency_record[i].max = lat->time; |
| 148 | return; |
| 149 | } |
| 150 | } |
| 151 | |
| 152 | i = firstnonnull; |
wuchi | b814751 | 2022-09-03 21:52:33 +0800 | [diff] [blame] | 153 | if (i >= MAXLR) |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 154 | return; |
| 155 | |
| 156 | /* Allocted a new one: */ |
| 157 | memcpy(&latency_record[i], lat, sizeof(struct latency_record)); |
| 158 | } |
| 159 | |
Arjan van de Ven | ad0b0fd55 | 2009-02-10 11:42:26 -0800 | [diff] [blame] | 160 | /** |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 161 | * __account_scheduler_latency - record an occurred latency |
Arjan van de Ven | ad0b0fd55 | 2009-02-10 11:42:26 -0800 | [diff] [blame] | 162 | * @tsk - the task struct of the task hitting the latency |
| 163 | * @usecs - the duration of the latency in microseconds |
| 164 | * @inter - 1 if the sleep was interruptible, 0 if uninterruptible |
| 165 | * |
| 166 | * This function is the main entry point for recording latency entries |
| 167 | * as called by the scheduler. |
| 168 | * |
| 169 | * This function has a few special cases to deal with normal 'non-latency' |
| 170 | * sleeps: specifically, interruptible sleep longer than 5 msec is skipped |
| 171 | * since this usually is caused by waiting for events via select() and co. |
| 172 | * |
| 173 | * Negative latencies (caused by time going backwards) are also explicitly |
| 174 | * skipped. |
| 175 | */ |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 176 | void __sched |
Arjan van de Ven | ad0b0fd55 | 2009-02-10 11:42:26 -0800 | [diff] [blame] | 177 | __account_scheduler_latency(struct task_struct *tsk, int usecs, int inter) |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 178 | { |
| 179 | unsigned long flags; |
| 180 | int i, q; |
| 181 | struct latency_record lat; |
| 182 | |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 183 | /* Long interruptible waits are generally user requested... */ |
| 184 | if (inter && usecs > 5000) |
| 185 | return; |
| 186 | |
Arjan van de Ven | ad0b0fd55 | 2009-02-10 11:42:26 -0800 | [diff] [blame] | 187 | /* Negative sleeps are time going backwards */ |
| 188 | /* Zero-time sleeps are non-interesting */ |
| 189 | if (usecs <= 0) |
| 190 | return; |
| 191 | |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 192 | memset(&lat, 0, sizeof(lat)); |
| 193 | lat.count = 1; |
| 194 | lat.time = usecs; |
| 195 | lat.max = usecs; |
Thomas Gleixner | f938772 | 2019-04-25 11:44:59 +0200 | [diff] [blame] | 196 | |
| 197 | stack_trace_save_tsk(tsk, lat.backtrace, LT_BACKTRACEDEPTH, 0); |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 198 | |
Thomas Gleixner | 757455d | 2009-09-17 17:31:09 +0200 | [diff] [blame] | 199 | raw_spin_lock_irqsave(&latency_lock, flags); |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 200 | |
| 201 | account_global_scheduler_latency(tsk, &lat); |
| 202 | |
Ken Chen | 3871525 | 2010-11-11 14:05:16 -0800 | [diff] [blame] | 203 | for (i = 0; i < tsk->latency_record_count; i++) { |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 204 | struct latency_record *mylat; |
| 205 | int same = 1; |
Dmitry Adamushko | 19fb518 | 2008-02-17 22:34:07 +0100 | [diff] [blame] | 206 | |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 207 | mylat = &tsk->latency_record[i]; |
Arjan van de Ven | ad0b0fd55 | 2009-02-10 11:42:26 -0800 | [diff] [blame] | 208 | for (q = 0; q < LT_BACKTRACEDEPTH; q++) { |
Dmitry Adamushko | 19fb518 | 2008-02-17 22:34:07 +0100 | [diff] [blame] | 209 | unsigned long record = lat.backtrace[q]; |
| 210 | |
| 211 | if (mylat->backtrace[q] != record) { |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 212 | same = 0; |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 213 | break; |
Dmitry Adamushko | 19fb518 | 2008-02-17 22:34:07 +0100 | [diff] [blame] | 214 | } |
| 215 | |
Thomas Gleixner | accddc4 | 2019-04-10 12:28:08 +0200 | [diff] [blame] | 216 | /* 0 entry is end of backtrace */ |
| 217 | if (!record) |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 218 | break; |
| 219 | } |
| 220 | if (same) { |
| 221 | mylat->count++; |
| 222 | mylat->time += lat.time; |
| 223 | if (lat.time > mylat->max) |
| 224 | mylat->max = lat.time; |
| 225 | goto out_unlock; |
| 226 | } |
| 227 | } |
| 228 | |
Ken Chen | 3871525 | 2010-11-11 14:05:16 -0800 | [diff] [blame] | 229 | /* |
| 230 | * short term hack; if we're > 32 we stop; future we recycle: |
| 231 | */ |
| 232 | if (tsk->latency_record_count >= LT_SAVECOUNT) |
| 233 | goto out_unlock; |
| 234 | |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 235 | /* Allocated a new one: */ |
Ken Chen | 3871525 | 2010-11-11 14:05:16 -0800 | [diff] [blame] | 236 | i = tsk->latency_record_count++; |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 237 | memcpy(&tsk->latency_record[i], &lat, sizeof(struct latency_record)); |
| 238 | |
| 239 | out_unlock: |
Thomas Gleixner | 757455d | 2009-09-17 17:31:09 +0200 | [diff] [blame] | 240 | raw_spin_unlock_irqrestore(&latency_lock, flags); |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 241 | } |
| 242 | |
| 243 | static int lstats_show(struct seq_file *m, void *v) |
| 244 | { |
| 245 | int i; |
| 246 | |
| 247 | seq_puts(m, "Latency Top version : v0.1\n"); |
| 248 | |
| 249 | for (i = 0; i < MAXLR; i++) { |
Joe Perches | 34e49d4 | 2011-01-12 17:00:30 -0800 | [diff] [blame] | 250 | struct latency_record *lr = &latency_record[i]; |
| 251 | |
| 252 | if (lr->backtrace[0]) { |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 253 | int q; |
Joe Perches | 34e49d4 | 2011-01-12 17:00:30 -0800 | [diff] [blame] | 254 | seq_printf(m, "%i %lu %lu", |
| 255 | lr->count, lr->time, lr->max); |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 256 | for (q = 0; q < LT_BACKTRACEDEPTH; q++) { |
Joe Perches | 34e49d4 | 2011-01-12 17:00:30 -0800 | [diff] [blame] | 257 | unsigned long bt = lr->backtrace[q]; |
Thomas Gleixner | accddc4 | 2019-04-10 12:28:08 +0200 | [diff] [blame] | 258 | |
Joe Perches | 34e49d4 | 2011-01-12 17:00:30 -0800 | [diff] [blame] | 259 | if (!bt) |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 260 | break; |
Thomas Gleixner | accddc4 | 2019-04-10 12:28:08 +0200 | [diff] [blame] | 261 | |
Joe Perches | 34e49d4 | 2011-01-12 17:00:30 -0800 | [diff] [blame] | 262 | seq_printf(m, " %ps", (void *)bt); |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 263 | } |
Fabian Frederick | eaa1809 | 2014-06-04 16:11:21 -0700 | [diff] [blame] | 264 | seq_puts(m, "\n"); |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 265 | } |
| 266 | } |
| 267 | return 0; |
| 268 | } |
| 269 | |
| 270 | static ssize_t |
| 271 | lstats_write(struct file *file, const char __user *buf, size_t count, |
| 272 | loff_t *offs) |
| 273 | { |
| 274 | clear_global_latency_tracing(); |
| 275 | |
| 276 | return count; |
| 277 | } |
| 278 | |
| 279 | static int lstats_open(struct inode *inode, struct file *filp) |
| 280 | { |
| 281 | return single_open(filp, lstats_show, NULL); |
| 282 | } |
| 283 | |
Alexey Dobriyan | 97a3253 | 2020-02-03 17:37:17 -0800 | [diff] [blame] | 284 | static const struct proc_ops lstats_proc_ops = { |
| 285 | .proc_open = lstats_open, |
| 286 | .proc_read = seq_read, |
| 287 | .proc_write = lstats_write, |
| 288 | .proc_lseek = seq_lseek, |
| 289 | .proc_release = single_release, |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 290 | }; |
| 291 | |
| 292 | static int __init init_lstats_procfs(void) |
| 293 | { |
Alexey Dobriyan | 97a3253 | 2020-02-03 17:37:17 -0800 | [diff] [blame] | 294 | proc_create("latency_stats", 0644, NULL, &lstats_proc_ops); |
liaohua | 988f11e | 2022-04-07 15:29:48 +0800 | [diff] [blame] | 295 | #ifdef CONFIG_SYSCTL |
| 296 | register_sysctl_init("kernel", latencytop_sysctl); |
| 297 | #endif |
Arjan van de Ven | 9745512 | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 298 | return 0; |
| 299 | } |
Arjan van de Ven | ad0b0fd55 | 2009-02-10 11:42:26 -0800 | [diff] [blame] | 300 | device_initcall(init_lstats_procfs); |