]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Generic hugetlb support. | |
3 | * (C) William Irwin, April 2004 | |
4 | */ | |
1da177e4 LT |
5 | #include <linux/list.h> |
6 | #include <linux/init.h> | |
7 | #include <linux/module.h> | |
8 | #include <linux/mm.h> | |
e1759c21 | 9 | #include <linux/seq_file.h> |
1da177e4 LT |
10 | #include <linux/sysctl.h> |
11 | #include <linux/highmem.h> | |
cddb8a5c | 12 | #include <linux/mmu_notifier.h> |
1da177e4 | 13 | #include <linux/nodemask.h> |
63551ae0 | 14 | #include <linux/pagemap.h> |
5da7ca86 | 15 | #include <linux/mempolicy.h> |
aea47ff3 | 16 | #include <linux/cpuset.h> |
3935baa9 | 17 | #include <linux/mutex.h> |
aa888a74 | 18 | #include <linux/bootmem.h> |
a3437870 | 19 | #include <linux/sysfs.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
0fe6e20b | 21 | #include <linux/rmap.h> |
fd6a03ed NH |
22 | #include <linux/swap.h> |
23 | #include <linux/swapops.h> | |
d6606683 | 24 | |
63551ae0 DG |
25 | #include <asm/page.h> |
26 | #include <asm/pgtable.h> | |
78a34ae2 | 27 | #include <asm/io.h> |
63551ae0 DG |
28 | |
29 | #include <linux/hugetlb.h> | |
9a305230 | 30 | #include <linux/node.h> |
7835e98b | 31 | #include "internal.h" |
1da177e4 LT |
32 | |
33 | const unsigned long hugetlb_zero = 0, hugetlb_infinity = ~0UL; | |
396faf03 MG |
34 | static gfp_t htlb_alloc_mask = GFP_HIGHUSER; |
35 | unsigned long hugepages_treat_as_movable; | |
a5516438 | 36 | |
e5ff2159 AK |
37 | static int max_hstate; |
38 | unsigned int default_hstate_idx; | |
39 | struct hstate hstates[HUGE_MAX_HSTATE]; | |
40 | ||
53ba51d2 JT |
41 | __initdata LIST_HEAD(huge_boot_pages); |
42 | ||
e5ff2159 AK |
43 | /* for command line parsing */ |
44 | static struct hstate * __initdata parsed_hstate; | |
45 | static unsigned long __initdata default_hstate_max_huge_pages; | |
e11bfbfc | 46 | static unsigned long __initdata default_hstate_size; |
e5ff2159 AK |
47 | |
48 | #define for_each_hstate(h) \ | |
49 | for ((h) = hstates; (h) < &hstates[max_hstate]; (h)++) | |
396faf03 | 50 | |
3935baa9 DG |
51 | /* |
52 | * Protects updates to hugepage_freelists, nr_huge_pages, and free_huge_pages | |
53 | */ | |
54 | static DEFINE_SPINLOCK(hugetlb_lock); | |
0bd0f9fb | 55 | |
96822904 AW |
56 | /* |
57 | * Region tracking -- allows tracking of reservations and instantiated pages | |
58 | * across the pages in a mapping. | |
84afd99b AW |
59 | * |
60 | * The region data structures are protected by a combination of the mmap_sem | |
61 | * and the hugetlb_instantion_mutex. To access or modify a region the caller | |
62 | * must either hold the mmap_sem for write, or the mmap_sem for read and | |
63 | * the hugetlb_instantiation mutex: | |
64 | * | |
65 | * down_write(&mm->mmap_sem); | |
66 | * or | |
67 | * down_read(&mm->mmap_sem); | |
68 | * mutex_lock(&hugetlb_instantiation_mutex); | |
96822904 AW |
69 | */ |
70 | struct file_region { | |
71 | struct list_head link; | |
72 | long from; | |
73 | long to; | |
74 | }; | |
75 | ||
76 | static long region_add(struct list_head *head, long f, long t) | |
77 | { | |
78 | struct file_region *rg, *nrg, *trg; | |
79 | ||
80 | /* Locate the region we are either in or before. */ | |
81 | list_for_each_entry(rg, head, link) | |
82 | if (f <= rg->to) | |
83 | break; | |
84 | ||
85 | /* Round our left edge to the current segment if it encloses us. */ | |
86 | if (f > rg->from) | |
87 | f = rg->from; | |
88 | ||
89 | /* Check for and consume any regions we now overlap with. */ | |
90 | nrg = rg; | |
91 | list_for_each_entry_safe(rg, trg, rg->link.prev, link) { | |
92 | if (&rg->link == head) | |
93 | break; | |
94 | if (rg->from > t) | |
95 | break; | |
96 | ||
97 | /* If this area reaches higher then extend our area to | |
98 | * include it completely. If this is not the first area | |
99 | * which we intend to reuse, free it. */ | |
100 | if (rg->to > t) | |
101 | t = rg->to; | |
102 | if (rg != nrg) { | |
103 | list_del(&rg->link); | |
104 | kfree(rg); | |
105 | } | |
106 | } | |
107 | nrg->from = f; | |
108 | nrg->to = t; | |
109 | return 0; | |
110 | } | |
111 | ||
112 | static long region_chg(struct list_head *head, long f, long t) | |
113 | { | |
114 | struct file_region *rg, *nrg; | |
115 | long chg = 0; | |
116 | ||
117 | /* Locate the region we are before or in. */ | |
118 | list_for_each_entry(rg, head, link) | |
119 | if (f <= rg->to) | |
120 | break; | |
121 | ||
122 | /* If we are below the current region then a new region is required. | |
123 | * Subtle, allocate a new region at the position but make it zero | |
124 | * size such that we can guarantee to record the reservation. */ | |
125 | if (&rg->link == head || t < rg->from) { | |
126 | nrg = kmalloc(sizeof(*nrg), GFP_KERNEL); | |
127 | if (!nrg) | |
128 | return -ENOMEM; | |
129 | nrg->from = f; | |
130 | nrg->to = f; | |
131 | INIT_LIST_HEAD(&nrg->link); | |
132 | list_add(&nrg->link, rg->link.prev); | |
133 | ||
134 | return t - f; | |
135 | } | |
136 | ||
137 | /* Round our left edge to the current segment if it encloses us. */ | |
138 | if (f > rg->from) | |
139 | f = rg->from; | |
140 | chg = t - f; | |
141 | ||
142 | /* Check for and consume any regions we now overlap with. */ | |
143 | list_for_each_entry(rg, rg->link.prev, link) { | |
144 | if (&rg->link == head) | |
145 | break; | |
146 | if (rg->from > t) | |
147 | return chg; | |
148 | ||
25985edc | 149 | /* We overlap with this area, if it extends further than |
96822904 AW |
150 | * us then we must extend ourselves. Account for its |
151 | * existing reservation. */ | |
152 | if (rg->to > t) { | |
153 | chg += rg->to - t; | |
154 | t = rg->to; | |
155 | } | |
156 | chg -= rg->to - rg->from; | |
157 | } | |
158 | return chg; | |
159 | } | |
160 | ||
161 | static long region_truncate(struct list_head *head, long end) | |
162 | { | |
163 | struct file_region *rg, *trg; | |
164 | long chg = 0; | |
165 | ||
166 | /* Locate the region we are either in or before. */ | |
167 | list_for_each_entry(rg, head, link) | |
168 | if (end <= rg->to) | |
169 | break; | |
170 | if (&rg->link == head) | |
171 | return 0; | |
172 | ||
173 | /* If we are in the middle of a region then adjust it. */ | |
174 | if (end > rg->from) { | |
175 | chg = rg->to - end; | |
176 | rg->to = end; | |
177 | rg = list_entry(rg->link.next, typeof(*rg), link); | |
178 | } | |
179 | ||
180 | /* Drop any remaining regions. */ | |
181 | list_for_each_entry_safe(rg, trg, rg->link.prev, link) { | |
182 | if (&rg->link == head) | |
183 | break; | |
184 | chg += rg->to - rg->from; | |
185 | list_del(&rg->link); | |
186 | kfree(rg); | |
187 | } | |
188 | return chg; | |
189 | } | |
190 | ||
84afd99b AW |
191 | static long region_count(struct list_head *head, long f, long t) |
192 | { | |
193 | struct file_region *rg; | |
194 | long chg = 0; | |
195 | ||
196 | /* Locate each segment we overlap with, and count that overlap. */ | |
197 | list_for_each_entry(rg, head, link) { | |
198 | int seg_from; | |
199 | int seg_to; | |
200 | ||
201 | if (rg->to <= f) | |
202 | continue; | |
203 | if (rg->from >= t) | |
204 | break; | |
205 | ||
206 | seg_from = max(rg->from, f); | |
207 | seg_to = min(rg->to, t); | |
208 | ||
209 | chg += seg_to - seg_from; | |
210 | } | |
211 | ||
212 | return chg; | |
213 | } | |
214 | ||
e7c4b0bf AW |
215 | /* |
216 | * Convert the address within this vma to the page offset within | |
217 | * the mapping, in pagecache page units; huge pages here. | |
218 | */ | |
a5516438 AK |
219 | static pgoff_t vma_hugecache_offset(struct hstate *h, |
220 | struct vm_area_struct *vma, unsigned long address) | |
e7c4b0bf | 221 | { |
a5516438 AK |
222 | return ((address - vma->vm_start) >> huge_page_shift(h)) + |
223 | (vma->vm_pgoff >> huge_page_order(h)); | |
e7c4b0bf AW |
224 | } |
225 | ||
0fe6e20b NH |
226 | pgoff_t linear_hugepage_index(struct vm_area_struct *vma, |
227 | unsigned long address) | |
228 | { | |
229 | return vma_hugecache_offset(hstate_vma(vma), vma, address); | |
230 | } | |
231 | ||
08fba699 MG |
232 | /* |
233 | * Return the size of the pages allocated when backing a VMA. In the majority | |
234 | * cases this will be same size as used by the page table entries. | |
235 | */ | |
236 | unsigned long vma_kernel_pagesize(struct vm_area_struct *vma) | |
237 | { | |
238 | struct hstate *hstate; | |
239 | ||
240 | if (!is_vm_hugetlb_page(vma)) | |
241 | return PAGE_SIZE; | |
242 | ||
243 | hstate = hstate_vma(vma); | |
244 | ||
245 | return 1UL << (hstate->order + PAGE_SHIFT); | |
246 | } | |
f340ca0f | 247 | EXPORT_SYMBOL_GPL(vma_kernel_pagesize); |
08fba699 | 248 | |
3340289d MG |
249 | /* |
250 | * Return the page size being used by the MMU to back a VMA. In the majority | |
251 | * of cases, the page size used by the kernel matches the MMU size. On | |
252 | * architectures where it differs, an architecture-specific version of this | |
253 | * function is required. | |
254 | */ | |
255 | #ifndef vma_mmu_pagesize | |
256 | unsigned long vma_mmu_pagesize(struct vm_area_struct *vma) | |
257 | { | |
258 | return vma_kernel_pagesize(vma); | |
259 | } | |
260 | #endif | |
261 | ||
84afd99b AW |
262 | /* |
263 | * Flags for MAP_PRIVATE reservations. These are stored in the bottom | |
264 | * bits of the reservation map pointer, which are always clear due to | |
265 | * alignment. | |
266 | */ | |
267 | #define HPAGE_RESV_OWNER (1UL << 0) | |
268 | #define HPAGE_RESV_UNMAPPED (1UL << 1) | |
04f2cbe3 | 269 | #define HPAGE_RESV_MASK (HPAGE_RESV_OWNER | HPAGE_RESV_UNMAPPED) |
84afd99b | 270 | |
a1e78772 MG |
271 | /* |
272 | * These helpers are used to track how many pages are reserved for | |
273 | * faults in a MAP_PRIVATE mapping. Only the process that called mmap() | |
274 | * is guaranteed to have their future faults succeed. | |
275 | * | |
276 | * With the exception of reset_vma_resv_huge_pages() which is called at fork(), | |
277 | * the reserve counters are updated with the hugetlb_lock held. It is safe | |
278 | * to reset the VMA at fork() time as it is not in use yet and there is no | |
279 | * chance of the global counters getting corrupted as a result of the values. | |
84afd99b AW |
280 | * |
281 | * The private mapping reservation is represented in a subtly different | |
282 | * manner to a shared mapping. A shared mapping has a region map associated | |
283 | * with the underlying file, this region map represents the backing file | |
284 | * pages which have ever had a reservation assigned which this persists even | |
285 | * after the page is instantiated. A private mapping has a region map | |
286 | * associated with the original mmap which is attached to all VMAs which | |
287 | * reference it, this region map represents those offsets which have consumed | |
288 | * reservation ie. where pages have been instantiated. | |
a1e78772 | 289 | */ |
e7c4b0bf AW |
290 | static unsigned long get_vma_private_data(struct vm_area_struct *vma) |
291 | { | |
292 | return (unsigned long)vma->vm_private_data; | |
293 | } | |
294 | ||
295 | static void set_vma_private_data(struct vm_area_struct *vma, | |
296 | unsigned long value) | |
297 | { | |
298 | vma->vm_private_data = (void *)value; | |
299 | } | |
300 | ||
84afd99b AW |
301 | struct resv_map { |
302 | struct kref refs; | |
303 | struct list_head regions; | |
304 | }; | |
305 | ||
2a4b3ded | 306 | static struct resv_map *resv_map_alloc(void) |
84afd99b AW |
307 | { |
308 | struct resv_map *resv_map = kmalloc(sizeof(*resv_map), GFP_KERNEL); | |
309 | if (!resv_map) | |
310 | return NULL; | |
311 | ||
312 | kref_init(&resv_map->refs); | |
313 | INIT_LIST_HEAD(&resv_map->regions); | |
314 | ||
315 | return resv_map; | |
316 | } | |
317 | ||
2a4b3ded | 318 | static void resv_map_release(struct kref *ref) |
84afd99b AW |
319 | { |
320 | struct resv_map *resv_map = container_of(ref, struct resv_map, refs); | |
321 | ||
322 | /* Clear out any active regions before we release the map. */ | |
323 | region_truncate(&resv_map->regions, 0); | |
324 | kfree(resv_map); | |
325 | } | |
326 | ||
327 | static struct resv_map *vma_resv_map(struct vm_area_struct *vma) | |
a1e78772 MG |
328 | { |
329 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); | |
f83a275d | 330 | if (!(vma->vm_flags & VM_MAYSHARE)) |
84afd99b AW |
331 | return (struct resv_map *)(get_vma_private_data(vma) & |
332 | ~HPAGE_RESV_MASK); | |
2a4b3ded | 333 | return NULL; |
a1e78772 MG |
334 | } |
335 | ||
84afd99b | 336 | static void set_vma_resv_map(struct vm_area_struct *vma, struct resv_map *map) |
a1e78772 MG |
337 | { |
338 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); | |
f83a275d | 339 | VM_BUG_ON(vma->vm_flags & VM_MAYSHARE); |
a1e78772 | 340 | |
84afd99b AW |
341 | set_vma_private_data(vma, (get_vma_private_data(vma) & |
342 | HPAGE_RESV_MASK) | (unsigned long)map); | |
04f2cbe3 MG |
343 | } |
344 | ||
345 | static void set_vma_resv_flags(struct vm_area_struct *vma, unsigned long flags) | |
346 | { | |
04f2cbe3 | 347 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); |
f83a275d | 348 | VM_BUG_ON(vma->vm_flags & VM_MAYSHARE); |
e7c4b0bf AW |
349 | |
350 | set_vma_private_data(vma, get_vma_private_data(vma) | flags); | |
04f2cbe3 MG |
351 | } |
352 | ||
353 | static int is_vma_resv_set(struct vm_area_struct *vma, unsigned long flag) | |
354 | { | |
355 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); | |
e7c4b0bf AW |
356 | |
357 | return (get_vma_private_data(vma) & flag) != 0; | |
a1e78772 MG |
358 | } |
359 | ||
360 | /* Decrement the reserved pages in the hugepage pool by one */ | |
a5516438 AK |
361 | static void decrement_hugepage_resv_vma(struct hstate *h, |
362 | struct vm_area_struct *vma) | |
a1e78772 | 363 | { |
c37f9fb1 AW |
364 | if (vma->vm_flags & VM_NORESERVE) |
365 | return; | |
366 | ||
f83a275d | 367 | if (vma->vm_flags & VM_MAYSHARE) { |
a1e78772 | 368 | /* Shared mappings always use reserves */ |
a5516438 | 369 | h->resv_huge_pages--; |
84afd99b | 370 | } else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { |
a1e78772 MG |
371 | /* |
372 | * Only the process that called mmap() has reserves for | |
373 | * private mappings. | |
374 | */ | |
a5516438 | 375 | h->resv_huge_pages--; |
a1e78772 MG |
376 | } |
377 | } | |
378 | ||
04f2cbe3 | 379 | /* Reset counters to 0 and clear all HPAGE_RESV_* flags */ |
a1e78772 MG |
380 | void reset_vma_resv_huge_pages(struct vm_area_struct *vma) |
381 | { | |
382 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); | |
f83a275d | 383 | if (!(vma->vm_flags & VM_MAYSHARE)) |
a1e78772 MG |
384 | vma->vm_private_data = (void *)0; |
385 | } | |
386 | ||
387 | /* Returns true if the VMA has associated reserve pages */ | |
7f09ca51 | 388 | static int vma_has_reserves(struct vm_area_struct *vma) |
a1e78772 | 389 | { |
f83a275d | 390 | if (vma->vm_flags & VM_MAYSHARE) |
7f09ca51 MG |
391 | return 1; |
392 | if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) | |
393 | return 1; | |
394 | return 0; | |
a1e78772 MG |
395 | } |
396 | ||
0ebabb41 NH |
397 | static void copy_gigantic_page(struct page *dst, struct page *src) |
398 | { | |
399 | int i; | |
400 | struct hstate *h = page_hstate(src); | |
401 | struct page *dst_base = dst; | |
402 | struct page *src_base = src; | |
403 | ||
404 | for (i = 0; i < pages_per_huge_page(h); ) { | |
405 | cond_resched(); | |
406 | copy_highpage(dst, src); | |
407 | ||
408 | i++; | |
409 | dst = mem_map_next(dst, dst_base, i); | |
410 | src = mem_map_next(src, src_base, i); | |
411 | } | |
412 | } | |
413 | ||
414 | void copy_huge_page(struct page *dst, struct page *src) | |
415 | { | |
416 | int i; | |
417 | struct hstate *h = page_hstate(src); | |
418 | ||
419 | if (unlikely(pages_per_huge_page(h) > MAX_ORDER_NR_PAGES)) { | |
420 | copy_gigantic_page(dst, src); | |
421 | return; | |
422 | } | |
423 | ||
424 | might_sleep(); | |
425 | for (i = 0; i < pages_per_huge_page(h); i++) { | |
426 | cond_resched(); | |
427 | copy_highpage(dst + i, src + i); | |
428 | } | |
429 | } | |
430 | ||
a5516438 | 431 | static void enqueue_huge_page(struct hstate *h, struct page *page) |
1da177e4 LT |
432 | { |
433 | int nid = page_to_nid(page); | |
a5516438 AK |
434 | list_add(&page->lru, &h->hugepage_freelists[nid]); |
435 | h->free_huge_pages++; | |
436 | h->free_huge_pages_node[nid]++; | |
1da177e4 LT |
437 | } |
438 | ||
bf50bab2 NH |
439 | static struct page *dequeue_huge_page_node(struct hstate *h, int nid) |
440 | { | |
441 | struct page *page; | |
442 | ||
443 | if (list_empty(&h->hugepage_freelists[nid])) | |
444 | return NULL; | |
445 | page = list_entry(h->hugepage_freelists[nid].next, struct page, lru); | |
446 | list_del(&page->lru); | |
a9869b83 | 447 | set_page_refcounted(page); |
bf50bab2 NH |
448 | h->free_huge_pages--; |
449 | h->free_huge_pages_node[nid]--; | |
450 | return page; | |
451 | } | |
452 | ||
a5516438 AK |
453 | static struct page *dequeue_huge_page_vma(struct hstate *h, |
454 | struct vm_area_struct *vma, | |
04f2cbe3 | 455 | unsigned long address, int avoid_reserve) |
1da177e4 | 456 | { |
1da177e4 | 457 | struct page *page = NULL; |
480eccf9 | 458 | struct mempolicy *mpol; |
19770b32 | 459 | nodemask_t *nodemask; |
c0ff7453 | 460 | struct zonelist *zonelist; |
dd1a239f MG |
461 | struct zone *zone; |
462 | struct zoneref *z; | |
1da177e4 | 463 | |
c0ff7453 MX |
464 | get_mems_allowed(); |
465 | zonelist = huge_zonelist(vma, address, | |
466 | htlb_alloc_mask, &mpol, &nodemask); | |
a1e78772 MG |
467 | /* |
468 | * A child process with MAP_PRIVATE mappings created by their parent | |
469 | * have no page reserves. This check ensures that reservations are | |
470 | * not "stolen". The child may still get SIGKILLed | |
471 | */ | |
7f09ca51 | 472 | if (!vma_has_reserves(vma) && |
a5516438 | 473 | h->free_huge_pages - h->resv_huge_pages == 0) |
c0ff7453 | 474 | goto err; |
a1e78772 | 475 | |
04f2cbe3 | 476 | /* If reserves cannot be used, ensure enough pages are in the pool */ |
a5516438 | 477 | if (avoid_reserve && h->free_huge_pages - h->resv_huge_pages == 0) |
6eab04a8 | 478 | goto err; |
04f2cbe3 | 479 | |
19770b32 MG |
480 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
481 | MAX_NR_ZONES - 1, nodemask) { | |
bf50bab2 NH |
482 | if (cpuset_zone_allowed_softwall(zone, htlb_alloc_mask)) { |
483 | page = dequeue_huge_page_node(h, zone_to_nid(zone)); | |
484 | if (page) { | |
485 | if (!avoid_reserve) | |
486 | decrement_hugepage_resv_vma(h, vma); | |
487 | break; | |
488 | } | |
3abf7afd | 489 | } |
1da177e4 | 490 | } |
c0ff7453 | 491 | err: |
52cd3b07 | 492 | mpol_cond_put(mpol); |
c0ff7453 | 493 | put_mems_allowed(); |
1da177e4 LT |
494 | return page; |
495 | } | |
496 | ||
a5516438 | 497 | static void update_and_free_page(struct hstate *h, struct page *page) |
6af2acb6 AL |
498 | { |
499 | int i; | |
a5516438 | 500 | |
18229df5 AW |
501 | VM_BUG_ON(h->order >= MAX_ORDER); |
502 | ||
a5516438 AK |
503 | h->nr_huge_pages--; |
504 | h->nr_huge_pages_node[page_to_nid(page)]--; | |
505 | for (i = 0; i < pages_per_huge_page(h); i++) { | |
6af2acb6 AL |
506 | page[i].flags &= ~(1 << PG_locked | 1 << PG_error | 1 << PG_referenced | |
507 | 1 << PG_dirty | 1 << PG_active | 1 << PG_reserved | | |
508 | 1 << PG_private | 1<< PG_writeback); | |
509 | } | |
510 | set_compound_page_dtor(page, NULL); | |
511 | set_page_refcounted(page); | |
7f2e9525 | 512 | arch_release_hugepage(page); |
a5516438 | 513 | __free_pages(page, huge_page_order(h)); |
6af2acb6 AL |
514 | } |
515 | ||
e5ff2159 AK |
516 | struct hstate *size_to_hstate(unsigned long size) |
517 | { | |
518 | struct hstate *h; | |
519 | ||
520 | for_each_hstate(h) { | |
521 | if (huge_page_size(h) == size) | |
522 | return h; | |
523 | } | |
524 | return NULL; | |
525 | } | |
526 | ||
27a85ef1 DG |
527 | static void free_huge_page(struct page *page) |
528 | { | |
a5516438 AK |
529 | /* |
530 | * Can't pass hstate in here because it is called from the | |
531 | * compound page destructor. | |
532 | */ | |
e5ff2159 | 533 | struct hstate *h = page_hstate(page); |
7893d1d5 | 534 | int nid = page_to_nid(page); |
c79fb75e | 535 | struct address_space *mapping; |
27a85ef1 | 536 | |
c79fb75e | 537 | mapping = (struct address_space *) page_private(page); |
e5df70ab | 538 | set_page_private(page, 0); |
23be7468 | 539 | page->mapping = NULL; |
7893d1d5 | 540 | BUG_ON(page_count(page)); |
0fe6e20b | 541 | BUG_ON(page_mapcount(page)); |
27a85ef1 DG |
542 | INIT_LIST_HEAD(&page->lru); |
543 | ||
544 | spin_lock(&hugetlb_lock); | |
aa888a74 | 545 | if (h->surplus_huge_pages_node[nid] && huge_page_order(h) < MAX_ORDER) { |
a5516438 AK |
546 | update_and_free_page(h, page); |
547 | h->surplus_huge_pages--; | |
548 | h->surplus_huge_pages_node[nid]--; | |
7893d1d5 | 549 | } else { |
a5516438 | 550 | enqueue_huge_page(h, page); |
7893d1d5 | 551 | } |
27a85ef1 | 552 | spin_unlock(&hugetlb_lock); |
c79fb75e | 553 | if (mapping) |
9a119c05 | 554 | hugetlb_put_quota(mapping, 1); |
27a85ef1 DG |
555 | } |
556 | ||
a5516438 | 557 | static void prep_new_huge_page(struct hstate *h, struct page *page, int nid) |
b7ba30c6 AK |
558 | { |
559 | set_compound_page_dtor(page, free_huge_page); | |
560 | spin_lock(&hugetlb_lock); | |
a5516438 AK |
561 | h->nr_huge_pages++; |
562 | h->nr_huge_pages_node[nid]++; | |
b7ba30c6 AK |
563 | spin_unlock(&hugetlb_lock); |
564 | put_page(page); /* free it into the hugepage allocator */ | |
565 | } | |
566 | ||
20a0307c WF |
567 | static void prep_compound_gigantic_page(struct page *page, unsigned long order) |
568 | { | |
569 | int i; | |
570 | int nr_pages = 1 << order; | |
571 | struct page *p = page + 1; | |
572 | ||
573 | /* we rely on prep_new_huge_page to set the destructor */ | |
574 | set_compound_order(page, order); | |
575 | __SetPageHead(page); | |
576 | for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) { | |
577 | __SetPageTail(p); | |
578 | p->first_page = page; | |
579 | } | |
580 | } | |
581 | ||
582 | int PageHuge(struct page *page) | |
583 | { | |
584 | compound_page_dtor *dtor; | |
585 | ||
586 | if (!PageCompound(page)) | |
587 | return 0; | |
588 | ||
589 | page = compound_head(page); | |
590 | dtor = get_compound_page_dtor(page); | |
591 | ||
592 | return dtor == free_huge_page; | |
593 | } | |
594 | ||
43131e14 NH |
595 | EXPORT_SYMBOL_GPL(PageHuge); |
596 | ||
a5516438 | 597 | static struct page *alloc_fresh_huge_page_node(struct hstate *h, int nid) |
1da177e4 | 598 | { |
1da177e4 | 599 | struct page *page; |
f96efd58 | 600 | |
aa888a74 AK |
601 | if (h->order >= MAX_ORDER) |
602 | return NULL; | |
603 | ||
6484eb3e | 604 | page = alloc_pages_exact_node(nid, |
551883ae NA |
605 | htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE| |
606 | __GFP_REPEAT|__GFP_NOWARN, | |
a5516438 | 607 | huge_page_order(h)); |
1da177e4 | 608 | if (page) { |
7f2e9525 | 609 | if (arch_prepare_hugepage(page)) { |
caff3a2c | 610 | __free_pages(page, huge_page_order(h)); |
7b8ee84d | 611 | return NULL; |
7f2e9525 | 612 | } |
a5516438 | 613 | prep_new_huge_page(h, page, nid); |
1da177e4 | 614 | } |
63b4613c NA |
615 | |
616 | return page; | |
617 | } | |
618 | ||
9a76db09 | 619 | /* |
6ae11b27 LS |
620 | * common helper functions for hstate_next_node_to_{alloc|free}. |
621 | * We may have allocated or freed a huge page based on a different | |
622 | * nodes_allowed previously, so h->next_node_to_{alloc|free} might | |
623 | * be outside of *nodes_allowed. Ensure that we use an allowed | |
624 | * node for alloc or free. | |
9a76db09 | 625 | */ |
6ae11b27 | 626 | static int next_node_allowed(int nid, nodemask_t *nodes_allowed) |
9a76db09 | 627 | { |
6ae11b27 | 628 | nid = next_node(nid, *nodes_allowed); |
9a76db09 | 629 | if (nid == MAX_NUMNODES) |
6ae11b27 | 630 | nid = first_node(*nodes_allowed); |
9a76db09 LS |
631 | VM_BUG_ON(nid >= MAX_NUMNODES); |
632 | ||
633 | return nid; | |
634 | } | |
635 | ||
6ae11b27 LS |
636 | static int get_valid_node_allowed(int nid, nodemask_t *nodes_allowed) |
637 | { | |
638 | if (!node_isset(nid, *nodes_allowed)) | |
639 | nid = next_node_allowed(nid, nodes_allowed); | |
640 | return nid; | |
641 | } | |
642 | ||
5ced66c9 | 643 | /* |
6ae11b27 LS |
644 | * returns the previously saved node ["this node"] from which to |
645 | * allocate a persistent huge page for the pool and advance the | |
646 | * next node from which to allocate, handling wrap at end of node | |
647 | * mask. | |
5ced66c9 | 648 | */ |
6ae11b27 LS |
649 | static int hstate_next_node_to_alloc(struct hstate *h, |
650 | nodemask_t *nodes_allowed) | |
5ced66c9 | 651 | { |
6ae11b27 LS |
652 | int nid; |
653 | ||
654 | VM_BUG_ON(!nodes_allowed); | |
655 | ||
656 | nid = get_valid_node_allowed(h->next_nid_to_alloc, nodes_allowed); | |
657 | h->next_nid_to_alloc = next_node_allowed(nid, nodes_allowed); | |
9a76db09 | 658 | |
9a76db09 | 659 | return nid; |
5ced66c9 AK |
660 | } |
661 | ||
6ae11b27 | 662 | static int alloc_fresh_huge_page(struct hstate *h, nodemask_t *nodes_allowed) |
63b4613c NA |
663 | { |
664 | struct page *page; | |
665 | int start_nid; | |
666 | int next_nid; | |
667 | int ret = 0; | |
668 | ||
6ae11b27 | 669 | start_nid = hstate_next_node_to_alloc(h, nodes_allowed); |
e8c5c824 | 670 | next_nid = start_nid; |
63b4613c NA |
671 | |
672 | do { | |
e8c5c824 | 673 | page = alloc_fresh_huge_page_node(h, next_nid); |
9a76db09 | 674 | if (page) { |
63b4613c | 675 | ret = 1; |
9a76db09 LS |
676 | break; |
677 | } | |
6ae11b27 | 678 | next_nid = hstate_next_node_to_alloc(h, nodes_allowed); |
9a76db09 | 679 | } while (next_nid != start_nid); |
63b4613c | 680 | |
3b116300 AL |
681 | if (ret) |
682 | count_vm_event(HTLB_BUDDY_PGALLOC); | |
683 | else | |
684 | count_vm_event(HTLB_BUDDY_PGALLOC_FAIL); | |
685 | ||
63b4613c | 686 | return ret; |
1da177e4 LT |
687 | } |
688 | ||
e8c5c824 | 689 | /* |
6ae11b27 LS |
690 | * helper for free_pool_huge_page() - return the previously saved |
691 | * node ["this node"] from which to free a huge page. Advance the | |
692 | * next node id whether or not we find a free huge page to free so | |
693 | * that the next attempt to free addresses the next node. | |
e8c5c824 | 694 | */ |
6ae11b27 | 695 | static int hstate_next_node_to_free(struct hstate *h, nodemask_t *nodes_allowed) |
e8c5c824 | 696 | { |
6ae11b27 LS |
697 | int nid; |
698 | ||
699 | VM_BUG_ON(!nodes_allowed); | |
700 | ||
701 | nid = get_valid_node_allowed(h->next_nid_to_free, nodes_allowed); | |
702 | h->next_nid_to_free = next_node_allowed(nid, nodes_allowed); | |
9a76db09 | 703 | |
9a76db09 | 704 | return nid; |
e8c5c824 LS |
705 | } |
706 | ||
707 | /* | |
708 | * Free huge page from pool from next node to free. | |
709 | * Attempt to keep persistent huge pages more or less | |
710 | * balanced over allowed nodes. | |
711 | * Called with hugetlb_lock locked. | |
712 | */ | |
6ae11b27 LS |
713 | static int free_pool_huge_page(struct hstate *h, nodemask_t *nodes_allowed, |
714 | bool acct_surplus) | |
e8c5c824 LS |
715 | { |
716 | int start_nid; | |
717 | int next_nid; | |
718 | int ret = 0; | |
719 | ||
6ae11b27 | 720 | start_nid = hstate_next_node_to_free(h, nodes_allowed); |
e8c5c824 LS |
721 | next_nid = start_nid; |
722 | ||
723 | do { | |
685f3457 LS |
724 | /* |
725 | * If we're returning unused surplus pages, only examine | |
726 | * nodes with surplus pages. | |
727 | */ | |
728 | if ((!acct_surplus || h->surplus_huge_pages_node[next_nid]) && | |
729 | !list_empty(&h->hugepage_freelists[next_nid])) { | |
e8c5c824 LS |
730 | struct page *page = |
731 | list_entry(h->hugepage_freelists[next_nid].next, | |
732 | struct page, lru); | |
733 | list_del(&page->lru); | |
734 | h->free_huge_pages--; | |
735 | h->free_huge_pages_node[next_nid]--; | |
685f3457 LS |
736 | if (acct_surplus) { |
737 | h->surplus_huge_pages--; | |
738 | h->surplus_huge_pages_node[next_nid]--; | |
739 | } | |
e8c5c824 LS |
740 | update_and_free_page(h, page); |
741 | ret = 1; | |
9a76db09 | 742 | break; |
e8c5c824 | 743 | } |
6ae11b27 | 744 | next_nid = hstate_next_node_to_free(h, nodes_allowed); |
9a76db09 | 745 | } while (next_nid != start_nid); |
e8c5c824 LS |
746 | |
747 | return ret; | |
748 | } | |
749 | ||
bf50bab2 | 750 | static struct page *alloc_buddy_huge_page(struct hstate *h, int nid) |
7893d1d5 AL |
751 | { |
752 | struct page *page; | |
bf50bab2 | 753 | unsigned int r_nid; |
7893d1d5 | 754 | |
aa888a74 AK |
755 | if (h->order >= MAX_ORDER) |
756 | return NULL; | |
757 | ||
d1c3fb1f NA |
758 | /* |
759 | * Assume we will successfully allocate the surplus page to | |
760 | * prevent racing processes from causing the surplus to exceed | |
761 | * overcommit | |
762 | * | |
763 | * This however introduces a different race, where a process B | |
764 | * tries to grow the static hugepage pool while alloc_pages() is | |
765 | * called by process A. B will only examine the per-node | |
766 | * counters in determining if surplus huge pages can be | |
767 | * converted to normal huge pages in adjust_pool_surplus(). A | |
768 | * won't be able to increment the per-node counter, until the | |
769 | * lock is dropped by B, but B doesn't drop hugetlb_lock until | |
770 | * no more huge pages can be converted from surplus to normal | |
771 | * state (and doesn't try to convert again). Thus, we have a | |
772 | * case where a surplus huge page exists, the pool is grown, and | |
773 | * the surplus huge page still exists after, even though it | |
774 | * should just have been converted to a normal huge page. This | |
775 | * does not leak memory, though, as the hugepage will be freed | |
776 | * once it is out of use. It also does not allow the counters to | |
777 | * go out of whack in adjust_pool_surplus() as we don't modify | |
778 | * the node values until we've gotten the hugepage and only the | |
779 | * per-node value is checked there. | |
780 | */ | |
781 | spin_lock(&hugetlb_lock); | |
a5516438 | 782 | if (h->surplus_huge_pages >= h->nr_overcommit_huge_pages) { |
d1c3fb1f NA |
783 | spin_unlock(&hugetlb_lock); |
784 | return NULL; | |
785 | } else { | |
a5516438 AK |
786 | h->nr_huge_pages++; |
787 | h->surplus_huge_pages++; | |
d1c3fb1f NA |
788 | } |
789 | spin_unlock(&hugetlb_lock); | |
790 | ||
bf50bab2 NH |
791 | if (nid == NUMA_NO_NODE) |
792 | page = alloc_pages(htlb_alloc_mask|__GFP_COMP| | |
793 | __GFP_REPEAT|__GFP_NOWARN, | |
794 | huge_page_order(h)); | |
795 | else | |
796 | page = alloc_pages_exact_node(nid, | |
797 | htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE| | |
798 | __GFP_REPEAT|__GFP_NOWARN, huge_page_order(h)); | |
d1c3fb1f | 799 | |
caff3a2c GS |
800 | if (page && arch_prepare_hugepage(page)) { |
801 | __free_pages(page, huge_page_order(h)); | |
802 | return NULL; | |
803 | } | |
804 | ||
d1c3fb1f | 805 | spin_lock(&hugetlb_lock); |
7893d1d5 | 806 | if (page) { |
bf50bab2 | 807 | r_nid = page_to_nid(page); |
7893d1d5 | 808 | set_compound_page_dtor(page, free_huge_page); |
d1c3fb1f NA |
809 | /* |
810 | * We incremented the global counters already | |
811 | */ | |
bf50bab2 NH |
812 | h->nr_huge_pages_node[r_nid]++; |
813 | h->surplus_huge_pages_node[r_nid]++; | |
3b116300 | 814 | __count_vm_event(HTLB_BUDDY_PGALLOC); |
d1c3fb1f | 815 | } else { |
a5516438 AK |
816 | h->nr_huge_pages--; |
817 | h->surplus_huge_pages--; | |
3b116300 | 818 | __count_vm_event(HTLB_BUDDY_PGALLOC_FAIL); |
7893d1d5 | 819 | } |
d1c3fb1f | 820 | spin_unlock(&hugetlb_lock); |
7893d1d5 AL |
821 | |
822 | return page; | |
823 | } | |
824 | ||
bf50bab2 NH |
825 | /* |
826 | * This allocation function is useful in the context where vma is irrelevant. | |
827 | * E.g. soft-offlining uses this function because it only cares physical | |
828 | * address of error page. | |
829 | */ | |
830 | struct page *alloc_huge_page_node(struct hstate *h, int nid) | |
831 | { | |
832 | struct page *page; | |
833 | ||
834 | spin_lock(&hugetlb_lock); | |
835 | page = dequeue_huge_page_node(h, nid); | |
836 | spin_unlock(&hugetlb_lock); | |
837 | ||
838 | if (!page) | |
839 | page = alloc_buddy_huge_page(h, nid); | |
840 | ||
841 | return page; | |
842 | } | |
843 | ||
e4e574b7 | 844 | /* |
25985edc | 845 | * Increase the hugetlb pool such that it can accommodate a reservation |
e4e574b7 AL |
846 | * of size 'delta'. |
847 | */ | |
a5516438 | 848 | static int gather_surplus_pages(struct hstate *h, int delta) |
e4e574b7 AL |
849 | { |
850 | struct list_head surplus_list; | |
851 | struct page *page, *tmp; | |
852 | int ret, i; | |
853 | int needed, allocated; | |
854 | ||
a5516438 | 855 | needed = (h->resv_huge_pages + delta) - h->free_huge_pages; |
ac09b3a1 | 856 | if (needed <= 0) { |
a5516438 | 857 | h->resv_huge_pages += delta; |
e4e574b7 | 858 | return 0; |
ac09b3a1 | 859 | } |
e4e574b7 AL |
860 | |
861 | allocated = 0; | |
862 | INIT_LIST_HEAD(&surplus_list); | |
863 | ||
864 | ret = -ENOMEM; | |
865 | retry: | |
866 | spin_unlock(&hugetlb_lock); | |
867 | for (i = 0; i < needed; i++) { | |
bf50bab2 | 868 | page = alloc_buddy_huge_page(h, NUMA_NO_NODE); |
a9869b83 | 869 | if (!page) |
e4e574b7 AL |
870 | /* |
871 | * We were not able to allocate enough pages to | |
872 | * satisfy the entire reservation so we free what | |
873 | * we've allocated so far. | |
874 | */ | |
e4e574b7 | 875 | goto free; |
e4e574b7 AL |
876 | |
877 | list_add(&page->lru, &surplus_list); | |
878 | } | |
879 | allocated += needed; | |
880 | ||
881 | /* | |
882 | * After retaking hugetlb_lock, we need to recalculate 'needed' | |
883 | * because either resv_huge_pages or free_huge_pages may have changed. | |
884 | */ | |
885 | spin_lock(&hugetlb_lock); | |
a5516438 AK |
886 | needed = (h->resv_huge_pages + delta) - |
887 | (h->free_huge_pages + allocated); | |
e4e574b7 AL |
888 | if (needed > 0) |
889 | goto retry; | |
890 | ||
891 | /* | |
892 | * The surplus_list now contains _at_least_ the number of extra pages | |
25985edc | 893 | * needed to accommodate the reservation. Add the appropriate number |
e4e574b7 | 894 | * of pages to the hugetlb pool and free the extras back to the buddy |
ac09b3a1 AL |
895 | * allocator. Commit the entire reservation here to prevent another |
896 | * process from stealing the pages as they are added to the pool but | |
897 | * before they are reserved. | |
e4e574b7 AL |
898 | */ |
899 | needed += allocated; | |
a5516438 | 900 | h->resv_huge_pages += delta; |
e4e574b7 | 901 | ret = 0; |
a9869b83 NH |
902 | |
903 | spin_unlock(&hugetlb_lock); | |
19fc3f0a | 904 | /* Free the needed pages to the hugetlb pool */ |
e4e574b7 | 905 | list_for_each_entry_safe(page, tmp, &surplus_list, lru) { |
19fc3f0a AL |
906 | if ((--needed) < 0) |
907 | break; | |
e4e574b7 | 908 | list_del(&page->lru); |
a9869b83 NH |
909 | /* |
910 | * This page is now managed by the hugetlb allocator and has | |
911 | * no users -- drop the buddy allocator's reference. | |
912 | */ | |
913 | put_page_testzero(page); | |
914 | VM_BUG_ON(page_count(page)); | |
a5516438 | 915 | enqueue_huge_page(h, page); |
19fc3f0a AL |
916 | } |
917 | ||
918 | /* Free unnecessary surplus pages to the buddy allocator */ | |
a9869b83 | 919 | free: |
19fc3f0a | 920 | if (!list_empty(&surplus_list)) { |
19fc3f0a AL |
921 | list_for_each_entry_safe(page, tmp, &surplus_list, lru) { |
922 | list_del(&page->lru); | |
a9869b83 | 923 | put_page(page); |
af767cbd | 924 | } |
e4e574b7 | 925 | } |
a9869b83 | 926 | spin_lock(&hugetlb_lock); |
e4e574b7 AL |
927 | |
928 | return ret; | |
929 | } | |
930 | ||
931 | /* | |
932 | * When releasing a hugetlb pool reservation, any surplus pages that were | |
933 | * allocated to satisfy the reservation must be explicitly freed if they were | |
934 | * never used. | |
685f3457 | 935 | * Called with hugetlb_lock held. |
e4e574b7 | 936 | */ |
a5516438 AK |
937 | static void return_unused_surplus_pages(struct hstate *h, |
938 | unsigned long unused_resv_pages) | |
e4e574b7 | 939 | { |
e4e574b7 AL |
940 | unsigned long nr_pages; |
941 | ||
ac09b3a1 | 942 | /* Uncommit the reservation */ |
a5516438 | 943 | h->resv_huge_pages -= unused_resv_pages; |
ac09b3a1 | 944 | |
aa888a74 AK |
945 | /* Cannot return gigantic pages currently */ |
946 | if (h->order >= MAX_ORDER) | |
947 | return; | |
948 | ||
a5516438 | 949 | nr_pages = min(unused_resv_pages, h->surplus_huge_pages); |
e4e574b7 | 950 | |
685f3457 LS |
951 | /* |
952 | * We want to release as many surplus pages as possible, spread | |
9b5e5d0f LS |
953 | * evenly across all nodes with memory. Iterate across these nodes |
954 | * until we can no longer free unreserved surplus pages. This occurs | |
955 | * when the nodes with surplus pages have no free pages. | |
956 | * free_pool_huge_page() will balance the the freed pages across the | |
957 | * on-line nodes with memory and will handle the hstate accounting. | |
685f3457 LS |
958 | */ |
959 | while (nr_pages--) { | |
9b5e5d0f | 960 | if (!free_pool_huge_page(h, &node_states[N_HIGH_MEMORY], 1)) |
685f3457 | 961 | break; |
e4e574b7 AL |
962 | } |
963 | } | |
964 | ||
c37f9fb1 AW |
965 | /* |
966 | * Determine if the huge page at addr within the vma has an associated | |
967 | * reservation. Where it does not we will need to logically increase | |
968 | * reservation and actually increase quota before an allocation can occur. | |
969 | * Where any new reservation would be required the reservation change is | |
970 | * prepared, but not committed. Once the page has been quota'd allocated | |
971 | * an instantiated the change should be committed via vma_commit_reservation. | |
972 | * No action is required on failure. | |
973 | */ | |
e2f17d94 | 974 | static long vma_needs_reservation(struct hstate *h, |
a5516438 | 975 | struct vm_area_struct *vma, unsigned long addr) |
c37f9fb1 AW |
976 | { |
977 | struct address_space *mapping = vma->vm_file->f_mapping; | |
978 | struct inode *inode = mapping->host; | |
979 | ||
f83a275d | 980 | if (vma->vm_flags & VM_MAYSHARE) { |
a5516438 | 981 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
c37f9fb1 AW |
982 | return region_chg(&inode->i_mapping->private_list, |
983 | idx, idx + 1); | |
984 | ||
84afd99b AW |
985 | } else if (!is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { |
986 | return 1; | |
c37f9fb1 | 987 | |
84afd99b | 988 | } else { |
e2f17d94 | 989 | long err; |
a5516438 | 990 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
84afd99b AW |
991 | struct resv_map *reservations = vma_resv_map(vma); |
992 | ||
993 | err = region_chg(&reservations->regions, idx, idx + 1); | |
994 | if (err < 0) | |
995 | return err; | |
996 | return 0; | |
997 | } | |
c37f9fb1 | 998 | } |
a5516438 AK |
999 | static void vma_commit_reservation(struct hstate *h, |
1000 | struct vm_area_struct *vma, unsigned long addr) | |
c37f9fb1 AW |
1001 | { |
1002 | struct address_space *mapping = vma->vm_file->f_mapping; | |
1003 | struct inode *inode = mapping->host; | |
1004 | ||
f83a275d | 1005 | if (vma->vm_flags & VM_MAYSHARE) { |
a5516438 | 1006 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
c37f9fb1 | 1007 | region_add(&inode->i_mapping->private_list, idx, idx + 1); |
84afd99b AW |
1008 | |
1009 | } else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { | |
a5516438 | 1010 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
84afd99b AW |
1011 | struct resv_map *reservations = vma_resv_map(vma); |
1012 | ||
1013 | /* Mark this page used in the map. */ | |
1014 | region_add(&reservations->regions, idx, idx + 1); | |
c37f9fb1 AW |
1015 | } |
1016 | } | |
1017 | ||
a1e78772 | 1018 | static struct page *alloc_huge_page(struct vm_area_struct *vma, |
04f2cbe3 | 1019 | unsigned long addr, int avoid_reserve) |
1da177e4 | 1020 | { |
a5516438 | 1021 | struct hstate *h = hstate_vma(vma); |
348ea204 | 1022 | struct page *page; |
a1e78772 MG |
1023 | struct address_space *mapping = vma->vm_file->f_mapping; |
1024 | struct inode *inode = mapping->host; | |
e2f17d94 | 1025 | long chg; |
a1e78772 MG |
1026 | |
1027 | /* | |
1028 | * Processes that did not create the mapping will have no reserves and | |
1029 | * will not have accounted against quota. Check that the quota can be | |
1030 | * made before satisfying the allocation | |
c37f9fb1 AW |
1031 | * MAP_NORESERVE mappings may also need pages and quota allocated |
1032 | * if no reserve mapping overlaps. | |
a1e78772 | 1033 | */ |
a5516438 | 1034 | chg = vma_needs_reservation(h, vma, addr); |
c37f9fb1 | 1035 | if (chg < 0) |
e0dcd8a0 | 1036 | return ERR_PTR(-VM_FAULT_OOM); |
c37f9fb1 | 1037 | if (chg) |
a1e78772 | 1038 | if (hugetlb_get_quota(inode->i_mapping, chg)) |
e0dcd8a0 | 1039 | return ERR_PTR(-VM_FAULT_SIGBUS); |
1da177e4 LT |
1040 | |
1041 | spin_lock(&hugetlb_lock); | |
a5516438 | 1042 | page = dequeue_huge_page_vma(h, vma, addr, avoid_reserve); |
1da177e4 | 1043 | spin_unlock(&hugetlb_lock); |
b45b5bd6 | 1044 | |
68842c9b | 1045 | if (!page) { |
bf50bab2 | 1046 | page = alloc_buddy_huge_page(h, NUMA_NO_NODE); |
68842c9b | 1047 | if (!page) { |
a1e78772 | 1048 | hugetlb_put_quota(inode->i_mapping, chg); |
4a6018f7 | 1049 | return ERR_PTR(-VM_FAULT_SIGBUS); |
68842c9b KC |
1050 | } |
1051 | } | |
348ea204 | 1052 | |
a1e78772 | 1053 | set_page_private(page, (unsigned long) mapping); |
90d8b7e6 | 1054 | |
a5516438 | 1055 | vma_commit_reservation(h, vma, addr); |
c37f9fb1 | 1056 | |
90d8b7e6 | 1057 | return page; |
b45b5bd6 DG |
1058 | } |
1059 | ||
91f47662 | 1060 | int __weak alloc_bootmem_huge_page(struct hstate *h) |
aa888a74 AK |
1061 | { |
1062 | struct huge_bootmem_page *m; | |
9b5e5d0f | 1063 | int nr_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); |
aa888a74 AK |
1064 | |
1065 | while (nr_nodes) { | |
1066 | void *addr; | |
1067 | ||
1068 | addr = __alloc_bootmem_node_nopanic( | |
6ae11b27 | 1069 | NODE_DATA(hstate_next_node_to_alloc(h, |
9b5e5d0f | 1070 | &node_states[N_HIGH_MEMORY])), |
aa888a74 AK |
1071 | huge_page_size(h), huge_page_size(h), 0); |
1072 | ||
1073 | if (addr) { | |
1074 | /* | |
1075 | * Use the beginning of the huge page to store the | |
1076 | * huge_bootmem_page struct (until gather_bootmem | |
1077 | * puts them into the mem_map). | |
1078 | */ | |
1079 | m = addr; | |
91f47662 | 1080 | goto found; |
aa888a74 | 1081 | } |
aa888a74 AK |
1082 | nr_nodes--; |
1083 | } | |
1084 | return 0; | |
1085 | ||
1086 | found: | |
1087 | BUG_ON((unsigned long)virt_to_phys(m) & (huge_page_size(h) - 1)); | |
1088 | /* Put them into a private list first because mem_map is not up yet */ | |
1089 | list_add(&m->list, &huge_boot_pages); | |
1090 | m->hstate = h; | |
1091 | return 1; | |
1092 | } | |
1093 | ||
18229df5 AW |
1094 | static void prep_compound_huge_page(struct page *page, int order) |
1095 | { | |
1096 | if (unlikely(order > (MAX_ORDER - 1))) | |
1097 | prep_compound_gigantic_page(page, order); | |
1098 | else | |
1099 | prep_compound_page(page, order); | |
1100 | } | |
1101 | ||
aa888a74 AK |
1102 | /* Put bootmem huge pages into the standard lists after mem_map is up */ |
1103 | static void __init gather_bootmem_prealloc(void) | |
1104 | { | |
1105 | struct huge_bootmem_page *m; | |
1106 | ||
1107 | list_for_each_entry(m, &huge_boot_pages, list) { | |
1108 | struct page *page = virt_to_page(m); | |
1109 | struct hstate *h = m->hstate; | |
1110 | __ClearPageReserved(page); | |
1111 | WARN_ON(page_count(page) != 1); | |
18229df5 | 1112 | prep_compound_huge_page(page, h->order); |
aa888a74 | 1113 | prep_new_huge_page(h, page, page_to_nid(page)); |
b0320c7b RA |
1114 | /* |
1115 | * If we had gigantic hugepages allocated at boot time, we need | |
1116 | * to restore the 'stolen' pages to totalram_pages in order to | |
1117 | * fix confusing memory reports from free(1) and another | |
1118 | * side-effects, like CommitLimit going negative. | |
1119 | */ | |
1120 | if (h->order > (MAX_ORDER - 1)) | |
1121 | totalram_pages += 1 << h->order; | |
aa888a74 AK |
1122 | } |
1123 | } | |
1124 | ||
8faa8b07 | 1125 | static void __init hugetlb_hstate_alloc_pages(struct hstate *h) |
1da177e4 LT |
1126 | { |
1127 | unsigned long i; | |
a5516438 | 1128 | |
e5ff2159 | 1129 | for (i = 0; i < h->max_huge_pages; ++i) { |
aa888a74 AK |
1130 | if (h->order >= MAX_ORDER) { |
1131 | if (!alloc_bootmem_huge_page(h)) | |
1132 | break; | |
9b5e5d0f LS |
1133 | } else if (!alloc_fresh_huge_page(h, |
1134 | &node_states[N_HIGH_MEMORY])) | |
1da177e4 | 1135 | break; |
1da177e4 | 1136 | } |
8faa8b07 | 1137 | h->max_huge_pages = i; |
e5ff2159 AK |
1138 | } |
1139 | ||
1140 | static void __init hugetlb_init_hstates(void) | |
1141 | { | |
1142 | struct hstate *h; | |
1143 | ||
1144 | for_each_hstate(h) { | |
8faa8b07 AK |
1145 | /* oversize hugepages were init'ed in early boot */ |
1146 | if (h->order < MAX_ORDER) | |
1147 | hugetlb_hstate_alloc_pages(h); | |
e5ff2159 AK |
1148 | } |
1149 | } | |
1150 | ||
4abd32db AK |
1151 | static char * __init memfmt(char *buf, unsigned long n) |
1152 | { | |
1153 | if (n >= (1UL << 30)) | |
1154 | sprintf(buf, "%lu GB", n >> 30); | |
1155 | else if (n >= (1UL << 20)) | |
1156 | sprintf(buf, "%lu MB", n >> 20); | |
1157 | else | |
1158 | sprintf(buf, "%lu KB", n >> 10); | |
1159 | return buf; | |
1160 | } | |
1161 | ||
e5ff2159 AK |
1162 | static void __init report_hugepages(void) |
1163 | { | |
1164 | struct hstate *h; | |
1165 | ||
1166 | for_each_hstate(h) { | |
4abd32db AK |
1167 | char buf[32]; |
1168 | printk(KERN_INFO "HugeTLB registered %s page size, " | |
1169 | "pre-allocated %ld pages\n", | |
1170 | memfmt(buf, huge_page_size(h)), | |
1171 | h->free_huge_pages); | |
e5ff2159 AK |
1172 | } |
1173 | } | |
1174 | ||
1da177e4 | 1175 | #ifdef CONFIG_HIGHMEM |
6ae11b27 LS |
1176 | static void try_to_free_low(struct hstate *h, unsigned long count, |
1177 | nodemask_t *nodes_allowed) | |
1da177e4 | 1178 | { |
4415cc8d CL |
1179 | int i; |
1180 | ||
aa888a74 AK |
1181 | if (h->order >= MAX_ORDER) |
1182 | return; | |
1183 | ||
6ae11b27 | 1184 | for_each_node_mask(i, *nodes_allowed) { |
1da177e4 | 1185 | struct page *page, *next; |
a5516438 AK |
1186 | struct list_head *freel = &h->hugepage_freelists[i]; |
1187 | list_for_each_entry_safe(page, next, freel, lru) { | |
1188 | if (count >= h->nr_huge_pages) | |
6b0c880d | 1189 | return; |
1da177e4 LT |
1190 | if (PageHighMem(page)) |
1191 | continue; | |
1192 | list_del(&page->lru); | |
e5ff2159 | 1193 | update_and_free_page(h, page); |
a5516438 AK |
1194 | h->free_huge_pages--; |
1195 | h->free_huge_pages_node[page_to_nid(page)]--; | |
1da177e4 LT |
1196 | } |
1197 | } | |
1198 | } | |
1199 | #else | |
6ae11b27 LS |
1200 | static inline void try_to_free_low(struct hstate *h, unsigned long count, |
1201 | nodemask_t *nodes_allowed) | |
1da177e4 LT |
1202 | { |
1203 | } | |
1204 | #endif | |
1205 | ||
20a0307c WF |
1206 | /* |
1207 | * Increment or decrement surplus_huge_pages. Keep node-specific counters | |
1208 | * balanced by operating on them in a round-robin fashion. | |
1209 | * Returns 1 if an adjustment was made. | |
1210 | */ | |
6ae11b27 LS |
1211 | static int adjust_pool_surplus(struct hstate *h, nodemask_t *nodes_allowed, |
1212 | int delta) | |
20a0307c | 1213 | { |
e8c5c824 | 1214 | int start_nid, next_nid; |
20a0307c WF |
1215 | int ret = 0; |
1216 | ||
1217 | VM_BUG_ON(delta != -1 && delta != 1); | |
20a0307c | 1218 | |
e8c5c824 | 1219 | if (delta < 0) |
6ae11b27 | 1220 | start_nid = hstate_next_node_to_alloc(h, nodes_allowed); |
e8c5c824 | 1221 | else |
6ae11b27 | 1222 | start_nid = hstate_next_node_to_free(h, nodes_allowed); |
e8c5c824 LS |
1223 | next_nid = start_nid; |
1224 | ||
1225 | do { | |
1226 | int nid = next_nid; | |
1227 | if (delta < 0) { | |
e8c5c824 LS |
1228 | /* |
1229 | * To shrink on this node, there must be a surplus page | |
1230 | */ | |
9a76db09 | 1231 | if (!h->surplus_huge_pages_node[nid]) { |
6ae11b27 LS |
1232 | next_nid = hstate_next_node_to_alloc(h, |
1233 | nodes_allowed); | |
e8c5c824 | 1234 | continue; |
9a76db09 | 1235 | } |
e8c5c824 LS |
1236 | } |
1237 | if (delta > 0) { | |
e8c5c824 LS |
1238 | /* |
1239 | * Surplus cannot exceed the total number of pages | |
1240 | */ | |
1241 | if (h->surplus_huge_pages_node[nid] >= | |
9a76db09 | 1242 | h->nr_huge_pages_node[nid]) { |
6ae11b27 LS |
1243 | next_nid = hstate_next_node_to_free(h, |
1244 | nodes_allowed); | |
e8c5c824 | 1245 | continue; |
9a76db09 | 1246 | } |
e8c5c824 | 1247 | } |
20a0307c WF |
1248 | |
1249 | h->surplus_huge_pages += delta; | |
1250 | h->surplus_huge_pages_node[nid] += delta; | |
1251 | ret = 1; | |
1252 | break; | |
e8c5c824 | 1253 | } while (next_nid != start_nid); |
20a0307c | 1254 | |
20a0307c WF |
1255 | return ret; |
1256 | } | |
1257 | ||
a5516438 | 1258 | #define persistent_huge_pages(h) (h->nr_huge_pages - h->surplus_huge_pages) |
6ae11b27 LS |
1259 | static unsigned long set_max_huge_pages(struct hstate *h, unsigned long count, |
1260 | nodemask_t *nodes_allowed) | |
1da177e4 | 1261 | { |
7893d1d5 | 1262 | unsigned long min_count, ret; |
1da177e4 | 1263 | |
aa888a74 AK |
1264 | if (h->order >= MAX_ORDER) |
1265 | return h->max_huge_pages; | |
1266 | ||
7893d1d5 AL |
1267 | /* |
1268 | * Increase the pool size | |
1269 | * First take pages out of surplus state. Then make up the | |
1270 | * remaining difference by allocating fresh huge pages. | |
d1c3fb1f NA |
1271 | * |
1272 | * We might race with alloc_buddy_huge_page() here and be unable | |
1273 | * to convert a surplus huge page to a normal huge page. That is | |
1274 | * not critical, though, it just means the overall size of the | |
1275 | * pool might be one hugepage larger than it needs to be, but | |
1276 | * within all the constraints specified by the sysctls. | |
7893d1d5 | 1277 | */ |
1da177e4 | 1278 | spin_lock(&hugetlb_lock); |
a5516438 | 1279 | while (h->surplus_huge_pages && count > persistent_huge_pages(h)) { |
6ae11b27 | 1280 | if (!adjust_pool_surplus(h, nodes_allowed, -1)) |
7893d1d5 AL |
1281 | break; |
1282 | } | |
1283 | ||
a5516438 | 1284 | while (count > persistent_huge_pages(h)) { |
7893d1d5 AL |
1285 | /* |
1286 | * If this allocation races such that we no longer need the | |
1287 | * page, free_huge_page will handle it by freeing the page | |
1288 | * and reducing the surplus. | |
1289 | */ | |
1290 | spin_unlock(&hugetlb_lock); | |
6ae11b27 | 1291 | ret = alloc_fresh_huge_page(h, nodes_allowed); |
7893d1d5 AL |
1292 | spin_lock(&hugetlb_lock); |
1293 | if (!ret) | |
1294 | goto out; | |
1295 | ||
536240f2 MG |
1296 | /* Bail for signals. Probably ctrl-c from user */ |
1297 | if (signal_pending(current)) | |
1298 | goto out; | |
7893d1d5 | 1299 | } |
7893d1d5 AL |
1300 | |
1301 | /* | |
1302 | * Decrease the pool size | |
1303 | * First return free pages to the buddy allocator (being careful | |
1304 | * to keep enough around to satisfy reservations). Then place | |
1305 | * pages into surplus state as needed so the pool will shrink | |
1306 | * to the desired size as pages become free. | |
d1c3fb1f NA |
1307 | * |
1308 | * By placing pages into the surplus state independent of the | |
1309 | * overcommit value, we are allowing the surplus pool size to | |
1310 | * exceed overcommit. There are few sane options here. Since | |
1311 | * alloc_buddy_huge_page() is checking the global counter, | |
1312 | * though, we'll note that we're not allowed to exceed surplus | |
1313 | * and won't grow the pool anywhere else. Not until one of the | |
1314 | * sysctls are changed, or the surplus pages go out of use. | |
7893d1d5 | 1315 | */ |
a5516438 | 1316 | min_count = h->resv_huge_pages + h->nr_huge_pages - h->free_huge_pages; |
6b0c880d | 1317 | min_count = max(count, min_count); |
6ae11b27 | 1318 | try_to_free_low(h, min_count, nodes_allowed); |
a5516438 | 1319 | while (min_count < persistent_huge_pages(h)) { |
6ae11b27 | 1320 | if (!free_pool_huge_page(h, nodes_allowed, 0)) |
1da177e4 | 1321 | break; |
1da177e4 | 1322 | } |
a5516438 | 1323 | while (count < persistent_huge_pages(h)) { |
6ae11b27 | 1324 | if (!adjust_pool_surplus(h, nodes_allowed, 1)) |
7893d1d5 AL |
1325 | break; |
1326 | } | |
1327 | out: | |
a5516438 | 1328 | ret = persistent_huge_pages(h); |
1da177e4 | 1329 | spin_unlock(&hugetlb_lock); |
7893d1d5 | 1330 | return ret; |
1da177e4 LT |
1331 | } |
1332 | ||
a3437870 NA |
1333 | #define HSTATE_ATTR_RO(_name) \ |
1334 | static struct kobj_attribute _name##_attr = __ATTR_RO(_name) | |
1335 | ||
1336 | #define HSTATE_ATTR(_name) \ | |
1337 | static struct kobj_attribute _name##_attr = \ | |
1338 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
1339 | ||
1340 | static struct kobject *hugepages_kobj; | |
1341 | static struct kobject *hstate_kobjs[HUGE_MAX_HSTATE]; | |
1342 | ||
9a305230 LS |
1343 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp); |
1344 | ||
1345 | static struct hstate *kobj_to_hstate(struct kobject *kobj, int *nidp) | |
a3437870 NA |
1346 | { |
1347 | int i; | |
9a305230 | 1348 | |
a3437870 | 1349 | for (i = 0; i < HUGE_MAX_HSTATE; i++) |
9a305230 LS |
1350 | if (hstate_kobjs[i] == kobj) { |
1351 | if (nidp) | |
1352 | *nidp = NUMA_NO_NODE; | |
a3437870 | 1353 | return &hstates[i]; |
9a305230 LS |
1354 | } |
1355 | ||
1356 | return kobj_to_node_hstate(kobj, nidp); | |
a3437870 NA |
1357 | } |
1358 | ||
06808b08 | 1359 | static ssize_t nr_hugepages_show_common(struct kobject *kobj, |
a3437870 NA |
1360 | struct kobj_attribute *attr, char *buf) |
1361 | { | |
9a305230 LS |
1362 | struct hstate *h; |
1363 | unsigned long nr_huge_pages; | |
1364 | int nid; | |
1365 | ||
1366 | h = kobj_to_hstate(kobj, &nid); | |
1367 | if (nid == NUMA_NO_NODE) | |
1368 | nr_huge_pages = h->nr_huge_pages; | |
1369 | else | |
1370 | nr_huge_pages = h->nr_huge_pages_node[nid]; | |
1371 | ||
1372 | return sprintf(buf, "%lu\n", nr_huge_pages); | |
a3437870 | 1373 | } |
adbe8726 | 1374 | |
06808b08 LS |
1375 | static ssize_t nr_hugepages_store_common(bool obey_mempolicy, |
1376 | struct kobject *kobj, struct kobj_attribute *attr, | |
1377 | const char *buf, size_t len) | |
a3437870 NA |
1378 | { |
1379 | int err; | |
9a305230 | 1380 | int nid; |
06808b08 | 1381 | unsigned long count; |
9a305230 | 1382 | struct hstate *h; |
bad44b5b | 1383 | NODEMASK_ALLOC(nodemask_t, nodes_allowed, GFP_KERNEL | __GFP_NORETRY); |
a3437870 | 1384 | |
06808b08 | 1385 | err = strict_strtoul(buf, 10, &count); |
73ae31e5 | 1386 | if (err) |
adbe8726 | 1387 | goto out; |
a3437870 | 1388 | |
9a305230 | 1389 | h = kobj_to_hstate(kobj, &nid); |
adbe8726 EM |
1390 | if (h->order >= MAX_ORDER) { |
1391 | err = -EINVAL; | |
1392 | goto out; | |
1393 | } | |
1394 | ||
9a305230 LS |
1395 | if (nid == NUMA_NO_NODE) { |
1396 | /* | |
1397 | * global hstate attribute | |
1398 | */ | |
1399 | if (!(obey_mempolicy && | |
1400 | init_nodemask_of_mempolicy(nodes_allowed))) { | |
1401 | NODEMASK_FREE(nodes_allowed); | |
1402 | nodes_allowed = &node_states[N_HIGH_MEMORY]; | |
1403 | } | |
1404 | } else if (nodes_allowed) { | |
1405 | /* | |
1406 | * per node hstate attribute: adjust count to global, | |
1407 | * but restrict alloc/free to the specified node. | |
1408 | */ | |
1409 | count += h->nr_huge_pages - h->nr_huge_pages_node[nid]; | |
1410 | init_nodemask_of_node(nodes_allowed, nid); | |
1411 | } else | |
1412 | nodes_allowed = &node_states[N_HIGH_MEMORY]; | |
1413 | ||
06808b08 | 1414 | h->max_huge_pages = set_max_huge_pages(h, count, nodes_allowed); |
a3437870 | 1415 | |
9b5e5d0f | 1416 | if (nodes_allowed != &node_states[N_HIGH_MEMORY]) |
06808b08 LS |
1417 | NODEMASK_FREE(nodes_allowed); |
1418 | ||
1419 | return len; | |
adbe8726 EM |
1420 | out: |
1421 | NODEMASK_FREE(nodes_allowed); | |
1422 | return err; | |
06808b08 LS |
1423 | } |
1424 | ||
1425 | static ssize_t nr_hugepages_show(struct kobject *kobj, | |
1426 | struct kobj_attribute *attr, char *buf) | |
1427 | { | |
1428 | return nr_hugepages_show_common(kobj, attr, buf); | |
1429 | } | |
1430 | ||
1431 | static ssize_t nr_hugepages_store(struct kobject *kobj, | |
1432 | struct kobj_attribute *attr, const char *buf, size_t len) | |
1433 | { | |
1434 | return nr_hugepages_store_common(false, kobj, attr, buf, len); | |
a3437870 NA |
1435 | } |
1436 | HSTATE_ATTR(nr_hugepages); | |
1437 | ||
06808b08 LS |
1438 | #ifdef CONFIG_NUMA |
1439 | ||
1440 | /* | |
1441 | * hstate attribute for optionally mempolicy-based constraint on persistent | |
1442 | * huge page alloc/free. | |
1443 | */ | |
1444 | static ssize_t nr_hugepages_mempolicy_show(struct kobject *kobj, | |
1445 | struct kobj_attribute *attr, char *buf) | |
1446 | { | |
1447 | return nr_hugepages_show_common(kobj, attr, buf); | |
1448 | } | |
1449 | ||
1450 | static ssize_t nr_hugepages_mempolicy_store(struct kobject *kobj, | |
1451 | struct kobj_attribute *attr, const char *buf, size_t len) | |
1452 | { | |
1453 | return nr_hugepages_store_common(true, kobj, attr, buf, len); | |
1454 | } | |
1455 | HSTATE_ATTR(nr_hugepages_mempolicy); | |
1456 | #endif | |
1457 | ||
1458 | ||
a3437870 NA |
1459 | static ssize_t nr_overcommit_hugepages_show(struct kobject *kobj, |
1460 | struct kobj_attribute *attr, char *buf) | |
1461 | { | |
9a305230 | 1462 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
a3437870 NA |
1463 | return sprintf(buf, "%lu\n", h->nr_overcommit_huge_pages); |
1464 | } | |
adbe8726 | 1465 | |
a3437870 NA |
1466 | static ssize_t nr_overcommit_hugepages_store(struct kobject *kobj, |
1467 | struct kobj_attribute *attr, const char *buf, size_t count) | |
1468 | { | |
1469 | int err; | |
1470 | unsigned long input; | |
9a305230 | 1471 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
a3437870 | 1472 | |
adbe8726 EM |
1473 | if (h->order >= MAX_ORDER) |
1474 | return -EINVAL; | |
1475 | ||
a3437870 NA |
1476 | err = strict_strtoul(buf, 10, &input); |
1477 | if (err) | |
73ae31e5 | 1478 | return err; |
a3437870 NA |
1479 | |
1480 | spin_lock(&hugetlb_lock); | |
1481 | h->nr_overcommit_huge_pages = input; | |
1482 | spin_unlock(&hugetlb_lock); | |
1483 | ||
1484 | return count; | |
1485 | } | |
1486 | HSTATE_ATTR(nr_overcommit_hugepages); | |
1487 | ||
1488 | static ssize_t free_hugepages_show(struct kobject *kobj, | |
1489 | struct kobj_attribute *attr, char *buf) | |
1490 | { | |
9a305230 LS |
1491 | struct hstate *h; |
1492 | unsigned long free_huge_pages; | |
1493 | int nid; | |
1494 | ||
1495 | h = kobj_to_hstate(kobj, &nid); | |
1496 | if (nid == NUMA_NO_NODE) | |
1497 | free_huge_pages = h->free_huge_pages; | |
1498 | else | |
1499 | free_huge_pages = h->free_huge_pages_node[nid]; | |
1500 | ||
1501 | return sprintf(buf, "%lu\n", free_huge_pages); | |
a3437870 NA |
1502 | } |
1503 | HSTATE_ATTR_RO(free_hugepages); | |
1504 | ||
1505 | static ssize_t resv_hugepages_show(struct kobject *kobj, | |
1506 | struct kobj_attribute *attr, char *buf) | |
1507 | { | |
9a305230 | 1508 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
a3437870 NA |
1509 | return sprintf(buf, "%lu\n", h->resv_huge_pages); |
1510 | } | |
1511 | HSTATE_ATTR_RO(resv_hugepages); | |
1512 | ||
1513 | static ssize_t surplus_hugepages_show(struct kobject *kobj, | |
1514 | struct kobj_attribute *attr, char *buf) | |
1515 | { | |
9a305230 LS |
1516 | struct hstate *h; |
1517 | unsigned long surplus_huge_pages; | |
1518 | int nid; | |
1519 | ||
1520 | h = kobj_to_hstate(kobj, &nid); | |
1521 | if (nid == NUMA_NO_NODE) | |
1522 | surplus_huge_pages = h->surplus_huge_pages; | |
1523 | else | |
1524 | surplus_huge_pages = h->surplus_huge_pages_node[nid]; | |
1525 | ||
1526 | return sprintf(buf, "%lu\n", surplus_huge_pages); | |
a3437870 NA |
1527 | } |
1528 | HSTATE_ATTR_RO(surplus_hugepages); | |
1529 | ||
1530 | static struct attribute *hstate_attrs[] = { | |
1531 | &nr_hugepages_attr.attr, | |
1532 | &nr_overcommit_hugepages_attr.attr, | |
1533 | &free_hugepages_attr.attr, | |
1534 | &resv_hugepages_attr.attr, | |
1535 | &surplus_hugepages_attr.attr, | |
06808b08 LS |
1536 | #ifdef CONFIG_NUMA |
1537 | &nr_hugepages_mempolicy_attr.attr, | |
1538 | #endif | |
a3437870 NA |
1539 | NULL, |
1540 | }; | |
1541 | ||
1542 | static struct attribute_group hstate_attr_group = { | |
1543 | .attrs = hstate_attrs, | |
1544 | }; | |
1545 | ||
094e9539 JM |
1546 | static int hugetlb_sysfs_add_hstate(struct hstate *h, struct kobject *parent, |
1547 | struct kobject **hstate_kobjs, | |
1548 | struct attribute_group *hstate_attr_group) | |
a3437870 NA |
1549 | { |
1550 | int retval; | |
9a305230 | 1551 | int hi = h - hstates; |
a3437870 | 1552 | |
9a305230 LS |
1553 | hstate_kobjs[hi] = kobject_create_and_add(h->name, parent); |
1554 | if (!hstate_kobjs[hi]) | |
a3437870 NA |
1555 | return -ENOMEM; |
1556 | ||
9a305230 | 1557 | retval = sysfs_create_group(hstate_kobjs[hi], hstate_attr_group); |
a3437870 | 1558 | if (retval) |
9a305230 | 1559 | kobject_put(hstate_kobjs[hi]); |
a3437870 NA |
1560 | |
1561 | return retval; | |
1562 | } | |
1563 | ||
1564 | static void __init hugetlb_sysfs_init(void) | |
1565 | { | |
1566 | struct hstate *h; | |
1567 | int err; | |
1568 | ||
1569 | hugepages_kobj = kobject_create_and_add("hugepages", mm_kobj); | |
1570 | if (!hugepages_kobj) | |
1571 | return; | |
1572 | ||
1573 | for_each_hstate(h) { | |
9a305230 LS |
1574 | err = hugetlb_sysfs_add_hstate(h, hugepages_kobj, |
1575 | hstate_kobjs, &hstate_attr_group); | |
a3437870 NA |
1576 | if (err) |
1577 | printk(KERN_ERR "Hugetlb: Unable to add hstate %s", | |
1578 | h->name); | |
1579 | } | |
1580 | } | |
1581 | ||
9a305230 LS |
1582 | #ifdef CONFIG_NUMA |
1583 | ||
1584 | /* | |
1585 | * node_hstate/s - associate per node hstate attributes, via their kobjects, | |
1586 | * with node sysdevs in node_devices[] using a parallel array. The array | |
1587 | * index of a node sysdev or _hstate == node id. | |
1588 | * This is here to avoid any static dependency of the node sysdev driver, in | |
1589 | * the base kernel, on the hugetlb module. | |
1590 | */ | |
1591 | struct node_hstate { | |
1592 | struct kobject *hugepages_kobj; | |
1593 | struct kobject *hstate_kobjs[HUGE_MAX_HSTATE]; | |
1594 | }; | |
1595 | struct node_hstate node_hstates[MAX_NUMNODES]; | |
1596 | ||
1597 | /* | |
1598 | * A subset of global hstate attributes for node sysdevs | |
1599 | */ | |
1600 | static struct attribute *per_node_hstate_attrs[] = { | |
1601 | &nr_hugepages_attr.attr, | |
1602 | &free_hugepages_attr.attr, | |
1603 | &surplus_hugepages_attr.attr, | |
1604 | NULL, | |
1605 | }; | |
1606 | ||
1607 | static struct attribute_group per_node_hstate_attr_group = { | |
1608 | .attrs = per_node_hstate_attrs, | |
1609 | }; | |
1610 | ||
1611 | /* | |
1612 | * kobj_to_node_hstate - lookup global hstate for node sysdev hstate attr kobj. | |
1613 | * Returns node id via non-NULL nidp. | |
1614 | */ | |
1615 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp) | |
1616 | { | |
1617 | int nid; | |
1618 | ||
1619 | for (nid = 0; nid < nr_node_ids; nid++) { | |
1620 | struct node_hstate *nhs = &node_hstates[nid]; | |
1621 | int i; | |
1622 | for (i = 0; i < HUGE_MAX_HSTATE; i++) | |
1623 | if (nhs->hstate_kobjs[i] == kobj) { | |
1624 | if (nidp) | |
1625 | *nidp = nid; | |
1626 | return &hstates[i]; | |
1627 | } | |
1628 | } | |
1629 | ||
1630 | BUG(); | |
1631 | return NULL; | |
1632 | } | |
1633 | ||
1634 | /* | |
1635 | * Unregister hstate attributes from a single node sysdev. | |
1636 | * No-op if no hstate attributes attached. | |
1637 | */ | |
1638 | void hugetlb_unregister_node(struct node *node) | |
1639 | { | |
1640 | struct hstate *h; | |
1641 | struct node_hstate *nhs = &node_hstates[node->sysdev.id]; | |
1642 | ||
1643 | if (!nhs->hugepages_kobj) | |
9b5e5d0f | 1644 | return; /* no hstate attributes */ |
9a305230 LS |
1645 | |
1646 | for_each_hstate(h) | |
1647 | if (nhs->hstate_kobjs[h - hstates]) { | |
1648 | kobject_put(nhs->hstate_kobjs[h - hstates]); | |
1649 | nhs->hstate_kobjs[h - hstates] = NULL; | |
1650 | } | |
1651 | ||
1652 | kobject_put(nhs->hugepages_kobj); | |
1653 | nhs->hugepages_kobj = NULL; | |
1654 | } | |
1655 | ||
1656 | /* | |
1657 | * hugetlb module exit: unregister hstate attributes from node sysdevs | |
1658 | * that have them. | |
1659 | */ | |
1660 | static void hugetlb_unregister_all_nodes(void) | |
1661 | { | |
1662 | int nid; | |
1663 | ||
1664 | /* | |
1665 | * disable node sysdev registrations. | |
1666 | */ | |
1667 | register_hugetlbfs_with_node(NULL, NULL); | |
1668 | ||
1669 | /* | |
1670 | * remove hstate attributes from any nodes that have them. | |
1671 | */ | |
1672 | for (nid = 0; nid < nr_node_ids; nid++) | |
1673 | hugetlb_unregister_node(&node_devices[nid]); | |
1674 | } | |
1675 | ||
1676 | /* | |
1677 | * Register hstate attributes for a single node sysdev. | |
1678 | * No-op if attributes already registered. | |
1679 | */ | |
1680 | void hugetlb_register_node(struct node *node) | |
1681 | { | |
1682 | struct hstate *h; | |
1683 | struct node_hstate *nhs = &node_hstates[node->sysdev.id]; | |
1684 | int err; | |
1685 | ||
1686 | if (nhs->hugepages_kobj) | |
1687 | return; /* already allocated */ | |
1688 | ||
1689 | nhs->hugepages_kobj = kobject_create_and_add("hugepages", | |
1690 | &node->sysdev.kobj); | |
1691 | if (!nhs->hugepages_kobj) | |
1692 | return; | |
1693 | ||
1694 | for_each_hstate(h) { | |
1695 | err = hugetlb_sysfs_add_hstate(h, nhs->hugepages_kobj, | |
1696 | nhs->hstate_kobjs, | |
1697 | &per_node_hstate_attr_group); | |
1698 | if (err) { | |
1699 | printk(KERN_ERR "Hugetlb: Unable to add hstate %s" | |
1700 | " for node %d\n", | |
1701 | h->name, node->sysdev.id); | |
1702 | hugetlb_unregister_node(node); | |
1703 | break; | |
1704 | } | |
1705 | } | |
1706 | } | |
1707 | ||
1708 | /* | |
9b5e5d0f LS |
1709 | * hugetlb init time: register hstate attributes for all registered node |
1710 | * sysdevs of nodes that have memory. All on-line nodes should have | |
1711 | * registered their associated sysdev by this time. | |
9a305230 LS |
1712 | */ |
1713 | static void hugetlb_register_all_nodes(void) | |
1714 | { | |
1715 | int nid; | |
1716 | ||
9b5e5d0f | 1717 | for_each_node_state(nid, N_HIGH_MEMORY) { |
9a305230 LS |
1718 | struct node *node = &node_devices[nid]; |
1719 | if (node->sysdev.id == nid) | |
1720 | hugetlb_register_node(node); | |
1721 | } | |
1722 | ||
1723 | /* | |
1724 | * Let the node sysdev driver know we're here so it can | |
1725 | * [un]register hstate attributes on node hotplug. | |
1726 | */ | |
1727 | register_hugetlbfs_with_node(hugetlb_register_node, | |
1728 | hugetlb_unregister_node); | |
1729 | } | |
1730 | #else /* !CONFIG_NUMA */ | |
1731 | ||
1732 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp) | |
1733 | { | |
1734 | BUG(); | |
1735 | if (nidp) | |
1736 | *nidp = -1; | |
1737 | return NULL; | |
1738 | } | |
1739 | ||
1740 | static void hugetlb_unregister_all_nodes(void) { } | |
1741 | ||
1742 | static void hugetlb_register_all_nodes(void) { } | |
1743 | ||
1744 | #endif | |
1745 | ||
a3437870 NA |
1746 | static void __exit hugetlb_exit(void) |
1747 | { | |
1748 | struct hstate *h; | |
1749 | ||
9a305230 LS |
1750 | hugetlb_unregister_all_nodes(); |
1751 | ||
a3437870 NA |
1752 | for_each_hstate(h) { |
1753 | kobject_put(hstate_kobjs[h - hstates]); | |
1754 | } | |
1755 | ||
1756 | kobject_put(hugepages_kobj); | |
1757 | } | |
1758 | module_exit(hugetlb_exit); | |
1759 | ||
1760 | static int __init hugetlb_init(void) | |
1761 | { | |
0ef89d25 BH |
1762 | /* Some platform decide whether they support huge pages at boot |
1763 | * time. On these, such as powerpc, HPAGE_SHIFT is set to 0 when | |
1764 | * there is no such support | |
1765 | */ | |
1766 | if (HPAGE_SHIFT == 0) | |
1767 | return 0; | |
a3437870 | 1768 | |
e11bfbfc NP |
1769 | if (!size_to_hstate(default_hstate_size)) { |
1770 | default_hstate_size = HPAGE_SIZE; | |
1771 | if (!size_to_hstate(default_hstate_size)) | |
1772 | hugetlb_add_hstate(HUGETLB_PAGE_ORDER); | |
a3437870 | 1773 | } |
e11bfbfc NP |
1774 | default_hstate_idx = size_to_hstate(default_hstate_size) - hstates; |
1775 | if (default_hstate_max_huge_pages) | |
1776 | default_hstate.max_huge_pages = default_hstate_max_huge_pages; | |
a3437870 NA |
1777 | |
1778 | hugetlb_init_hstates(); | |
1779 | ||
aa888a74 AK |
1780 | gather_bootmem_prealloc(); |
1781 | ||
a3437870 NA |
1782 | report_hugepages(); |
1783 | ||
1784 | hugetlb_sysfs_init(); | |
1785 | ||
9a305230 LS |
1786 | hugetlb_register_all_nodes(); |
1787 | ||
a3437870 NA |
1788 | return 0; |
1789 | } | |
1790 | module_init(hugetlb_init); | |
1791 | ||
1792 | /* Should be called on processing a hugepagesz=... option */ | |
1793 | void __init hugetlb_add_hstate(unsigned order) | |
1794 | { | |
1795 | struct hstate *h; | |
8faa8b07 AK |
1796 | unsigned long i; |
1797 | ||
a3437870 NA |
1798 | if (size_to_hstate(PAGE_SIZE << order)) { |
1799 | printk(KERN_WARNING "hugepagesz= specified twice, ignoring\n"); | |
1800 | return; | |
1801 | } | |
1802 | BUG_ON(max_hstate >= HUGE_MAX_HSTATE); | |
1803 | BUG_ON(order == 0); | |
1804 | h = &hstates[max_hstate++]; | |
1805 | h->order = order; | |
1806 | h->mask = ~((1ULL << (order + PAGE_SHIFT)) - 1); | |
8faa8b07 AK |
1807 | h->nr_huge_pages = 0; |
1808 | h->free_huge_pages = 0; | |
1809 | for (i = 0; i < MAX_NUMNODES; ++i) | |
1810 | INIT_LIST_HEAD(&h->hugepage_freelists[i]); | |
9b5e5d0f LS |
1811 | h->next_nid_to_alloc = first_node(node_states[N_HIGH_MEMORY]); |
1812 | h->next_nid_to_free = first_node(node_states[N_HIGH_MEMORY]); | |
a3437870 NA |
1813 | snprintf(h->name, HSTATE_NAME_LEN, "hugepages-%lukB", |
1814 | huge_page_size(h)/1024); | |
8faa8b07 | 1815 | |
a3437870 NA |
1816 | parsed_hstate = h; |
1817 | } | |
1818 | ||
e11bfbfc | 1819 | static int __init hugetlb_nrpages_setup(char *s) |
a3437870 NA |
1820 | { |
1821 | unsigned long *mhp; | |
8faa8b07 | 1822 | static unsigned long *last_mhp; |
a3437870 NA |
1823 | |
1824 | /* | |
1825 | * !max_hstate means we haven't parsed a hugepagesz= parameter yet, | |
1826 | * so this hugepages= parameter goes to the "default hstate". | |
1827 | */ | |
1828 | if (!max_hstate) | |
1829 | mhp = &default_hstate_max_huge_pages; | |
1830 | else | |
1831 | mhp = &parsed_hstate->max_huge_pages; | |
1832 | ||
8faa8b07 AK |
1833 | if (mhp == last_mhp) { |
1834 | printk(KERN_WARNING "hugepages= specified twice without " | |
1835 | "interleaving hugepagesz=, ignoring\n"); | |
1836 | return 1; | |
1837 | } | |
1838 | ||
a3437870 NA |
1839 | if (sscanf(s, "%lu", mhp) <= 0) |
1840 | *mhp = 0; | |
1841 | ||
8faa8b07 AK |
1842 | /* |
1843 | * Global state is always initialized later in hugetlb_init. | |
1844 | * But we need to allocate >= MAX_ORDER hstates here early to still | |
1845 | * use the bootmem allocator. | |
1846 | */ | |
1847 | if (max_hstate && parsed_hstate->order >= MAX_ORDER) | |
1848 | hugetlb_hstate_alloc_pages(parsed_hstate); | |
1849 | ||
1850 | last_mhp = mhp; | |
1851 | ||
a3437870 NA |
1852 | return 1; |
1853 | } | |
e11bfbfc NP |
1854 | __setup("hugepages=", hugetlb_nrpages_setup); |
1855 | ||
1856 | static int __init hugetlb_default_setup(char *s) | |
1857 | { | |
1858 | default_hstate_size = memparse(s, &s); | |
1859 | return 1; | |
1860 | } | |
1861 | __setup("default_hugepagesz=", hugetlb_default_setup); | |
a3437870 | 1862 | |
8a213460 NA |
1863 | static unsigned int cpuset_mems_nr(unsigned int *array) |
1864 | { | |
1865 | int node; | |
1866 | unsigned int nr = 0; | |
1867 | ||
1868 | for_each_node_mask(node, cpuset_current_mems_allowed) | |
1869 | nr += array[node]; | |
1870 | ||
1871 | return nr; | |
1872 | } | |
1873 | ||
1874 | #ifdef CONFIG_SYSCTL | |
06808b08 LS |
1875 | static int hugetlb_sysctl_handler_common(bool obey_mempolicy, |
1876 | struct ctl_table *table, int write, | |
1877 | void __user *buffer, size_t *length, loff_t *ppos) | |
1da177e4 | 1878 | { |
e5ff2159 AK |
1879 | struct hstate *h = &default_hstate; |
1880 | unsigned long tmp; | |
08d4a246 | 1881 | int ret; |
e5ff2159 | 1882 | |
c033a93c | 1883 | tmp = h->max_huge_pages; |
e5ff2159 | 1884 | |
adbe8726 EM |
1885 | if (write && h->order >= MAX_ORDER) |
1886 | return -EINVAL; | |
1887 | ||
e5ff2159 AK |
1888 | table->data = &tmp; |
1889 | table->maxlen = sizeof(unsigned long); | |
08d4a246 MH |
1890 | ret = proc_doulongvec_minmax(table, write, buffer, length, ppos); |
1891 | if (ret) | |
1892 | goto out; | |
e5ff2159 | 1893 | |
06808b08 | 1894 | if (write) { |
bad44b5b DR |
1895 | NODEMASK_ALLOC(nodemask_t, nodes_allowed, |
1896 | GFP_KERNEL | __GFP_NORETRY); | |
06808b08 LS |
1897 | if (!(obey_mempolicy && |
1898 | init_nodemask_of_mempolicy(nodes_allowed))) { | |
1899 | NODEMASK_FREE(nodes_allowed); | |
1900 | nodes_allowed = &node_states[N_HIGH_MEMORY]; | |
1901 | } | |
1902 | h->max_huge_pages = set_max_huge_pages(h, tmp, nodes_allowed); | |
1903 | ||
1904 | if (nodes_allowed != &node_states[N_HIGH_MEMORY]) | |
1905 | NODEMASK_FREE(nodes_allowed); | |
1906 | } | |
08d4a246 MH |
1907 | out: |
1908 | return ret; | |
1da177e4 | 1909 | } |
396faf03 | 1910 | |
06808b08 LS |
1911 | int hugetlb_sysctl_handler(struct ctl_table *table, int write, |
1912 | void __user *buffer, size_t *length, loff_t *ppos) | |
1913 | { | |
1914 | ||
1915 | return hugetlb_sysctl_handler_common(false, table, write, | |
1916 | buffer, length, ppos); | |
1917 | } | |
1918 | ||
1919 | #ifdef CONFIG_NUMA | |
1920 | int hugetlb_mempolicy_sysctl_handler(struct ctl_table *table, int write, | |
1921 | void __user *buffer, size_t *length, loff_t *ppos) | |
1922 | { | |
1923 | return hugetlb_sysctl_handler_common(true, table, write, | |
1924 | buffer, length, ppos); | |
1925 | } | |
1926 | #endif /* CONFIG_NUMA */ | |
1927 | ||
396faf03 | 1928 | int hugetlb_treat_movable_handler(struct ctl_table *table, int write, |
8d65af78 | 1929 | void __user *buffer, |
396faf03 MG |
1930 | size_t *length, loff_t *ppos) |
1931 | { | |
8d65af78 | 1932 | proc_dointvec(table, write, buffer, length, ppos); |
396faf03 MG |
1933 | if (hugepages_treat_as_movable) |
1934 | htlb_alloc_mask = GFP_HIGHUSER_MOVABLE; | |
1935 | else | |
1936 | htlb_alloc_mask = GFP_HIGHUSER; | |
1937 | return 0; | |
1938 | } | |
1939 | ||
a3d0c6aa | 1940 | int hugetlb_overcommit_handler(struct ctl_table *table, int write, |
8d65af78 | 1941 | void __user *buffer, |
a3d0c6aa NA |
1942 | size_t *length, loff_t *ppos) |
1943 | { | |
a5516438 | 1944 | struct hstate *h = &default_hstate; |
e5ff2159 | 1945 | unsigned long tmp; |
08d4a246 | 1946 | int ret; |
e5ff2159 | 1947 | |
c033a93c | 1948 | tmp = h->nr_overcommit_huge_pages; |
e5ff2159 | 1949 | |
adbe8726 EM |
1950 | if (write && h->order >= MAX_ORDER) |
1951 | return -EINVAL; | |
1952 | ||
e5ff2159 AK |
1953 | table->data = &tmp; |
1954 | table->maxlen = sizeof(unsigned long); | |
08d4a246 MH |
1955 | ret = proc_doulongvec_minmax(table, write, buffer, length, ppos); |
1956 | if (ret) | |
1957 | goto out; | |
e5ff2159 AK |
1958 | |
1959 | if (write) { | |
1960 | spin_lock(&hugetlb_lock); | |
1961 | h->nr_overcommit_huge_pages = tmp; | |
1962 | spin_unlock(&hugetlb_lock); | |
1963 | } | |
08d4a246 MH |
1964 | out: |
1965 | return ret; | |
a3d0c6aa NA |
1966 | } |
1967 | ||
1da177e4 LT |
1968 | #endif /* CONFIG_SYSCTL */ |
1969 | ||
e1759c21 | 1970 | void hugetlb_report_meminfo(struct seq_file *m) |
1da177e4 | 1971 | { |
a5516438 | 1972 | struct hstate *h = &default_hstate; |
e1759c21 | 1973 | seq_printf(m, |
4f98a2fe RR |
1974 | "HugePages_Total: %5lu\n" |
1975 | "HugePages_Free: %5lu\n" | |
1976 | "HugePages_Rsvd: %5lu\n" | |
1977 | "HugePages_Surp: %5lu\n" | |
1978 | "Hugepagesize: %8lu kB\n", | |
a5516438 AK |
1979 | h->nr_huge_pages, |
1980 | h->free_huge_pages, | |
1981 | h->resv_huge_pages, | |
1982 | h->surplus_huge_pages, | |
1983 | 1UL << (huge_page_order(h) + PAGE_SHIFT - 10)); | |
1da177e4 LT |
1984 | } |
1985 | ||
1986 | int hugetlb_report_node_meminfo(int nid, char *buf) | |
1987 | { | |
a5516438 | 1988 | struct hstate *h = &default_hstate; |
1da177e4 LT |
1989 | return sprintf(buf, |
1990 | "Node %d HugePages_Total: %5u\n" | |
a1de0919 NA |
1991 | "Node %d HugePages_Free: %5u\n" |
1992 | "Node %d HugePages_Surp: %5u\n", | |
a5516438 AK |
1993 | nid, h->nr_huge_pages_node[nid], |
1994 | nid, h->free_huge_pages_node[nid], | |
1995 | nid, h->surplus_huge_pages_node[nid]); | |
1da177e4 LT |
1996 | } |
1997 | ||
1da177e4 LT |
1998 | /* Return the number pages of memory we physically have, in PAGE_SIZE units. */ |
1999 | unsigned long hugetlb_total_pages(void) | |
2000 | { | |
a5516438 AK |
2001 | struct hstate *h = &default_hstate; |
2002 | return h->nr_huge_pages * pages_per_huge_page(h); | |
1da177e4 | 2003 | } |
1da177e4 | 2004 | |
a5516438 | 2005 | static int hugetlb_acct_memory(struct hstate *h, long delta) |
fc1b8a73 MG |
2006 | { |
2007 | int ret = -ENOMEM; | |
2008 | ||
2009 | spin_lock(&hugetlb_lock); | |
2010 | /* | |
2011 | * When cpuset is configured, it breaks the strict hugetlb page | |
2012 | * reservation as the accounting is done on a global variable. Such | |
2013 | * reservation is completely rubbish in the presence of cpuset because | |
2014 | * the reservation is not checked against page availability for the | |
2015 | * current cpuset. Application can still potentially OOM'ed by kernel | |
2016 | * with lack of free htlb page in cpuset that the task is in. | |
2017 | * Attempt to enforce strict accounting with cpuset is almost | |
2018 | * impossible (or too ugly) because cpuset is too fluid that | |
2019 | * task or memory node can be dynamically moved between cpusets. | |
2020 | * | |
2021 | * The change of semantics for shared hugetlb mapping with cpuset is | |
2022 | * undesirable. However, in order to preserve some of the semantics, | |
2023 | * we fall back to check against current free page availability as | |
2024 | * a best attempt and hopefully to minimize the impact of changing | |
2025 | * semantics that cpuset has. | |
2026 | */ | |
2027 | if (delta > 0) { | |
a5516438 | 2028 | if (gather_surplus_pages(h, delta) < 0) |
fc1b8a73 MG |
2029 | goto out; |
2030 | ||
a5516438 AK |
2031 | if (delta > cpuset_mems_nr(h->free_huge_pages_node)) { |
2032 | return_unused_surplus_pages(h, delta); | |
fc1b8a73 MG |
2033 | goto out; |
2034 | } | |
2035 | } | |
2036 | ||
2037 | ret = 0; | |
2038 | if (delta < 0) | |
a5516438 | 2039 | return_unused_surplus_pages(h, (unsigned long) -delta); |
fc1b8a73 MG |
2040 | |
2041 | out: | |
2042 | spin_unlock(&hugetlb_lock); | |
2043 | return ret; | |
2044 | } | |
2045 | ||
84afd99b AW |
2046 | static void hugetlb_vm_op_open(struct vm_area_struct *vma) |
2047 | { | |
2048 | struct resv_map *reservations = vma_resv_map(vma); | |
2049 | ||
2050 | /* | |
2051 | * This new VMA should share its siblings reservation map if present. | |
2052 | * The VMA will only ever have a valid reservation map pointer where | |
2053 | * it is being copied for another still existing VMA. As that VMA | |
25985edc | 2054 | * has a reference to the reservation map it cannot disappear until |
84afd99b AW |
2055 | * after this open call completes. It is therefore safe to take a |
2056 | * new reference here without additional locking. | |
2057 | */ | |
2058 | if (reservations) | |
2059 | kref_get(&reservations->refs); | |
2060 | } | |
2061 | ||
a1e78772 MG |
2062 | static void hugetlb_vm_op_close(struct vm_area_struct *vma) |
2063 | { | |
a5516438 | 2064 | struct hstate *h = hstate_vma(vma); |
84afd99b AW |
2065 | struct resv_map *reservations = vma_resv_map(vma); |
2066 | unsigned long reserve; | |
2067 | unsigned long start; | |
2068 | unsigned long end; | |
2069 | ||
2070 | if (reservations) { | |
a5516438 AK |
2071 | start = vma_hugecache_offset(h, vma, vma->vm_start); |
2072 | end = vma_hugecache_offset(h, vma, vma->vm_end); | |
84afd99b AW |
2073 | |
2074 | reserve = (end - start) - | |
2075 | region_count(&reservations->regions, start, end); | |
2076 | ||
2077 | kref_put(&reservations->refs, resv_map_release); | |
2078 | ||
7251ff78 | 2079 | if (reserve) { |
a5516438 | 2080 | hugetlb_acct_memory(h, -reserve); |
7251ff78 AL |
2081 | hugetlb_put_quota(vma->vm_file->f_mapping, reserve); |
2082 | } | |
84afd99b | 2083 | } |
a1e78772 MG |
2084 | } |
2085 | ||
1da177e4 LT |
2086 | /* |
2087 | * We cannot handle pagefaults against hugetlb pages at all. They cause | |
2088 | * handle_mm_fault() to try to instantiate regular-sized pages in the | |
2089 | * hugegpage VMA. do_page_fault() is supposed to trap this, so BUG is we get | |
2090 | * this far. | |
2091 | */ | |
d0217ac0 | 2092 | static int hugetlb_vm_op_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
2093 | { |
2094 | BUG(); | |
d0217ac0 | 2095 | return 0; |
1da177e4 LT |
2096 | } |
2097 | ||
f0f37e2f | 2098 | const struct vm_operations_struct hugetlb_vm_ops = { |
d0217ac0 | 2099 | .fault = hugetlb_vm_op_fault, |
84afd99b | 2100 | .open = hugetlb_vm_op_open, |
a1e78772 | 2101 | .close = hugetlb_vm_op_close, |
1da177e4 LT |
2102 | }; |
2103 | ||
1e8f889b DG |
2104 | static pte_t make_huge_pte(struct vm_area_struct *vma, struct page *page, |
2105 | int writable) | |
63551ae0 DG |
2106 | { |
2107 | pte_t entry; | |
2108 | ||
1e8f889b | 2109 | if (writable) { |
63551ae0 DG |
2110 | entry = |
2111 | pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot))); | |
2112 | } else { | |
7f2e9525 | 2113 | entry = huge_pte_wrprotect(mk_pte(page, vma->vm_page_prot)); |
63551ae0 DG |
2114 | } |
2115 | entry = pte_mkyoung(entry); | |
2116 | entry = pte_mkhuge(entry); | |
2117 | ||
2118 | return entry; | |
2119 | } | |
2120 | ||
1e8f889b DG |
2121 | static void set_huge_ptep_writable(struct vm_area_struct *vma, |
2122 | unsigned long address, pte_t *ptep) | |
2123 | { | |
2124 | pte_t entry; | |
2125 | ||
7f2e9525 GS |
2126 | entry = pte_mkwrite(pte_mkdirty(huge_ptep_get(ptep))); |
2127 | if (huge_ptep_set_access_flags(vma, address, ptep, entry, 1)) { | |
4b3073e1 | 2128 | update_mmu_cache(vma, address, ptep); |
8dab5241 | 2129 | } |
1e8f889b DG |
2130 | } |
2131 | ||
2132 | ||
63551ae0 DG |
2133 | int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src, |
2134 | struct vm_area_struct *vma) | |
2135 | { | |
2136 | pte_t *src_pte, *dst_pte, entry; | |
2137 | struct page *ptepage; | |
1c59827d | 2138 | unsigned long addr; |
1e8f889b | 2139 | int cow; |
a5516438 AK |
2140 | struct hstate *h = hstate_vma(vma); |
2141 | unsigned long sz = huge_page_size(h); | |
1e8f889b DG |
2142 | |
2143 | cow = (vma->vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; | |
63551ae0 | 2144 | |
a5516438 | 2145 | for (addr = vma->vm_start; addr < vma->vm_end; addr += sz) { |
c74df32c HD |
2146 | src_pte = huge_pte_offset(src, addr); |
2147 | if (!src_pte) | |
2148 | continue; | |
a5516438 | 2149 | dst_pte = huge_pte_alloc(dst, addr, sz); |
63551ae0 DG |
2150 | if (!dst_pte) |
2151 | goto nomem; | |
c5c99429 LW |
2152 | |
2153 | /* If the pagetables are shared don't copy or take references */ | |
2154 | if (dst_pte == src_pte) | |
2155 | continue; | |
2156 | ||
c74df32c | 2157 | spin_lock(&dst->page_table_lock); |
46478758 | 2158 | spin_lock_nested(&src->page_table_lock, SINGLE_DEPTH_NESTING); |
7f2e9525 | 2159 | if (!huge_pte_none(huge_ptep_get(src_pte))) { |
1e8f889b | 2160 | if (cow) |
7f2e9525 GS |
2161 | huge_ptep_set_wrprotect(src, addr, src_pte); |
2162 | entry = huge_ptep_get(src_pte); | |
1c59827d HD |
2163 | ptepage = pte_page(entry); |
2164 | get_page(ptepage); | |
0fe6e20b | 2165 | page_dup_rmap(ptepage); |
1c59827d HD |
2166 | set_huge_pte_at(dst, addr, dst_pte, entry); |
2167 | } | |
2168 | spin_unlock(&src->page_table_lock); | |
c74df32c | 2169 | spin_unlock(&dst->page_table_lock); |
63551ae0 DG |
2170 | } |
2171 | return 0; | |
2172 | ||
2173 | nomem: | |
2174 | return -ENOMEM; | |
2175 | } | |
2176 | ||
290408d4 NH |
2177 | static int is_hugetlb_entry_migration(pte_t pte) |
2178 | { | |
2179 | swp_entry_t swp; | |
2180 | ||
2181 | if (huge_pte_none(pte) || pte_present(pte)) | |
2182 | return 0; | |
2183 | swp = pte_to_swp_entry(pte); | |
2184 | if (non_swap_entry(swp) && is_migration_entry(swp)) { | |
2185 | return 1; | |
2186 | } else | |
2187 | return 0; | |
2188 | } | |
2189 | ||
fd6a03ed NH |
2190 | static int is_hugetlb_entry_hwpoisoned(pte_t pte) |
2191 | { | |
2192 | swp_entry_t swp; | |
2193 | ||
2194 | if (huge_pte_none(pte) || pte_present(pte)) | |
2195 | return 0; | |
2196 | swp = pte_to_swp_entry(pte); | |
2197 | if (non_swap_entry(swp) && is_hwpoison_entry(swp)) { | |
2198 | return 1; | |
2199 | } else | |
2200 | return 0; | |
2201 | } | |
2202 | ||
502717f4 | 2203 | void __unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start, |
04f2cbe3 | 2204 | unsigned long end, struct page *ref_page) |
63551ae0 DG |
2205 | { |
2206 | struct mm_struct *mm = vma->vm_mm; | |
2207 | unsigned long address; | |
c7546f8f | 2208 | pte_t *ptep; |
63551ae0 DG |
2209 | pte_t pte; |
2210 | struct page *page; | |
fe1668ae | 2211 | struct page *tmp; |
a5516438 AK |
2212 | struct hstate *h = hstate_vma(vma); |
2213 | unsigned long sz = huge_page_size(h); | |
2214 | ||
c0a499c2 | 2215 | /* |
3d48ae45 | 2216 | * A page gathering list, protected by per file i_mmap_mutex. The |
c0a499c2 KC |
2217 | * lock is used to avoid list corruption from multiple unmapping |
2218 | * of the same page since we are using page->lru. | |
2219 | */ | |
fe1668ae | 2220 | LIST_HEAD(page_list); |
63551ae0 DG |
2221 | |
2222 | WARN_ON(!is_vm_hugetlb_page(vma)); | |
a5516438 AK |
2223 | BUG_ON(start & ~huge_page_mask(h)); |
2224 | BUG_ON(end & ~huge_page_mask(h)); | |
63551ae0 | 2225 | |
cddb8a5c | 2226 | mmu_notifier_invalidate_range_start(mm, start, end); |
508034a3 | 2227 | spin_lock(&mm->page_table_lock); |
a5516438 | 2228 | for (address = start; address < end; address += sz) { |
c7546f8f | 2229 | ptep = huge_pte_offset(mm, address); |
4c887265 | 2230 | if (!ptep) |
c7546f8f DG |
2231 | continue; |
2232 | ||
39dde65c KC |
2233 | if (huge_pmd_unshare(mm, &address, ptep)) |
2234 | continue; | |
2235 | ||
04f2cbe3 MG |
2236 | /* |
2237 | * If a reference page is supplied, it is because a specific | |
2238 | * page is being unmapped, not a range. Ensure the page we | |
2239 | * are about to unmap is the actual page of interest. | |
2240 | */ | |
2241 | if (ref_page) { | |
2242 | pte = huge_ptep_get(ptep); | |
2243 | if (huge_pte_none(pte)) | |
2244 | continue; | |
2245 | page = pte_page(pte); | |
2246 | if (page != ref_page) | |
2247 | continue; | |
2248 | ||
2249 | /* | |
2250 | * Mark the VMA as having unmapped its page so that | |
2251 | * future faults in this VMA will fail rather than | |
2252 | * looking like data was lost | |
2253 | */ | |
2254 | set_vma_resv_flags(vma, HPAGE_RESV_UNMAPPED); | |
2255 | } | |
2256 | ||
c7546f8f | 2257 | pte = huge_ptep_get_and_clear(mm, address, ptep); |
7f2e9525 | 2258 | if (huge_pte_none(pte)) |
63551ae0 | 2259 | continue; |
c7546f8f | 2260 | |
fd6a03ed NH |
2261 | /* |
2262 | * HWPoisoned hugepage is already unmapped and dropped reference | |
2263 | */ | |
2264 | if (unlikely(is_hugetlb_entry_hwpoisoned(pte))) | |
2265 | continue; | |
2266 | ||
63551ae0 | 2267 | page = pte_page(pte); |
6649a386 KC |
2268 | if (pte_dirty(pte)) |
2269 | set_page_dirty(page); | |
fe1668ae | 2270 | list_add(&page->lru, &page_list); |
63551ae0 | 2271 | } |
1da177e4 | 2272 | spin_unlock(&mm->page_table_lock); |
508034a3 | 2273 | flush_tlb_range(vma, start, end); |
cddb8a5c | 2274 | mmu_notifier_invalidate_range_end(mm, start, end); |
fe1668ae | 2275 | list_for_each_entry_safe(page, tmp, &page_list, lru) { |
0fe6e20b | 2276 | page_remove_rmap(page); |
fe1668ae KC |
2277 | list_del(&page->lru); |
2278 | put_page(page); | |
2279 | } | |
1da177e4 | 2280 | } |
63551ae0 | 2281 | |
502717f4 | 2282 | void unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start, |
04f2cbe3 | 2283 | unsigned long end, struct page *ref_page) |
502717f4 | 2284 | { |
3d48ae45 | 2285 | mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); |
a137e1cc | 2286 | __unmap_hugepage_range(vma, start, end, ref_page); |
3d48ae45 | 2287 | mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); |
502717f4 KC |
2288 | } |
2289 | ||
04f2cbe3 MG |
2290 | /* |
2291 | * This is called when the original mapper is failing to COW a MAP_PRIVATE | |
2292 | * mappping it owns the reserve page for. The intention is to unmap the page | |
2293 | * from other VMAs and let the children be SIGKILLed if they are faulting the | |
2294 | * same region. | |
2295 | */ | |
2a4b3ded HH |
2296 | static int unmap_ref_private(struct mm_struct *mm, struct vm_area_struct *vma, |
2297 | struct page *page, unsigned long address) | |
04f2cbe3 | 2298 | { |
7526674d | 2299 | struct hstate *h = hstate_vma(vma); |
04f2cbe3 MG |
2300 | struct vm_area_struct *iter_vma; |
2301 | struct address_space *mapping; | |
2302 | struct prio_tree_iter iter; | |
2303 | pgoff_t pgoff; | |
2304 | ||
2305 | /* | |
2306 | * vm_pgoff is in PAGE_SIZE units, hence the different calculation | |
2307 | * from page cache lookup which is in HPAGE_SIZE units. | |
2308 | */ | |
7526674d | 2309 | address = address & huge_page_mask(h); |
04f2cbe3 MG |
2310 | pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) |
2311 | + (vma->vm_pgoff >> PAGE_SHIFT); | |
2312 | mapping = (struct address_space *)page_private(page); | |
2313 | ||
4eb2b1dc MG |
2314 | /* |
2315 | * Take the mapping lock for the duration of the table walk. As | |
2316 | * this mapping should be shared between all the VMAs, | |
2317 | * __unmap_hugepage_range() is called as the lock is already held | |
2318 | */ | |
3d48ae45 | 2319 | mutex_lock(&mapping->i_mmap_mutex); |
04f2cbe3 MG |
2320 | vma_prio_tree_foreach(iter_vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
2321 | /* Do not unmap the current VMA */ | |
2322 | if (iter_vma == vma) | |
2323 | continue; | |
2324 | ||
2325 | /* | |
2326 | * Unmap the page from other VMAs without their own reserves. | |
2327 | * They get marked to be SIGKILLed if they fault in these | |
2328 | * areas. This is because a future no-page fault on this VMA | |
2329 | * could insert a zeroed page instead of the data existing | |
2330 | * from the time of fork. This would look like data corruption | |
2331 | */ | |
2332 | if (!is_vma_resv_set(iter_vma, HPAGE_RESV_OWNER)) | |
4eb2b1dc | 2333 | __unmap_hugepage_range(iter_vma, |
7526674d | 2334 | address, address + huge_page_size(h), |
04f2cbe3 MG |
2335 | page); |
2336 | } | |
3d48ae45 | 2337 | mutex_unlock(&mapping->i_mmap_mutex); |
04f2cbe3 MG |
2338 | |
2339 | return 1; | |
2340 | } | |
2341 | ||
0fe6e20b NH |
2342 | /* |
2343 | * Hugetlb_cow() should be called with page lock of the original hugepage held. | |
2344 | */ | |
1e8f889b | 2345 | static int hugetlb_cow(struct mm_struct *mm, struct vm_area_struct *vma, |
04f2cbe3 MG |
2346 | unsigned long address, pte_t *ptep, pte_t pte, |
2347 | struct page *pagecache_page) | |
1e8f889b | 2348 | { |
a5516438 | 2349 | struct hstate *h = hstate_vma(vma); |
1e8f889b | 2350 | struct page *old_page, *new_page; |
79ac6ba4 | 2351 | int avoidcopy; |
04f2cbe3 | 2352 | int outside_reserve = 0; |
1e8f889b DG |
2353 | |
2354 | old_page = pte_page(pte); | |
2355 | ||
04f2cbe3 | 2356 | retry_avoidcopy: |
1e8f889b DG |
2357 | /* If no-one else is actually using this page, avoid the copy |
2358 | * and just make the page writable */ | |
0fe6e20b | 2359 | avoidcopy = (page_mapcount(old_page) == 1); |
1e8f889b | 2360 | if (avoidcopy) { |
56c9cfb1 NH |
2361 | if (PageAnon(old_page)) |
2362 | page_move_anon_rmap(old_page, vma, address); | |
1e8f889b | 2363 | set_huge_ptep_writable(vma, address, ptep); |
83c54070 | 2364 | return 0; |
1e8f889b DG |
2365 | } |
2366 | ||
04f2cbe3 MG |
2367 | /* |
2368 | * If the process that created a MAP_PRIVATE mapping is about to | |
2369 | * perform a COW due to a shared page count, attempt to satisfy | |
2370 | * the allocation without using the existing reserves. The pagecache | |
2371 | * page is used to determine if the reserve at this address was | |
2372 | * consumed or not. If reserves were used, a partial faulted mapping | |
2373 | * at the time of fork() could consume its reserves on COW instead | |
2374 | * of the full address range. | |
2375 | */ | |
f83a275d | 2376 | if (!(vma->vm_flags & VM_MAYSHARE) && |
04f2cbe3 MG |
2377 | is_vma_resv_set(vma, HPAGE_RESV_OWNER) && |
2378 | old_page != pagecache_page) | |
2379 | outside_reserve = 1; | |
2380 | ||
1e8f889b | 2381 | page_cache_get(old_page); |
b76c8cfb LW |
2382 | |
2383 | /* Drop page_table_lock as buddy allocator may be called */ | |
2384 | spin_unlock(&mm->page_table_lock); | |
04f2cbe3 | 2385 | new_page = alloc_huge_page(vma, address, outside_reserve); |
1e8f889b | 2386 | |
2fc39cec | 2387 | if (IS_ERR(new_page)) { |
1e8f889b | 2388 | page_cache_release(old_page); |
04f2cbe3 MG |
2389 | |
2390 | /* | |
2391 | * If a process owning a MAP_PRIVATE mapping fails to COW, | |
2392 | * it is due to references held by a child and an insufficient | |
2393 | * huge page pool. To guarantee the original mappers | |
2394 | * reliability, unmap the page from child processes. The child | |
2395 | * may get SIGKILLed if it later faults. | |
2396 | */ | |
2397 | if (outside_reserve) { | |
2398 | BUG_ON(huge_pte_none(pte)); | |
2399 | if (unmap_ref_private(mm, vma, old_page, address)) { | |
2400 | BUG_ON(page_count(old_page) != 1); | |
2401 | BUG_ON(huge_pte_none(pte)); | |
b76c8cfb | 2402 | spin_lock(&mm->page_table_lock); |
04f2cbe3 MG |
2403 | goto retry_avoidcopy; |
2404 | } | |
2405 | WARN_ON_ONCE(1); | |
2406 | } | |
2407 | ||
b76c8cfb LW |
2408 | /* Caller expects lock to be held */ |
2409 | spin_lock(&mm->page_table_lock); | |
2fc39cec | 2410 | return -PTR_ERR(new_page); |
1e8f889b DG |
2411 | } |
2412 | ||
0fe6e20b NH |
2413 | /* |
2414 | * When the original hugepage is shared one, it does not have | |
2415 | * anon_vma prepared. | |
2416 | */ | |
44e2aa93 DN |
2417 | if (unlikely(anon_vma_prepare(vma))) { |
2418 | /* Caller expects lock to be held */ | |
2419 | spin_lock(&mm->page_table_lock); | |
0fe6e20b | 2420 | return VM_FAULT_OOM; |
44e2aa93 | 2421 | } |
0fe6e20b | 2422 | |
47ad8475 AA |
2423 | copy_user_huge_page(new_page, old_page, address, vma, |
2424 | pages_per_huge_page(h)); | |
0ed361de | 2425 | __SetPageUptodate(new_page); |
1e8f889b | 2426 | |
b76c8cfb LW |
2427 | /* |
2428 | * Retake the page_table_lock to check for racing updates | |
2429 | * before the page tables are altered | |
2430 | */ | |
2431 | spin_lock(&mm->page_table_lock); | |
a5516438 | 2432 | ptep = huge_pte_offset(mm, address & huge_page_mask(h)); |
7f2e9525 | 2433 | if (likely(pte_same(huge_ptep_get(ptep), pte))) { |
1e8f889b | 2434 | /* Break COW */ |
3edd4fc9 DD |
2435 | mmu_notifier_invalidate_range_start(mm, |
2436 | address & huge_page_mask(h), | |
2437 | (address & huge_page_mask(h)) + huge_page_size(h)); | |
8fe627ec | 2438 | huge_ptep_clear_flush(vma, address, ptep); |
1e8f889b DG |
2439 | set_huge_pte_at(mm, address, ptep, |
2440 | make_huge_pte(vma, new_page, 1)); | |
0fe6e20b | 2441 | page_remove_rmap(old_page); |
cd67f0d2 | 2442 | hugepage_add_new_anon_rmap(new_page, vma, address); |
1e8f889b DG |
2443 | /* Make the old page be freed below */ |
2444 | new_page = old_page; | |
3edd4fc9 DD |
2445 | mmu_notifier_invalidate_range_end(mm, |
2446 | address & huge_page_mask(h), | |
2447 | (address & huge_page_mask(h)) + huge_page_size(h)); | |
1e8f889b DG |
2448 | } |
2449 | page_cache_release(new_page); | |
2450 | page_cache_release(old_page); | |
83c54070 | 2451 | return 0; |
1e8f889b DG |
2452 | } |
2453 | ||
04f2cbe3 | 2454 | /* Return the pagecache page at a given address within a VMA */ |
a5516438 AK |
2455 | static struct page *hugetlbfs_pagecache_page(struct hstate *h, |
2456 | struct vm_area_struct *vma, unsigned long address) | |
04f2cbe3 MG |
2457 | { |
2458 | struct address_space *mapping; | |
e7c4b0bf | 2459 | pgoff_t idx; |
04f2cbe3 MG |
2460 | |
2461 | mapping = vma->vm_file->f_mapping; | |
a5516438 | 2462 | idx = vma_hugecache_offset(h, vma, address); |
04f2cbe3 MG |
2463 | |
2464 | return find_lock_page(mapping, idx); | |
2465 | } | |
2466 | ||
3ae77f43 HD |
2467 | /* |
2468 | * Return whether there is a pagecache page to back given address within VMA. | |
2469 | * Caller follow_hugetlb_page() holds page_table_lock so we cannot lock_page. | |
2470 | */ | |
2471 | static bool hugetlbfs_pagecache_present(struct hstate *h, | |
2a15efc9 HD |
2472 | struct vm_area_struct *vma, unsigned long address) |
2473 | { | |
2474 | struct address_space *mapping; | |
2475 | pgoff_t idx; | |
2476 | struct page *page; | |
2477 | ||
2478 | mapping = vma->vm_file->f_mapping; | |
2479 | idx = vma_hugecache_offset(h, vma, address); | |
2480 | ||
2481 | page = find_get_page(mapping, idx); | |
2482 | if (page) | |
2483 | put_page(page); | |
2484 | return page != NULL; | |
2485 | } | |
2486 | ||
a1ed3dda | 2487 | static int hugetlb_no_page(struct mm_struct *mm, struct vm_area_struct *vma, |
788c7df4 | 2488 | unsigned long address, pte_t *ptep, unsigned int flags) |
ac9b9c66 | 2489 | { |
a5516438 | 2490 | struct hstate *h = hstate_vma(vma); |
ac9b9c66 | 2491 | int ret = VM_FAULT_SIGBUS; |
e7c4b0bf | 2492 | pgoff_t idx; |
4c887265 | 2493 | unsigned long size; |
4c887265 AL |
2494 | struct page *page; |
2495 | struct address_space *mapping; | |
1e8f889b | 2496 | pte_t new_pte; |
4c887265 | 2497 | |
04f2cbe3 MG |
2498 | /* |
2499 | * Currently, we are forced to kill the process in the event the | |
2500 | * original mapper has unmapped pages from the child due to a failed | |
25985edc | 2501 | * COW. Warn that such a situation has occurred as it may not be obvious |
04f2cbe3 MG |
2502 | */ |
2503 | if (is_vma_resv_set(vma, HPAGE_RESV_UNMAPPED)) { | |
2504 | printk(KERN_WARNING | |
2505 | "PID %d killed due to inadequate hugepage pool\n", | |
2506 | current->pid); | |
2507 | return ret; | |
2508 | } | |
2509 | ||
4c887265 | 2510 | mapping = vma->vm_file->f_mapping; |
a5516438 | 2511 | idx = vma_hugecache_offset(h, vma, address); |
4c887265 AL |
2512 | |
2513 | /* | |
2514 | * Use page lock to guard against racing truncation | |
2515 | * before we get page_table_lock. | |
2516 | */ | |
6bda666a CL |
2517 | retry: |
2518 | page = find_lock_page(mapping, idx); | |
2519 | if (!page) { | |
a5516438 | 2520 | size = i_size_read(mapping->host) >> huge_page_shift(h); |
ebed4bfc HD |
2521 | if (idx >= size) |
2522 | goto out; | |
04f2cbe3 | 2523 | page = alloc_huge_page(vma, address, 0); |
2fc39cec AL |
2524 | if (IS_ERR(page)) { |
2525 | ret = -PTR_ERR(page); | |
6bda666a CL |
2526 | goto out; |
2527 | } | |
47ad8475 | 2528 | clear_huge_page(page, address, pages_per_huge_page(h)); |
0ed361de | 2529 | __SetPageUptodate(page); |
ac9b9c66 | 2530 | |
f83a275d | 2531 | if (vma->vm_flags & VM_MAYSHARE) { |
6bda666a | 2532 | int err; |
45c682a6 | 2533 | struct inode *inode = mapping->host; |
6bda666a CL |
2534 | |
2535 | err = add_to_page_cache(page, mapping, idx, GFP_KERNEL); | |
2536 | if (err) { | |
2537 | put_page(page); | |
6bda666a CL |
2538 | if (err == -EEXIST) |
2539 | goto retry; | |
2540 | goto out; | |
2541 | } | |
45c682a6 KC |
2542 | |
2543 | spin_lock(&inode->i_lock); | |
a5516438 | 2544 | inode->i_blocks += blocks_per_huge_page(h); |
45c682a6 | 2545 | spin_unlock(&inode->i_lock); |
0fe6e20b | 2546 | page_dup_rmap(page); |
23be7468 | 2547 | } else { |
6bda666a | 2548 | lock_page(page); |
0fe6e20b NH |
2549 | if (unlikely(anon_vma_prepare(vma))) { |
2550 | ret = VM_FAULT_OOM; | |
2551 | goto backout_unlocked; | |
2552 | } | |
2553 | hugepage_add_new_anon_rmap(page, vma, address); | |
23be7468 | 2554 | } |
0fe6e20b | 2555 | } else { |
998b4382 NH |
2556 | /* |
2557 | * If memory error occurs between mmap() and fault, some process | |
2558 | * don't have hwpoisoned swap entry for errored virtual address. | |
2559 | * So we need to block hugepage fault by PG_hwpoison bit check. | |
2560 | */ | |
2561 | if (unlikely(PageHWPoison(page))) { | |
aa50d3a7 AK |
2562 | ret = VM_FAULT_HWPOISON | |
2563 | VM_FAULT_SET_HINDEX(h - hstates); | |
998b4382 NH |
2564 | goto backout_unlocked; |
2565 | } | |
0fe6e20b | 2566 | page_dup_rmap(page); |
6bda666a | 2567 | } |
1e8f889b | 2568 | |
57303d80 AW |
2569 | /* |
2570 | * If we are going to COW a private mapping later, we examine the | |
2571 | * pending reservations for this page now. This will ensure that | |
2572 | * any allocations necessary to record that reservation occur outside | |
2573 | * the spinlock. | |
2574 | */ | |
788c7df4 | 2575 | if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) |
2b26736c AW |
2576 | if (vma_needs_reservation(h, vma, address) < 0) { |
2577 | ret = VM_FAULT_OOM; | |
2578 | goto backout_unlocked; | |
2579 | } | |
57303d80 | 2580 | |
ac9b9c66 | 2581 | spin_lock(&mm->page_table_lock); |
a5516438 | 2582 | size = i_size_read(mapping->host) >> huge_page_shift(h); |
4c887265 AL |
2583 | if (idx >= size) |
2584 | goto backout; | |
2585 | ||
83c54070 | 2586 | ret = 0; |
7f2e9525 | 2587 | if (!huge_pte_none(huge_ptep_get(ptep))) |
4c887265 AL |
2588 | goto backout; |
2589 | ||
1e8f889b DG |
2590 | new_pte = make_huge_pte(vma, page, ((vma->vm_flags & VM_WRITE) |
2591 | && (vma->vm_flags & VM_SHARED))); | |
2592 | set_huge_pte_at(mm, address, ptep, new_pte); | |
2593 | ||
788c7df4 | 2594 | if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) { |
1e8f889b | 2595 | /* Optimization, do the COW without a second fault */ |
04f2cbe3 | 2596 | ret = hugetlb_cow(mm, vma, address, ptep, new_pte, page); |
1e8f889b DG |
2597 | } |
2598 | ||
ac9b9c66 | 2599 | spin_unlock(&mm->page_table_lock); |
4c887265 AL |
2600 | unlock_page(page); |
2601 | out: | |
ac9b9c66 | 2602 | return ret; |
4c887265 AL |
2603 | |
2604 | backout: | |
2605 | spin_unlock(&mm->page_table_lock); | |
2b26736c | 2606 | backout_unlocked: |
4c887265 AL |
2607 | unlock_page(page); |
2608 | put_page(page); | |
2609 | goto out; | |
ac9b9c66 HD |
2610 | } |
2611 | ||
86e5216f | 2612 | int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
788c7df4 | 2613 | unsigned long address, unsigned int flags) |
86e5216f AL |
2614 | { |
2615 | pte_t *ptep; | |
2616 | pte_t entry; | |
1e8f889b | 2617 | int ret; |
0fe6e20b | 2618 | struct page *page = NULL; |
57303d80 | 2619 | struct page *pagecache_page = NULL; |
3935baa9 | 2620 | static DEFINE_MUTEX(hugetlb_instantiation_mutex); |
a5516438 | 2621 | struct hstate *h = hstate_vma(vma); |
86e5216f | 2622 | |
fd6a03ed NH |
2623 | ptep = huge_pte_offset(mm, address); |
2624 | if (ptep) { | |
2625 | entry = huge_ptep_get(ptep); | |
290408d4 NH |
2626 | if (unlikely(is_hugetlb_entry_migration(entry))) { |
2627 | migration_entry_wait(mm, (pmd_t *)ptep, address); | |
2628 | return 0; | |
2629 | } else if (unlikely(is_hugetlb_entry_hwpoisoned(entry))) | |
aa50d3a7 AK |
2630 | return VM_FAULT_HWPOISON_LARGE | |
2631 | VM_FAULT_SET_HINDEX(h - hstates); | |
fd6a03ed NH |
2632 | } |
2633 | ||
a5516438 | 2634 | ptep = huge_pte_alloc(mm, address, huge_page_size(h)); |
86e5216f AL |
2635 | if (!ptep) |
2636 | return VM_FAULT_OOM; | |
2637 | ||
3935baa9 DG |
2638 | /* |
2639 | * Serialize hugepage allocation and instantiation, so that we don't | |
2640 | * get spurious allocation failures if two CPUs race to instantiate | |
2641 | * the same page in the page cache. | |
2642 | */ | |
2643 | mutex_lock(&hugetlb_instantiation_mutex); | |
7f2e9525 GS |
2644 | entry = huge_ptep_get(ptep); |
2645 | if (huge_pte_none(entry)) { | |
788c7df4 | 2646 | ret = hugetlb_no_page(mm, vma, address, ptep, flags); |
b4d1d99f | 2647 | goto out_mutex; |
3935baa9 | 2648 | } |
86e5216f | 2649 | |
83c54070 | 2650 | ret = 0; |
1e8f889b | 2651 | |
57303d80 AW |
2652 | /* |
2653 | * If we are going to COW the mapping later, we examine the pending | |
2654 | * reservations for this page now. This will ensure that any | |
2655 | * allocations necessary to record that reservation occur outside the | |
2656 | * spinlock. For private mappings, we also lookup the pagecache | |
2657 | * page now as it is used to determine if a reservation has been | |
2658 | * consumed. | |
2659 | */ | |
788c7df4 | 2660 | if ((flags & FAULT_FLAG_WRITE) && !pte_write(entry)) { |
2b26736c AW |
2661 | if (vma_needs_reservation(h, vma, address) < 0) { |
2662 | ret = VM_FAULT_OOM; | |
b4d1d99f | 2663 | goto out_mutex; |
2b26736c | 2664 | } |
57303d80 | 2665 | |
f83a275d | 2666 | if (!(vma->vm_flags & VM_MAYSHARE)) |
57303d80 AW |
2667 | pagecache_page = hugetlbfs_pagecache_page(h, |
2668 | vma, address); | |
2669 | } | |
2670 | ||
56c9cfb1 NH |
2671 | /* |
2672 | * hugetlb_cow() requires page locks of pte_page(entry) and | |
2673 | * pagecache_page, so here we need take the former one | |
2674 | * when page != pagecache_page or !pagecache_page. | |
2675 | * Note that locking order is always pagecache_page -> page, | |
2676 | * so no worry about deadlock. | |
2677 | */ | |
2678 | page = pte_page(entry); | |
2679 | if (page != pagecache_page) | |
0fe6e20b | 2680 | lock_page(page); |
0fe6e20b | 2681 | |
1e8f889b DG |
2682 | spin_lock(&mm->page_table_lock); |
2683 | /* Check for a racing update before calling hugetlb_cow */ | |
b4d1d99f DG |
2684 | if (unlikely(!pte_same(entry, huge_ptep_get(ptep)))) |
2685 | goto out_page_table_lock; | |
2686 | ||
2687 | ||
788c7df4 | 2688 | if (flags & FAULT_FLAG_WRITE) { |
b4d1d99f | 2689 | if (!pte_write(entry)) { |
57303d80 AW |
2690 | ret = hugetlb_cow(mm, vma, address, ptep, entry, |
2691 | pagecache_page); | |
b4d1d99f DG |
2692 | goto out_page_table_lock; |
2693 | } | |
2694 | entry = pte_mkdirty(entry); | |
2695 | } | |
2696 | entry = pte_mkyoung(entry); | |
788c7df4 HD |
2697 | if (huge_ptep_set_access_flags(vma, address, ptep, entry, |
2698 | flags & FAULT_FLAG_WRITE)) | |
4b3073e1 | 2699 | update_mmu_cache(vma, address, ptep); |
b4d1d99f DG |
2700 | |
2701 | out_page_table_lock: | |
1e8f889b | 2702 | spin_unlock(&mm->page_table_lock); |
57303d80 AW |
2703 | |
2704 | if (pagecache_page) { | |
2705 | unlock_page(pagecache_page); | |
2706 | put_page(pagecache_page); | |
2707 | } | |
1f64d69c DN |
2708 | if (page != pagecache_page) |
2709 | unlock_page(page); | |
57303d80 | 2710 | |
b4d1d99f | 2711 | out_mutex: |
3935baa9 | 2712 | mutex_unlock(&hugetlb_instantiation_mutex); |
1e8f889b DG |
2713 | |
2714 | return ret; | |
86e5216f AL |
2715 | } |
2716 | ||
ceb86879 AK |
2717 | /* Can be overriden by architectures */ |
2718 | __attribute__((weak)) struct page * | |
2719 | follow_huge_pud(struct mm_struct *mm, unsigned long address, | |
2720 | pud_t *pud, int write) | |
2721 | { | |
2722 | BUG(); | |
2723 | return NULL; | |
2724 | } | |
2725 | ||
63551ae0 DG |
2726 | int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2727 | struct page **pages, struct vm_area_struct **vmas, | |
5b23dbe8 | 2728 | unsigned long *position, int *length, int i, |
2a15efc9 | 2729 | unsigned int flags) |
63551ae0 | 2730 | { |
d5d4b0aa KC |
2731 | unsigned long pfn_offset; |
2732 | unsigned long vaddr = *position; | |
63551ae0 | 2733 | int remainder = *length; |
a5516438 | 2734 | struct hstate *h = hstate_vma(vma); |
63551ae0 | 2735 | |
1c59827d | 2736 | spin_lock(&mm->page_table_lock); |
63551ae0 | 2737 | while (vaddr < vma->vm_end && remainder) { |
4c887265 | 2738 | pte_t *pte; |
2a15efc9 | 2739 | int absent; |
4c887265 | 2740 | struct page *page; |
63551ae0 | 2741 | |
4c887265 AL |
2742 | /* |
2743 | * Some archs (sparc64, sh*) have multiple pte_ts to | |
2a15efc9 | 2744 | * each hugepage. We have to make sure we get the |
4c887265 AL |
2745 | * first, for the page indexing below to work. |
2746 | */ | |
a5516438 | 2747 | pte = huge_pte_offset(mm, vaddr & huge_page_mask(h)); |
2a15efc9 HD |
2748 | absent = !pte || huge_pte_none(huge_ptep_get(pte)); |
2749 | ||
2750 | /* | |
2751 | * When coredumping, it suits get_dump_page if we just return | |
3ae77f43 HD |
2752 | * an error where there's an empty slot with no huge pagecache |
2753 | * to back it. This way, we avoid allocating a hugepage, and | |
2754 | * the sparse dumpfile avoids allocating disk blocks, but its | |
2755 | * huge holes still show up with zeroes where they need to be. | |
2a15efc9 | 2756 | */ |
3ae77f43 HD |
2757 | if (absent && (flags & FOLL_DUMP) && |
2758 | !hugetlbfs_pagecache_present(h, vma, vaddr)) { | |
2a15efc9 HD |
2759 | remainder = 0; |
2760 | break; | |
2761 | } | |
63551ae0 | 2762 | |
2a15efc9 HD |
2763 | if (absent || |
2764 | ((flags & FOLL_WRITE) && !pte_write(huge_ptep_get(pte)))) { | |
4c887265 | 2765 | int ret; |
63551ae0 | 2766 | |
4c887265 | 2767 | spin_unlock(&mm->page_table_lock); |
2a15efc9 HD |
2768 | ret = hugetlb_fault(mm, vma, vaddr, |
2769 | (flags & FOLL_WRITE) ? FAULT_FLAG_WRITE : 0); | |
4c887265 | 2770 | spin_lock(&mm->page_table_lock); |
a89182c7 | 2771 | if (!(ret & VM_FAULT_ERROR)) |
4c887265 | 2772 | continue; |
63551ae0 | 2773 | |
4c887265 | 2774 | remainder = 0; |
4c887265 AL |
2775 | break; |
2776 | } | |
2777 | ||
a5516438 | 2778 | pfn_offset = (vaddr & ~huge_page_mask(h)) >> PAGE_SHIFT; |
7f2e9525 | 2779 | page = pte_page(huge_ptep_get(pte)); |
d5d4b0aa | 2780 | same_page: |
d6692183 | 2781 | if (pages) { |
2a15efc9 | 2782 | pages[i] = mem_map_offset(page, pfn_offset); |
4b2e38ad | 2783 | get_page(pages[i]); |
d6692183 | 2784 | } |
63551ae0 DG |
2785 | |
2786 | if (vmas) | |
2787 | vmas[i] = vma; | |
2788 | ||
2789 | vaddr += PAGE_SIZE; | |
d5d4b0aa | 2790 | ++pfn_offset; |
63551ae0 DG |
2791 | --remainder; |
2792 | ++i; | |
d5d4b0aa | 2793 | if (vaddr < vma->vm_end && remainder && |
a5516438 | 2794 | pfn_offset < pages_per_huge_page(h)) { |
d5d4b0aa KC |
2795 | /* |
2796 | * We use pfn_offset to avoid touching the pageframes | |
2797 | * of this compound page. | |
2798 | */ | |
2799 | goto same_page; | |
2800 | } | |
63551ae0 | 2801 | } |
1c59827d | 2802 | spin_unlock(&mm->page_table_lock); |
63551ae0 DG |
2803 | *length = remainder; |
2804 | *position = vaddr; | |
2805 | ||
2a15efc9 | 2806 | return i ? i : -EFAULT; |
63551ae0 | 2807 | } |
8f860591 ZY |
2808 | |
2809 | void hugetlb_change_protection(struct vm_area_struct *vma, | |
2810 | unsigned long address, unsigned long end, pgprot_t newprot) | |
2811 | { | |
2812 | struct mm_struct *mm = vma->vm_mm; | |
2813 | unsigned long start = address; | |
2814 | pte_t *ptep; | |
2815 | pte_t pte; | |
a5516438 | 2816 | struct hstate *h = hstate_vma(vma); |
8f860591 ZY |
2817 | |
2818 | BUG_ON(address >= end); | |
2819 | flush_cache_range(vma, address, end); | |
2820 | ||
3d48ae45 | 2821 | mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); |
8f860591 | 2822 | spin_lock(&mm->page_table_lock); |
a5516438 | 2823 | for (; address < end; address += huge_page_size(h)) { |
8f860591 ZY |
2824 | ptep = huge_pte_offset(mm, address); |
2825 | if (!ptep) | |
2826 | continue; | |
39dde65c KC |
2827 | if (huge_pmd_unshare(mm, &address, ptep)) |
2828 | continue; | |
7f2e9525 | 2829 | if (!huge_pte_none(huge_ptep_get(ptep))) { |
8f860591 ZY |
2830 | pte = huge_ptep_get_and_clear(mm, address, ptep); |
2831 | pte = pte_mkhuge(pte_modify(pte, newprot)); | |
2832 | set_huge_pte_at(mm, address, ptep, pte); | |
8f860591 ZY |
2833 | } |
2834 | } | |
2835 | spin_unlock(&mm->page_table_lock); | |
3d48ae45 | 2836 | mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); |
8f860591 ZY |
2837 | |
2838 | flush_tlb_range(vma, start, end); | |
2839 | } | |
2840 | ||
a1e78772 MG |
2841 | int hugetlb_reserve_pages(struct inode *inode, |
2842 | long from, long to, | |
5a6fe125 | 2843 | struct vm_area_struct *vma, |
ca16d140 | 2844 | vm_flags_t vm_flags) |
e4e574b7 | 2845 | { |
17c9d12e | 2846 | long ret, chg; |
a5516438 | 2847 | struct hstate *h = hstate_inode(inode); |
e4e574b7 | 2848 | |
17c9d12e MG |
2849 | /* |
2850 | * Only apply hugepage reservation if asked. At fault time, an | |
2851 | * attempt will be made for VM_NORESERVE to allocate a page | |
2852 | * and filesystem quota without using reserves | |
2853 | */ | |
ca16d140 | 2854 | if (vm_flags & VM_NORESERVE) |
17c9d12e MG |
2855 | return 0; |
2856 | ||
a1e78772 MG |
2857 | /* |
2858 | * Shared mappings base their reservation on the number of pages that | |
2859 | * are already allocated on behalf of the file. Private mappings need | |
2860 | * to reserve the full area even if read-only as mprotect() may be | |
2861 | * called to make the mapping read-write. Assume !vma is a shm mapping | |
2862 | */ | |
f83a275d | 2863 | if (!vma || vma->vm_flags & VM_MAYSHARE) |
a1e78772 | 2864 | chg = region_chg(&inode->i_mapping->private_list, from, to); |
17c9d12e MG |
2865 | else { |
2866 | struct resv_map *resv_map = resv_map_alloc(); | |
2867 | if (!resv_map) | |
2868 | return -ENOMEM; | |
2869 | ||
a1e78772 | 2870 | chg = to - from; |
84afd99b | 2871 | |
17c9d12e MG |
2872 | set_vma_resv_map(vma, resv_map); |
2873 | set_vma_resv_flags(vma, HPAGE_RESV_OWNER); | |
2874 | } | |
2875 | ||
e4e574b7 AL |
2876 | if (chg < 0) |
2877 | return chg; | |
8a630112 | 2878 | |
17c9d12e | 2879 | /* There must be enough filesystem quota for the mapping */ |
90d8b7e6 AL |
2880 | if (hugetlb_get_quota(inode->i_mapping, chg)) |
2881 | return -ENOSPC; | |
5a6fe125 MG |
2882 | |
2883 | /* | |
17c9d12e MG |
2884 | * Check enough hugepages are available for the reservation. |
2885 | * Hand back the quota if there are not | |
5a6fe125 | 2886 | */ |
a5516438 | 2887 | ret = hugetlb_acct_memory(h, chg); |
68842c9b KC |
2888 | if (ret < 0) { |
2889 | hugetlb_put_quota(inode->i_mapping, chg); | |
a43a8c39 | 2890 | return ret; |
68842c9b | 2891 | } |
17c9d12e MG |
2892 | |
2893 | /* | |
2894 | * Account for the reservations made. Shared mappings record regions | |
2895 | * that have reservations as they are shared by multiple VMAs. | |
2896 | * When the last VMA disappears, the region map says how much | |
2897 | * the reservation was and the page cache tells how much of | |
2898 | * the reservation was consumed. Private mappings are per-VMA and | |
2899 | * only the consumed reservations are tracked. When the VMA | |
2900 | * disappears, the original reservation is the VMA size and the | |
2901 | * consumed reservations are stored in the map. Hence, nothing | |
2902 | * else has to be done for private mappings here | |
2903 | */ | |
f83a275d | 2904 | if (!vma || vma->vm_flags & VM_MAYSHARE) |
a1e78772 | 2905 | region_add(&inode->i_mapping->private_list, from, to); |
a43a8c39 KC |
2906 | return 0; |
2907 | } | |
2908 | ||
2909 | void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed) | |
2910 | { | |
a5516438 | 2911 | struct hstate *h = hstate_inode(inode); |
a43a8c39 | 2912 | long chg = region_truncate(&inode->i_mapping->private_list, offset); |
45c682a6 KC |
2913 | |
2914 | spin_lock(&inode->i_lock); | |
e4c6f8be | 2915 | inode->i_blocks -= (blocks_per_huge_page(h) * freed); |
45c682a6 KC |
2916 | spin_unlock(&inode->i_lock); |
2917 | ||
90d8b7e6 | 2918 | hugetlb_put_quota(inode->i_mapping, (chg - freed)); |
a5516438 | 2919 | hugetlb_acct_memory(h, -(chg - freed)); |
a43a8c39 | 2920 | } |
93f70f90 | 2921 | |
d5bd9106 AK |
2922 | #ifdef CONFIG_MEMORY_FAILURE |
2923 | ||
6de2b1aa NH |
2924 | /* Should be called in hugetlb_lock */ |
2925 | static int is_hugepage_on_freelist(struct page *hpage) | |
2926 | { | |
2927 | struct page *page; | |
2928 | struct page *tmp; | |
2929 | struct hstate *h = page_hstate(hpage); | |
2930 | int nid = page_to_nid(hpage); | |
2931 | ||
2932 | list_for_each_entry_safe(page, tmp, &h->hugepage_freelists[nid], lru) | |
2933 | if (page == hpage) | |
2934 | return 1; | |
2935 | return 0; | |
2936 | } | |
2937 | ||
93f70f90 NH |
2938 | /* |
2939 | * This function is called from memory failure code. | |
2940 | * Assume the caller holds page lock of the head page. | |
2941 | */ | |
6de2b1aa | 2942 | int dequeue_hwpoisoned_huge_page(struct page *hpage) |
93f70f90 NH |
2943 | { |
2944 | struct hstate *h = page_hstate(hpage); | |
2945 | int nid = page_to_nid(hpage); | |
6de2b1aa | 2946 | int ret = -EBUSY; |
93f70f90 NH |
2947 | |
2948 | spin_lock(&hugetlb_lock); | |
6de2b1aa NH |
2949 | if (is_hugepage_on_freelist(hpage)) { |
2950 | list_del(&hpage->lru); | |
8c6c2ecb | 2951 | set_page_refcounted(hpage); |
6de2b1aa NH |
2952 | h->free_huge_pages--; |
2953 | h->free_huge_pages_node[nid]--; | |
2954 | ret = 0; | |
2955 | } | |
93f70f90 | 2956 | spin_unlock(&hugetlb_lock); |
6de2b1aa | 2957 | return ret; |
93f70f90 | 2958 | } |
6de2b1aa | 2959 | #endif |