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