]> git.sesse.net Git - bcachefs-tools-debian/blobdiff - libbcachefs/util.c
Merge pull request #211 from oz123/master
[bcachefs-tools-debian] / libbcachefs / util.c
index fa8537502fedfea6bc2928eb87f8044d1f5abf19..539735033947b91f4bd26768d51ca3049512dd45 100644 (file)
@@ -1,3 +1,4 @@
+// SPDX-License-Identifier: GPL-2.0
 /*
  * random utiility code, for bcache but in theory not specific to bcache
  *
 
 #include <linux/bio.h>
 #include <linux/blkdev.h>
+#include <linux/console.h>
 #include <linux/ctype.h>
 #include <linux/debugfs.h>
+#include <linux/eytzinger.h>
 #include <linux/freezer.h>
 #include <linux/kthread.h>
 #include <linux/log2.h>
 #include <linux/math64.h>
+#include <linux/percpu.h>
+#include <linux/preempt.h>
 #include <linux/random.h>
 #include <linux/seq_file.h>
 #include <linux/string.h>
 #include <linux/types.h>
 #include <linux/sched/clock.h>
+#include <linux/mean_and_variance.h>
 
 #include "util.h"
 
-#define simple_strtoint(c, end, base)  simple_strtol(c, end, base)
-#define simple_strtouint(c, end, base) simple_strtoul(c, end, base)
+static const char si_units[] = "?kMGTPEZY";
 
-#define STRTO_H(name, type)                                    \
-int bch2_ ## name ## _h(const char *cp, type *res)             \
-{                                                              \
-       int u = 0;                                              \
-       char *e;                                                \
-       type i = simple_ ## name(cp, &e, 10);                   \
-                                                               \
-       switch (tolower(*e)) {                                  \
-       default:                                                \
-               return -EINVAL;                                 \
-       case 'y':                                               \
-       case 'z':                                               \
-               u++;                                            \
-       case 'e':                                               \
-               u++;                                            \
-       case 'p':                                               \
-               u++;                                            \
-       case 't':                                               \
-               u++;                                            \
-       case 'g':                                               \
-               u++;                                            \
-       case 'm':                                               \
-               u++;                                            \
-       case 'k':                                               \
-               u++;                                            \
-               if (e++ == cp)                                  \
-                       return -EINVAL;                         \
-       case '\n':                                              \
-       case '\0':                                              \
-               if (*e == '\n')                                 \
-                       e++;                                    \
-       }                                                       \
-                                                               \
-       if (*e)                                                 \
-               return -EINVAL;                                 \
-                                                               \
-       while (u--) {                                           \
-               if ((type) ~0 > 0 &&                            \
-                   (type) ~0 / 1024 <= i)                      \
-                       return -EINVAL;                         \
-               if ((i > 0 && ANYSINT_MAX(type) / 1024 < i) ||  \
-                   (i < 0 && -ANYSINT_MAX(type) / 1024 > i))   \
-                       return -EINVAL;                         \
-               i *= 1024;                                      \
-       }                                                       \
-                                                               \
-       *res = i;                                               \
-       return 0;                                               \
-}                                                              \
-
-STRTO_H(strtoint, int)
-STRTO_H(strtouint, unsigned int)
-STRTO_H(strtoll, long long)
-STRTO_H(strtoull, unsigned long long)
+/* string_get_size units: */
+static const char *const units_2[] = {
+       "B", "KiB", "MiB", "GiB", "TiB", "PiB", "EiB", "ZiB", "YiB"
+};
+static const char *const units_10[] = {
+       "B", "kB", "MB", "GB", "TB", "PB", "EB", "ZB", "YB"
+};
 
-ssize_t bch2_hprint(char *buf, s64 v)
+static int parse_u64(const char *cp, u64 *res)
 {
-       static const char units[] = "?kMGTPEZY";
-       char dec[4] = "";
-       int u, t = 0;
+       const char *start = cp;
+       u64 v = 0;
 
-       for (u = 0; v >= 1024 || v <= -1024; u++) {
-               t = v & ~(~0U << 10);
-               v >>= 10;
-       }
+       if (!isdigit(*cp))
+               return -EINVAL;
 
-       if (!u)
-               return sprintf(buf, "%lli", v);
+       do {
+               if (v > U64_MAX / 10)
+                       return -ERANGE;
+               v *= 10;
+               if (v > U64_MAX - (*cp - '0'))
+                       return -ERANGE;
+               v += *cp - '0';
+               cp++;
+       } while (isdigit(*cp));
 
-       /*
-        * 103 is magic: t is in the range [-1023, 1023] and we want
-        * to turn it into [-9, 9]
-        */
-       if (v < 100 && v > -100)
-               scnprintf(dec, sizeof(dec), ".%i", t / 103);
+       *res = v;
+       return cp - start;
+}
+
+static int bch2_pow(u64 n, u64 p, u64 *res)
+{
+       *res = 1;
 
-       return sprintf(buf, "%lli%s%c", v, dec, units[u]);
+       while (p--) {
+               if (*res > div_u64(U64_MAX, n))
+                       return -ERANGE;
+               *res *= n;
+       }
+       return 0;
 }
 
-ssize_t bch2_scnprint_string_list(char *buf, size_t size,
-                                 const char * const list[],
-                                 size_t selected)
+static int parse_unit_suffix(const char *cp, u64 *res)
 {
-       char *out = buf;
-       size_t i;
+       const char *start = cp;
+       u64 base = 1024;
+       unsigned u;
+       int ret;
+
+       if (*cp == ' ')
+               cp++;
 
-       if (size)
-               *out = '\0';
+       for (u = 1; u < strlen(si_units); u++)
+               if (*cp == si_units[u]) {
+                       cp++;
+                       goto got_unit;
+               }
+
+       for (u = 0; u < ARRAY_SIZE(units_2); u++)
+               if (!strncmp(cp, units_2[u], strlen(units_2[u]))) {
+                       cp += strlen(units_2[u]);
+                       goto got_unit;
+               }
 
-       for (i = 0; list[i]; i++)
-               out += scnprintf(out, buf + size - out,
-                                i == selected ? "[%s] " : "%s ", list[i]);
+       for (u = 0; u < ARRAY_SIZE(units_10); u++)
+               if (!strncmp(cp, units_10[u], strlen(units_10[u]))) {
+                       cp += strlen(units_10[u]);
+                       base = 1000;
+                       goto got_unit;
+               }
 
-       if (out != buf)
-               *--out = '\0';
+       *res = 1;
+       return 0;
+got_unit:
+       ret = bch2_pow(base, u, res);
+       if (ret)
+               return ret;
 
-       return out - buf;
+       return cp - start;
 }
 
-ssize_t bch2_read_string_list(const char *buf, const char * const list[])
+#define parse_or_ret(cp, _f)                   \
+do {                                           \
+       int _ret = _f;                          \
+       if (_ret < 0)                           \
+               return _ret;                    \
+       cp += _ret;                             \
+} while (0)
+
+static int __bch2_strtou64_h(const char *cp, u64 *res)
 {
-       size_t i, len;
+       const char *start = cp;
+       u64 v = 0, b, f_n = 0, f_d = 1;
+       int ret;
 
-       buf = skip_spaces(buf);
+       parse_or_ret(cp, parse_u64(cp, &v));
 
-       len = strlen(buf);
-       while (len && isspace(buf[len - 1]))
-               --len;
+       if (*cp == '.') {
+               cp++;
+               ret = parse_u64(cp, &f_n);
+               if (ret < 0)
+                       return ret;
+               cp += ret;
 
-       for (i = 0; list[i]; i++)
-               if (strlen(list[i]) == len &&
-                   !memcmp(buf, list[i], len))
-                       break;
+               ret = bch2_pow(10, ret, &f_d);
+               if (ret)
+                       return ret;
+       }
+
+       parse_or_ret(cp, parse_unit_suffix(cp, &b));
+
+       if (v > div_u64(U64_MAX, b))
+               return -ERANGE;
+       v *= b;
+
+       if (f_n > div_u64(U64_MAX, b))
+               return -ERANGE;
 
-       return list[i] ? i : -EINVAL;
+       f_n = div_u64(f_n * b, f_d);
+       if (v + f_n < v)
+               return -ERANGE;
+       v += f_n;
+
+       *res = v;
+       return cp - start;
 }
 
-ssize_t bch2_scnprint_flag_list(char *buf, size_t size,
-                               const char * const list[], u64 flags)
+static int __bch2_strtoh(const char *cp, u64 *res,
+                        u64 t_max, bool t_signed)
 {
-       char *out = buf, *end = buf + size;
-       unsigned bit, nr = 0;
+       bool positive = *cp != '-';
+       u64 v = 0;
+
+       if (*cp == '+' || *cp == '-')
+               cp++;
 
-       while (list[nr])
-               nr++;
+       parse_or_ret(cp, __bch2_strtou64_h(cp, &v));
 
-       if (size)
-               *out = '\0';
+       if (*cp == '\n')
+               cp++;
+       if (*cp)
+               return -EINVAL;
+
+       if (positive) {
+               if (v > t_max)
+                       return -ERANGE;
+       } else {
+               if (v && !t_signed)
+                       return -ERANGE;
 
-       while (flags && (bit = __ffs(flags)) < nr) {
-               out += scnprintf(out, end - out, "%s,", list[bit]);
-               flags ^= 1 << bit;
+               if (v > t_max + 1)
+                       return -ERANGE;
+               v = -v;
        }
 
-       if (out != buf)
-               *--out = '\0';
+       *res = v;
+       return 0;
+}
 
-       return out - buf;
+#define STRTO_H(name, type)                                    \
+int bch2_ ## name ## _h(const char *cp, type *res)             \
+{                                                              \
+       u64 v = 0;                                              \
+       int ret = __bch2_strtoh(cp, &v, ANYSINT_MAX(type),      \
+                       ANYSINT_MAX(type) != ((type) ~0ULL));   \
+       *res = v;                                               \
+       return ret;                                             \
 }
 
+STRTO_H(strtoint, int)
+STRTO_H(strtouint, unsigned int)
+STRTO_H(strtoll, long long)
+STRTO_H(strtoull, unsigned long long)
+STRTO_H(strtou64, u64)
+
 u64 bch2_read_flag_list(char *opt, const char * const list[])
 {
        u64 ret = 0;
-       char *p, *s, *d = kstrndup(opt, PAGE_SIZE - 1, GFP_KERNEL);
+       char *p, *s, *d = kstrdup(opt, GFP_KERNEL);
 
        if (!d)
                return -ENOMEM;
@@ -175,7 +215,8 @@ u64 bch2_read_flag_list(char *opt, const char * const list[])
        s = strim(d);
 
        while ((p = strsep(&s, ","))) {
-               int flag = bch2_read_string_list(p, list);
+               int flag = match_string(list, -1, p);
+
                if (flag < 0) {
                        ret = -1;
                        break;
@@ -200,66 +241,257 @@ bool bch2_is_zero(const void *_p, size_t n)
        return true;
 }
 
-void bch2_time_stats_clear(struct time_stats *stats)
+void bch2_prt_u64_base2_nbits(struct printbuf *out, u64 v, unsigned nr_bits)
 {
-       spin_lock(&stats->lock);
+       while (nr_bits)
+               prt_char(out, '0' + ((v >> --nr_bits) & 1));
+}
 
-       stats->count = 0;
-       stats->last_duration = 0;
-       stats->max_duration = 0;
-       stats->average_duration = 0;
-       stats->average_frequency = 0;
-       stats->last = 0;
+void bch2_prt_u64_base2(struct printbuf *out, u64 v)
+{
+       bch2_prt_u64_base2_nbits(out, v, fls64(v) ?: 1);
+}
 
-       spin_unlock(&stats->lock);
+void bch2_print_string_as_lines(const char *prefix, const char *lines)
+{
+       const char *p;
+
+       if (!lines) {
+               printk("%s (null)\n", prefix);
+               return;
+       }
+
+       console_lock();
+       while (1) {
+               p = strchrnul(lines, '\n');
+               printk("%s%.*s\n", prefix, (int) (p - lines), lines);
+               if (!*p)
+                       break;
+               lines = p + 1;
+       }
+       console_unlock();
 }
 
-void __bch2_time_stats_update(struct time_stats *stats, u64 start_time)
+int bch2_save_backtrace(bch_stacktrace *stack, struct task_struct *task, unsigned skipnr,
+                       gfp_t gfp)
 {
-       u64 now, duration, last;
+#ifdef CONFIG_STACKTRACE
+       unsigned nr_entries = 0;
 
-       stats->count++;
+       stack->nr = 0;
+       int ret = darray_make_room_gfp(stack, 32, gfp);
+       if (ret)
+               return ret;
 
-       now             = local_clock();
-       duration        = time_after64(now, start_time)
-               ? now - start_time : 0;
-       last            = time_after64(now, stats->last)
-               ? now - stats->last : 0;
+       if (!down_read_trylock(&task->signal->exec_update_lock))
+               return -1;
 
-       stats->last_duration = duration;
-       stats->max_duration = max(stats->max_duration, duration);
+       do {
+               nr_entries = stack_trace_save_tsk(task, stack->data, stack->size, skipnr + 1);
+       } while (nr_entries == stack->size &&
+                !(ret = darray_make_room(stack, stack->size * 2)));
 
-       if (stats->last) {
-               stats->average_duration = ewma_add(stats->average_duration,
-                                                  duration << 8, 3);
+       stack->nr = nr_entries;
+       up_read(&task->signal->exec_update_lock);
 
-               if (stats->average_frequency)
-                       stats->average_frequency =
-                               ewma_add(stats->average_frequency,
-                                        last << 8, 3);
-               else
-                       stats->average_frequency  = last << 8;
-       } else {
-               stats->average_duration = duration << 8;
+       return ret;
+#else
+       return 0;
+#endif
+}
+
+void bch2_prt_backtrace(struct printbuf *out, bch_stacktrace *stack)
+{
+       darray_for_each(*stack, i) {
+               prt_printf(out, "[<0>] %pB", (void *) *i);
+               prt_newline(out);
        }
+}
+
+int bch2_prt_task_backtrace(struct printbuf *out, struct task_struct *task, unsigned skipnr, gfp_t gfp)
+{
+       bch_stacktrace stack = { 0 };
+       int ret = bch2_save_backtrace(&stack, task, skipnr + 1, gfp);
+
+       bch2_prt_backtrace(out, &stack);
+       darray_exit(&stack);
+       return ret;
+}
 
-       stats->last = now ?: 1;
+#ifndef __KERNEL__
+#include <time.h>
+void bch2_prt_datetime(struct printbuf *out, time64_t sec)
+{
+       time_t t = sec;
+       char buf[64];
+       ctime_r(&t, buf);
+       strim(buf);
+       prt_str(out, buf);
+}
+#else
+void bch2_prt_datetime(struct printbuf *out, time64_t sec)
+{
+       char buf[64];
+       snprintf(buf, sizeof(buf), "%ptT", &sec);
+       prt_u64(out, sec);
 }
+#endif
+
+void bch2_pr_time_units(struct printbuf *out, u64 ns)
+{
+       const struct time_unit *u = pick_time_units(ns);
+
+       prt_printf(out, "%llu %s", div_u64(ns, u->nsecs), u->name);
+}
+
+static void bch2_pr_time_units_aligned(struct printbuf *out, u64 ns)
+{
+       const struct time_unit *u = pick_time_units(ns);
+
+       prt_printf(out, "%llu ", div64_u64(ns, u->nsecs));
+       prt_tab_rjust(out);
+       prt_printf(out, "%s", u->name);
+}
+
+static inline void pr_name_and_units(struct printbuf *out, const char *name, u64 ns)
+{
+       prt_str(out, name);
+       prt_tab(out);
+       bch2_pr_time_units_aligned(out, ns);
+       prt_newline(out);
+}
+
+#define TABSTOP_SIZE 12
 
-void bch2_time_stats_update(struct time_stats *stats, u64 start_time)
+void bch2_time_stats_to_text(struct printbuf *out, struct time_stats *stats)
 {
-       spin_lock(&stats->lock);
-       __bch2_time_stats_update(stats, start_time);
-       spin_unlock(&stats->lock);
+       struct quantiles *quantiles = time_stats_to_quantiles(stats);
+       s64 f_mean = 0, d_mean = 0;
+       u64 f_stddev = 0, d_stddev = 0;
+
+       if (stats->buffer) {
+               int cpu;
+
+               spin_lock_irq(&stats->lock);
+               for_each_possible_cpu(cpu)
+                       __time_stats_clear_buffer(stats, per_cpu_ptr(stats->buffer, cpu));
+               spin_unlock_irq(&stats->lock);
+       }
+
+       /*
+        * avoid divide by zero
+        */
+       if (stats->freq_stats.n) {
+               f_mean = mean_and_variance_get_mean(stats->freq_stats);
+               f_stddev = mean_and_variance_get_stddev(stats->freq_stats);
+               d_mean = mean_and_variance_get_mean(stats->duration_stats);
+               d_stddev = mean_and_variance_get_stddev(stats->duration_stats);
+       }
+
+       printbuf_tabstop_push(out, out->indent + TABSTOP_SIZE);
+       prt_printf(out, "count:");
+       prt_tab(out);
+       prt_printf(out, "%llu ",
+                        stats->duration_stats.n);
+       printbuf_tabstop_pop(out);
+       prt_newline(out);
+
+       printbuf_tabstops_reset(out);
+
+       printbuf_tabstop_push(out, out->indent + 20);
+       printbuf_tabstop_push(out, TABSTOP_SIZE + 2);
+       printbuf_tabstop_push(out, 0);
+       printbuf_tabstop_push(out, TABSTOP_SIZE + 2);
+
+       prt_tab(out);
+       prt_printf(out, "since mount");
+       prt_tab_rjust(out);
+       prt_tab(out);
+       prt_printf(out, "recent");
+       prt_tab_rjust(out);
+       prt_newline(out);
+
+       printbuf_tabstops_reset(out);
+       printbuf_tabstop_push(out, out->indent + 20);
+       printbuf_tabstop_push(out, TABSTOP_SIZE);
+       printbuf_tabstop_push(out, 2);
+       printbuf_tabstop_push(out, TABSTOP_SIZE);
+
+       prt_printf(out, "duration of events");
+       prt_newline(out);
+       printbuf_indent_add(out, 2);
+
+       pr_name_and_units(out, "min:", stats->min_duration);
+       pr_name_and_units(out, "max:", stats->max_duration);
+       pr_name_and_units(out, "total:", stats->total_duration);
+
+       prt_printf(out, "mean:");
+       prt_tab(out);
+       bch2_pr_time_units_aligned(out, d_mean);
+       prt_tab(out);
+       bch2_pr_time_units_aligned(out, mean_and_variance_weighted_get_mean(stats->duration_stats_weighted, TIME_STATS_MV_WEIGHT));
+       prt_newline(out);
+
+       prt_printf(out, "stddev:");
+       prt_tab(out);
+       bch2_pr_time_units_aligned(out, d_stddev);
+       prt_tab(out);
+       bch2_pr_time_units_aligned(out, mean_and_variance_weighted_get_stddev(stats->duration_stats_weighted, TIME_STATS_MV_WEIGHT));
+
+       printbuf_indent_sub(out, 2);
+       prt_newline(out);
+
+       prt_printf(out, "time between events");
+       prt_newline(out);
+       printbuf_indent_add(out, 2);
+
+       pr_name_and_units(out, "min:", stats->min_freq);
+       pr_name_and_units(out, "max:", stats->max_freq);
+
+       prt_printf(out, "mean:");
+       prt_tab(out);
+       bch2_pr_time_units_aligned(out, f_mean);
+       prt_tab(out);
+       bch2_pr_time_units_aligned(out, mean_and_variance_weighted_get_mean(stats->freq_stats_weighted, TIME_STATS_MV_WEIGHT));
+       prt_newline(out);
+
+       prt_printf(out, "stddev:");
+       prt_tab(out);
+       bch2_pr_time_units_aligned(out, f_stddev);
+       prt_tab(out);
+       bch2_pr_time_units_aligned(out, mean_and_variance_weighted_get_stddev(stats->freq_stats_weighted, TIME_STATS_MV_WEIGHT));
+
+       printbuf_indent_sub(out, 2);
+       prt_newline(out);
+
+       printbuf_tabstops_reset(out);
+
+       if (quantiles) {
+               int i = eytzinger0_first(NR_QUANTILES);
+               const struct time_unit *u =
+                       pick_time_units(quantiles->entries[i].m);
+               u64 last_q = 0;
+
+               prt_printf(out, "quantiles (%s):\t", u->name);
+               eytzinger0_for_each(i, NR_QUANTILES) {
+                       bool is_last = eytzinger0_next(i, NR_QUANTILES) == -1;
+
+                       u64 q = max(quantiles->entries[i].m, last_q);
+                       prt_printf(out, "%llu ", div_u64(q, u->nsecs));
+                       if (is_last)
+                               prt_newline(out);
+                       last_q = q;
+               }
+       }
 }
 
+/* ratelimit: */
+
 /**
  * bch2_ratelimit_delay() - return how long to delay until the next time to do
- * some work
- *
- * @d - the struct bch_ratelimit to update
- *
- * Returns the amount of time to delay by, in jiffies
+ *             some work
+ * @d:         the struct bch_ratelimit to update
+ * Returns:    the amount of time to delay by, in jiffies
  */
 u64 bch2_ratelimit_delay(struct bch_ratelimit *d)
 {
@@ -272,9 +504,8 @@ u64 bch2_ratelimit_delay(struct bch_ratelimit *d)
 
 /**
  * bch2_ratelimit_increment() - increment @d by the amount of work done
- *
- * @d - the struct bch_ratelimit to update
- * @done - the amount of work done, in arbitrary units
+ * @d:         the struct bch_ratelimit to update
+ * @done:      the amount of work done, in arbitrary units
  */
 void bch2_ratelimit_increment(struct bch_ratelimit *d, u64 done)
 {
@@ -289,26 +520,7 @@ void bch2_ratelimit_increment(struct bch_ratelimit *d, u64 done)
                d->next = now - NSEC_PER_SEC * 2;
 }
 
-int bch2_ratelimit_wait_freezable_stoppable(struct bch_ratelimit *d)
-{
-       bool kthread = (current->flags & PF_KTHREAD) != 0;
-
-       while (1) {
-               u64 delay = bch2_ratelimit_delay(d);
-
-               if (delay)
-                       set_current_state(TASK_INTERRUPTIBLE);
-
-               if (kthread && kthread_should_stop())
-                       return 1;
-
-               if (!delay)
-                       return 0;
-
-               schedule_timeout(delay);
-               try_to_freeze();
-       }
-}
+/* pd controller: */
 
 /*
  * Updates pd_controller. Attempts to scale inputed values to units per second.
@@ -372,64 +584,82 @@ void bch2_pd_controller_init(struct bch_pd_controller *pd)
        pd->backpressure        = 1;
 }
 
-size_t bch2_pd_controller_print_debug(struct bch_pd_controller *pd, char *buf)
-{
-       /* 2^64 - 1 is 20 digits, plus null byte */
-       char rate[21];
-       char actual[21];
-       char target[21];
-       char proportional[21];
-       char derivative[21];
-       char change[21];
-       s64 next_io;
-
-       bch2_hprint(rate,       pd->rate.rate);
-       bch2_hprint(actual,     pd->last_actual);
-       bch2_hprint(target,     pd->last_target);
-       bch2_hprint(proportional, pd->last_proportional);
-       bch2_hprint(derivative, pd->last_derivative);
-       bch2_hprint(change,     pd->last_change);
-
-       next_io = div64_s64(pd->rate.next - local_clock(), NSEC_PER_MSEC);
-
-       return sprintf(buf,
-                      "rate:\t\t%s/sec\n"
-                      "target:\t\t%s\n"
-                      "actual:\t\t%s\n"
-                      "proportional:\t%s\n"
-                      "derivative:\t%s\n"
-                      "change:\t\t%s/sec\n"
-                      "next io:\t%llims\n",
-                      rate, target, actual, proportional,
-                      derivative, change, next_io);
-}
-
-void bch2_bio_map(struct bio *bio, void *base)
-{
-       size_t size = bio->bi_iter.bi_size;
-       struct bio_vec *bv = bio->bi_io_vec;
-
-       BUG_ON(!bio->bi_iter.bi_size);
-       BUG_ON(bio->bi_vcnt);
-
-       bv->bv_offset = base ? offset_in_page(base) : 0;
-       goto start;
-
-       for (; size; bio->bi_vcnt++, bv++) {
-               bv->bv_offset   = 0;
-start:         bv->bv_len      = min_t(size_t, PAGE_SIZE - bv->bv_offset,
-                                       size);
-               BUG_ON(bio->bi_vcnt >= bio->bi_max_vecs);
-               if (base) {
-                       bv->bv_page = is_vmalloc_addr(base)
+void bch2_pd_controller_debug_to_text(struct printbuf *out, struct bch_pd_controller *pd)
+{
+       if (!out->nr_tabstops)
+               printbuf_tabstop_push(out, 20);
+
+       prt_printf(out, "rate:");
+       prt_tab(out);
+       prt_human_readable_s64(out, pd->rate.rate);
+       prt_newline(out);
+
+       prt_printf(out, "target:");
+       prt_tab(out);
+       prt_human_readable_u64(out, pd->last_target);
+       prt_newline(out);
+
+       prt_printf(out, "actual:");
+       prt_tab(out);
+       prt_human_readable_u64(out, pd->last_actual);
+       prt_newline(out);
+
+       prt_printf(out, "proportional:");
+       prt_tab(out);
+       prt_human_readable_s64(out, pd->last_proportional);
+       prt_newline(out);
+
+       prt_printf(out, "derivative:");
+       prt_tab(out);
+       prt_human_readable_s64(out, pd->last_derivative);
+       prt_newline(out);
+
+       prt_printf(out, "change:");
+       prt_tab(out);
+       prt_human_readable_s64(out, pd->last_change);
+       prt_newline(out);
+
+       prt_printf(out, "next io:");
+       prt_tab(out);
+       prt_printf(out, "%llims", div64_s64(pd->rate.next - local_clock(), NSEC_PER_MSEC));
+       prt_newline(out);
+}
+
+/* misc: */
+
+void bch2_bio_map(struct bio *bio, void *base, size_t size)
+{
+       while (size) {
+               struct page *page = is_vmalloc_addr(base)
                                ? vmalloc_to_page(base)
                                : virt_to_page(base);
+               unsigned offset = offset_in_page(base);
+               unsigned len = min_t(size_t, PAGE_SIZE - offset, size);
+
+               BUG_ON(!bio_add_page(bio, page, len, offset));
+               size -= len;
+               base += len;
+       }
+}
+
+int bch2_bio_alloc_pages(struct bio *bio, size_t size, gfp_t gfp_mask)
+{
+       while (size) {
+               struct page *page = alloc_pages(gfp_mask, 0);
+               unsigned len = min_t(size_t, PAGE_SIZE, size);
+
+               if (!page)
+                       return -ENOMEM;
 
-                       base += bv->bv_len;
+               if (unlikely(!bio_add_page(bio, page, len, 0))) {
+                       __free_page(page);
+                       break;
                }
 
-               size -= bv->bv_len;
+               size -= len;
        }
+
+       return 0;
 }
 
 size_t bch2_rand_range(size_t max)
@@ -447,15 +677,16 @@ size_t bch2_rand_range(size_t max)
        return rand;
 }
 
-void memcpy_to_bio(struct bio *dst, struct bvec_iter dst_iter, void *src)
+void memcpy_to_bio(struct bio *dst, struct bvec_iter dst_iter, const void *src)
 {
        struct bio_vec bv;
        struct bvec_iter iter;
 
        __bio_for_each_segment(bv, dst, iter, dst_iter) {
-               void *dstp = kmap_atomic(bv.bv_page);
+               void *dstp = kmap_local_page(bv.bv_page);
+
                memcpy(dstp + bv.bv_offset, src, bv.bv_len);
-               kunmap_atomic(dstp);
+               kunmap_local(dstp);
 
                src += bv.bv_len;
        }
@@ -467,187 +698,15 @@ void memcpy_from_bio(void *dst, struct bio *src, struct bvec_iter src_iter)
        struct bvec_iter iter;
 
        __bio_for_each_segment(bv, src, iter, src_iter) {
-               void *srcp = kmap_atomic(bv.bv_page);
+               void *srcp = kmap_local_page(bv.bv_page);
+
                memcpy(dst, srcp + bv.bv_offset, bv.bv_len);
-               kunmap_atomic(srcp);
+               kunmap_local(srcp);
 
                dst += bv.bv_len;
        }
 }
 
-size_t bch_scnmemcpy(char *buf, size_t size, const char *src, size_t len)
-{
-       size_t n;
-
-       if (!size)
-               return 0;
-
-       n = min(size - 1, len);
-       memcpy(buf, src, n);
-       buf[n] = '\0';
-
-       return n;
-}
-
-#include "eytzinger.h"
-
-static int alignment_ok(const void *base, size_t align)
-{
-       return IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ||
-               ((unsigned long)base & (align - 1)) == 0;
-}
-
-static void u32_swap(void *a, void *b, size_t size)
-{
-       u32 t = *(u32 *)a;
-       *(u32 *)a = *(u32 *)b;
-       *(u32 *)b = t;
-}
-
-static void u64_swap(void *a, void *b, size_t size)
-{
-       u64 t = *(u64 *)a;
-       *(u64 *)a = *(u64 *)b;
-       *(u64 *)b = t;
-}
-
-static void generic_swap(void *a, void *b, size_t size)
-{
-       char t;
-
-       do {
-               t = *(char *)a;
-               *(char *)a++ = *(char *)b;
-               *(char *)b++ = t;
-       } while (--size > 0);
-}
-
-static inline int do_cmp(void *base, size_t n, size_t size,
-                        int (*cmp_func)(const void *, const void *, size_t),
-                        size_t l, size_t r)
-{
-       return cmp_func(base + inorder_to_eytzinger0(l, n) * size,
-                       base + inorder_to_eytzinger0(r, n) * size,
-                       size);
-}
-
-static inline void do_swap(void *base, size_t n, size_t size,
-                          void (*swap_func)(void *, void *, size_t),
-                          size_t l, size_t r)
-{
-       swap_func(base + inorder_to_eytzinger0(l, n) * size,
-                 base + inorder_to_eytzinger0(r, n) * size,
-                 size);
-}
-
-void eytzinger0_sort(void *base, size_t n, size_t size,
-                    int (*cmp_func)(const void *, const void *, size_t),
-                    void (*swap_func)(void *, void *, size_t))
-{
-       int i, c, r;
-
-       if (!swap_func) {
-               if (size == 4 && alignment_ok(base, 4))
-                       swap_func = u32_swap;
-               else if (size == 8 && alignment_ok(base, 8))
-                       swap_func = u64_swap;
-               else
-                       swap_func = generic_swap;
-       }
-
-       /* heapify */
-       for (i = n / 2 - 1; i >= 0; --i) {
-               for (r = i; r * 2 + 1 < n; r = c) {
-                       c = r * 2 + 1;
-
-                       if (c + 1 < n &&
-                           do_cmp(base, n, size, cmp_func, c, c + 1) < 0)
-                               c++;
-
-                       if (do_cmp(base, n, size, cmp_func, r, c) >= 0)
-                               break;
-
-                       do_swap(base, n, size, swap_func, r, c);
-               }
-       }
-
-       /* sort */
-       for (i = n - 1; i > 0; --i) {
-               do_swap(base, n, size, swap_func, 0, i);
-
-               for (r = 0; r * 2 + 1 < i; r = c) {
-                       c = r * 2 + 1;
-
-                       if (c + 1 < i &&
-                           do_cmp(base, n, size, cmp_func, c, c + 1) < 0)
-                               c++;
-
-                       if (do_cmp(base, n, size, cmp_func, r, c) >= 0)
-                               break;
-
-                       do_swap(base, n, size, swap_func, r, c);
-               }
-       }
-}
-
-void sort_cmp_size(void *base, size_t num, size_t size,
-         int (*cmp_func)(const void *, const void *, size_t),
-         void (*swap_func)(void *, void *, size_t size))
-{
-       /* pre-scale counters for performance */
-       int i = (num/2 - 1) * size, n = num * size, c, r;
-
-       if (!swap_func) {
-               if (size == 4 && alignment_ok(base, 4))
-                       swap_func = u32_swap;
-               else if (size == 8 && alignment_ok(base, 8))
-                       swap_func = u64_swap;
-               else
-                       swap_func = generic_swap;
-       }
-
-       /* heapify */
-       for ( ; i >= 0; i -= size) {
-               for (r = i; r * 2 + size < n; r  = c) {
-                       c = r * 2 + size;
-                       if (c < n - size &&
-                           cmp_func(base + c, base + c + size, size) < 0)
-                               c += size;
-                       if (cmp_func(base + r, base + c, size) >= 0)
-                               break;
-                       swap_func(base + r, base + c, size);
-               }
-       }
-
-       /* sort */
-       for (i = n - size; i > 0; i -= size) {
-               swap_func(base, base + i, size);
-               for (r = 0; r * 2 + size < i; r = c) {
-                       c = r * 2 + size;
-                       if (c < i - size &&
-                           cmp_func(base + c, base + c + size, size) < 0)
-                               c += size;
-                       if (cmp_func(base + r, base + c, size) >= 0)
-                               break;
-                       swap_func(base + r, base + c, size);
-               }
-       }
-}
-
-void mempool_free_vp(void *element, void *pool_data)
-{
-       size_t size = (size_t) pool_data;
-
-       vpfree(element, size);
-}
-
-void *mempool_alloc_vp(gfp_t gfp_mask, void *pool_data)
-{
-       size_t size = (size_t) pool_data;
-
-       return vpmalloc(size, gfp_mask);
-}
-
 #if 0
 void eytzinger1_test(void)
 {
@@ -773,3 +832,65 @@ void eytzinger0_find_test(void)
        kfree(test_array);
 }
 #endif
+
+/*
+ * Accumulate percpu counters onto one cpu's copy - only valid when access
+ * against any percpu counter is guarded against
+ */
+u64 *bch2_acc_percpu_u64s(u64 __percpu *p, unsigned nr)
+{
+       u64 *ret;
+       int cpu;
+
+       /* access to pcpu vars has to be blocked by other locking */
+       preempt_disable();
+       ret = this_cpu_ptr(p);
+       preempt_enable();
+
+       for_each_possible_cpu(cpu) {
+               u64 *i = per_cpu_ptr(p, cpu);
+
+               if (i != ret) {
+                       acc_u64s(ret, i, nr);
+                       memset(i, 0, nr * sizeof(u64));
+               }
+       }
+
+       return ret;
+}
+
+void bch2_darray_str_exit(darray_str *d)
+{
+       darray_for_each(*d, i)
+               kfree(*i);
+       darray_exit(d);
+}
+
+int bch2_split_devs(const char *_dev_name, darray_str *ret)
+{
+       darray_init(ret);
+
+       char *dev_name, *s, *orig;
+
+       dev_name = orig = kstrdup(_dev_name, GFP_KERNEL);
+       if (!dev_name)
+               return -ENOMEM;
+
+       while ((s = strsep(&dev_name, ":"))) {
+               char *p = kstrdup(s, GFP_KERNEL);
+               if (!p)
+                       goto err;
+
+               if (darray_push(ret, p)) {
+                       kfree(p);
+                       goto err;
+               }
+       }
+
+       kfree(orig);
+       return 0;
+err:
+       bch2_darray_str_exit(ret);
+       kfree(orig);
+       return -ENOMEM;
+}