Back to home page

OSCL-LXR

 
 

    


0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /* delayacct.c - per-task delay accounting
0003  *
0004  * Copyright (C) Shailabh Nagar, IBM Corp. 2006
0005  */
0006 
0007 #include <linux/sched.h>
0008 #include <linux/sched/task.h>
0009 #include <linux/sched/cputime.h>
0010 #include <linux/sched/clock.h>
0011 #include <linux/slab.h>
0012 #include <linux/taskstats.h>
0013 #include <linux/sysctl.h>
0014 #include <linux/delayacct.h>
0015 #include <linux/module.h>
0016 
0017 DEFINE_STATIC_KEY_FALSE(delayacct_key);
0018 int delayacct_on __read_mostly; /* Delay accounting turned on/off */
0019 struct kmem_cache *delayacct_cache;
0020 
0021 static void set_delayacct(bool enabled)
0022 {
0023     if (enabled) {
0024         static_branch_enable(&delayacct_key);
0025         delayacct_on = 1;
0026     } else {
0027         delayacct_on = 0;
0028         static_branch_disable(&delayacct_key);
0029     }
0030 }
0031 
0032 static int __init delayacct_setup_enable(char *str)
0033 {
0034     delayacct_on = 1;
0035     return 1;
0036 }
0037 __setup("delayacct", delayacct_setup_enable);
0038 
0039 void delayacct_init(void)
0040 {
0041     delayacct_cache = KMEM_CACHE(task_delay_info, SLAB_PANIC|SLAB_ACCOUNT);
0042     delayacct_tsk_init(&init_task);
0043     set_delayacct(delayacct_on);
0044 }
0045 
0046 #ifdef CONFIG_PROC_SYSCTL
0047 static int sysctl_delayacct(struct ctl_table *table, int write, void *buffer,
0048              size_t *lenp, loff_t *ppos)
0049 {
0050     int state = delayacct_on;
0051     struct ctl_table t;
0052     int err;
0053 
0054     if (write && !capable(CAP_SYS_ADMIN))
0055         return -EPERM;
0056 
0057     t = *table;
0058     t.data = &state;
0059     err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
0060     if (err < 0)
0061         return err;
0062     if (write)
0063         set_delayacct(state);
0064     return err;
0065 }
0066 
0067 static struct ctl_table kern_delayacct_table[] = {
0068     {
0069         .procname       = "task_delayacct",
0070         .data           = NULL,
0071         .maxlen         = sizeof(unsigned int),
0072         .mode           = 0644,
0073         .proc_handler   = sysctl_delayacct,
0074         .extra1         = SYSCTL_ZERO,
0075         .extra2         = SYSCTL_ONE,
0076     },
0077     { }
0078 };
0079 
0080 static __init int kernel_delayacct_sysctls_init(void)
0081 {
0082     register_sysctl_init("kernel", kern_delayacct_table);
0083     return 0;
0084 }
0085 late_initcall(kernel_delayacct_sysctls_init);
0086 #endif
0087 
0088 void __delayacct_tsk_init(struct task_struct *tsk)
0089 {
0090     tsk->delays = kmem_cache_zalloc(delayacct_cache, GFP_KERNEL);
0091     if (tsk->delays)
0092         raw_spin_lock_init(&tsk->delays->lock);
0093 }
0094 
0095 /*
0096  * Finish delay accounting for a statistic using its timestamps (@start),
0097  * accumalator (@total) and @count
0098  */
0099 static void delayacct_end(raw_spinlock_t *lock, u64 *start, u64 *total, u32 *count)
0100 {
0101     s64 ns = local_clock() - *start;
0102     unsigned long flags;
0103 
0104     if (ns > 0) {
0105         raw_spin_lock_irqsave(lock, flags);
0106         *total += ns;
0107         (*count)++;
0108         raw_spin_unlock_irqrestore(lock, flags);
0109     }
0110 }
0111 
0112 void __delayacct_blkio_start(void)
0113 {
0114     current->delays->blkio_start = local_clock();
0115 }
0116 
0117 /*
0118  * We cannot rely on the `current` macro, as we haven't yet switched back to
0119  * the process being woken.
0120  */
0121 void __delayacct_blkio_end(struct task_struct *p)
0122 {
0123     delayacct_end(&p->delays->lock,
0124               &p->delays->blkio_start,
0125               &p->delays->blkio_delay,
0126               &p->delays->blkio_count);
0127 }
0128 
0129 int delayacct_add_tsk(struct taskstats *d, struct task_struct *tsk)
0130 {
0131     u64 utime, stime, stimescaled, utimescaled;
0132     unsigned long long t2, t3;
0133     unsigned long flags, t1;
0134     s64 tmp;
0135 
0136     task_cputime(tsk, &utime, &stime);
0137     tmp = (s64)d->cpu_run_real_total;
0138     tmp += utime + stime;
0139     d->cpu_run_real_total = (tmp < (s64)d->cpu_run_real_total) ? 0 : tmp;
0140 
0141     task_cputime_scaled(tsk, &utimescaled, &stimescaled);
0142     tmp = (s64)d->cpu_scaled_run_real_total;
0143     tmp += utimescaled + stimescaled;
0144     d->cpu_scaled_run_real_total =
0145         (tmp < (s64)d->cpu_scaled_run_real_total) ? 0 : tmp;
0146 
0147     /*
0148      * No locking available for sched_info (and too expensive to add one)
0149      * Mitigate by taking snapshot of values
0150      */
0151     t1 = tsk->sched_info.pcount;
0152     t2 = tsk->sched_info.run_delay;
0153     t3 = tsk->se.sum_exec_runtime;
0154 
0155     d->cpu_count += t1;
0156 
0157     tmp = (s64)d->cpu_delay_total + t2;
0158     d->cpu_delay_total = (tmp < (s64)d->cpu_delay_total) ? 0 : tmp;
0159 
0160     tmp = (s64)d->cpu_run_virtual_total + t3;
0161     d->cpu_run_virtual_total =
0162         (tmp < (s64)d->cpu_run_virtual_total) ? 0 : tmp;
0163 
0164     if (!tsk->delays)
0165         return 0;
0166 
0167     /* zero XXX_total, non-zero XXX_count implies XXX stat overflowed */
0168 
0169     raw_spin_lock_irqsave(&tsk->delays->lock, flags);
0170     tmp = d->blkio_delay_total + tsk->delays->blkio_delay;
0171     d->blkio_delay_total = (tmp < d->blkio_delay_total) ? 0 : tmp;
0172     tmp = d->swapin_delay_total + tsk->delays->swapin_delay;
0173     d->swapin_delay_total = (tmp < d->swapin_delay_total) ? 0 : tmp;
0174     tmp = d->freepages_delay_total + tsk->delays->freepages_delay;
0175     d->freepages_delay_total = (tmp < d->freepages_delay_total) ? 0 : tmp;
0176     tmp = d->thrashing_delay_total + tsk->delays->thrashing_delay;
0177     d->thrashing_delay_total = (tmp < d->thrashing_delay_total) ? 0 : tmp;
0178     tmp = d->compact_delay_total + tsk->delays->compact_delay;
0179     d->compact_delay_total = (tmp < d->compact_delay_total) ? 0 : tmp;
0180     tmp = d->wpcopy_delay_total + tsk->delays->wpcopy_delay;
0181     d->wpcopy_delay_total = (tmp < d->wpcopy_delay_total) ? 0 : tmp;
0182     d->blkio_count += tsk->delays->blkio_count;
0183     d->swapin_count += tsk->delays->swapin_count;
0184     d->freepages_count += tsk->delays->freepages_count;
0185     d->thrashing_count += tsk->delays->thrashing_count;
0186     d->compact_count += tsk->delays->compact_count;
0187     d->wpcopy_count += tsk->delays->wpcopy_count;
0188     raw_spin_unlock_irqrestore(&tsk->delays->lock, flags);
0189 
0190     return 0;
0191 }
0192 
0193 __u64 __delayacct_blkio_ticks(struct task_struct *tsk)
0194 {
0195     __u64 ret;
0196     unsigned long flags;
0197 
0198     raw_spin_lock_irqsave(&tsk->delays->lock, flags);
0199     ret = nsec_to_clock_t(tsk->delays->blkio_delay);
0200     raw_spin_unlock_irqrestore(&tsk->delays->lock, flags);
0201     return ret;
0202 }
0203 
0204 void __delayacct_freepages_start(void)
0205 {
0206     current->delays->freepages_start = local_clock();
0207 }
0208 
0209 void __delayacct_freepages_end(void)
0210 {
0211     delayacct_end(&current->delays->lock,
0212               &current->delays->freepages_start,
0213               &current->delays->freepages_delay,
0214               &current->delays->freepages_count);
0215 }
0216 
0217 void __delayacct_thrashing_start(void)
0218 {
0219     current->delays->thrashing_start = local_clock();
0220 }
0221 
0222 void __delayacct_thrashing_end(void)
0223 {
0224     delayacct_end(&current->delays->lock,
0225               &current->delays->thrashing_start,
0226               &current->delays->thrashing_delay,
0227               &current->delays->thrashing_count);
0228 }
0229 
0230 void __delayacct_swapin_start(void)
0231 {
0232     current->delays->swapin_start = local_clock();
0233 }
0234 
0235 void __delayacct_swapin_end(void)
0236 {
0237     delayacct_end(&current->delays->lock,
0238               &current->delays->swapin_start,
0239               &current->delays->swapin_delay,
0240               &current->delays->swapin_count);
0241 }
0242 
0243 void __delayacct_compact_start(void)
0244 {
0245     current->delays->compact_start = local_clock();
0246 }
0247 
0248 void __delayacct_compact_end(void)
0249 {
0250     delayacct_end(&current->delays->lock,
0251               &current->delays->compact_start,
0252               &current->delays->compact_delay,
0253               &current->delays->compact_count);
0254 }
0255 
0256 void __delayacct_wpcopy_start(void)
0257 {
0258     current->delays->wpcopy_start = local_clock();
0259 }
0260 
0261 void __delayacct_wpcopy_end(void)
0262 {
0263     delayacct_end(&current->delays->lock,
0264               &current->delays->wpcopy_start,
0265               &current->delays->wpcopy_delay,
0266               &current->delays->wpcopy_count);
0267 }