projects/linux·blob master · refs

1136 lines · 30.2 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 * tracing_map - lock-free map for tracing
4 *
5 * Copyright (C) 2015 Tom Zanussi <[email protected]>
6 *
7 * tracing_map implementation inspired by lock-free map algorithms
8 * originated by Dr. Cliff Click:
9 *
10 * http://www.azulsystems.com/blog/cliff/2007-03-26-non-blocking-hashtable
11 * http://www.azulsystems.com/events/javaone_2007/2007_LockFreeHash.pdf
12 */
13
14#include <linux/vmalloc.h>
15#include <linux/jhash.h>
16#include <linux/slab.h>
17#include <linux/sort.h>
18#include <linux/kmemleak.h>
19
20#include "tracing_map.h"
21#include "trace.h"
22
23/*
24 * NOTE: For a detailed description of the data structures used by
25 * these functions (such as tracing_map_elt) please see the overview
26 * of tracing_map data structures at the beginning of tracing_map.h.
27 */
28
29/**
30 * tracing_map_update_sum - Add a value to a tracing_map_elt's sum field
31 * @elt: The tracing_map_elt
32 * @i: The index of the given sum associated with the tracing_map_elt
33 * @n: The value to add to the sum
34 *
35 * Add n to sum i associated with the specified tracing_map_elt
36 * instance. The index i is the index returned by the call to
37 * tracing_map_add_sum_field() when the tracing map was set up.
38 */
39void tracing_map_update_sum(struct tracing_map_elt *elt, unsigned int i, u64 n)
40{
41 atomic64_add(n, &elt->fields[i].sum);
42}
43
44/**
45 * tracing_map_read_sum - Return the value of a tracing_map_elt's sum field
46 * @elt: The tracing_map_elt
47 * @i: The index of the given sum associated with the tracing_map_elt
48 *
49 * Retrieve the value of the sum i associated with the specified
50 * tracing_map_elt instance. The index i is the index returned by the
51 * call to tracing_map_add_sum_field() when the tracing map was set
52 * up.
53 *
54 * Return: The sum associated with field i for elt.
55 */
56u64 tracing_map_read_sum(struct tracing_map_elt *elt, unsigned int i)
57{
58 return (u64)atomic64_read(&elt->fields[i].sum);
59}
60
61/**
62 * tracing_map_set_var - Assign a tracing_map_elt's variable field
63 * @elt: The tracing_map_elt
64 * @i: The index of the given variable associated with the tracing_map_elt
65 * @n: The value to assign
66 *
67 * Assign n to variable i associated with the specified tracing_map_elt
68 * instance. The index i is the index returned by the call to
69 * tracing_map_add_var() when the tracing map was set up.
70 */
71void tracing_map_set_var(struct tracing_map_elt *elt, unsigned int i, u64 n)
72{
73 atomic64_set(&elt->vars[i], n);
74 elt->var_set[i] = true;
75}
76
77/**
78 * tracing_map_var_set - Return whether or not a variable has been set
79 * @elt: The tracing_map_elt
80 * @i: The index of the given variable associated with the tracing_map_elt
81 *
82 * Return true if the variable has been set, false otherwise. The
83 * index i is the index returned by the call to tracing_map_add_var()
84 * when the tracing map was set up.
85 */
86bool tracing_map_var_set(struct tracing_map_elt *elt, unsigned int i)
87{
88 return elt->var_set[i];
89}
90
91/**
92 * tracing_map_read_var - Return the value of a tracing_map_elt's variable field
93 * @elt: The tracing_map_elt
94 * @i: The index of the given variable associated with the tracing_map_elt
95 *
96 * Retrieve the value of the variable i associated with the specified
97 * tracing_map_elt instance. The index i is the index returned by the
98 * call to tracing_map_add_var() when the tracing map was set
99 * up.
100 *
101 * Return: The variable value associated with field i for elt.
102 */
103u64 tracing_map_read_var(struct tracing_map_elt *elt, unsigned int i)
104{
105 return (u64)atomic64_read(&elt->vars[i]);
106}
107
108/**
109 * tracing_map_read_var_once - Return and reset a tracing_map_elt's variable field
110 * @elt: The tracing_map_elt
111 * @i: The index of the given variable associated with the tracing_map_elt
112 *
113 * Retrieve the value of the variable i associated with the specified
114 * tracing_map_elt instance, and reset the variable to the 'not set'
115 * state. The index i is the index returned by the call to
116 * tracing_map_add_var() when the tracing map was set up. The reset
117 * essentially makes the variable a read-once variable if it's only
118 * accessed using this function.
119 *
120 * Return: The variable value associated with field i for elt.
121 */
122u64 tracing_map_read_var_once(struct tracing_map_elt *elt, unsigned int i)
123{
124 elt->var_set[i] = false;
125 return (u64)atomic64_read(&elt->vars[i]);
126}
127
128int tracing_map_cmp_string(void *val_a, void *val_b)
129{
130 char *a = val_a;
131 char *b = val_b;
132
133 return strcmp(a, b);
134}
135
136int tracing_map_cmp_none(void *val_a, void *val_b)
137{
138 return 0;
139}
140
141static int tracing_map_cmp_atomic64(void *val_a, void *val_b)
142{
143 u64 a = atomic64_read((atomic64_t *)val_a);
144 u64 b = atomic64_read((atomic64_t *)val_b);
145
146 return (a > b) ? 1 : ((a < b) ? -1 : 0);
147}
148
149#define DEFINE_TRACING_MAP_CMP_FN(type) \
150static int tracing_map_cmp_##type(void *val_a, void *val_b) \
151{ \
152 type a = (type)(*(u64 *)val_a); \
153 type b = (type)(*(u64 *)val_b); \
154 \
155 return (a > b) ? 1 : ((a < b) ? -1 : 0); \
156}
157
158DEFINE_TRACING_MAP_CMP_FN(s64);
159DEFINE_TRACING_MAP_CMP_FN(u64);
160DEFINE_TRACING_MAP_CMP_FN(s32);
161DEFINE_TRACING_MAP_CMP_FN(u32);
162DEFINE_TRACING_MAP_CMP_FN(s16);
163DEFINE_TRACING_MAP_CMP_FN(u16);
164DEFINE_TRACING_MAP_CMP_FN(s8);
165DEFINE_TRACING_MAP_CMP_FN(u8);
166
167tracing_map_cmp_fn_t tracing_map_cmp_num(int field_size,
168 int field_is_signed)
169{
170 tracing_map_cmp_fn_t fn = tracing_map_cmp_none;
171
172 switch (field_size) {
173 case 8:
174 if (field_is_signed)
175 fn = tracing_map_cmp_s64;
176 else
177 fn = tracing_map_cmp_u64;
178 break;
179 case 4:
180 if (field_is_signed)
181 fn = tracing_map_cmp_s32;
182 else
183 fn = tracing_map_cmp_u32;
184 break;
185 case 2:
186 if (field_is_signed)
187 fn = tracing_map_cmp_s16;
188 else
189 fn = tracing_map_cmp_u16;
190 break;
191 case 1:
192 if (field_is_signed)
193 fn = tracing_map_cmp_s8;
194 else
195 fn = tracing_map_cmp_u8;
196 break;
197 }
198
199 return fn;
200}
201
202static int tracing_map_add_field(struct tracing_map *map,
203 tracing_map_cmp_fn_t cmp_fn)
204{
205 int ret = -EINVAL;
206
207 if (map->n_fields < TRACING_MAP_FIELDS_MAX) {
208 ret = map->n_fields;
209 map->fields[map->n_fields++].cmp_fn = cmp_fn;
210 }
211
212 return ret;
213}
214
215/**
216 * tracing_map_add_sum_field - Add a field describing a tracing_map sum
217 * @map: The tracing_map
218 *
219 * Add a sum field to the key and return the index identifying it in
220 * the map and associated tracing_map_elts. This is the index used
221 * for instance to update a sum for a particular tracing_map_elt using
222 * tracing_map_update_sum() or reading it via tracing_map_read_sum().
223 *
224 * Return: The index identifying the field in the map and associated
225 * tracing_map_elts, or -EINVAL on error.
226 */
227int tracing_map_add_sum_field(struct tracing_map *map)
228{
229 return tracing_map_add_field(map, tracing_map_cmp_atomic64);
230}
231
232/**
233 * tracing_map_add_var - Add a field describing a tracing_map var
234 * @map: The tracing_map
235 *
236 * Add a var to the map and return the index identifying it in the map
237 * and associated tracing_map_elts. This is the index used for
238 * instance to update a var for a particular tracing_map_elt using
239 * tracing_map_update_var() or reading it via tracing_map_read_var().
240 *
241 * Return: The index identifying the var in the map and associated
242 * tracing_map_elts, or -EINVAL on error.
243 */
244int tracing_map_add_var(struct tracing_map *map)
245{
246 int ret = -EINVAL;
247
248 if (map->n_vars < TRACING_MAP_VARS_MAX)
249 ret = map->n_vars++;
250
251 return ret;
252}
253
254/**
255 * tracing_map_add_key_field - Add a field describing a tracing_map key
256 * @map: The tracing_map
257 * @offset: The offset within the key
258 * @cmp_fn: The comparison function that will be used to sort on the key
259 *
260 * Let the map know there is a key and that if it's used as a sort key
261 * to use cmp_fn.
262 *
263 * A key can be a subset of a compound key; for that purpose, the
264 * offset param is used to describe where within the compound key
265 * the key referenced by this key field resides.
266 *
267 * Return: The index identifying the field in the map and associated
268 * tracing_map_elts, or -EINVAL on error.
269 */
270int tracing_map_add_key_field(struct tracing_map *map,
271 unsigned int offset,
272 tracing_map_cmp_fn_t cmp_fn)
273
274{
275 int idx = tracing_map_add_field(map, cmp_fn);
276
277 if (idx < 0)
278 return idx;
279
280 map->fields[idx].offset = offset;
281
282 map->key_idx[map->n_keys++] = idx;
283
284 return idx;
285}
286
287static void tracing_map_array_clear(struct tracing_map_array *a)
288{
289 unsigned int i;
290
291 for (i = 0; i < a->n_pages; i++)
292 memset(a->pages[i], 0, PAGE_SIZE);
293}
294
295static void tracing_map_array_free(struct tracing_map_array *a)
296{
297 unsigned int i;
298
299 if (!a)
300 return;
301
302 for (i = 0; i < a->n_pages; i++) {
303 if (!a->pages[i])
304 break;
305 kmemleak_free(a->pages[i]);
306 free_page((unsigned long)a->pages[i]);
307 }
308
309 kfree(a);
310}
311
312static struct tracing_map_array *tracing_map_array_alloc(unsigned int n_elts,
313 unsigned int entry_size)
314{
315 struct tracing_map_array *a;
316 unsigned int entry_size_shift;
317 unsigned int entries_per_page;
318 unsigned int n_pages;
319 unsigned int i;
320
321 entry_size_shift = fls(roundup_pow_of_two(entry_size) - 1);
322 entries_per_page = PAGE_SIZE / (1 << entry_size_shift);
323 n_pages = max(1, n_elts / entries_per_page);
324
325 a = kzalloc_flex(*a, pages, n_pages);
326 if (!a)
327 return NULL;
328
329 a->entry_size_shift = entry_size_shift;
330 a->entries_per_page = entries_per_page;
331 a->n_pages = n_pages;
332 a->entry_shift = fls(a->entries_per_page) - 1;
333 a->entry_mask = (1 << a->entry_shift) - 1;
334
335 for (i = 0; i < a->n_pages; i++) {
336 a->pages[i] = (void *)get_zeroed_page(GFP_KERNEL);
337 if (!a->pages[i])
338 goto free;
339 kmemleak_alloc(a->pages[i], PAGE_SIZE, 1, GFP_KERNEL);
340 }
341 out:
342 return a;
343 free:
344 tracing_map_array_free(a);
345 a = NULL;
346
347 goto out;
348}
349
350static void tracing_map_elt_clear(struct tracing_map_elt *elt)
351{
352 unsigned i;
353
354 for (i = 0; i < elt->map->n_fields; i++)
355 if (elt->fields[i].cmp_fn == tracing_map_cmp_atomic64)
356 atomic64_set(&elt->fields[i].sum, 0);
357
358 for (i = 0; i < elt->map->n_vars; i++) {
359 atomic64_set(&elt->vars[i], 0);
360 elt->var_set[i] = false;
361 }
362
363 if (elt->map->ops && elt->map->ops->elt_clear)
364 elt->map->ops->elt_clear(elt);
365}
366
367static void tracing_map_elt_init_fields(struct tracing_map_elt *elt)
368{
369 unsigned int i;
370
371 tracing_map_elt_clear(elt);
372
373 for (i = 0; i < elt->map->n_fields; i++) {
374 elt->fields[i].cmp_fn = elt->map->fields[i].cmp_fn;
375
376 if (elt->fields[i].cmp_fn != tracing_map_cmp_atomic64)
377 elt->fields[i].offset = elt->map->fields[i].offset;
378 }
379}
380
381static void __tracing_map_elt_free(struct tracing_map_elt *elt)
382{
383 if (!elt)
384 return;
385
386 kfree(elt->fields);
387 kfree(elt->vars);
388 kfree(elt->var_set);
389 kfree(elt->key);
390 kfree(elt);
391}
392
393static void tracing_map_elt_free(struct tracing_map_elt *elt)
394{
395 if (!elt)
396 return;
397
398 /* Only objects initialized with alloc_elt() should be passed to free_elt().*/
399 if (elt->map->ops && elt->map->ops->elt_free)
400 elt->map->ops->elt_free(elt);
401 __tracing_map_elt_free(elt);
402}
403
404static struct tracing_map_elt *tracing_map_elt_alloc(struct tracing_map *map)
405{
406 struct tracing_map_elt *elt;
407 int err = 0;
408
409 elt = kzalloc_obj(*elt);
410 if (!elt)
411 return ERR_PTR(-ENOMEM);
412
413 elt->map = map;
414
415 elt->key = kzalloc(map->key_size, GFP_KERNEL);
416 if (!elt->key) {
417 err = -ENOMEM;
418 goto free;
419 }
420
421 elt->fields = kzalloc_objs(*elt->fields, map->n_fields);
422 if (!elt->fields) {
423 err = -ENOMEM;
424 goto free;
425 }
426
427 elt->vars = kzalloc_objs(*elt->vars, map->n_vars);
428 if (!elt->vars) {
429 err = -ENOMEM;
430 goto free;
431 }
432
433 elt->var_set = kzalloc_objs(*elt->var_set, map->n_vars);
434 if (!elt->var_set) {
435 err = -ENOMEM;
436 goto free;
437 }
438
439 tracing_map_elt_init_fields(elt);
440
441 if (map->ops && map->ops->elt_alloc) {
442 err = map->ops->elt_alloc(elt);
443 if (err)
444 goto free;
445 }
446 return elt;
447 free:
448 __tracing_map_elt_free(elt);
449
450 return ERR_PTR(err);
451}
452
453static struct tracing_map_elt *get_free_elt(struct tracing_map *map)
454{
455 struct tracing_map_elt *elt = NULL;
456 int idx;
457
458 idx = atomic_fetch_add_unless(&map->next_elt, 1, map->max_elts);
459 if (idx < map->max_elts) {
460 elt = *(TRACING_MAP_ELT(map->elts, idx));
461 if (map->ops && map->ops->elt_init)
462 map->ops->elt_init(elt);
463 }
464
465 return elt;
466}
467
468static void tracing_map_free_elts(struct tracing_map *map)
469{
470 unsigned int i;
471
472 if (!map->elts)
473 return;
474
475 for (i = 0; i < map->max_elts; i++) {
476 tracing_map_elt_free(*(TRACING_MAP_ELT(map->elts, i)));
477 *(TRACING_MAP_ELT(map->elts, i)) = NULL;
478 }
479
480 tracing_map_array_free(map->elts);
481 map->elts = NULL;
482}
483
484static int tracing_map_alloc_elts(struct tracing_map *map)
485{
486 unsigned int i;
487
488 map->elts = tracing_map_array_alloc(map->max_elts,
489 sizeof(struct tracing_map_elt *));
490 if (!map->elts)
491 return -ENOMEM;
492
493 for (i = 0; i < map->max_elts; i++) {
494 *(TRACING_MAP_ELT(map->elts, i)) = tracing_map_elt_alloc(map);
495 if (IS_ERR(*(TRACING_MAP_ELT(map->elts, i)))) {
496 *(TRACING_MAP_ELT(map->elts, i)) = NULL;
497 tracing_map_free_elts(map);
498
499 return -ENOMEM;
500 }
501 }
502
503 return 0;
504}
505
506static inline bool keys_match(void *key, void *test_key, unsigned key_size)
507{
508 bool match = true;
509
510 if (memcmp(key, test_key, key_size))
511 match = false;
512
513 return match;
514}
515
516static inline struct tracing_map_elt *
517__tracing_map_insert(struct tracing_map *map, void *key, bool lookup_only)
518{
519 u32 idx, key_hash, test_key;
520 int dup_try = 0;
521 struct tracing_map_entry *entry;
522 struct tracing_map_elt *val;
523
524 key_hash = jhash(key, map->key_size, 0);
525 if (key_hash == 0)
526 key_hash = 1;
527 idx = key_hash >> (32 - (map->map_bits + 1));
528
529 while (1) {
530 idx &= (map->map_size - 1);
531 entry = TRACING_MAP_ENTRY(map->map, idx);
532 test_key = entry->key;
533
534 if (test_key && test_key == key_hash) {
535 val = READ_ONCE(entry->val);
536 if (val &&
537 keys_match(key, val->key, map->key_size)) {
538 if (!lookup_only)
539 atomic64_inc(&map->hits);
540 return val;
541 } else if (unlikely(!val)) {
542 /*
543 * The key is present. But, val (pointer to elt
544 * struct) is still NULL. which means some other
545 * thread is in the process of inserting an
546 * element.
547 *
548 * On top of that, it's key_hash is same as the
549 * one being inserted right now. So, it's
550 * possible that the element has the same
551 * key as well.
552 */
553
554 dup_try++;
555 if (dup_try > map->map_size) {
556 atomic64_inc(&map->drops);
557 break;
558 }
559 continue;
560 }
561 }
562
563 if (!test_key) {
564 if (lookup_only)
565 break;
566
567 if (!cmpxchg(&entry->key, 0, key_hash)) {
568 struct tracing_map_elt *elt;
569
570 elt = get_free_elt(map);
571 if (!elt) {
572 atomic64_inc(&map->drops);
573 entry->key = 0;
574 break;
575 }
576
577 memcpy(elt->key, key, map->key_size);
578 /*
579 * Ensure the initialization is visible and
580 * publish the elt.
581 */
582 smp_wmb();
583 WRITE_ONCE(entry->val, elt);
584 atomic64_inc(&map->hits);
585
586 return entry->val;
587 } else {
588 /*
589 * cmpxchg() failed. Loop around once
590 * more to check what key was inserted.
591 */
592 dup_try++;
593 continue;
594 }
595 }
596
597 idx++;
598 }
599
600 return NULL;
601}
602
603/**
604 * tracing_map_insert - Insert key and/or retrieve val from a tracing_map
605 * @map: The tracing_map to insert into
606 * @key: The key to insert
607 *
608 * Inserts a key into a tracing_map and creates and returns a new
609 * tracing_map_elt for it, or if the key has already been inserted by
610 * a previous call, returns the tracing_map_elt already associated
611 * with it. When the map was created, the number of elements to be
612 * allocated for the map was specified (internally maintained as
613 * 'max_elts' in struct tracing_map), and that number of
614 * tracing_map_elts was created by tracing_map_init(). This is the
615 * pre-allocated pool of tracing_map_elts that tracing_map_insert()
616 * will allocate from when adding new keys. Once that pool is
617 * exhausted, tracing_map_insert() is useless and will return NULL to
618 * signal that state. There are two user-visible tracing_map
619 * variables, 'hits' and 'drops', which are updated by this function.
620 * Every time an element is either successfully inserted or retrieved,
621 * the 'hits' value is incremented. Every time an element insertion
622 * fails, the 'drops' value is incremented.
623 *
624 * This is a lock-free tracing map insertion function implementing a
625 * modified form of Cliff Click's basic insertion algorithm. It
626 * requires the table size be a power of two. To prevent any
627 * possibility of an infinite loop we always make the internal table
628 * size double the size of the requested table size (max_elts * 2).
629 * Likewise, we never reuse a slot or resize or delete elements - when
630 * we've reached max_elts entries, we simply return NULL once we've
631 * run out of entries. Readers can at any point in time traverse the
632 * tracing map and safely access the key/val pairs.
633 *
634 * Return: the tracing_map_elt pointer val associated with the key.
635 * If this was a newly inserted key, the val will be a newly allocated
636 * and associated tracing_map_elt pointer val. If the key wasn't
637 * found and the pool of tracing_map_elts has been exhausted, NULL is
638 * returned and no further insertions will succeed.
639 */
640struct tracing_map_elt *tracing_map_insert(struct tracing_map *map, void *key)
641{
642 return __tracing_map_insert(map, key, false);
643}
644
645/**
646 * tracing_map_lookup - Retrieve val from a tracing_map
647 * @map: The tracing_map to perform the lookup on
648 * @key: The key to look up
649 *
650 * Looks up key in tracing_map and if found returns the matching
651 * tracing_map_elt. This is a lock-free lookup; see
652 * tracing_map_insert() for details on tracing_map and how it works.
653 * Every time an element is retrieved, the 'hits' value is
654 * incremented. There is one user-visible tracing_map variable,
655 * 'hits', which is updated by this function. Every time an element
656 * is successfully retrieved, the 'hits' value is incremented. The
657 * 'drops' value is never updated by this function.
658 *
659 * Return: the tracing_map_elt pointer val associated with the key.
660 * If the key wasn't found, NULL is returned.
661 */
662struct tracing_map_elt *tracing_map_lookup(struct tracing_map *map, void *key)
663{
664 return __tracing_map_insert(map, key, true);
665}
666
667/**
668 * tracing_map_destroy - Destroy a tracing_map
669 * @map: The tracing_map to destroy
670 *
671 * Frees a tracing_map along with its associated array of
672 * tracing_map_elts.
673 *
674 * Callers should make sure there are no readers or writers actively
675 * reading or inserting into the map before calling this.
676 */
677void tracing_map_destroy(struct tracing_map *map)
678{
679 if (!map)
680 return;
681
682 tracing_map_free_elts(map);
683
684 tracing_map_array_free(map->map);
685 kfree(map);
686}
687
688/**
689 * tracing_map_clear - Clear a tracing_map
690 * @map: The tracing_map to clear
691 *
692 * Resets the tracing map to a cleared or initial state. The
693 * tracing_map_elts are all cleared, and the array of struct
694 * tracing_map_entry is reset to an initialized state.
695 *
696 * Callers should make sure there are no writers actively inserting
697 * into the map before calling this.
698 */
699void tracing_map_clear(struct tracing_map *map)
700{
701 unsigned int i;
702
703 atomic_set(&map->next_elt, 0);
704 atomic64_set(&map->hits, 0);
705 atomic64_set(&map->drops, 0);
706
707 tracing_map_array_clear(map->map);
708
709 for (i = 0; i < map->max_elts; i++)
710 tracing_map_elt_clear(*(TRACING_MAP_ELT(map->elts, i)));
711}
712
713static void set_sort_key(struct tracing_map *map,
714 struct tracing_map_sort_key *sort_key)
715{
716 map->sort_key = *sort_key;
717}
718
719/**
720 * tracing_map_create - Create a lock-free map and element pool
721 * @map_bits: The size of the map (2 ** map_bits)
722 * @key_size: The size of the key for the map in bytes
723 * @ops: Optional client-defined tracing_map_ops instance
724 * @private_data: Client data associated with the map
725 *
726 * Creates and sets up a map to contain 2 ** map_bits number of
727 * elements (internally maintained as 'max_elts' in struct
728 * tracing_map). Before using, map fields should be added to the map
729 * with tracing_map_add_sum_field() and tracing_map_add_key_field().
730 * tracing_map_init() should then be called to allocate the array of
731 * tracing_map_elts, in order to avoid allocating anything in the map
732 * insertion path. The user-specified map size reflects the maximum
733 * number of elements that can be contained in the table requested by
734 * the user - internally we double that in order to keep the table
735 * sparse and keep collisions manageable.
736 *
737 * A tracing_map is a special-purpose map designed to aggregate or
738 * 'sum' one or more values associated with a specific object of type
739 * tracing_map_elt, which is attached by the map to a given key.
740 *
741 * tracing_map_create() sets up the map itself, and provides
742 * operations for inserting tracing_map_elts, but doesn't allocate the
743 * tracing_map_elts themselves, or provide a means for describing the
744 * keys or sums associated with the tracing_map_elts. All
745 * tracing_map_elts for a given map have the same set of sums and
746 * keys, which are defined by the client using the functions
747 * tracing_map_add_key_field() and tracing_map_add_sum_field(). Once
748 * the fields are defined, the pool of elements allocated for the map
749 * can be created, which occurs when the client code calls
750 * tracing_map_init().
751 *
752 * When tracing_map_init() returns, tracing_map_elt elements can be
753 * inserted into the map using tracing_map_insert(). When called,
754 * tracing_map_insert() grabs a free tracing_map_elt from the pool, or
755 * finds an existing match in the map and in either case returns it.
756 * The client can then use tracing_map_update_sum() and
757 * tracing_map_read_sum() to update or read a given sum field for the
758 * tracing_map_elt.
759 *
760 * The client can at any point retrieve and traverse the current set
761 * of inserted tracing_map_elts in a tracing_map, via
762 * tracing_map_sort_entries(). Sorting can be done on any field,
763 * including keys.
764 *
765 * See tracing_map.h for a description of tracing_map_ops.
766 *
767 * Return: the tracing_map pointer if successful, ERR_PTR if not.
768 */
769struct tracing_map *tracing_map_create(unsigned int map_bits,
770 unsigned int key_size,
771 const struct tracing_map_ops *ops,
772 void *private_data)
773{
774 struct tracing_map *map;
775 unsigned int i;
776
777 if (map_bits < TRACING_MAP_BITS_MIN ||
778 map_bits > TRACING_MAP_BITS_MAX)
779 return ERR_PTR(-EINVAL);
780
781 map = kzalloc_obj(*map);
782 if (!map)
783 return ERR_PTR(-ENOMEM);
784
785 map->map_bits = map_bits;
786 map->max_elts = (1 << map_bits);
787 atomic_set(&map->next_elt, 0);
788
789 map->map_size = (1 << (map_bits + 1));
790 map->ops = ops;
791
792 map->private_data = private_data;
793
794 map->map = tracing_map_array_alloc(map->map_size,
795 sizeof(struct tracing_map_entry));
796 if (!map->map)
797 goto free;
798
799 map->key_size = key_size;
800 for (i = 0; i < TRACING_MAP_KEYS_MAX; i++)
801 map->key_idx[i] = -1;
802 out:
803 return map;
804 free:
805 tracing_map_destroy(map);
806 map = ERR_PTR(-ENOMEM);
807
808 goto out;
809}
810
811/**
812 * tracing_map_init - Allocate and clear a map's tracing_map_elts
813 * @map: The tracing_map to initialize
814 *
815 * Allocates a clears a pool of tracing_map_elts equal to the
816 * user-specified size of 2 ** map_bits (internally maintained as
817 * 'max_elts' in struct tracing_map). Before using, the map fields
818 * should be added to the map with tracing_map_add_sum_field() and
819 * tracing_map_add_key_field(). tracing_map_init() should then be
820 * called to allocate the array of tracing_map_elts, in order to avoid
821 * allocating anything in the map insertion path. The user-specified
822 * map size reflects the max number of elements requested by the user
823 * - internally we double that in order to keep the table sparse and
824 * keep collisions manageable.
825 *
826 * See tracing_map.h for a description of tracing_map_ops.
827 *
828 * Return: the tracing_map pointer if successful, ERR_PTR if not.
829 */
830int tracing_map_init(struct tracing_map *map)
831{
832 int err;
833
834 if (map->n_fields < 2)
835 return -EINVAL; /* need at least 1 key and 1 val */
836
837 err = tracing_map_alloc_elts(map);
838 if (err)
839 return err;
840
841 tracing_map_clear(map);
842
843 return err;
844}
845
846static int cmp_entries_dup(const void *A, const void *B)
847{
848 const struct tracing_map_sort_entry *a, *b;
849
850 a = *(const struct tracing_map_sort_entry **)A;
851 b = *(const struct tracing_map_sort_entry **)B;
852
853 return memcmp(a->key, b->key, a->elt->map->key_size);
854}
855
856static int cmp_entries_sum(const void *A, const void *B)
857{
858 const struct tracing_map_elt *elt_a, *elt_b;
859 const struct tracing_map_sort_entry *a, *b;
860 struct tracing_map_sort_key *sort_key;
861 struct tracing_map_field *field;
862 tracing_map_cmp_fn_t cmp_fn;
863 void *val_a, *val_b;
864 int ret = 0;
865
866 a = *(const struct tracing_map_sort_entry **)A;
867 b = *(const struct tracing_map_sort_entry **)B;
868
869 elt_a = a->elt;
870 elt_b = b->elt;
871
872 sort_key = &elt_a->map->sort_key;
873
874 field = &elt_a->fields[sort_key->field_idx];
875 cmp_fn = field->cmp_fn;
876
877 val_a = &elt_a->fields[sort_key->field_idx].sum;
878 val_b = &elt_b->fields[sort_key->field_idx].sum;
879
880 ret = cmp_fn(val_a, val_b);
881 if (sort_key->descending)
882 ret = -ret;
883
884 return ret;
885}
886
887static int cmp_entries_key(const void *A, const void *B)
888{
889 const struct tracing_map_elt *elt_a, *elt_b;
890 const struct tracing_map_sort_entry *a, *b;
891 struct tracing_map_sort_key *sort_key;
892 struct tracing_map_field *field;
893 tracing_map_cmp_fn_t cmp_fn;
894 void *val_a, *val_b;
895 int ret = 0;
896
897 a = *(const struct tracing_map_sort_entry **)A;
898 b = *(const struct tracing_map_sort_entry **)B;
899
900 elt_a = a->elt;
901 elt_b = b->elt;
902
903 sort_key = &elt_a->map->sort_key;
904
905 field = &elt_a->fields[sort_key->field_idx];
906
907 cmp_fn = field->cmp_fn;
908
909 val_a = elt_a->key + field->offset;
910 val_b = elt_b->key + field->offset;
911
912 ret = cmp_fn(val_a, val_b);
913 if (sort_key->descending)
914 ret = -ret;
915
916 return ret;
917}
918
919static void destroy_sort_entry(struct tracing_map_sort_entry *entry)
920{
921 if (!entry)
922 return;
923
924 if (entry->elt_copied)
925 tracing_map_elt_free(entry->elt);
926
927 kfree(entry);
928}
929
930/**
931 * tracing_map_destroy_sort_entries - Destroy an array of sort entries
932 * @entries: The entries to destroy
933 * @n_entries: The number of entries in the array
934 *
935 * Destroy the elements returned by a tracing_map_sort_entries() call.
936 */
937void tracing_map_destroy_sort_entries(struct tracing_map_sort_entry **entries,
938 unsigned int n_entries)
939{
940 unsigned int i;
941
942 for (i = 0; i < n_entries; i++)
943 destroy_sort_entry(entries[i]);
944
945 vfree(entries);
946}
947
948static struct tracing_map_sort_entry *
949create_sort_entry(void *key, struct tracing_map_elt *elt)
950{
951 struct tracing_map_sort_entry *sort_entry;
952
953 sort_entry = kzalloc_obj(*sort_entry);
954 if (!sort_entry)
955 return NULL;
956
957 sort_entry->key = key;
958 sort_entry->elt = elt;
959
960 return sort_entry;
961}
962
963static void detect_dups(struct tracing_map_sort_entry **sort_entries,
964 int n_entries, unsigned int key_size)
965{
966 unsigned int total_dups = 0;
967 int i;
968 void *key;
969
970 if (n_entries < 2)
971 return;
972
973 sort(sort_entries, n_entries, sizeof(struct tracing_map_sort_entry *),
974 (int (*)(const void *, const void *))cmp_entries_dup, NULL);
975
976 key = sort_entries[0]->key;
977 for (i = 1; i < n_entries; i++) {
978 if (!memcmp(sort_entries[i]->key, key, key_size)) {
979 total_dups++;
980 continue;
981 }
982 key = sort_entries[i]->key;
983 }
984
985 WARN_ONCE(total_dups > 0,
986 "Duplicates detected: %d\n", total_dups);
987}
988
989static bool is_key(struct tracing_map *map, unsigned int field_idx)
990{
991 unsigned int i;
992
993 for (i = 0; i < map->n_keys; i++)
994 if (map->key_idx[i] == field_idx)
995 return true;
996 return false;
997}
998
999static void sort_secondary(struct tracing_map *map,
1000 const struct tracing_map_sort_entry **entries,
1001 unsigned int n_entries,
1002 struct tracing_map_sort_key *primary_key,
1003 struct tracing_map_sort_key *secondary_key)
1004{
1005 int (*primary_fn)(const void *, const void *);
1006 int (*secondary_fn)(const void *, const void *);
1007 unsigned i, start = 0, n_sub = 1;
1008
1009 if (is_key(map, primary_key->field_idx))
1010 primary_fn = cmp_entries_key;
1011 else
1012 primary_fn = cmp_entries_sum;
1013
1014 if (is_key(map, secondary_key->field_idx))
1015 secondary_fn = cmp_entries_key;
1016 else
1017 secondary_fn = cmp_entries_sum;
1018
1019 for (i = 0; i < n_entries - 1; i++) {
1020 const struct tracing_map_sort_entry **a = &entries[i];
1021 const struct tracing_map_sort_entry **b = &entries[i + 1];
1022
1023 if (primary_fn(a, b) == 0) {
1024 n_sub++;
1025 if (i < n_entries - 2)
1026 continue;
1027 }
1028
1029 if (n_sub < 2) {
1030 start = i + 1;
1031 n_sub = 1;
1032 continue;
1033 }
1034
1035 set_sort_key(map, secondary_key);
1036 sort(&entries[start], n_sub,
1037 sizeof(struct tracing_map_sort_entry *),
1038 (int (*)(const void *, const void *))secondary_fn, NULL);
1039 set_sort_key(map, primary_key);
1040
1041 start = i + 1;
1042 n_sub = 1;
1043 }
1044}
1045
1046/**
1047 * tracing_map_sort_entries - Sort the current set of tracing_map_elts in a map
1048 * @map: The tracing_map
1049 * @sort_keys: The sort key to use for sorting
1050 * @n_sort_keys: hitcount, always have at least one
1051 * @sort_entries: outval: pointer to allocated and sorted array of entries
1052 *
1053 * tracing_map_sort_entries() sorts the current set of entries in the
1054 * map and returns the list of tracing_map_sort_entries containing
1055 * them to the client in the sort_entries param. The client can
1056 * access the struct tracing_map_elt element of interest directly as
1057 * the 'elt' field of a returned struct tracing_map_sort_entry object.
1058 *
1059 * The sort_key has only two fields: idx and descending. 'idx' refers
1060 * to the index of the field added via tracing_map_add_sum_field() or
1061 * tracing_map_add_key_field() when the tracing_map was initialized.
1062 * 'descending' is a flag that if set reverses the sort order, which
1063 * by default is ascending.
1064 *
1065 * The client should not hold on to the returned array but should use
1066 * it and call tracing_map_destroy_sort_entries() when done.
1067 *
1068 * Return: the number of sort_entries in the struct tracing_map_sort_entry
1069 * array, negative on error
1070 */
1071int tracing_map_sort_entries(struct tracing_map *map,
1072 struct tracing_map_sort_key *sort_keys,
1073 unsigned int n_sort_keys,
1074 struct tracing_map_sort_entry ***sort_entries)
1075{
1076 int (*cmp_entries_fn)(const void *, const void *);
1077 struct tracing_map_sort_entry *sort_entry, **entries;
1078 int i, n_entries, ret;
1079
1080 entries = vmalloc_array(map->max_elts, sizeof(sort_entry));
1081 if (!entries)
1082 return -ENOMEM;
1083
1084 for (i = 0, n_entries = 0; i < map->map_size; i++) {
1085 struct tracing_map_entry *entry;
1086
1087 entry = TRACING_MAP_ENTRY(map->map, i);
1088
1089 if (!entry->key || !entry->val)
1090 continue;
1091
1092 entries[n_entries] = create_sort_entry(entry->val->key,
1093 entry->val);
1094 if (!entries[n_entries++]) {
1095 ret = -ENOMEM;
1096 goto free;
1097 }
1098 }
1099
1100 if (n_entries == 0) {
1101 ret = 0;
1102 goto free;
1103 }
1104
1105 if (n_entries == 1) {
1106 *sort_entries = entries;
1107 return 1;
1108 }
1109
1110 detect_dups(entries, n_entries, map->key_size);
1111
1112 if (is_key(map, sort_keys[0].field_idx))
1113 cmp_entries_fn = cmp_entries_key;
1114 else
1115 cmp_entries_fn = cmp_entries_sum;
1116
1117 set_sort_key(map, &sort_keys[0]);
1118
1119 sort(entries, n_entries, sizeof(struct tracing_map_sort_entry *),
1120 (int (*)(const void *, const void *))cmp_entries_fn, NULL);
1121
1122 if (n_sort_keys > 1)
1123 sort_secondary(map,
1124 (const struct tracing_map_sort_entry **)entries,
1125 n_entries,
1126 &sort_keys[0],
1127 &sort_keys[1]);
1128
1129 *sort_entries = entries;
1130
1131 return n_entries;
1132 free:
1133 tracing_map_destroy_sort_entries(entries, n_entries);
1134
1135 return ret;
1136}