diff options
author | Tejun Heo <tj@kernel.org> | 2009-07-21 08:18:35 -0400 |
---|---|---|
committer | Tejun Heo <tj@kernel.org> | 2009-10-02 00:29:29 -0400 |
commit | 23fb064bb96f001ecb8682129f7ee1bc1ca691bc (patch) | |
tree | ae9173b25aa69cda1b974c630334ffb61cee7ebe | |
parent | 52594762a39dfb6338c9d0906ca21dd9ae9453be (diff) |
percpu: kill legacy percpu allocator
With ia64 converted, there's no arch left which still uses legacy
percpu allocator. Kill it.
Signed-off-by: Tejun Heo <tj@kernel.org>
Delightedly-acked-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
-rw-r--r-- | include/linux/percpu.h | 24 | ||||
-rw-r--r-- | kernel/module.c | 150 | ||||
-rw-r--r-- | mm/Makefile | 4 | ||||
-rw-r--r-- | mm/allocpercpu.c | 177 | ||||
-rw-r--r-- | mm/percpu.c | 2 |
5 files changed, 0 insertions, 357 deletions
diff --git a/include/linux/percpu.h b/include/linux/percpu.h index 878836ca999c..5baf5b8788fb 100644 --- a/include/linux/percpu.h +++ b/include/linux/percpu.h | |||
@@ -34,8 +34,6 @@ | |||
34 | 34 | ||
35 | #ifdef CONFIG_SMP | 35 | #ifdef CONFIG_SMP |
36 | 36 | ||
37 | #ifndef CONFIG_HAVE_LEGACY_PER_CPU_AREA | ||
38 | |||
39 | /* minimum unit size, also is the maximum supported allocation size */ | 37 | /* minimum unit size, also is the maximum supported allocation size */ |
40 | #define PCPU_MIN_UNIT_SIZE PFN_ALIGN(64 << 10) | 38 | #define PCPU_MIN_UNIT_SIZE PFN_ALIGN(64 << 10) |
41 | 39 | ||
@@ -130,28 +128,6 @@ extern int __init pcpu_page_first_chunk(size_t reserved_size, | |||
130 | #define per_cpu_ptr(ptr, cpu) SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu))) | 128 | #define per_cpu_ptr(ptr, cpu) SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu))) |
131 | 129 | ||
132 | extern void *__alloc_reserved_percpu(size_t size, size_t align); | 130 | extern void *__alloc_reserved_percpu(size_t size, size_t align); |
133 | |||
134 | #else /* CONFIG_HAVE_LEGACY_PER_CPU_AREA */ | ||
135 | |||
136 | struct percpu_data { | ||
137 | void *ptrs[1]; | ||
138 | }; | ||
139 | |||
140 | /* pointer disguising messes up the kmemleak objects tracking */ | ||
141 | #ifndef CONFIG_DEBUG_KMEMLEAK | ||
142 | #define __percpu_disguise(pdata) (struct percpu_data *)~(unsigned long)(pdata) | ||
143 | #else | ||
144 | #define __percpu_disguise(pdata) (struct percpu_data *)(pdata) | ||
145 | #endif | ||
146 | |||
147 | #define per_cpu_ptr(ptr, cpu) \ | ||
148 | ({ \ | ||
149 | struct percpu_data *__p = __percpu_disguise(ptr); \ | ||
150 | (__typeof__(ptr))__p->ptrs[(cpu)]; \ | ||
151 | }) | ||
152 | |||
153 | #endif /* CONFIG_HAVE_LEGACY_PER_CPU_AREA */ | ||
154 | |||
155 | extern void *__alloc_percpu(size_t size, size_t align); | 131 | extern void *__alloc_percpu(size_t size, size_t align); |
156 | extern void free_percpu(void *__pdata); | 132 | extern void free_percpu(void *__pdata); |
157 | 133 | ||
diff --git a/kernel/module.c b/kernel/module.c index 8b7d8805819d..64787cddeb5e 100644 --- a/kernel/module.c +++ b/kernel/module.c | |||
@@ -370,8 +370,6 @@ EXPORT_SYMBOL_GPL(find_module); | |||
370 | 370 | ||
371 | #ifdef CONFIG_SMP | 371 | #ifdef CONFIG_SMP |
372 | 372 | ||
373 | #ifndef CONFIG_HAVE_LEGACY_PER_CPU_AREA | ||
374 | |||
375 | static void *percpu_modalloc(unsigned long size, unsigned long align, | 373 | static void *percpu_modalloc(unsigned long size, unsigned long align, |
376 | const char *name) | 374 | const char *name) |
377 | { | 375 | { |
@@ -395,154 +393,6 @@ static void percpu_modfree(void *freeme) | |||
395 | free_percpu(freeme); | 393 | free_percpu(freeme); |
396 | } | 394 | } |
397 | 395 | ||
398 | #else /* ... CONFIG_HAVE_LEGACY_PER_CPU_AREA */ | ||
399 | |||
400 | /* Number of blocks used and allocated. */ | ||
401 | static unsigned int pcpu_num_used, pcpu_num_allocated; | ||
402 | /* Size of each block. -ve means used. */ | ||
403 | static int *pcpu_size; | ||
404 | |||
405 | static int split_block(unsigned int i, unsigned short size) | ||
406 | { | ||
407 | /* Reallocation required? */ | ||
408 | if (pcpu_num_used + 1 > pcpu_num_allocated) { | ||
409 | int *new; | ||
410 | |||
411 | new = krealloc(pcpu_size, sizeof(new[0])*pcpu_num_allocated*2, | ||
412 | GFP_KERNEL); | ||
413 | if (!new) | ||
414 | return 0; | ||
415 | |||
416 | pcpu_num_allocated *= 2; | ||
417 | pcpu_size = new; | ||
418 | } | ||
419 | |||
420 | /* Insert a new subblock */ | ||
421 | memmove(&pcpu_size[i+1], &pcpu_size[i], | ||
422 | sizeof(pcpu_size[0]) * (pcpu_num_used - i)); | ||
423 | pcpu_num_used++; | ||
424 | |||
425 | pcpu_size[i+1] -= size; | ||
426 | pcpu_size[i] = size; | ||
427 | return 1; | ||
428 | } | ||
429 | |||
430 | static inline unsigned int block_size(int val) | ||
431 | { | ||
432 | if (val < 0) | ||
433 | return -val; | ||
434 | return val; | ||
435 | } | ||
436 | |||
437 | static void *percpu_modalloc(unsigned long size, unsigned long align, | ||
438 | const char *name) | ||
439 | { | ||
440 | unsigned long extra; | ||
441 | unsigned int i; | ||
442 | void *ptr; | ||
443 | int cpu; | ||
444 | |||
445 | if (align > PAGE_SIZE) { | ||
446 | printk(KERN_WARNING "%s: per-cpu alignment %li > %li\n", | ||
447 | name, align, PAGE_SIZE); | ||
448 | align = PAGE_SIZE; | ||
449 | } | ||
450 | |||
451 | ptr = __per_cpu_start; | ||
452 | for (i = 0; i < pcpu_num_used; ptr += block_size(pcpu_size[i]), i++) { | ||
453 | /* Extra for alignment requirement. */ | ||
454 | extra = ALIGN((unsigned long)ptr, align) - (unsigned long)ptr; | ||
455 | BUG_ON(i == 0 && extra != 0); | ||
456 | |||
457 | if (pcpu_size[i] < 0 || pcpu_size[i] < extra + size) | ||
458 | continue; | ||
459 | |||
460 | /* Transfer extra to previous block. */ | ||
461 | if (pcpu_size[i-1] < 0) | ||
462 | pcpu_size[i-1] -= extra; | ||
463 | else | ||
464 | pcpu_size[i-1] += extra; | ||
465 | pcpu_size[i] -= extra; | ||
466 | ptr += extra; | ||
467 | |||
468 | /* Split block if warranted */ | ||
469 | if (pcpu_size[i] - size > sizeof(unsigned long)) | ||
470 | if (!split_block(i, size)) | ||
471 | return NULL; | ||
472 | |||
473 | /* add the per-cpu scanning areas */ | ||
474 | for_each_possible_cpu(cpu) | ||
475 | kmemleak_alloc(ptr + per_cpu_offset(cpu), size, 0, | ||
476 | GFP_KERNEL); | ||
477 | |||
478 | /* Mark allocated */ | ||
479 | pcpu_size[i] = -pcpu_size[i]; | ||
480 | return ptr; | ||
481 | } | ||
482 | |||
483 | printk(KERN_WARNING "Could not allocate %lu bytes percpu data\n", | ||
484 | size); | ||
485 | return NULL; | ||
486 | } | ||
487 | |||
488 | static void percpu_modfree(void *freeme) | ||
489 | { | ||
490 | unsigned int i; | ||
491 | void *ptr = __per_cpu_start + block_size(pcpu_size[0]); | ||
492 | int cpu; | ||
493 | |||
494 | /* First entry is core kernel percpu data. */ | ||
495 | for (i = 1; i < pcpu_num_used; ptr += block_size(pcpu_size[i]), i++) { | ||
496 | if (ptr == freeme) { | ||
497 | pcpu_size[i] = -pcpu_size[i]; | ||
498 | goto free; | ||
499 | } | ||
500 | } | ||
501 | BUG(); | ||
502 | |||
503 | free: | ||
504 | /* remove the per-cpu scanning areas */ | ||
505 | for_each_possible_cpu(cpu) | ||
506 | kmemleak_free(freeme + per_cpu_offset(cpu)); | ||
507 | |||
508 | /* Merge with previous? */ | ||
509 | if (pcpu_size[i-1] >= 0) { | ||
510 | pcpu_size[i-1] += pcpu_size[i]; | ||
511 | pcpu_num_used--; | ||
512 | memmove(&pcpu_size[i], &pcpu_size[i+1], | ||
513 | (pcpu_num_used - i) * sizeof(pcpu_size[0])); | ||
514 | i--; | ||
515 | } | ||
516 | /* Merge with next? */ | ||
517 | if (i+1 < pcpu_num_used && pcpu_size[i+1] >= 0) { | ||
518 | pcpu_size[i] += pcpu_size[i+1]; | ||
519 | pcpu_num_used--; | ||
520 | memmove(&pcpu_size[i+1], &pcpu_size[i+2], | ||
521 | (pcpu_num_used - (i+1)) * sizeof(pcpu_size[0])); | ||
522 | } | ||
523 | } | ||
524 | |||
525 | static int percpu_modinit(void) | ||
526 | { | ||
527 | pcpu_num_used = 2; | ||
528 | pcpu_num_allocated = 2; | ||
529 | pcpu_size = kmalloc(sizeof(pcpu_size[0]) * pcpu_num_allocated, | ||
530 | GFP_KERNEL); | ||
531 | /* Static in-kernel percpu data (used). */ | ||
532 | pcpu_size[0] = -(__per_cpu_end-__per_cpu_start); | ||
533 | /* Free room. */ | ||
534 | pcpu_size[1] = PERCPU_ENOUGH_ROOM + pcpu_size[0]; | ||
535 | if (pcpu_size[1] < 0) { | ||
536 | printk(KERN_ERR "No per-cpu room for modules.\n"); | ||
537 | pcpu_num_used = 1; | ||
538 | } | ||
539 | |||
540 | return 0; | ||
541 | } | ||
542 | __initcall(percpu_modinit); | ||
543 | |||
544 | #endif /* CONFIG_HAVE_LEGACY_PER_CPU_AREA */ | ||
545 | |||
546 | static unsigned int find_pcpusec(Elf_Ehdr *hdr, | 396 | static unsigned int find_pcpusec(Elf_Ehdr *hdr, |
547 | Elf_Shdr *sechdrs, | 397 | Elf_Shdr *sechdrs, |
548 | const char *secstrings) | 398 | const char *secstrings) |
diff --git a/mm/Makefile b/mm/Makefile index ebf849042ed3..82131d0f8d85 100644 --- a/mm/Makefile +++ b/mm/Makefile | |||
@@ -34,11 +34,7 @@ obj-$(CONFIG_FAILSLAB) += failslab.o | |||
34 | obj-$(CONFIG_MEMORY_HOTPLUG) += memory_hotplug.o | 34 | obj-$(CONFIG_MEMORY_HOTPLUG) += memory_hotplug.o |
35 | obj-$(CONFIG_FS_XIP) += filemap_xip.o | 35 | obj-$(CONFIG_FS_XIP) += filemap_xip.o |
36 | obj-$(CONFIG_MIGRATION) += migrate.o | 36 | obj-$(CONFIG_MIGRATION) += migrate.o |
37 | ifndef CONFIG_HAVE_LEGACY_PER_CPU_AREA | ||
38 | obj-$(CONFIG_SMP) += percpu.o | 37 | obj-$(CONFIG_SMP) += percpu.o |
39 | else | ||
40 | obj-$(CONFIG_SMP) += allocpercpu.o | ||
41 | endif | ||
42 | obj-$(CONFIG_QUICKLIST) += quicklist.o | 38 | obj-$(CONFIG_QUICKLIST) += quicklist.o |
43 | obj-$(CONFIG_CGROUP_MEM_RES_CTLR) += memcontrol.o page_cgroup.o | 39 | obj-$(CONFIG_CGROUP_MEM_RES_CTLR) += memcontrol.o page_cgroup.o |
44 | obj-$(CONFIG_MEMORY_FAILURE) += memory-failure.o | 40 | obj-$(CONFIG_MEMORY_FAILURE) += memory-failure.o |
diff --git a/mm/allocpercpu.c b/mm/allocpercpu.c deleted file mode 100644 index df34ceae0c67..000000000000 --- a/mm/allocpercpu.c +++ /dev/null | |||
@@ -1,177 +0,0 @@ | |||
1 | /* | ||
2 | * linux/mm/allocpercpu.c | ||
3 | * | ||
4 | * Separated from slab.c August 11, 2006 Christoph Lameter | ||
5 | */ | ||
6 | #include <linux/mm.h> | ||
7 | #include <linux/module.h> | ||
8 | #include <linux/bootmem.h> | ||
9 | #include <asm/sections.h> | ||
10 | |||
11 | #ifndef cache_line_size | ||
12 | #define cache_line_size() L1_CACHE_BYTES | ||
13 | #endif | ||
14 | |||
15 | /** | ||
16 | * percpu_depopulate - depopulate per-cpu data for given cpu | ||
17 | * @__pdata: per-cpu data to depopulate | ||
18 | * @cpu: depopulate per-cpu data for this cpu | ||
19 | * | ||
20 | * Depopulating per-cpu data for a cpu going offline would be a typical | ||
21 | * use case. You need to register a cpu hotplug handler for that purpose. | ||
22 | */ | ||
23 | static void percpu_depopulate(void *__pdata, int cpu) | ||
24 | { | ||
25 | struct percpu_data *pdata = __percpu_disguise(__pdata); | ||
26 | |||
27 | kfree(pdata->ptrs[cpu]); | ||
28 | pdata->ptrs[cpu] = NULL; | ||
29 | } | ||
30 | |||
31 | /** | ||
32 | * percpu_depopulate_mask - depopulate per-cpu data for some cpu's | ||
33 | * @__pdata: per-cpu data to depopulate | ||
34 | * @mask: depopulate per-cpu data for cpu's selected through mask bits | ||
35 | */ | ||
36 | static void __percpu_depopulate_mask(void *__pdata, const cpumask_t *mask) | ||
37 | { | ||
38 | int cpu; | ||
39 | for_each_cpu_mask_nr(cpu, *mask) | ||
40 | percpu_depopulate(__pdata, cpu); | ||
41 | } | ||
42 | |||
43 | #define percpu_depopulate_mask(__pdata, mask) \ | ||
44 | __percpu_depopulate_mask((__pdata), &(mask)) | ||
45 | |||
46 | /** | ||
47 | * percpu_populate - populate per-cpu data for given cpu | ||
48 | * @__pdata: per-cpu data to populate further | ||
49 | * @size: size of per-cpu object | ||
50 | * @gfp: may sleep or not etc. | ||
51 | * @cpu: populate per-data for this cpu | ||
52 | * | ||
53 | * Populating per-cpu data for a cpu coming online would be a typical | ||
54 | * use case. You need to register a cpu hotplug handler for that purpose. | ||
55 | * Per-cpu object is populated with zeroed buffer. | ||
56 | */ | ||
57 | static void *percpu_populate(void *__pdata, size_t size, gfp_t gfp, int cpu) | ||
58 | { | ||
59 | struct percpu_data *pdata = __percpu_disguise(__pdata); | ||
60 | int node = cpu_to_node(cpu); | ||
61 | |||
62 | /* | ||
63 | * We should make sure each CPU gets private memory. | ||
64 | */ | ||
65 | size = roundup(size, cache_line_size()); | ||
66 | |||
67 | BUG_ON(pdata->ptrs[cpu]); | ||
68 | if (node_online(node)) | ||
69 | pdata->ptrs[cpu] = kmalloc_node(size, gfp|__GFP_ZERO, node); | ||
70 | else | ||
71 | pdata->ptrs[cpu] = kzalloc(size, gfp); | ||
72 | return pdata->ptrs[cpu]; | ||
73 | } | ||
74 | |||
75 | /** | ||
76 | * percpu_populate_mask - populate per-cpu data for more cpu's | ||
77 | * @__pdata: per-cpu data to populate further | ||
78 | * @size: size of per-cpu object | ||
79 | * @gfp: may sleep or not etc. | ||
80 | * @mask: populate per-cpu data for cpu's selected through mask bits | ||
81 | * | ||
82 | * Per-cpu objects are populated with zeroed buffers. | ||
83 | */ | ||
84 | static int __percpu_populate_mask(void *__pdata, size_t size, gfp_t gfp, | ||
85 | cpumask_t *mask) | ||
86 | { | ||
87 | cpumask_t populated; | ||
88 | int cpu; | ||
89 | |||
90 | cpus_clear(populated); | ||
91 | for_each_cpu_mask_nr(cpu, *mask) | ||
92 | if (unlikely(!percpu_populate(__pdata, size, gfp, cpu))) { | ||
93 | __percpu_depopulate_mask(__pdata, &populated); | ||
94 | return -ENOMEM; | ||
95 | } else | ||
96 | cpu_set(cpu, populated); | ||
97 | return 0; | ||
98 | } | ||
99 | |||
100 | #define percpu_populate_mask(__pdata, size, gfp, mask) \ | ||
101 | __percpu_populate_mask((__pdata), (size), (gfp), &(mask)) | ||
102 | |||
103 | /** | ||
104 | * alloc_percpu - initial setup of per-cpu data | ||
105 | * @size: size of per-cpu object | ||
106 | * @align: alignment | ||
107 | * | ||
108 | * Allocate dynamic percpu area. Percpu objects are populated with | ||
109 | * zeroed buffers. | ||
110 | */ | ||
111 | void *__alloc_percpu(size_t size, size_t align) | ||
112 | { | ||
113 | /* | ||
114 | * We allocate whole cache lines to avoid false sharing | ||
115 | */ | ||
116 | size_t sz = roundup(nr_cpu_ids * sizeof(void *), cache_line_size()); | ||
117 | void *pdata = kzalloc(sz, GFP_KERNEL); | ||
118 | void *__pdata = __percpu_disguise(pdata); | ||
119 | |||
120 | /* | ||
121 | * Can't easily make larger alignment work with kmalloc. WARN | ||
122 | * on it. Larger alignment should only be used for module | ||
123 | * percpu sections on SMP for which this path isn't used. | ||
124 | */ | ||
125 | WARN_ON_ONCE(align > SMP_CACHE_BYTES); | ||
126 | |||
127 | if (unlikely(!pdata)) | ||
128 | return NULL; | ||
129 | if (likely(!__percpu_populate_mask(__pdata, size, GFP_KERNEL, | ||
130 | &cpu_possible_map))) | ||
131 | return __pdata; | ||
132 | kfree(pdata); | ||
133 | return NULL; | ||
134 | } | ||
135 | EXPORT_SYMBOL_GPL(__alloc_percpu); | ||
136 | |||
137 | /** | ||
138 | * free_percpu - final cleanup of per-cpu data | ||
139 | * @__pdata: object to clean up | ||
140 | * | ||
141 | * We simply clean up any per-cpu object left. No need for the client to | ||
142 | * track and specify through a bis mask which per-cpu objects are to free. | ||
143 | */ | ||
144 | void free_percpu(void *__pdata) | ||
145 | { | ||
146 | if (unlikely(!__pdata)) | ||
147 | return; | ||
148 | __percpu_depopulate_mask(__pdata, cpu_possible_mask); | ||
149 | kfree(__percpu_disguise(__pdata)); | ||
150 | } | ||
151 | EXPORT_SYMBOL_GPL(free_percpu); | ||
152 | |||
153 | /* | ||
154 | * Generic percpu area setup. | ||
155 | */ | ||
156 | #ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA | ||
157 | unsigned long __per_cpu_offset[NR_CPUS] __read_mostly; | ||
158 | |||
159 | EXPORT_SYMBOL(__per_cpu_offset); | ||
160 | |||
161 | void __init setup_per_cpu_areas(void) | ||
162 | { | ||
163 | unsigned long size, i; | ||
164 | char *ptr; | ||
165 | unsigned long nr_possible_cpus = num_possible_cpus(); | ||
166 | |||
167 | /* Copy section for each CPU (we discard the original) */ | ||
168 | size = ALIGN(PERCPU_ENOUGH_ROOM, PAGE_SIZE); | ||
169 | ptr = alloc_bootmem_pages(size * nr_possible_cpus); | ||
170 | |||
171 | for_each_possible_cpu(i) { | ||
172 | __per_cpu_offset[i] = ptr - __per_cpu_start; | ||
173 | memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start); | ||
174 | ptr += size; | ||
175 | } | ||
176 | } | ||
177 | #endif /* CONFIG_HAVE_SETUP_PER_CPU_AREA */ | ||
diff --git a/mm/percpu.c b/mm/percpu.c index 4a048abad043..e4e08b87b77e 100644 --- a/mm/percpu.c +++ b/mm/percpu.c | |||
@@ -46,8 +46,6 @@ | |||
46 | * | 46 | * |
47 | * To use this allocator, arch code should do the followings. | 47 | * To use this allocator, arch code should do the followings. |
48 | * | 48 | * |
49 | * - drop CONFIG_HAVE_LEGACY_PER_CPU_AREA | ||
50 | * | ||
51 | * - define __addr_to_pcpu_ptr() and __pcpu_ptr_to_addr() to translate | 49 | * - define __addr_to_pcpu_ptr() and __pcpu_ptr_to_addr() to translate |
52 | * regular address to percpu pointer and back if they need to be | 50 | * regular address to percpu pointer and back if they need to be |
53 | * different from the default | 51 | * different from the default |