Skip to content

Commit 4fa8d68

Browse files
kkdwivediAlexei Starovoitov
kkdwivedi
authored and
Alexei Starovoitov
committed
bpf: Convert hashtab.c to rqspinlock
Convert hashtab.c from raw_spinlock to rqspinlock, and drop the hashed per-cpu counter crud from the code base which is no longer necessary. Closes: https://lore.kernel.org/bpf/675302fd.050a0220.2477f.0004.GAE@google.com Closes: https://lore.kernel.org/bpf/000000000000b3e63e061eed3f6b@google.com Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com> Link: https://lore.kernel.org/r/20250316040541.108729-20-memxor@gmail.com Signed-off-by: Alexei Starovoitov <ast@kernel.org>
1 parent a6884f6 commit 4fa8d68

File tree

1 file changed

+32
-70
lines changed

1 file changed

+32
-70
lines changed

‎kernel/bpf/hashtab.c

Lines changed: 32 additions & 70 deletions
Original file line numberDiff line numberDiff line change
@@ -16,6 +16,7 @@
1616
#include "bpf_lru_list.h"
1717
#include "map_in_map.h"
1818
#include <linux/bpf_mem_alloc.h>
19+
#include <asm/rqspinlock.h>
1920

2021
#define HTAB_CREATE_FLAG_MASK \
2122
(BPF_F_NO_PREALLOC | BPF_F_NO_COMMON_LRU | BPF_F_NUMA_NODE | \
@@ -78,7 +79,7 @@
7879
*/
7980
struct bucket {
8081
struct hlist_nulls_head head;
81-
raw_spinlock_t raw_lock;
82+
rqspinlock_t raw_lock;
8283
};
8384

8485
#define HASHTAB_MAP_LOCK_COUNT 8
@@ -104,8 +105,6 @@ struct bpf_htab {
104105
u32 n_buckets; /* number of hash buckets */
105106
u32 elem_size; /* size of each element in bytes */
106107
u32 hashrnd;
107-
struct lock_class_key lockdep_key;
108-
int __percpu *map_locked[HASHTAB_MAP_LOCK_COUNT];
109108
};
110109

111110
/* each htab element is struct htab_elem + key + value */
@@ -140,45 +139,26 @@ static void htab_init_buckets(struct bpf_htab *htab)
140139

141140
for (i = 0; i < htab->n_buckets; i++) {
142141
INIT_HLIST_NULLS_HEAD(&htab->buckets[i].head, i);
143-
raw_spin_lock_init(&htab->buckets[i].raw_lock);
144-
lockdep_set_class(&htab->buckets[i].raw_lock,
145-
&htab->lockdep_key);
142+
raw_res_spin_lock_init(&htab->buckets[i].raw_lock);
146143
cond_resched();
147144
}
148145
}
149146

150-
static inline int htab_lock_bucket(const struct bpf_htab *htab,
151-
struct bucket *b, u32 hash,
152-
unsigned long *pflags)
147+
static inline int htab_lock_bucket(struct bucket *b, unsigned long *pflags)
153148
{
154149
unsigned long flags;
150+
int ret;
155151

156-
hash = hash & min_t(u32, HASHTAB_MAP_LOCK_MASK, htab->n_buckets - 1);
157-
158-
preempt_disable();
159-
local_irq_save(flags);
160-
if (unlikely(__this_cpu_inc_return(*(htab->map_locked[hash])) != 1)) {
161-
__this_cpu_dec(*(htab->map_locked[hash]));
162-
local_irq_restore(flags);
163-
preempt_enable();
164-
return -EBUSY;
165-
}
166-
167-
raw_spin_lock(&b->raw_lock);
152+
ret = raw_res_spin_lock_irqsave(&b->raw_lock, flags);
153+
if (ret)
154+
return ret;
168155
*pflags = flags;
169-
170156
return 0;
171157
}
172158

173-
static inline void htab_unlock_bucket(const struct bpf_htab *htab,
174-
struct bucket *b, u32 hash,
175-
unsigned long flags)
159+
static inline void htab_unlock_bucket(struct bucket *b, unsigned long flags)
176160
{
177-
hash = hash & min_t(u32, HASHTAB_MAP_LOCK_MASK, htab->n_buckets - 1);
178-
raw_spin_unlock(&b->raw_lock);
179-
__this_cpu_dec(*(htab->map_locked[hash]));
180-
local_irq_restore(flags);
181-
preempt_enable();
161+
raw_res_spin_unlock_irqrestore(&b->raw_lock, flags);
182162
}
183163

184164
static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node);
@@ -483,14 +463,12 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
483463
bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
484464
bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
485465
struct bpf_htab *htab;
486-
int err, i;
466+
int err;
487467

488468
htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
489469
if (!htab)
490470
return ERR_PTR(-ENOMEM);
491471

492-
lockdep_register_key(&htab->lockdep_key);
493-
494472
bpf_map_init_from_attr(&htab->map, attr);
495473

496474
if (percpu_lru) {
@@ -536,15 +514,6 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
536514
if (!htab->buckets)
537515
goto free_elem_count;
538516

539-
for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++) {
540-
htab->map_locked[i] = bpf_map_alloc_percpu(&htab->map,
541-
sizeof(int),
542-
sizeof(int),
543-
GFP_USER);
544-
if (!htab->map_locked[i])
545-
goto free_map_locked;
546-
}
547-
548517
if (htab->map.map_flags & BPF_F_ZERO_SEED)
549518
htab->hashrnd = 0;
550519
else
@@ -607,15 +576,12 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
607576
free_map_locked:
608577
if (htab->use_percpu_counter)
609578
percpu_counter_destroy(&htab->pcount);
610-
for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++)
611-
free_percpu(htab->map_locked[i]);
612579
bpf_map_area_free(htab->buckets);
613580
bpf_mem_alloc_destroy(&htab->pcpu_ma);
614581
bpf_mem_alloc_destroy(&htab->ma);
615582
free_elem_count:
616583
bpf_map_free_elem_count(&htab->map);
617584
free_htab:
618-
lockdep_unregister_key(&htab->lockdep_key);
619585
bpf_map_area_free(htab);
620586
return ERR_PTR(err);
621587
}
@@ -820,7 +786,7 @@ static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node)
820786
b = __select_bucket(htab, tgt_l->hash);
821787
head = &b->head;
822788

823-
ret = htab_lock_bucket(htab, b, tgt_l->hash, &flags);
789+
ret = htab_lock_bucket(b, &flags);
824790
if (ret)
825791
return false;
826792

@@ -831,7 +797,7 @@ static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node)
831797
break;
832798
}
833799

834-
htab_unlock_bucket(htab, b, tgt_l->hash, flags);
800+
htab_unlock_bucket(b, flags);
835801

836802
if (l == tgt_l)
837803
check_and_free_fields(htab, l);
@@ -1150,7 +1116,7 @@ static long htab_map_update_elem(struct bpf_map *map, void *key, void *value,
11501116
*/
11511117
}
11521118

1153-
ret = htab_lock_bucket(htab, b, hash, &flags);
1119+
ret = htab_lock_bucket(b, &flags);
11541120
if (ret)
11551121
return ret;
11561122

@@ -1201,7 +1167,7 @@ static long htab_map_update_elem(struct bpf_map *map, void *key, void *value,
12011167
check_and_free_fields(htab, l_old);
12021168
}
12031169
}
1204-
htab_unlock_bucket(htab, b, hash, flags);
1170+
htab_unlock_bucket(b, flags);
12051171
if (l_old) {
12061172
if (old_map_ptr)
12071173
map->ops->map_fd_put_ptr(map, old_map_ptr, true);
@@ -1210,7 +1176,7 @@ static long htab_map_update_elem(struct bpf_map *map, void *key, void *value,
12101176
}
12111177
return 0;
12121178
err:
1213-
htab_unlock_bucket(htab, b, hash, flags);
1179+
htab_unlock_bucket(b, flags);
12141180
return ret;
12151181
}
12161182

@@ -1257,7 +1223,7 @@ static long htab_lru_map_update_elem(struct bpf_map *map, void *key, void *value
12571223
copy_map_value(&htab->map,
12581224
l_new->key + round_up(map->key_size, 8), value);
12591225

1260-
ret = htab_lock_bucket(htab, b, hash, &flags);
1226+
ret = htab_lock_bucket(b, &flags);
12611227
if (ret)
12621228
goto err_lock_bucket;
12631229

@@ -1278,7 +1244,7 @@ static long htab_lru_map_update_elem(struct bpf_map *map, void *key, void *value
12781244
ret = 0;
12791245

12801246
err:
1281-
htab_unlock_bucket(htab, b, hash, flags);
1247+
htab_unlock_bucket(b, flags);
12821248

12831249
err_lock_bucket:
12841250
if (ret)
@@ -1315,7 +1281,7 @@ static long __htab_percpu_map_update_elem(struct bpf_map *map, void *key,
13151281
b = __select_bucket(htab, hash);
13161282
head = &b->head;
13171283

1318-
ret = htab_lock_bucket(htab, b, hash, &flags);
1284+
ret = htab_lock_bucket(b, &flags);
13191285
if (ret)
13201286
return ret;
13211287

@@ -1340,7 +1306,7 @@ static long __htab_percpu_map_update_elem(struct bpf_map *map, void *key,
13401306
}
13411307
ret = 0;
13421308
err:
1343-
htab_unlock_bucket(htab, b, hash, flags);
1309+
htab_unlock_bucket(b, flags);
13441310
return ret;
13451311
}
13461312

@@ -1381,7 +1347,7 @@ static long __htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
13811347
return -ENOMEM;
13821348
}
13831349

1384-
ret = htab_lock_bucket(htab, b, hash, &flags);
1350+
ret = htab_lock_bucket(b, &flags);
13851351
if (ret)
13861352
goto err_lock_bucket;
13871353

@@ -1405,7 +1371,7 @@ static long __htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
14051371
}
14061372
ret = 0;
14071373
err:
1408-
htab_unlock_bucket(htab, b, hash, flags);
1374+
htab_unlock_bucket(b, flags);
14091375
err_lock_bucket:
14101376
if (l_new) {
14111377
bpf_map_dec_elem_count(&htab->map);
@@ -1447,7 +1413,7 @@ static long htab_map_delete_elem(struct bpf_map *map, void *key)
14471413
b = __select_bucket(htab, hash);
14481414
head = &b->head;
14491415

1450-
ret = htab_lock_bucket(htab, b, hash, &flags);
1416+
ret = htab_lock_bucket(b, &flags);
14511417
if (ret)
14521418
return ret;
14531419

@@ -1457,7 +1423,7 @@ static long htab_map_delete_elem(struct bpf_map *map, void *key)
14571423
else
14581424
ret = -ENOENT;
14591425

1460-
htab_unlock_bucket(htab, b, hash, flags);
1426+
htab_unlock_bucket(b, flags);
14611427

14621428
if (l)
14631429
free_htab_elem(htab, l);
@@ -1483,7 +1449,7 @@ static long htab_lru_map_delete_elem(struct bpf_map *map, void *key)
14831449
b = __select_bucket(htab, hash);
14841450
head = &b->head;
14851451

1486-
ret = htab_lock_bucket(htab, b, hash, &flags);
1452+
ret = htab_lock_bucket(b, &flags);
14871453
if (ret)
14881454
return ret;
14891455

@@ -1494,7 +1460,7 @@ static long htab_lru_map_delete_elem(struct bpf_map *map, void *key)
14941460
else
14951461
ret = -ENOENT;
14961462

1497-
htab_unlock_bucket(htab, b, hash, flags);
1463+
htab_unlock_bucket(b, flags);
14981464
if (l)
14991465
htab_lru_push_free(htab, l);
15001466
return ret;
@@ -1561,7 +1527,6 @@ static void htab_map_free_timers_and_wq(struct bpf_map *map)
15611527
static void htab_map_free(struct bpf_map *map)
15621528
{
15631529
struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1564-
int i;
15651530

15661531
/* bpf_free_used_maps() or close(map_fd) will trigger this map_free callback.
15671532
* bpf_free_used_maps() is called after bpf prog is no longer executing.
@@ -1586,9 +1551,6 @@ static void htab_map_free(struct bpf_map *map)
15861551
bpf_mem_alloc_destroy(&htab->ma);
15871552
if (htab->use_percpu_counter)
15881553
percpu_counter_destroy(&htab->pcount);
1589-
for (i = 0; i < HASHTAB_MAP_LOCK_COUNT; i++)
1590-
free_percpu(htab->map_locked[i]);
1591-
lockdep_unregister_key(&htab->lockdep_key);
15921554
bpf_map_area_free(htab);
15931555
}
15941556

@@ -1631,7 +1593,7 @@ static int __htab_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
16311593
b = __select_bucket(htab, hash);
16321594
head = &b->head;
16331595

1634-
ret = htab_lock_bucket(htab, b, hash, &bflags);
1596+
ret = htab_lock_bucket(b, &bflags);
16351597
if (ret)
16361598
return ret;
16371599

@@ -1668,7 +1630,7 @@ static int __htab_map_lookup_and_delete_elem(struct bpf_map *map, void *key,
16681630
hlist_nulls_del_rcu(&l->hash_node);
16691631

16701632
out_unlock:
1671-
htab_unlock_bucket(htab, b, hash, bflags);
1633+
htab_unlock_bucket(b, bflags);
16721634

16731635
if (l) {
16741636
if (is_lru_map)
@@ -1790,7 +1752,7 @@ __htab_map_lookup_and_delete_batch(struct bpf_map *map,
17901752
head = &b->head;
17911753
/* do not grab the lock unless need it (bucket_cnt > 0). */
17921754
if (locked) {
1793-
ret = htab_lock_bucket(htab, b, batch, &flags);
1755+
ret = htab_lock_bucket(b, &flags);
17941756
if (ret) {
17951757
rcu_read_unlock();
17961758
bpf_enable_instrumentation();
@@ -1813,7 +1775,7 @@ __htab_map_lookup_and_delete_batch(struct bpf_map *map,
18131775
/* Note that since bucket_cnt > 0 here, it is implicit
18141776
* that the locked was grabbed, so release it.
18151777
*/
1816-
htab_unlock_bucket(htab, b, batch, flags);
1778+
htab_unlock_bucket(b, flags);
18171779
rcu_read_unlock();
18181780
bpf_enable_instrumentation();
18191781
goto after_loop;
@@ -1824,7 +1786,7 @@ __htab_map_lookup_and_delete_batch(struct bpf_map *map,
18241786
/* Note that since bucket_cnt > 0 here, it is implicit
18251787
* that the locked was grabbed, so release it.
18261788
*/
1827-
htab_unlock_bucket(htab, b, batch, flags);
1789+
htab_unlock_bucket(b, flags);
18281790
rcu_read_unlock();
18291791
bpf_enable_instrumentation();
18301792
kvfree(keys);
@@ -1887,7 +1849,7 @@ __htab_map_lookup_and_delete_batch(struct bpf_map *map,
18871849
dst_val += value_size;
18881850
}
18891851

1890-
htab_unlock_bucket(htab, b, batch, flags);
1852+
htab_unlock_bucket(b, flags);
18911853
locked = false;
18921854

18931855
while (node_to_free) {

0 commit comments

Comments
 (0)