1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
|
From 8749642c432e7e3023df6495c97bc6065b1f72d3 Mon Sep 17 00:00:00 2001
From: Peter Zijlstra <a.p.zijlstra@chello.nl>
Date: Fri, 3 Jul 2009 08:44:43 -0500
Subject: [PATCH 092/274] mm: More lock breaks in slab.c
Handle __free_pages outside of the locked regions. This reduces the
lock contention on the percpu slab locks in -rt significantly.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
mm/slab.c | 90 ++++++++++++++++++++++++++++++++++++++++++++++---------------
1 file changed, 68 insertions(+), 22 deletions(-)
diff --git a/mm/slab.c b/mm/slab.c
index 7266165..cdc4dc3 100644
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -727,6 +727,7 @@ static void slab_set_debugobj_lock_classes(struct kmem_cache *cachep)
#endif
static DEFINE_PER_CPU(struct delayed_work, slab_reap_work);
+static DEFINE_PER_CPU(struct list_head, slab_free_list);
static DEFINE_LOCAL_IRQ_LOCK(slab_lock);
#ifndef CONFIG_PREEMPT_RT_BASE
@@ -742,14 +743,39 @@ slab_on_each_cpu(void (*func)(void *arg, int this_cpu), void *arg)
{
unsigned int i;
- for_each_online_cpu(i) {
- spin_lock_irq(&per_cpu(slab_lock, i).lock);
+ for_each_online_cpu(i)
func(arg, i);
- spin_unlock_irq(&per_cpu(slab_lock, i).lock);
- }
}
#endif
+static void free_delayed(struct list_head *h)
+{
+ while(!list_empty(h)) {
+ struct page *page = list_first_entry(h, struct page, lru);
+
+ list_del(&page->lru);
+ __free_pages(page, page->index);
+ }
+}
+
+static void unlock_l3_and_free_delayed(spinlock_t *list_lock)
+{
+ LIST_HEAD(tmp);
+
+ list_splice_init(&__get_cpu_var(slab_free_list), &tmp);
+ local_spin_unlock_irq(slab_lock, list_lock);
+ free_delayed(&tmp);
+}
+
+static void unlock_slab_and_free_delayed(unsigned long flags)
+{
+ LIST_HEAD(tmp);
+
+ list_splice_init(&__get_cpu_var(slab_free_list), &tmp);
+ local_unlock_irqrestore(slab_lock, flags);
+ free_delayed(&tmp);
+}
+
static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
{
return cachep->array[smp_processor_id()];
@@ -1245,7 +1271,7 @@ static void __cpuinit cpuup_canceled(long cpu)
free_block(cachep, nc->entry, nc->avail, node);
if (!cpumask_empty(mask)) {
- local_spin_unlock_irq(slab_lock, &l3->list_lock);
+ unlock_l3_and_free_delayed(&l3->list_lock);
goto free_array_cache;
}
@@ -1259,7 +1285,7 @@ static void __cpuinit cpuup_canceled(long cpu)
alien = l3->alien;
l3->alien = NULL;
- local_spin_unlock_irq(slab_lock, &l3->list_lock);
+ unlock_l3_and_free_delayed(&l3->list_lock);
kfree(shared);
if (alien) {
@@ -1540,6 +1566,8 @@ void __init kmem_cache_init(void)
use_alien_caches = 0;
local_irq_lock_init(slab_lock);
+ for_each_possible_cpu(i)
+ INIT_LIST_HEAD(&per_cpu(slab_free_list, i));
for (i = 0; i < NUM_INIT_LISTS; i++) {
kmem_list3_init(&initkmem_list3[i]);
@@ -1871,12 +1899,14 @@ static void *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, int nodeid)
/*
* Interface to system's page release.
*/
-static void kmem_freepages(struct kmem_cache *cachep, void *addr)
+static void kmem_freepages(struct kmem_cache *cachep, void *addr, bool delayed)
{
unsigned long i = (1 << cachep->gfporder);
- struct page *page = virt_to_page(addr);
+ struct page *page, *basepage = virt_to_page(addr);
const unsigned long nr_freed = i;
+ page = basepage;
+
kmemcheck_free_shadow(page, cachep->gfporder);
if (cachep->flags & SLAB_RECLAIM_ACCOUNT)
@@ -1892,7 +1922,13 @@ static void kmem_freepages(struct kmem_cache *cachep, void *addr)
}
if (current->reclaim_state)
current->reclaim_state->reclaimed_slab += nr_freed;
- free_pages((unsigned long)addr, cachep->gfporder);
+
+ if (!delayed) {
+ free_pages((unsigned long)addr, cachep->gfporder);
+ } else {
+ basepage->index = cachep->gfporder;
+ list_add(&basepage->lru, &__get_cpu_var(slab_free_list));
+ }
}
static void kmem_rcu_free(struct rcu_head *head)
@@ -1900,7 +1936,7 @@ static void kmem_rcu_free(struct rcu_head *head)
struct slab_rcu *slab_rcu = (struct slab_rcu *)head;
struct kmem_cache *cachep = slab_rcu->cachep;
- kmem_freepages(cachep, slab_rcu->addr);
+ kmem_freepages(cachep, slab_rcu->addr, false);
if (OFF_SLAB(cachep))
kmem_cache_free(cachep->slabp_cache, slab_rcu);
}
@@ -2119,7 +2155,8 @@ static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slab
* Before calling the slab must have been unlinked from the cache. The
* cache-lock is not held/needed.
*/
-static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp)
+static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp,
+ bool delayed)
{
void *addr = slabp->s_mem - slabp->colouroff;
@@ -2132,7 +2169,7 @@ static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp)
slab_rcu->addr = addr;
call_rcu(&slab_rcu->head, kmem_rcu_free);
} else {
- kmem_freepages(cachep, addr);
+ kmem_freepages(cachep, addr, delayed);
if (OFF_SLAB(cachep))
kmem_cache_free(cachep->slabp_cache, slabp);
}
@@ -2654,9 +2691,15 @@ static void do_drain(void *arg)
__do_drain(arg, smp_processor_id());
}
#else
-static void do_drain(void *arg, int this_cpu)
+static void do_drain(void *arg, int cpu)
{
- __do_drain(arg, this_cpu);
+ LIST_HEAD(tmp);
+
+ spin_lock_irq(&per_cpu(slab_lock, cpu).lock);
+ __do_drain(arg, cpu);
+ list_splice_init(&per_cpu(slab_free_list, cpu), &tmp);
+ spin_unlock_irq(&per_cpu(slab_lock, cpu).lock);
+ free_delayed(&tmp);
}
#endif
@@ -2714,7 +2757,7 @@ static int drain_freelist(struct kmem_cache *cache,
*/
l3->free_objects -= cache->num;
local_spin_unlock_irq(slab_lock, &l3->list_lock);
- slab_destroy(cache, slabp);
+ slab_destroy(cache, slabp, false);
nr_freed++;
}
out:
@@ -3049,7 +3092,7 @@ static int cache_grow(struct kmem_cache *cachep,
spin_unlock(&l3->list_lock);
return 1;
opps1:
- kmem_freepages(cachep, objp);
+ kmem_freepages(cachep, objp, false);
failed:
if (local_flags & __GFP_WAIT)
local_lock_irq(slab_lock);
@@ -3703,7 +3746,7 @@ static void free_block(struct kmem_cache *cachep, void **objpp, int nr_objects,
* a different cache, refer to comments before
* alloc_slabmgmt.
*/
- slab_destroy(cachep, slabp);
+ slab_destroy(cachep, slabp, true);
} else {
list_add(&slabp->list, &l3->slabs_free);
}
@@ -3970,7 +4013,7 @@ void kmem_cache_free(struct kmem_cache *cachep, void *objp)
debug_check_no_obj_freed(objp, obj_size(cachep));
local_lock_irqsave(slab_lock, flags);
__cache_free(cachep, objp, __builtin_return_address(0));
- local_unlock_irqrestore(slab_lock, flags);
+ unlock_slab_and_free_delayed(flags);
trace_kmem_cache_free(_RET_IP_, objp);
}
@@ -4000,7 +4043,7 @@ void kfree(const void *objp)
debug_check_no_obj_freed(objp, obj_size(c));
local_lock_irqsave(slab_lock, flags);
__cache_free(c, (void *)objp, __builtin_return_address(0));
- local_unlock_irqrestore(slab_lock, flags);
+ unlock_slab_and_free_delayed(flags);
}
EXPORT_SYMBOL(kfree);
@@ -4056,7 +4099,8 @@ static int alloc_kmemlist(struct kmem_cache *cachep, gfp_t gfp)
}
l3->free_limit = (1 + nr_cpus_node(node)) *
cachep->batchcount + cachep->num;
- local_spin_unlock_irq(slab_lock, &l3->list_lock);
+ unlock_l3_and_free_delayed(&l3->list_lock);
+
kfree(shared);
free_alien_cache(new_alien);
continue;
@@ -4122,7 +4166,9 @@ static void do_ccupdate_local(void *info)
#else
static void do_ccupdate_local(void *info, int cpu)
{
+ spin_lock_irq(&per_cpu(slab_lock, cpu).lock);
__do_ccupdate_local(info, cpu);
+ spin_unlock_irq(&per_cpu(slab_lock, cpu).lock);
}
#endif
@@ -4164,8 +4210,8 @@ static int do_tune_cpucache(struct kmem_cache *cachep, int limit,
local_spin_lock_irq(slab_lock,
&cachep->nodelists[cpu_to_mem(i)]->list_lock);
free_block(cachep, ccold->entry, ccold->avail, cpu_to_mem(i));
- local_spin_unlock_irq(slab_lock,
- &cachep->nodelists[cpu_to_mem(i)]->list_lock);
+
+ unlock_l3_and_free_delayed(&cachep->nodelists[cpu_to_mem(i)]->list_lock);
kfree(ccold);
}
kfree(new);
--
1.7.10.4
|
