aboutsummaryrefslogtreecommitdiffstats
path: root/mm/vmalloc.c
diff options
context:
space:
mode:
Diffstat (limited to 'mm/vmalloc.c')
-rw-r--r--mm/vmalloc.c994
1 files changed, 858 insertions, 136 deletions
diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index bba06c41fc59..65ae576030da 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -8,6 +8,7 @@
8 * Numa awareness, Christoph Lameter, SGI, June 2005 8 * Numa awareness, Christoph Lameter, SGI, June 2005
9 */ 9 */
10 10
11#include <linux/vmalloc.h>
11#include <linux/mm.h> 12#include <linux/mm.h>
12#include <linux/module.h> 13#include <linux/module.h>
13#include <linux/highmem.h> 14#include <linux/highmem.h>
@@ -16,18 +17,18 @@
16#include <linux/interrupt.h> 17#include <linux/interrupt.h>
17#include <linux/seq_file.h> 18#include <linux/seq_file.h>
18#include <linux/debugobjects.h> 19#include <linux/debugobjects.h>
19#include <linux/vmalloc.h>
20#include <linux/kallsyms.h> 20#include <linux/kallsyms.h>
21#include <linux/list.h>
22#include <linux/rbtree.h>
23#include <linux/radix-tree.h>
24#include <linux/rcupdate.h>
21 25
26#include <asm/atomic.h>
22#include <asm/uaccess.h> 27#include <asm/uaccess.h>
23#include <asm/tlbflush.h> 28#include <asm/tlbflush.h>
24 29
25 30
26DEFINE_RWLOCK(vmlist_lock); 31/*** Page table manipulation functions ***/
27struct vm_struct *vmlist;
28
29static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
30 int node, void *caller);
31 32
32static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end) 33static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
33{ 34{
@@ -40,8 +41,7 @@ static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
40 } while (pte++, addr += PAGE_SIZE, addr != end); 41 } while (pte++, addr += PAGE_SIZE, addr != end);
41} 42}
42 43
43static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr, 44static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
44 unsigned long end)
45{ 45{
46 pmd_t *pmd; 46 pmd_t *pmd;
47 unsigned long next; 47 unsigned long next;
@@ -55,8 +55,7 @@ static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
55 } while (pmd++, addr = next, addr != end); 55 } while (pmd++, addr = next, addr != end);
56} 56}
57 57
58static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr, 58static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
59 unsigned long end)
60{ 59{
61 pud_t *pud; 60 pud_t *pud;
62 unsigned long next; 61 unsigned long next;
@@ -70,12 +69,10 @@ static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
70 } while (pud++, addr = next, addr != end); 69 } while (pud++, addr = next, addr != end);
71} 70}
72 71
73void unmap_kernel_range(unsigned long addr, unsigned long size) 72static void vunmap_page_range(unsigned long addr, unsigned long end)
74{ 73{
75 pgd_t *pgd; 74 pgd_t *pgd;
76 unsigned long next; 75 unsigned long next;
77 unsigned long start = addr;
78 unsigned long end = addr + size;
79 76
80 BUG_ON(addr >= end); 77 BUG_ON(addr >= end);
81 pgd = pgd_offset_k(addr); 78 pgd = pgd_offset_k(addr);
@@ -86,35 +83,36 @@ void unmap_kernel_range(unsigned long addr, unsigned long size)
86 continue; 83 continue;
87 vunmap_pud_range(pgd, addr, next); 84 vunmap_pud_range(pgd, addr, next);
88 } while (pgd++, addr = next, addr != end); 85 } while (pgd++, addr = next, addr != end);
89 flush_tlb_kernel_range(start, end);
90}
91
92static void unmap_vm_area(struct vm_struct *area)
93{
94 unmap_kernel_range((unsigned long)area->addr, area->size);
95} 86}
96 87
97static int vmap_pte_range(pmd_t *pmd, unsigned long addr, 88static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
98 unsigned long end, pgprot_t prot, struct page ***pages) 89 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
99{ 90{
100 pte_t *pte; 91 pte_t *pte;
101 92
93 /*
94 * nr is a running index into the array which helps higher level
95 * callers keep track of where we're up to.
96 */
97
102 pte = pte_alloc_kernel(pmd, addr); 98 pte = pte_alloc_kernel(pmd, addr);
103 if (!pte) 99 if (!pte)
104 return -ENOMEM; 100 return -ENOMEM;
105 do { 101 do {
106 struct page *page = **pages; 102 struct page *page = pages[*nr];
107 WARN_ON(!pte_none(*pte)); 103
108 if (!page) 104 if (WARN_ON(!pte_none(*pte)))
105 return -EBUSY;
106 if (WARN_ON(!page))
109 return -ENOMEM; 107 return -ENOMEM;
110 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot)); 108 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
111 (*pages)++; 109 (*nr)++;
112 } while (pte++, addr += PAGE_SIZE, addr != end); 110 } while (pte++, addr += PAGE_SIZE, addr != end);
113 return 0; 111 return 0;
114} 112}
115 113
116static inline int vmap_pmd_range(pud_t *pud, unsigned long addr, 114static int vmap_pmd_range(pud_t *pud, unsigned long addr,
117 unsigned long end, pgprot_t prot, struct page ***pages) 115 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
118{ 116{
119 pmd_t *pmd; 117 pmd_t *pmd;
120 unsigned long next; 118 unsigned long next;
@@ -124,14 +122,14 @@ static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
124 return -ENOMEM; 122 return -ENOMEM;
125 do { 123 do {
126 next = pmd_addr_end(addr, end); 124 next = pmd_addr_end(addr, end);
127 if (vmap_pte_range(pmd, addr, next, prot, pages)) 125 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
128 return -ENOMEM; 126 return -ENOMEM;
129 } while (pmd++, addr = next, addr != end); 127 } while (pmd++, addr = next, addr != end);
130 return 0; 128 return 0;
131} 129}
132 130
133static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr, 131static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
134 unsigned long end, pgprot_t prot, struct page ***pages) 132 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
135{ 133{
136 pud_t *pud; 134 pud_t *pud;
137 unsigned long next; 135 unsigned long next;
@@ -141,57 +139,78 @@ static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
141 return -ENOMEM; 139 return -ENOMEM;
142 do { 140 do {
143 next = pud_addr_end(addr, end); 141 next = pud_addr_end(addr, end);
144 if (vmap_pmd_range(pud, addr, next, prot, pages)) 142 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
145 return -ENOMEM; 143 return -ENOMEM;
146 } while (pud++, addr = next, addr != end); 144 } while (pud++, addr = next, addr != end);
147 return 0; 145 return 0;
148} 146}
149 147
150int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages) 148/*
149 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
150 * will have pfns corresponding to the "pages" array.
151 *
152 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
153 */
154static int vmap_page_range(unsigned long addr, unsigned long end,
155 pgprot_t prot, struct page **pages)
151{ 156{
152 pgd_t *pgd; 157 pgd_t *pgd;
153 unsigned long next; 158 unsigned long next;
154 unsigned long addr = (unsigned long) area->addr; 159 int err = 0;
155 unsigned long end = addr + area->size - PAGE_SIZE; 160 int nr = 0;
156 int err;
157 161
158 BUG_ON(addr >= end); 162 BUG_ON(addr >= end);
159 pgd = pgd_offset_k(addr); 163 pgd = pgd_offset_k(addr);
160 do { 164 do {
161 next = pgd_addr_end(addr, end); 165 next = pgd_addr_end(addr, end);
162 err = vmap_pud_range(pgd, addr, next, prot, pages); 166 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
163 if (err) 167 if (err)
164 break; 168 break;
165 } while (pgd++, addr = next, addr != end); 169 } while (pgd++, addr = next, addr != end);
166 flush_cache_vmap((unsigned long) area->addr, end); 170 flush_cache_vmap(addr, end);
167 return err; 171
172 if (unlikely(err))
173 return err;
174 return nr;
175}
176
177static inline int is_vmalloc_or_module_addr(const void *x)
178{
179 /*
180 * x86-64 and sparc64 put modules in a special place,
181 * and fall back on vmalloc() if that fails. Others
182 * just put it in the vmalloc space.
183 */
184#if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
185 unsigned long addr = (unsigned long)x;
186 if (addr >= MODULES_VADDR && addr < MODULES_END)
187 return 1;
188#endif
189 return is_vmalloc_addr(x);
168} 190}
169EXPORT_SYMBOL_GPL(map_vm_area);
170 191
171/* 192/*
172 * Map a vmalloc()-space virtual address to the physical page. 193 * Walk a vmap address to the struct page it maps.
173 */ 194 */
174struct page *vmalloc_to_page(const void *vmalloc_addr) 195struct page *vmalloc_to_page(const void *vmalloc_addr)
175{ 196{
176 unsigned long addr = (unsigned long) vmalloc_addr; 197 unsigned long addr = (unsigned long) vmalloc_addr;
177 struct page *page = NULL; 198 struct page *page = NULL;
178 pgd_t *pgd = pgd_offset_k(addr); 199 pgd_t *pgd = pgd_offset_k(addr);
179 pud_t *pud;
180 pmd_t *pmd;
181 pte_t *ptep, pte;
182 200
183 /* 201 /*
184 * XXX we might need to change this if we add VIRTUAL_BUG_ON for 202 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
185 * architectures that do not vmalloc module space 203 * architectures that do not vmalloc module space
186 */ 204 */
187 VIRTUAL_BUG_ON(!is_vmalloc_addr(vmalloc_addr) && 205 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
188 !is_module_address(addr));
189 206
190 if (!pgd_none(*pgd)) { 207 if (!pgd_none(*pgd)) {
191 pud = pud_offset(pgd, addr); 208 pud_t *pud = pud_offset(pgd, addr);
192 if (!pud_none(*pud)) { 209 if (!pud_none(*pud)) {
193 pmd = pmd_offset(pud, addr); 210 pmd_t *pmd = pmd_offset(pud, addr);
194 if (!pmd_none(*pmd)) { 211 if (!pmd_none(*pmd)) {
212 pte_t *ptep, pte;
213
195 ptep = pte_offset_map(pmd, addr); 214 ptep = pte_offset_map(pmd, addr);
196 pte = *ptep; 215 pte = *ptep;
197 if (pte_present(pte)) 216 if (pte_present(pte))
@@ -213,13 +232,751 @@ unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
213} 232}
214EXPORT_SYMBOL(vmalloc_to_pfn); 233EXPORT_SYMBOL(vmalloc_to_pfn);
215 234
216static struct vm_struct * 235
217__get_vm_area_node(unsigned long size, unsigned long flags, unsigned long start, 236/*** Global kva allocator ***/
218 unsigned long end, int node, gfp_t gfp_mask, void *caller) 237
238#define VM_LAZY_FREE 0x01
239#define VM_LAZY_FREEING 0x02
240#define VM_VM_AREA 0x04
241
242struct vmap_area {
243 unsigned long va_start;
244 unsigned long va_end;
245 unsigned long flags;
246 struct rb_node rb_node; /* address sorted rbtree */
247 struct list_head list; /* address sorted list */
248 struct list_head purge_list; /* "lazy purge" list */
249 void *private;
250 struct rcu_head rcu_head;
251};
252
253static DEFINE_SPINLOCK(vmap_area_lock);
254static struct rb_root vmap_area_root = RB_ROOT;
255static LIST_HEAD(vmap_area_list);
256
257static struct vmap_area *__find_vmap_area(unsigned long addr)
219{ 258{
220 struct vm_struct **p, *tmp, *area; 259 struct rb_node *n = vmap_area_root.rb_node;
221 unsigned long align = 1; 260
261 while (n) {
262 struct vmap_area *va;
263
264 va = rb_entry(n, struct vmap_area, rb_node);
265 if (addr < va->va_start)
266 n = n->rb_left;
267 else if (addr > va->va_start)
268 n = n->rb_right;
269 else
270 return va;
271 }
272
273 return NULL;
274}
275
276static void __insert_vmap_area(struct vmap_area *va)
277{
278 struct rb_node **p = &vmap_area_root.rb_node;
279 struct rb_node *parent = NULL;
280 struct rb_node *tmp;
281
282 while (*p) {
283 struct vmap_area *tmp;
284
285 parent = *p;
286 tmp = rb_entry(parent, struct vmap_area, rb_node);
287 if (va->va_start < tmp->va_end)
288 p = &(*p)->rb_left;
289 else if (va->va_end > tmp->va_start)
290 p = &(*p)->rb_right;
291 else
292 BUG();
293 }
294
295 rb_link_node(&va->rb_node, parent, p);
296 rb_insert_color(&va->rb_node, &vmap_area_root);
297
298 /* address-sort this list so it is usable like the vmlist */
299 tmp = rb_prev(&va->rb_node);
300 if (tmp) {
301 struct vmap_area *prev;
302 prev = rb_entry(tmp, struct vmap_area, rb_node);
303 list_add_rcu(&va->list, &prev->list);
304 } else
305 list_add_rcu(&va->list, &vmap_area_list);
306}
307
308static void purge_vmap_area_lazy(void);
309
310/*
311 * Allocate a region of KVA of the specified size and alignment, within the
312 * vstart and vend.
313 */
314static struct vmap_area *alloc_vmap_area(unsigned long size,
315 unsigned long align,
316 unsigned long vstart, unsigned long vend,
317 int node, gfp_t gfp_mask)
318{
319 struct vmap_area *va;
320 struct rb_node *n;
222 unsigned long addr; 321 unsigned long addr;
322 int purged = 0;
323
324 BUG_ON(size & ~PAGE_MASK);
325
326 addr = ALIGN(vstart, align);
327
328 va = kmalloc_node(sizeof(struct vmap_area),
329 gfp_mask & GFP_RECLAIM_MASK, node);
330 if (unlikely(!va))
331 return ERR_PTR(-ENOMEM);
332
333retry:
334 spin_lock(&vmap_area_lock);
335 /* XXX: could have a last_hole cache */
336 n = vmap_area_root.rb_node;
337 if (n) {
338 struct vmap_area *first = NULL;
339
340 do {
341 struct vmap_area *tmp;
342 tmp = rb_entry(n, struct vmap_area, rb_node);
343 if (tmp->va_end >= addr) {
344 if (!first && tmp->va_start < addr + size)
345 first = tmp;
346 n = n->rb_left;
347 } else {
348 first = tmp;
349 n = n->rb_right;
350 }
351 } while (n);
352
353 if (!first)
354 goto found;
355
356 if (first->va_end < addr) {
357 n = rb_next(&first->rb_node);
358 if (n)
359 first = rb_entry(n, struct vmap_area, rb_node);
360 else
361 goto found;
362 }
363
364 while (addr + size >= first->va_start && addr + size <= vend) {
365 addr = ALIGN(first->va_end + PAGE_SIZE, align);
366
367 n = rb_next(&first->rb_node);
368 if (n)
369 first = rb_entry(n, struct vmap_area, rb_node);
370 else
371 goto found;
372 }
373 }
374found:
375 if (addr + size > vend) {
376 spin_unlock(&vmap_area_lock);
377 if (!purged) {
378 purge_vmap_area_lazy();
379 purged = 1;
380 goto retry;
381 }
382 if (printk_ratelimit())
383 printk(KERN_WARNING "vmap allocation failed: "
384 "use vmalloc=<size> to increase size.\n");
385 return ERR_PTR(-EBUSY);
386 }
387
388 BUG_ON(addr & (align-1));
389
390 va->va_start = addr;
391 va->va_end = addr + size;
392 va->flags = 0;
393 __insert_vmap_area(va);
394 spin_unlock(&vmap_area_lock);
395
396 return va;
397}
398
399static void rcu_free_va(struct rcu_head *head)
400{
401 struct vmap_area *va = container_of(head, struct vmap_area, rcu_head);
402
403 kfree(va);
404}
405
406static void __free_vmap_area(struct vmap_area *va)
407{
408 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
409 rb_erase(&va->rb_node, &vmap_area_root);
410 RB_CLEAR_NODE(&va->rb_node);
411 list_del_rcu(&va->list);
412
413 call_rcu(&va->rcu_head, rcu_free_va);
414}
415
416/*
417 * Free a region of KVA allocated by alloc_vmap_area
418 */
419static void free_vmap_area(struct vmap_area *va)
420{
421 spin_lock(&vmap_area_lock);
422 __free_vmap_area(va);
423 spin_unlock(&vmap_area_lock);
424}
425
426/*
427 * Clear the pagetable entries of a given vmap_area
428 */
429static void unmap_vmap_area(struct vmap_area *va)
430{
431 vunmap_page_range(va->va_start, va->va_end);
432}
433
434/*
435 * lazy_max_pages is the maximum amount of virtual address space we gather up
436 * before attempting to purge with a TLB flush.
437 *
438 * There is a tradeoff here: a larger number will cover more kernel page tables
439 * and take slightly longer to purge, but it will linearly reduce the number of
440 * global TLB flushes that must be performed. It would seem natural to scale
441 * this number up linearly with the number of CPUs (because vmapping activity
442 * could also scale linearly with the number of CPUs), however it is likely
443 * that in practice, workloads might be constrained in other ways that mean
444 * vmap activity will not scale linearly with CPUs. Also, I want to be
445 * conservative and not introduce a big latency on huge systems, so go with
446 * a less aggressive log scale. It will still be an improvement over the old
447 * code, and it will be simple to change the scale factor if we find that it
448 * becomes a problem on bigger systems.
449 */
450static unsigned long lazy_max_pages(void)
451{
452 unsigned int log;
453
454 log = fls(num_online_cpus());
455
456 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
457}
458
459static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
460
461/*
462 * Purges all lazily-freed vmap areas.
463 *
464 * If sync is 0 then don't purge if there is already a purge in progress.
465 * If force_flush is 1, then flush kernel TLBs between *start and *end even
466 * if we found no lazy vmap areas to unmap (callers can use this to optimise
467 * their own TLB flushing).
468 * Returns with *start = min(*start, lowest purged address)
469 * *end = max(*end, highest purged address)
470 */
471static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
472 int sync, int force_flush)
473{
474 static DEFINE_SPINLOCK(purge_lock);
475 LIST_HEAD(valist);
476 struct vmap_area *va;
477 int nr = 0;
478
479 /*
480 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
481 * should not expect such behaviour. This just simplifies locking for
482 * the case that isn't actually used at the moment anyway.
483 */
484 if (!sync && !force_flush) {
485 if (!spin_trylock(&purge_lock))
486 return;
487 } else
488 spin_lock(&purge_lock);
489
490 rcu_read_lock();
491 list_for_each_entry_rcu(va, &vmap_area_list, list) {
492 if (va->flags & VM_LAZY_FREE) {
493 if (va->va_start < *start)
494 *start = va->va_start;
495 if (va->va_end > *end)
496 *end = va->va_end;
497 nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
498 unmap_vmap_area(va);
499 list_add_tail(&va->purge_list, &valist);
500 va->flags |= VM_LAZY_FREEING;
501 va->flags &= ~VM_LAZY_FREE;
502 }
503 }
504 rcu_read_unlock();
505
506 if (nr) {
507 BUG_ON(nr > atomic_read(&vmap_lazy_nr));
508 atomic_sub(nr, &vmap_lazy_nr);
509 }
510
511 if (nr || force_flush)
512 flush_tlb_kernel_range(*start, *end);
513
514 if (nr) {
515 spin_lock(&vmap_area_lock);
516 list_for_each_entry(va, &valist, purge_list)
517 __free_vmap_area(va);
518 spin_unlock(&vmap_area_lock);
519 }
520 spin_unlock(&purge_lock);
521}
522
523/*
524 * Kick off a purge of the outstanding lazy areas.
525 */
526static void purge_vmap_area_lazy(void)
527{
528 unsigned long start = ULONG_MAX, end = 0;
529
530 __purge_vmap_area_lazy(&start, &end, 0, 0);
531}
532
533/*
534 * Free and unmap a vmap area
535 */
536static void free_unmap_vmap_area(struct vmap_area *va)
537{
538 va->flags |= VM_LAZY_FREE;
539 atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
540 if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
541 purge_vmap_area_lazy();
542}
543
544static struct vmap_area *find_vmap_area(unsigned long addr)
545{
546 struct vmap_area *va;
547
548 spin_lock(&vmap_area_lock);
549 va = __find_vmap_area(addr);
550 spin_unlock(&vmap_area_lock);
551
552 return va;
553}
554
555static void free_unmap_vmap_area_addr(unsigned long addr)
556{
557 struct vmap_area *va;
558
559 va = find_vmap_area(addr);
560 BUG_ON(!va);
561 free_unmap_vmap_area(va);
562}
563
564
565/*** Per cpu kva allocator ***/
566
567/*
568 * vmap space is limited especially on 32 bit architectures. Ensure there is
569 * room for at least 16 percpu vmap blocks per CPU.
570 */
571/*
572 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
573 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
574 * instead (we just need a rough idea)
575 */
576#if BITS_PER_LONG == 32
577#define VMALLOC_SPACE (128UL*1024*1024)
578#else
579#define VMALLOC_SPACE (128UL*1024*1024*1024)
580#endif
581
582#define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
583#define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
584#define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
585#define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
586#define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
587#define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
588#define VMAP_BBMAP_BITS VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
589 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
590 VMALLOC_PAGES / NR_CPUS / 16))
591
592#define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
593
594struct vmap_block_queue {
595 spinlock_t lock;
596 struct list_head free;
597 struct list_head dirty;
598 unsigned int nr_dirty;
599};
600
601struct vmap_block {
602 spinlock_t lock;
603 struct vmap_area *va;
604 struct vmap_block_queue *vbq;
605 unsigned long free, dirty;
606 DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
607 DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
608 union {
609 struct {
610 struct list_head free_list;
611 struct list_head dirty_list;
612 };
613 struct rcu_head rcu_head;
614 };
615};
616
617/* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
618static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
619
620/*
621 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
622 * in the free path. Could get rid of this if we change the API to return a
623 * "cookie" from alloc, to be passed to free. But no big deal yet.
624 */
625static DEFINE_SPINLOCK(vmap_block_tree_lock);
626static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
627
628/*
629 * We should probably have a fallback mechanism to allocate virtual memory
630 * out of partially filled vmap blocks. However vmap block sizing should be
631 * fairly reasonable according to the vmalloc size, so it shouldn't be a
632 * big problem.
633 */
634
635static unsigned long addr_to_vb_idx(unsigned long addr)
636{
637 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
638 addr /= VMAP_BLOCK_SIZE;
639 return addr;
640}
641
642static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
643{
644 struct vmap_block_queue *vbq;
645 struct vmap_block *vb;
646 struct vmap_area *va;
647 unsigned long vb_idx;
648 int node, err;
649
650 node = numa_node_id();
651
652 vb = kmalloc_node(sizeof(struct vmap_block),
653 gfp_mask & GFP_RECLAIM_MASK, node);
654 if (unlikely(!vb))
655 return ERR_PTR(-ENOMEM);
656
657 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
658 VMALLOC_START, VMALLOC_END,
659 node, gfp_mask);
660 if (unlikely(IS_ERR(va))) {
661 kfree(vb);
662 return ERR_PTR(PTR_ERR(va));
663 }
664
665 err = radix_tree_preload(gfp_mask);
666 if (unlikely(err)) {
667 kfree(vb);
668 free_vmap_area(va);
669 return ERR_PTR(err);
670 }
671
672 spin_lock_init(&vb->lock);
673 vb->va = va;
674 vb->free = VMAP_BBMAP_BITS;
675 vb->dirty = 0;
676 bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
677 bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
678 INIT_LIST_HEAD(&vb->free_list);
679 INIT_LIST_HEAD(&vb->dirty_list);
680
681 vb_idx = addr_to_vb_idx(va->va_start);
682 spin_lock(&vmap_block_tree_lock);
683 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
684 spin_unlock(&vmap_block_tree_lock);
685 BUG_ON(err);
686 radix_tree_preload_end();
687
688 vbq = &get_cpu_var(vmap_block_queue);
689 vb->vbq = vbq;
690 spin_lock(&vbq->lock);
691 list_add(&vb->free_list, &vbq->free);
692 spin_unlock(&vbq->lock);
693 put_cpu_var(vmap_cpu_blocks);
694
695 return vb;
696}
697
698static void rcu_free_vb(struct rcu_head *head)
699{
700 struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head);
701
702 kfree(vb);
703}
704
705static void free_vmap_block(struct vmap_block *vb)
706{
707 struct vmap_block *tmp;
708 unsigned long vb_idx;
709
710 spin_lock(&vb->vbq->lock);
711 if (!list_empty(&vb->free_list))
712 list_del(&vb->free_list);
713 if (!list_empty(&vb->dirty_list))
714 list_del(&vb->dirty_list);
715 spin_unlock(&vb->vbq->lock);
716
717 vb_idx = addr_to_vb_idx(vb->va->va_start);
718 spin_lock(&vmap_block_tree_lock);
719 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
720 spin_unlock(&vmap_block_tree_lock);
721 BUG_ON(tmp != vb);
722
723 free_unmap_vmap_area(vb->va);
724 call_rcu(&vb->rcu_head, rcu_free_vb);
725}
726
727static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
728{
729 struct vmap_block_queue *vbq;
730 struct vmap_block *vb;
731 unsigned long addr = 0;
732 unsigned int order;
733
734 BUG_ON(size & ~PAGE_MASK);
735 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
736 order = get_order(size);
737
738again:
739 rcu_read_lock();
740 vbq = &get_cpu_var(vmap_block_queue);
741 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
742 int i;
743
744 spin_lock(&vb->lock);
745 i = bitmap_find_free_region(vb->alloc_map,
746 VMAP_BBMAP_BITS, order);
747
748 if (i >= 0) {
749 addr = vb->va->va_start + (i << PAGE_SHIFT);
750 BUG_ON(addr_to_vb_idx(addr) !=
751 addr_to_vb_idx(vb->va->va_start));
752 vb->free -= 1UL << order;
753 if (vb->free == 0) {
754 spin_lock(&vbq->lock);
755 list_del_init(&vb->free_list);
756 spin_unlock(&vbq->lock);
757 }
758 spin_unlock(&vb->lock);
759 break;
760 }
761 spin_unlock(&vb->lock);
762 }
763 put_cpu_var(vmap_cpu_blocks);
764 rcu_read_unlock();
765
766 if (!addr) {
767 vb = new_vmap_block(gfp_mask);
768 if (IS_ERR(vb))
769 return vb;
770 goto again;
771 }
772
773 return (void *)addr;
774}
775
776static void vb_free(const void *addr, unsigned long size)
777{
778 unsigned long offset;
779 unsigned long vb_idx;
780 unsigned int order;
781 struct vmap_block *vb;
782
783 BUG_ON(size & ~PAGE_MASK);
784 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
785 order = get_order(size);
786
787 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
788
789 vb_idx = addr_to_vb_idx((unsigned long)addr);
790 rcu_read_lock();
791 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
792 rcu_read_unlock();
793 BUG_ON(!vb);
794
795 spin_lock(&vb->lock);
796 bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order);
797 if (!vb->dirty) {
798 spin_lock(&vb->vbq->lock);
799 list_add(&vb->dirty_list, &vb->vbq->dirty);
800 spin_unlock(&vb->vbq->lock);
801 }
802 vb->dirty += 1UL << order;
803 if (vb->dirty == VMAP_BBMAP_BITS) {
804 BUG_ON(vb->free || !list_empty(&vb->free_list));
805 spin_unlock(&vb->lock);
806 free_vmap_block(vb);
807 } else
808 spin_unlock(&vb->lock);
809}
810
811/**
812 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
813 *
814 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
815 * to amortize TLB flushing overheads. What this means is that any page you
816 * have now, may, in a former life, have been mapped into kernel virtual
817 * address by the vmap layer and so there might be some CPUs with TLB entries
818 * still referencing that page (additional to the regular 1:1 kernel mapping).
819 *
820 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
821 * be sure that none of the pages we have control over will have any aliases
822 * from the vmap layer.
823 */
824void vm_unmap_aliases(void)
825{
826 unsigned long start = ULONG_MAX, end = 0;
827 int cpu;
828 int flush = 0;
829
830 for_each_possible_cpu(cpu) {
831 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
832 struct vmap_block *vb;
833
834 rcu_read_lock();
835 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
836 int i;
837
838 spin_lock(&vb->lock);
839 i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
840 while (i < VMAP_BBMAP_BITS) {
841 unsigned long s, e;
842 int j;
843 j = find_next_zero_bit(vb->dirty_map,
844 VMAP_BBMAP_BITS, i);
845
846 s = vb->va->va_start + (i << PAGE_SHIFT);
847 e = vb->va->va_start + (j << PAGE_SHIFT);
848 vunmap_page_range(s, e);
849 flush = 1;
850
851 if (s < start)
852 start = s;
853 if (e > end)
854 end = e;
855
856 i = j;
857 i = find_next_bit(vb->dirty_map,
858 VMAP_BBMAP_BITS, i);
859 }
860 spin_unlock(&vb->lock);
861 }
862 rcu_read_unlock();
863 }
864
865 __purge_vmap_area_lazy(&start, &end, 1, flush);
866}
867EXPORT_SYMBOL_GPL(vm_unmap_aliases);
868
869/**
870 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
871 * @mem: the pointer returned by vm_map_ram
872 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
873 */
874void vm_unmap_ram(const void *mem, unsigned int count)
875{
876 unsigned long size = count << PAGE_SHIFT;
877 unsigned long addr = (unsigned long)mem;
878
879 BUG_ON(!addr);
880 BUG_ON(addr < VMALLOC_START);
881 BUG_ON(addr > VMALLOC_END);
882 BUG_ON(addr & (PAGE_SIZE-1));
883
884 debug_check_no_locks_freed(mem, size);
885
886 if (likely(count <= VMAP_MAX_ALLOC))
887 vb_free(mem, size);
888 else
889 free_unmap_vmap_area_addr(addr);
890}
891EXPORT_SYMBOL(vm_unmap_ram);
892
893/**
894 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
895 * @pages: an array of pointers to the pages to be mapped
896 * @count: number of pages
897 * @node: prefer to allocate data structures on this node
898 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
899 * @returns: a pointer to the address that has been mapped, or NULL on failure
900 */
901void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
902{
903 unsigned long size = count << PAGE_SHIFT;
904 unsigned long addr;
905 void *mem;
906
907 if (likely(count <= VMAP_MAX_ALLOC)) {
908 mem = vb_alloc(size, GFP_KERNEL);
909 if (IS_ERR(mem))
910 return NULL;
911 addr = (unsigned long)mem;
912 } else {
913 struct vmap_area *va;
914 va = alloc_vmap_area(size, PAGE_SIZE,
915 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
916 if (IS_ERR(va))
917 return NULL;
918
919 addr = va->va_start;
920 mem = (void *)addr;
921 }
922 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
923 vm_unmap_ram(mem, count);
924 return NULL;
925 }
926 return mem;
927}
928EXPORT_SYMBOL(vm_map_ram);
929
930void __init vmalloc_init(void)
931{
932 int i;
933
934 for_each_possible_cpu(i) {
935 struct vmap_block_queue *vbq;
936
937 vbq = &per_cpu(vmap_block_queue, i);
938 spin_lock_init(&vbq->lock);
939 INIT_LIST_HEAD(&vbq->free);
940 INIT_LIST_HEAD(&vbq->dirty);
941 vbq->nr_dirty = 0;
942 }
943}
944
945void unmap_kernel_range(unsigned long addr, unsigned long size)
946{
947 unsigned long end = addr + size;
948 vunmap_page_range(addr, end);
949 flush_tlb_kernel_range(addr, end);
950}
951
952int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
953{
954 unsigned long addr = (unsigned long)area->addr;
955 unsigned long end = addr + area->size - PAGE_SIZE;
956 int err;
957
958 err = vmap_page_range(addr, end, prot, *pages);
959 if (err > 0) {
960 *pages += err;
961 err = 0;
962 }
963
964 return err;
965}
966EXPORT_SYMBOL_GPL(map_vm_area);
967
968/*** Old vmalloc interfaces ***/
969DEFINE_RWLOCK(vmlist_lock);
970struct vm_struct *vmlist;
971
972static struct vm_struct *__get_vm_area_node(unsigned long size,
973 unsigned long flags, unsigned long start, unsigned long end,
974 int node, gfp_t gfp_mask, void *caller)
975{
976 static struct vmap_area *va;
977 struct vm_struct *area;
978 struct vm_struct *tmp, **p;
979 unsigned long align = 1;
223 980
224 BUG_ON(in_interrupt()); 981 BUG_ON(in_interrupt());
225 if (flags & VM_IOREMAP) { 982 if (flags & VM_IOREMAP) {
@@ -232,13 +989,12 @@ __get_vm_area_node(unsigned long size, unsigned long flags, unsigned long start,
232 989
233 align = 1ul << bit; 990 align = 1ul << bit;
234 } 991 }
235 addr = ALIGN(start, align); 992
236 size = PAGE_ALIGN(size); 993 size = PAGE_ALIGN(size);
237 if (unlikely(!size)) 994 if (unlikely(!size))
238 return NULL; 995 return NULL;
239 996
240 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node); 997 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
241
242 if (unlikely(!area)) 998 if (unlikely(!area))
243 return NULL; 999 return NULL;
244 1000
@@ -247,48 +1003,32 @@ __get_vm_area_node(unsigned long size, unsigned long flags, unsigned long start,
247 */ 1003 */
248 size += PAGE_SIZE; 1004 size += PAGE_SIZE;
249 1005
250 write_lock(&vmlist_lock); 1006 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
251 for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) { 1007 if (IS_ERR(va)) {
252 if ((unsigned long)tmp->addr < addr) { 1008 kfree(area);
253 if((unsigned long)tmp->addr + tmp->size >= addr) 1009 return NULL;
254 addr = ALIGN(tmp->size +
255 (unsigned long)tmp->addr, align);
256 continue;
257 }
258 if ((size + addr) < addr)
259 goto out;
260 if (size + addr <= (unsigned long)tmp->addr)
261 goto found;
262 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
263 if (addr > end - size)
264 goto out;
265 } 1010 }
266 if ((size + addr) < addr)
267 goto out;
268 if (addr > end - size)
269 goto out;
270
271found:
272 area->next = *p;
273 *p = area;
274 1011
275 area->flags = flags; 1012 area->flags = flags;
276 area->addr = (void *)addr; 1013 area->addr = (void *)va->va_start;
277 area->size = size; 1014 area->size = size;
278 area->pages = NULL; 1015 area->pages = NULL;
279 area->nr_pages = 0; 1016 area->nr_pages = 0;
280 area->phys_addr = 0; 1017 area->phys_addr = 0;
281 area->caller = caller; 1018 area->caller = caller;
1019 va->private = area;
1020 va->flags |= VM_VM_AREA;
1021
1022 write_lock(&vmlist_lock);
1023 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1024 if (tmp->addr >= area->addr)
1025 break;
1026 }
1027 area->next = *p;
1028 *p = area;
282 write_unlock(&vmlist_lock); 1029 write_unlock(&vmlist_lock);
283 1030
284 return area; 1031 return area;
285
286out:
287 write_unlock(&vmlist_lock);
288 kfree(area);
289 if (printk_ratelimit())
290 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
291 return NULL;
292} 1032}
293 1033
294struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags, 1034struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
@@ -328,39 +1068,15 @@ struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
328 gfp_mask, __builtin_return_address(0)); 1068 gfp_mask, __builtin_return_address(0));
329} 1069}
330 1070
331/* Caller must hold vmlist_lock */ 1071static struct vm_struct *find_vm_area(const void *addr)
332static struct vm_struct *__find_vm_area(const void *addr)
333{ 1072{
334 struct vm_struct *tmp; 1073 struct vmap_area *va;
335
336 for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
337 if (tmp->addr == addr)
338 break;
339 }
340
341 return tmp;
342}
343 1074
344/* Caller must hold vmlist_lock */ 1075 va = find_vmap_area((unsigned long)addr);
345static struct vm_struct *__remove_vm_area(const void *addr) 1076 if (va && va->flags & VM_VM_AREA)
346{ 1077 return va->private;
347 struct vm_struct **p, *tmp;
348 1078
349 for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
350 if (tmp->addr == addr)
351 goto found;
352 }
353 return NULL; 1079 return NULL;
354
355found:
356 unmap_vm_area(tmp);
357 *p = tmp->next;
358
359 /*
360 * Remove the guard page.
361 */
362 tmp->size -= PAGE_SIZE;
363 return tmp;
364} 1080}
365 1081
366/** 1082/**
@@ -373,11 +1089,24 @@ found:
373 */ 1089 */
374struct vm_struct *remove_vm_area(const void *addr) 1090struct vm_struct *remove_vm_area(const void *addr)
375{ 1091{
376 struct vm_struct *v; 1092 struct vmap_area *va;
377 write_lock(&vmlist_lock); 1093
378 v = __remove_vm_area(addr); 1094 va = find_vmap_area((unsigned long)addr);
379 write_unlock(&vmlist_lock); 1095 if (va && va->flags & VM_VM_AREA) {
380 return v; 1096 struct vm_struct *vm = va->private;
1097 struct vm_struct *tmp, **p;
1098 free_unmap_vmap_area(va);
1099 vm->size -= PAGE_SIZE;
1100
1101 write_lock(&vmlist_lock);
1102 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1103 ;
1104 *p = tmp->next;
1105 write_unlock(&vmlist_lock);
1106
1107 return vm;
1108 }
1109 return NULL;
381} 1110}
382 1111
383static void __vunmap(const void *addr, int deallocate_pages) 1112static void __vunmap(const void *addr, int deallocate_pages)
@@ -487,6 +1216,8 @@ void *vmap(struct page **pages, unsigned int count,
487} 1216}
488EXPORT_SYMBOL(vmap); 1217EXPORT_SYMBOL(vmap);
489 1218
1219static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1220 int node, void *caller);
490static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, 1221static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
491 pgprot_t prot, int node, void *caller) 1222 pgprot_t prot, int node, void *caller)
492{ 1223{
@@ -613,10 +1344,8 @@ void *vmalloc_user(unsigned long size)
613 1344
614 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL); 1345 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
615 if (ret) { 1346 if (ret) {
616 write_lock(&vmlist_lock); 1347 area = find_vm_area(ret);
617 area = __find_vm_area(ret);
618 area->flags |= VM_USERMAP; 1348 area->flags |= VM_USERMAP;
619 write_unlock(&vmlist_lock);
620 } 1349 }
621 return ret; 1350 return ret;
622} 1351}
@@ -696,10 +1425,8 @@ void *vmalloc_32_user(unsigned long size)
696 1425
697 ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL); 1426 ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
698 if (ret) { 1427 if (ret) {
699 write_lock(&vmlist_lock); 1428 area = find_vm_area(ret);
700 area = __find_vm_area(ret);
701 area->flags |= VM_USERMAP; 1429 area->flags |= VM_USERMAP;
702 write_unlock(&vmlist_lock);
703 } 1430 }
704 return ret; 1431 return ret;
705} 1432}
@@ -800,26 +1527,25 @@ int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
800 struct vm_struct *area; 1527 struct vm_struct *area;
801 unsigned long uaddr = vma->vm_start; 1528 unsigned long uaddr = vma->vm_start;
802 unsigned long usize = vma->vm_end - vma->vm_start; 1529 unsigned long usize = vma->vm_end - vma->vm_start;
803 int ret;
804 1530
805 if ((PAGE_SIZE-1) & (unsigned long)addr) 1531 if ((PAGE_SIZE-1) & (unsigned long)addr)
806 return -EINVAL; 1532 return -EINVAL;
807 1533
808 read_lock(&vmlist_lock); 1534 area = find_vm_area(addr);
809 area = __find_vm_area(addr);
810 if (!area) 1535 if (!area)
811 goto out_einval_locked; 1536 return -EINVAL;
812 1537
813 if (!(area->flags & VM_USERMAP)) 1538 if (!(area->flags & VM_USERMAP))
814 goto out_einval_locked; 1539 return -EINVAL;
815 1540
816 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE) 1541 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
817 goto out_einval_locked; 1542 return -EINVAL;
818 read_unlock(&vmlist_lock);
819 1543
820 addr += pgoff << PAGE_SHIFT; 1544 addr += pgoff << PAGE_SHIFT;
821 do { 1545 do {
822 struct page *page = vmalloc_to_page(addr); 1546 struct page *page = vmalloc_to_page(addr);
1547 int ret;
1548
823 ret = vm_insert_page(vma, uaddr, page); 1549 ret = vm_insert_page(vma, uaddr, page);
824 if (ret) 1550 if (ret)
825 return ret; 1551 return ret;
@@ -832,11 +1558,7 @@ int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
832 /* Prevent "things" like memory migration? VM_flags need a cleanup... */ 1558 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
833 vma->vm_flags |= VM_RESERVED; 1559 vma->vm_flags |= VM_RESERVED;
834 1560
835 return ret; 1561 return 0;
836
837out_einval_locked:
838 read_unlock(&vmlist_lock);
839 return -EINVAL;
840} 1562}
841EXPORT_SYMBOL(remap_vmalloc_range); 1563EXPORT_SYMBOL(remap_vmalloc_range);
842 1564
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-09 01:19:16 -0500