Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Generic hugetlb support. | |
6d49e352 | 3 | * (C) Nadia Yvette Chambers, April 2004 |
1da177e4 | 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> | |
24669e58 | 27 | #include <asm/tlb.h> |
63551ae0 | 28 | |
24669e58 | 29 | #include <linux/io.h> |
63551ae0 | 30 | #include <linux/hugetlb.h> |
9dd540e2 | 31 | #include <linux/hugetlb_cgroup.h> |
9a305230 | 32 | #include <linux/node.h> |
7835e98b | 33 | #include "internal.h" |
1da177e4 LT |
34 | |
35 | const unsigned long hugetlb_zero = 0, hugetlb_infinity = ~0UL; | |
396faf03 MG |
36 | static gfp_t htlb_alloc_mask = GFP_HIGHUSER; |
37 | unsigned long hugepages_treat_as_movable; | |
a5516438 | 38 | |
c3f38a38 | 39 | int hugetlb_max_hstate __read_mostly; |
e5ff2159 AK |
40 | unsigned int default_hstate_idx; |
41 | struct hstate hstates[HUGE_MAX_HSTATE]; | |
42 | ||
53ba51d2 JT |
43 | __initdata LIST_HEAD(huge_boot_pages); |
44 | ||
e5ff2159 AK |
45 | /* for command line parsing */ |
46 | static struct hstate * __initdata parsed_hstate; | |
47 | static unsigned long __initdata default_hstate_max_huge_pages; | |
e11bfbfc | 48 | static unsigned long __initdata default_hstate_size; |
e5ff2159 | 49 | |
3935baa9 DG |
50 | /* |
51 | * Protects updates to hugepage_freelists, nr_huge_pages, and free_huge_pages | |
52 | */ | |
c3f38a38 | 53 | DEFINE_SPINLOCK(hugetlb_lock); |
0bd0f9fb | 54 | |
90481622 DG |
55 | static inline void unlock_or_release_subpool(struct hugepage_subpool *spool) |
56 | { | |
57 | bool free = (spool->count == 0) && (spool->used_hpages == 0); | |
58 | ||
59 | spin_unlock(&spool->lock); | |
60 | ||
61 | /* If no pages are used, and no other handles to the subpool | |
62 | * remain, free the subpool the subpool remain */ | |
63 | if (free) | |
64 | kfree(spool); | |
65 | } | |
66 | ||
67 | struct hugepage_subpool *hugepage_new_subpool(long nr_blocks) | |
68 | { | |
69 | struct hugepage_subpool *spool; | |
70 | ||
71 | spool = kmalloc(sizeof(*spool), GFP_KERNEL); | |
72 | if (!spool) | |
73 | return NULL; | |
74 | ||
75 | spin_lock_init(&spool->lock); | |
76 | spool->count = 1; | |
77 | spool->max_hpages = nr_blocks; | |
78 | spool->used_hpages = 0; | |
79 | ||
80 | return spool; | |
81 | } | |
82 | ||
83 | void hugepage_put_subpool(struct hugepage_subpool *spool) | |
84 | { | |
85 | spin_lock(&spool->lock); | |
86 | BUG_ON(!spool->count); | |
87 | spool->count--; | |
88 | unlock_or_release_subpool(spool); | |
89 | } | |
90 | ||
91 | static int hugepage_subpool_get_pages(struct hugepage_subpool *spool, | |
92 | long delta) | |
93 | { | |
94 | int ret = 0; | |
95 | ||
96 | if (!spool) | |
97 | return 0; | |
98 | ||
99 | spin_lock(&spool->lock); | |
100 | if ((spool->used_hpages + delta) <= spool->max_hpages) { | |
101 | spool->used_hpages += delta; | |
102 | } else { | |
103 | ret = -ENOMEM; | |
104 | } | |
105 | spin_unlock(&spool->lock); | |
106 | ||
107 | return ret; | |
108 | } | |
109 | ||
110 | static void hugepage_subpool_put_pages(struct hugepage_subpool *spool, | |
111 | long delta) | |
112 | { | |
113 | if (!spool) | |
114 | return; | |
115 | ||
116 | spin_lock(&spool->lock); | |
117 | spool->used_hpages -= delta; | |
118 | /* If hugetlbfs_put_super couldn't free spool due to | |
119 | * an outstanding quota reference, free it now. */ | |
120 | unlock_or_release_subpool(spool); | |
121 | } | |
122 | ||
123 | static inline struct hugepage_subpool *subpool_inode(struct inode *inode) | |
124 | { | |
125 | return HUGETLBFS_SB(inode->i_sb)->spool; | |
126 | } | |
127 | ||
128 | static inline struct hugepage_subpool *subpool_vma(struct vm_area_struct *vma) | |
129 | { | |
496ad9aa | 130 | return subpool_inode(file_inode(vma->vm_file)); |
90481622 DG |
131 | } |
132 | ||
96822904 AW |
133 | /* |
134 | * Region tracking -- allows tracking of reservations and instantiated pages | |
135 | * across the pages in a mapping. | |
84afd99b AW |
136 | * |
137 | * The region data structures are protected by a combination of the mmap_sem | |
c748c262 | 138 | * and the hugetlb_instantiation_mutex. To access or modify a region the caller |
84afd99b | 139 | * must either hold the mmap_sem for write, or the mmap_sem for read and |
c748c262 | 140 | * the hugetlb_instantiation_mutex: |
84afd99b | 141 | * |
32f84528 | 142 | * down_write(&mm->mmap_sem); |
84afd99b | 143 | * or |
32f84528 CF |
144 | * down_read(&mm->mmap_sem); |
145 | * mutex_lock(&hugetlb_instantiation_mutex); | |
96822904 AW |
146 | */ |
147 | struct file_region { | |
148 | struct list_head link; | |
149 | long from; | |
150 | long to; | |
151 | }; | |
152 | ||
153 | static long region_add(struct list_head *head, long f, long t) | |
154 | { | |
155 | struct file_region *rg, *nrg, *trg; | |
156 | ||
157 | /* Locate the region we are either in or before. */ | |
158 | list_for_each_entry(rg, head, link) | |
159 | if (f <= rg->to) | |
160 | break; | |
161 | ||
162 | /* Round our left edge to the current segment if it encloses us. */ | |
163 | if (f > rg->from) | |
164 | f = rg->from; | |
165 | ||
166 | /* Check for and consume any regions we now overlap with. */ | |
167 | nrg = rg; | |
168 | list_for_each_entry_safe(rg, trg, rg->link.prev, link) { | |
169 | if (&rg->link == head) | |
170 | break; | |
171 | if (rg->from > t) | |
172 | break; | |
173 | ||
174 | /* If this area reaches higher then extend our area to | |
175 | * include it completely. If this is not the first area | |
176 | * which we intend to reuse, free it. */ | |
177 | if (rg->to > t) | |
178 | t = rg->to; | |
179 | if (rg != nrg) { | |
180 | list_del(&rg->link); | |
181 | kfree(rg); | |
182 | } | |
183 | } | |
184 | nrg->from = f; | |
185 | nrg->to = t; | |
186 | return 0; | |
187 | } | |
188 | ||
189 | static long region_chg(struct list_head *head, long f, long t) | |
190 | { | |
191 | struct file_region *rg, *nrg; | |
192 | long chg = 0; | |
193 | ||
194 | /* Locate the region we are before or in. */ | |
195 | list_for_each_entry(rg, head, link) | |
196 | if (f <= rg->to) | |
197 | break; | |
198 | ||
199 | /* If we are below the current region then a new region is required. | |
200 | * Subtle, allocate a new region at the position but make it zero | |
201 | * size such that we can guarantee to record the reservation. */ | |
202 | if (&rg->link == head || t < rg->from) { | |
203 | nrg = kmalloc(sizeof(*nrg), GFP_KERNEL); | |
204 | if (!nrg) | |
205 | return -ENOMEM; | |
206 | nrg->from = f; | |
207 | nrg->to = f; | |
208 | INIT_LIST_HEAD(&nrg->link); | |
209 | list_add(&nrg->link, rg->link.prev); | |
210 | ||
211 | return t - f; | |
212 | } | |
213 | ||
214 | /* Round our left edge to the current segment if it encloses us. */ | |
215 | if (f > rg->from) | |
216 | f = rg->from; | |
217 | chg = t - f; | |
218 | ||
219 | /* Check for and consume any regions we now overlap with. */ | |
220 | list_for_each_entry(rg, rg->link.prev, link) { | |
221 | if (&rg->link == head) | |
222 | break; | |
223 | if (rg->from > t) | |
224 | return chg; | |
225 | ||
25985edc | 226 | /* We overlap with this area, if it extends further than |
96822904 AW |
227 | * us then we must extend ourselves. Account for its |
228 | * existing reservation. */ | |
229 | if (rg->to > t) { | |
230 | chg += rg->to - t; | |
231 | t = rg->to; | |
232 | } | |
233 | chg -= rg->to - rg->from; | |
234 | } | |
235 | return chg; | |
236 | } | |
237 | ||
238 | static long region_truncate(struct list_head *head, long end) | |
239 | { | |
240 | struct file_region *rg, *trg; | |
241 | long chg = 0; | |
242 | ||
243 | /* Locate the region we are either in or before. */ | |
244 | list_for_each_entry(rg, head, link) | |
245 | if (end <= rg->to) | |
246 | break; | |
247 | if (&rg->link == head) | |
248 | return 0; | |
249 | ||
250 | /* If we are in the middle of a region then adjust it. */ | |
251 | if (end > rg->from) { | |
252 | chg = rg->to - end; | |
253 | rg->to = end; | |
254 | rg = list_entry(rg->link.next, typeof(*rg), link); | |
255 | } | |
256 | ||
257 | /* Drop any remaining regions. */ | |
258 | list_for_each_entry_safe(rg, trg, rg->link.prev, link) { | |
259 | if (&rg->link == head) | |
260 | break; | |
261 | chg += rg->to - rg->from; | |
262 | list_del(&rg->link); | |
263 | kfree(rg); | |
264 | } | |
265 | return chg; | |
266 | } | |
267 | ||
84afd99b AW |
268 | static long region_count(struct list_head *head, long f, long t) |
269 | { | |
270 | struct file_region *rg; | |
271 | long chg = 0; | |
272 | ||
273 | /* Locate each segment we overlap with, and count that overlap. */ | |
274 | list_for_each_entry(rg, head, link) { | |
f2135a4a WSH |
275 | long seg_from; |
276 | long seg_to; | |
84afd99b AW |
277 | |
278 | if (rg->to <= f) | |
279 | continue; | |
280 | if (rg->from >= t) | |
281 | break; | |
282 | ||
283 | seg_from = max(rg->from, f); | |
284 | seg_to = min(rg->to, t); | |
285 | ||
286 | chg += seg_to - seg_from; | |
287 | } | |
288 | ||
289 | return chg; | |
290 | } | |
291 | ||
e7c4b0bf AW |
292 | /* |
293 | * Convert the address within this vma to the page offset within | |
294 | * the mapping, in pagecache page units; huge pages here. | |
295 | */ | |
a5516438 AK |
296 | static pgoff_t vma_hugecache_offset(struct hstate *h, |
297 | struct vm_area_struct *vma, unsigned long address) | |
e7c4b0bf | 298 | { |
a5516438 AK |
299 | return ((address - vma->vm_start) >> huge_page_shift(h)) + |
300 | (vma->vm_pgoff >> huge_page_order(h)); | |
e7c4b0bf AW |
301 | } |
302 | ||
0fe6e20b NH |
303 | pgoff_t linear_hugepage_index(struct vm_area_struct *vma, |
304 | unsigned long address) | |
305 | { | |
306 | return vma_hugecache_offset(hstate_vma(vma), vma, address); | |
307 | } | |
308 | ||
08fba699 MG |
309 | /* |
310 | * Return the size of the pages allocated when backing a VMA. In the majority | |
311 | * cases this will be same size as used by the page table entries. | |
312 | */ | |
313 | unsigned long vma_kernel_pagesize(struct vm_area_struct *vma) | |
314 | { | |
315 | struct hstate *hstate; | |
316 | ||
317 | if (!is_vm_hugetlb_page(vma)) | |
318 | return PAGE_SIZE; | |
319 | ||
320 | hstate = hstate_vma(vma); | |
321 | ||
2415cf12 | 322 | return 1UL << huge_page_shift(hstate); |
08fba699 | 323 | } |
f340ca0f | 324 | EXPORT_SYMBOL_GPL(vma_kernel_pagesize); |
08fba699 | 325 | |
3340289d MG |
326 | /* |
327 | * Return the page size being used by the MMU to back a VMA. In the majority | |
328 | * of cases, the page size used by the kernel matches the MMU size. On | |
329 | * architectures where it differs, an architecture-specific version of this | |
330 | * function is required. | |
331 | */ | |
332 | #ifndef vma_mmu_pagesize | |
333 | unsigned long vma_mmu_pagesize(struct vm_area_struct *vma) | |
334 | { | |
335 | return vma_kernel_pagesize(vma); | |
336 | } | |
337 | #endif | |
338 | ||
84afd99b AW |
339 | /* |
340 | * Flags for MAP_PRIVATE reservations. These are stored in the bottom | |
341 | * bits of the reservation map pointer, which are always clear due to | |
342 | * alignment. | |
343 | */ | |
344 | #define HPAGE_RESV_OWNER (1UL << 0) | |
345 | #define HPAGE_RESV_UNMAPPED (1UL << 1) | |
04f2cbe3 | 346 | #define HPAGE_RESV_MASK (HPAGE_RESV_OWNER | HPAGE_RESV_UNMAPPED) |
84afd99b | 347 | |
a1e78772 MG |
348 | /* |
349 | * These helpers are used to track how many pages are reserved for | |
350 | * faults in a MAP_PRIVATE mapping. Only the process that called mmap() | |
351 | * is guaranteed to have their future faults succeed. | |
352 | * | |
353 | * With the exception of reset_vma_resv_huge_pages() which is called at fork(), | |
354 | * the reserve counters are updated with the hugetlb_lock held. It is safe | |
355 | * to reset the VMA at fork() time as it is not in use yet and there is no | |
356 | * chance of the global counters getting corrupted as a result of the values. | |
84afd99b AW |
357 | * |
358 | * The private mapping reservation is represented in a subtly different | |
359 | * manner to a shared mapping. A shared mapping has a region map associated | |
360 | * with the underlying file, this region map represents the backing file | |
361 | * pages which have ever had a reservation assigned which this persists even | |
362 | * after the page is instantiated. A private mapping has a region map | |
363 | * associated with the original mmap which is attached to all VMAs which | |
364 | * reference it, this region map represents those offsets which have consumed | |
365 | * reservation ie. where pages have been instantiated. | |
a1e78772 | 366 | */ |
e7c4b0bf AW |
367 | static unsigned long get_vma_private_data(struct vm_area_struct *vma) |
368 | { | |
369 | return (unsigned long)vma->vm_private_data; | |
370 | } | |
371 | ||
372 | static void set_vma_private_data(struct vm_area_struct *vma, | |
373 | unsigned long value) | |
374 | { | |
375 | vma->vm_private_data = (void *)value; | |
376 | } | |
377 | ||
84afd99b AW |
378 | struct resv_map { |
379 | struct kref refs; | |
380 | struct list_head regions; | |
381 | }; | |
382 | ||
2a4b3ded | 383 | static struct resv_map *resv_map_alloc(void) |
84afd99b AW |
384 | { |
385 | struct resv_map *resv_map = kmalloc(sizeof(*resv_map), GFP_KERNEL); | |
386 | if (!resv_map) | |
387 | return NULL; | |
388 | ||
389 | kref_init(&resv_map->refs); | |
390 | INIT_LIST_HEAD(&resv_map->regions); | |
391 | ||
392 | return resv_map; | |
393 | } | |
394 | ||
2a4b3ded | 395 | static void resv_map_release(struct kref *ref) |
84afd99b AW |
396 | { |
397 | struct resv_map *resv_map = container_of(ref, struct resv_map, refs); | |
398 | ||
399 | /* Clear out any active regions before we release the map. */ | |
400 | region_truncate(&resv_map->regions, 0); | |
401 | kfree(resv_map); | |
402 | } | |
403 | ||
404 | static struct resv_map *vma_resv_map(struct vm_area_struct *vma) | |
a1e78772 MG |
405 | { |
406 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); | |
f83a275d | 407 | if (!(vma->vm_flags & VM_MAYSHARE)) |
84afd99b AW |
408 | return (struct resv_map *)(get_vma_private_data(vma) & |
409 | ~HPAGE_RESV_MASK); | |
2a4b3ded | 410 | return NULL; |
a1e78772 MG |
411 | } |
412 | ||
84afd99b | 413 | static void set_vma_resv_map(struct vm_area_struct *vma, struct resv_map *map) |
a1e78772 MG |
414 | { |
415 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); | |
f83a275d | 416 | VM_BUG_ON(vma->vm_flags & VM_MAYSHARE); |
a1e78772 | 417 | |
84afd99b AW |
418 | set_vma_private_data(vma, (get_vma_private_data(vma) & |
419 | HPAGE_RESV_MASK) | (unsigned long)map); | |
04f2cbe3 MG |
420 | } |
421 | ||
422 | static void set_vma_resv_flags(struct vm_area_struct *vma, unsigned long flags) | |
423 | { | |
04f2cbe3 | 424 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); |
f83a275d | 425 | VM_BUG_ON(vma->vm_flags & VM_MAYSHARE); |
e7c4b0bf AW |
426 | |
427 | set_vma_private_data(vma, get_vma_private_data(vma) | flags); | |
04f2cbe3 MG |
428 | } |
429 | ||
430 | static int is_vma_resv_set(struct vm_area_struct *vma, unsigned long flag) | |
431 | { | |
432 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); | |
e7c4b0bf AW |
433 | |
434 | return (get_vma_private_data(vma) & flag) != 0; | |
a1e78772 MG |
435 | } |
436 | ||
04f2cbe3 | 437 | /* Reset counters to 0 and clear all HPAGE_RESV_* flags */ |
a1e78772 MG |
438 | void reset_vma_resv_huge_pages(struct vm_area_struct *vma) |
439 | { | |
440 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); | |
f83a275d | 441 | if (!(vma->vm_flags & VM_MAYSHARE)) |
a1e78772 MG |
442 | vma->vm_private_data = (void *)0; |
443 | } | |
444 | ||
445 | /* Returns true if the VMA has associated reserve pages */ | |
af0ed73e | 446 | static int vma_has_reserves(struct vm_area_struct *vma, long chg) |
a1e78772 | 447 | { |
af0ed73e JK |
448 | if (vma->vm_flags & VM_NORESERVE) { |
449 | /* | |
450 | * This address is already reserved by other process(chg == 0), | |
451 | * so, we should decrement reserved count. Without decrementing, | |
452 | * reserve count remains after releasing inode, because this | |
453 | * allocated page will go into page cache and is regarded as | |
454 | * coming from reserved pool in releasing step. Currently, we | |
455 | * don't have any other solution to deal with this situation | |
456 | * properly, so add work-around here. | |
457 | */ | |
458 | if (vma->vm_flags & VM_MAYSHARE && chg == 0) | |
459 | return 1; | |
460 | else | |
461 | return 0; | |
462 | } | |
a63884e9 JK |
463 | |
464 | /* Shared mappings always use reserves */ | |
f83a275d | 465 | if (vma->vm_flags & VM_MAYSHARE) |
7f09ca51 | 466 | return 1; |
a63884e9 JK |
467 | |
468 | /* | |
469 | * Only the process that called mmap() has reserves for | |
470 | * private mappings. | |
471 | */ | |
7f09ca51 MG |
472 | if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) |
473 | return 1; | |
a63884e9 | 474 | |
7f09ca51 | 475 | return 0; |
a1e78772 MG |
476 | } |
477 | ||
0ebabb41 NH |
478 | static void copy_gigantic_page(struct page *dst, struct page *src) |
479 | { | |
480 | int i; | |
481 | struct hstate *h = page_hstate(src); | |
482 | struct page *dst_base = dst; | |
483 | struct page *src_base = src; | |
484 | ||
485 | for (i = 0; i < pages_per_huge_page(h); ) { | |
486 | cond_resched(); | |
487 | copy_highpage(dst, src); | |
488 | ||
489 | i++; | |
490 | dst = mem_map_next(dst, dst_base, i); | |
491 | src = mem_map_next(src, src_base, i); | |
492 | } | |
493 | } | |
494 | ||
495 | void copy_huge_page(struct page *dst, struct page *src) | |
496 | { | |
497 | int i; | |
498 | struct hstate *h = page_hstate(src); | |
499 | ||
500 | if (unlikely(pages_per_huge_page(h) > MAX_ORDER_NR_PAGES)) { | |
501 | copy_gigantic_page(dst, src); | |
502 | return; | |
503 | } | |
504 | ||
505 | might_sleep(); | |
506 | for (i = 0; i < pages_per_huge_page(h); i++) { | |
507 | cond_resched(); | |
508 | copy_highpage(dst + i, src + i); | |
509 | } | |
510 | } | |
511 | ||
a5516438 | 512 | static void enqueue_huge_page(struct hstate *h, struct page *page) |
1da177e4 LT |
513 | { |
514 | int nid = page_to_nid(page); | |
0edaecfa | 515 | list_move(&page->lru, &h->hugepage_freelists[nid]); |
a5516438 AK |
516 | h->free_huge_pages++; |
517 | h->free_huge_pages_node[nid]++; | |
1da177e4 LT |
518 | } |
519 | ||
bf50bab2 NH |
520 | static struct page *dequeue_huge_page_node(struct hstate *h, int nid) |
521 | { | |
522 | struct page *page; | |
523 | ||
524 | if (list_empty(&h->hugepage_freelists[nid])) | |
525 | return NULL; | |
526 | page = list_entry(h->hugepage_freelists[nid].next, struct page, lru); | |
0edaecfa | 527 | list_move(&page->lru, &h->hugepage_activelist); |
a9869b83 | 528 | set_page_refcounted(page); |
bf50bab2 NH |
529 | h->free_huge_pages--; |
530 | h->free_huge_pages_node[nid]--; | |
531 | return page; | |
532 | } | |
533 | ||
a5516438 AK |
534 | static struct page *dequeue_huge_page_vma(struct hstate *h, |
535 | struct vm_area_struct *vma, | |
af0ed73e JK |
536 | unsigned long address, int avoid_reserve, |
537 | long chg) | |
1da177e4 | 538 | { |
b1c12cbc | 539 | struct page *page = NULL; |
480eccf9 | 540 | struct mempolicy *mpol; |
19770b32 | 541 | nodemask_t *nodemask; |
c0ff7453 | 542 | struct zonelist *zonelist; |
dd1a239f MG |
543 | struct zone *zone; |
544 | struct zoneref *z; | |
cc9a6c87 | 545 | unsigned int cpuset_mems_cookie; |
1da177e4 | 546 | |
a1e78772 MG |
547 | /* |
548 | * A child process with MAP_PRIVATE mappings created by their parent | |
549 | * have no page reserves. This check ensures that reservations are | |
550 | * not "stolen". The child may still get SIGKILLed | |
551 | */ | |
af0ed73e | 552 | if (!vma_has_reserves(vma, chg) && |
a5516438 | 553 | h->free_huge_pages - h->resv_huge_pages == 0) |
c0ff7453 | 554 | goto err; |
a1e78772 | 555 | |
04f2cbe3 | 556 | /* If reserves cannot be used, ensure enough pages are in the pool */ |
a5516438 | 557 | if (avoid_reserve && h->free_huge_pages - h->resv_huge_pages == 0) |
6eab04a8 | 558 | goto err; |
04f2cbe3 | 559 | |
9966c4bb JK |
560 | retry_cpuset: |
561 | cpuset_mems_cookie = get_mems_allowed(); | |
562 | zonelist = huge_zonelist(vma, address, | |
563 | htlb_alloc_mask, &mpol, &nodemask); | |
564 | ||
19770b32 MG |
565 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
566 | MAX_NR_ZONES - 1, nodemask) { | |
bf50bab2 NH |
567 | if (cpuset_zone_allowed_softwall(zone, htlb_alloc_mask)) { |
568 | page = dequeue_huge_page_node(h, zone_to_nid(zone)); | |
569 | if (page) { | |
af0ed73e JK |
570 | if (avoid_reserve) |
571 | break; | |
572 | if (!vma_has_reserves(vma, chg)) | |
573 | break; | |
574 | ||
07443a85 | 575 | SetPagePrivate(page); |
af0ed73e | 576 | h->resv_huge_pages--; |
bf50bab2 NH |
577 | break; |
578 | } | |
3abf7afd | 579 | } |
1da177e4 | 580 | } |
cc9a6c87 | 581 | |
52cd3b07 | 582 | mpol_cond_put(mpol); |
cc9a6c87 MG |
583 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
584 | goto retry_cpuset; | |
1da177e4 | 585 | return page; |
cc9a6c87 MG |
586 | |
587 | err: | |
cc9a6c87 | 588 | return NULL; |
1da177e4 LT |
589 | } |
590 | ||
a5516438 | 591 | static void update_and_free_page(struct hstate *h, struct page *page) |
6af2acb6 AL |
592 | { |
593 | int i; | |
a5516438 | 594 | |
18229df5 AW |
595 | VM_BUG_ON(h->order >= MAX_ORDER); |
596 | ||
a5516438 AK |
597 | h->nr_huge_pages--; |
598 | h->nr_huge_pages_node[page_to_nid(page)]--; | |
599 | for (i = 0; i < pages_per_huge_page(h); i++) { | |
32f84528 CF |
600 | page[i].flags &= ~(1 << PG_locked | 1 << PG_error | |
601 | 1 << PG_referenced | 1 << PG_dirty | | |
602 | 1 << PG_active | 1 << PG_reserved | | |
603 | 1 << PG_private | 1 << PG_writeback); | |
6af2acb6 | 604 | } |
9dd540e2 | 605 | VM_BUG_ON(hugetlb_cgroup_from_page(page)); |
6af2acb6 AL |
606 | set_compound_page_dtor(page, NULL); |
607 | set_page_refcounted(page); | |
7f2e9525 | 608 | arch_release_hugepage(page); |
a5516438 | 609 | __free_pages(page, huge_page_order(h)); |
6af2acb6 AL |
610 | } |
611 | ||
e5ff2159 AK |
612 | struct hstate *size_to_hstate(unsigned long size) |
613 | { | |
614 | struct hstate *h; | |
615 | ||
616 | for_each_hstate(h) { | |
617 | if (huge_page_size(h) == size) | |
618 | return h; | |
619 | } | |
620 | return NULL; | |
621 | } | |
622 | ||
27a85ef1 DG |
623 | static void free_huge_page(struct page *page) |
624 | { | |
a5516438 AK |
625 | /* |
626 | * Can't pass hstate in here because it is called from the | |
627 | * compound page destructor. | |
628 | */ | |
e5ff2159 | 629 | struct hstate *h = page_hstate(page); |
7893d1d5 | 630 | int nid = page_to_nid(page); |
90481622 DG |
631 | struct hugepage_subpool *spool = |
632 | (struct hugepage_subpool *)page_private(page); | |
07443a85 | 633 | bool restore_reserve; |
27a85ef1 | 634 | |
e5df70ab | 635 | set_page_private(page, 0); |
23be7468 | 636 | page->mapping = NULL; |
7893d1d5 | 637 | BUG_ON(page_count(page)); |
0fe6e20b | 638 | BUG_ON(page_mapcount(page)); |
07443a85 | 639 | restore_reserve = PagePrivate(page); |
27a85ef1 DG |
640 | |
641 | spin_lock(&hugetlb_lock); | |
6d76dcf4 AK |
642 | hugetlb_cgroup_uncharge_page(hstate_index(h), |
643 | pages_per_huge_page(h), page); | |
07443a85 JK |
644 | if (restore_reserve) |
645 | h->resv_huge_pages++; | |
646 | ||
aa888a74 | 647 | if (h->surplus_huge_pages_node[nid] && huge_page_order(h) < MAX_ORDER) { |
0edaecfa AK |
648 | /* remove the page from active list */ |
649 | list_del(&page->lru); | |
a5516438 AK |
650 | update_and_free_page(h, page); |
651 | h->surplus_huge_pages--; | |
652 | h->surplus_huge_pages_node[nid]--; | |
7893d1d5 | 653 | } else { |
5d3a551c | 654 | arch_clear_hugepage_flags(page); |
a5516438 | 655 | enqueue_huge_page(h, page); |
7893d1d5 | 656 | } |
27a85ef1 | 657 | spin_unlock(&hugetlb_lock); |
90481622 | 658 | hugepage_subpool_put_pages(spool, 1); |
27a85ef1 DG |
659 | } |
660 | ||
a5516438 | 661 | static void prep_new_huge_page(struct hstate *h, struct page *page, int nid) |
b7ba30c6 | 662 | { |
0edaecfa | 663 | INIT_LIST_HEAD(&page->lru); |
b7ba30c6 AK |
664 | set_compound_page_dtor(page, free_huge_page); |
665 | spin_lock(&hugetlb_lock); | |
9dd540e2 | 666 | set_hugetlb_cgroup(page, NULL); |
a5516438 AK |
667 | h->nr_huge_pages++; |
668 | h->nr_huge_pages_node[nid]++; | |
b7ba30c6 AK |
669 | spin_unlock(&hugetlb_lock); |
670 | put_page(page); /* free it into the hugepage allocator */ | |
671 | } | |
672 | ||
20a0307c WF |
673 | static void prep_compound_gigantic_page(struct page *page, unsigned long order) |
674 | { | |
675 | int i; | |
676 | int nr_pages = 1 << order; | |
677 | struct page *p = page + 1; | |
678 | ||
679 | /* we rely on prep_new_huge_page to set the destructor */ | |
680 | set_compound_order(page, order); | |
681 | __SetPageHead(page); | |
682 | for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) { | |
683 | __SetPageTail(p); | |
58a84aa9 | 684 | set_page_count(p, 0); |
20a0307c WF |
685 | p->first_page = page; |
686 | } | |
687 | } | |
688 | ||
7795912c AM |
689 | /* |
690 | * PageHuge() only returns true for hugetlbfs pages, but not for normal or | |
691 | * transparent huge pages. See the PageTransHuge() documentation for more | |
692 | * details. | |
693 | */ | |
20a0307c WF |
694 | int PageHuge(struct page *page) |
695 | { | |
696 | compound_page_dtor *dtor; | |
697 | ||
698 | if (!PageCompound(page)) | |
699 | return 0; | |
700 | ||
701 | page = compound_head(page); | |
702 | dtor = get_compound_page_dtor(page); | |
703 | ||
704 | return dtor == free_huge_page; | |
705 | } | |
43131e14 NH |
706 | EXPORT_SYMBOL_GPL(PageHuge); |
707 | ||
13d60f4b ZY |
708 | pgoff_t __basepage_index(struct page *page) |
709 | { | |
710 | struct page *page_head = compound_head(page); | |
711 | pgoff_t index = page_index(page_head); | |
712 | unsigned long compound_idx; | |
713 | ||
714 | if (!PageHuge(page_head)) | |
715 | return page_index(page); | |
716 | ||
717 | if (compound_order(page_head) >= MAX_ORDER) | |
718 | compound_idx = page_to_pfn(page) - page_to_pfn(page_head); | |
719 | else | |
720 | compound_idx = page - page_head; | |
721 | ||
722 | return (index << compound_order(page_head)) + compound_idx; | |
723 | } | |
724 | ||
a5516438 | 725 | static struct page *alloc_fresh_huge_page_node(struct hstate *h, int nid) |
1da177e4 | 726 | { |
1da177e4 | 727 | struct page *page; |
f96efd58 | 728 | |
aa888a74 AK |
729 | if (h->order >= MAX_ORDER) |
730 | return NULL; | |
731 | ||
6484eb3e | 732 | page = alloc_pages_exact_node(nid, |
551883ae NA |
733 | htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE| |
734 | __GFP_REPEAT|__GFP_NOWARN, | |
a5516438 | 735 | huge_page_order(h)); |
1da177e4 | 736 | if (page) { |
7f2e9525 | 737 | if (arch_prepare_hugepage(page)) { |
caff3a2c | 738 | __free_pages(page, huge_page_order(h)); |
7b8ee84d | 739 | return NULL; |
7f2e9525 | 740 | } |
a5516438 | 741 | prep_new_huge_page(h, page, nid); |
1da177e4 | 742 | } |
63b4613c NA |
743 | |
744 | return page; | |
745 | } | |
746 | ||
9a76db09 | 747 | /* |
6ae11b27 LS |
748 | * common helper functions for hstate_next_node_to_{alloc|free}. |
749 | * We may have allocated or freed a huge page based on a different | |
750 | * nodes_allowed previously, so h->next_node_to_{alloc|free} might | |
751 | * be outside of *nodes_allowed. Ensure that we use an allowed | |
752 | * node for alloc or free. | |
9a76db09 | 753 | */ |
6ae11b27 | 754 | static int next_node_allowed(int nid, nodemask_t *nodes_allowed) |
9a76db09 | 755 | { |
6ae11b27 | 756 | nid = next_node(nid, *nodes_allowed); |
9a76db09 | 757 | if (nid == MAX_NUMNODES) |
6ae11b27 | 758 | nid = first_node(*nodes_allowed); |
9a76db09 LS |
759 | VM_BUG_ON(nid >= MAX_NUMNODES); |
760 | ||
761 | return nid; | |
762 | } | |
763 | ||
6ae11b27 LS |
764 | static int get_valid_node_allowed(int nid, nodemask_t *nodes_allowed) |
765 | { | |
766 | if (!node_isset(nid, *nodes_allowed)) | |
767 | nid = next_node_allowed(nid, nodes_allowed); | |
768 | return nid; | |
769 | } | |
770 | ||
5ced66c9 | 771 | /* |
6ae11b27 LS |
772 | * returns the previously saved node ["this node"] from which to |
773 | * allocate a persistent huge page for the pool and advance the | |
774 | * next node from which to allocate, handling wrap at end of node | |
775 | * mask. | |
5ced66c9 | 776 | */ |
6ae11b27 LS |
777 | static int hstate_next_node_to_alloc(struct hstate *h, |
778 | nodemask_t *nodes_allowed) | |
5ced66c9 | 779 | { |
6ae11b27 LS |
780 | int nid; |
781 | ||
782 | VM_BUG_ON(!nodes_allowed); | |
783 | ||
784 | nid = get_valid_node_allowed(h->next_nid_to_alloc, nodes_allowed); | |
785 | h->next_nid_to_alloc = next_node_allowed(nid, nodes_allowed); | |
9a76db09 | 786 | |
9a76db09 | 787 | return nid; |
5ced66c9 AK |
788 | } |
789 | ||
e8c5c824 | 790 | /* |
6ae11b27 LS |
791 | * helper for free_pool_huge_page() - return the previously saved |
792 | * node ["this node"] from which to free a huge page. Advance the | |
793 | * next node id whether or not we find a free huge page to free so | |
794 | * that the next attempt to free addresses the next node. | |
e8c5c824 | 795 | */ |
6ae11b27 | 796 | static int hstate_next_node_to_free(struct hstate *h, nodemask_t *nodes_allowed) |
e8c5c824 | 797 | { |
6ae11b27 LS |
798 | int nid; |
799 | ||
800 | VM_BUG_ON(!nodes_allowed); | |
801 | ||
802 | nid = get_valid_node_allowed(h->next_nid_to_free, nodes_allowed); | |
803 | h->next_nid_to_free = next_node_allowed(nid, nodes_allowed); | |
9a76db09 | 804 | |
9a76db09 | 805 | return nid; |
e8c5c824 LS |
806 | } |
807 | ||
b2261026 JK |
808 | #define for_each_node_mask_to_alloc(hs, nr_nodes, node, mask) \ |
809 | for (nr_nodes = nodes_weight(*mask); \ | |
810 | nr_nodes > 0 && \ | |
811 | ((node = hstate_next_node_to_alloc(hs, mask)) || 1); \ | |
812 | nr_nodes--) | |
813 | ||
814 | #define for_each_node_mask_to_free(hs, nr_nodes, node, mask) \ | |
815 | for (nr_nodes = nodes_weight(*mask); \ | |
816 | nr_nodes > 0 && \ | |
817 | ((node = hstate_next_node_to_free(hs, mask)) || 1); \ | |
818 | nr_nodes--) | |
819 | ||
820 | static int alloc_fresh_huge_page(struct hstate *h, nodemask_t *nodes_allowed) | |
821 | { | |
822 | struct page *page; | |
823 | int nr_nodes, node; | |
824 | int ret = 0; | |
825 | ||
826 | for_each_node_mask_to_alloc(h, nr_nodes, node, nodes_allowed) { | |
827 | page = alloc_fresh_huge_page_node(h, node); | |
828 | if (page) { | |
829 | ret = 1; | |
830 | break; | |
831 | } | |
832 | } | |
833 | ||
834 | if (ret) | |
835 | count_vm_event(HTLB_BUDDY_PGALLOC); | |
836 | else | |
837 | count_vm_event(HTLB_BUDDY_PGALLOC_FAIL); | |
838 | ||
839 | return ret; | |
840 | } | |
841 | ||
e8c5c824 LS |
842 | /* |
843 | * Free huge page from pool from next node to free. | |
844 | * Attempt to keep persistent huge pages more or less | |
845 | * balanced over allowed nodes. | |
846 | * Called with hugetlb_lock locked. | |
847 | */ | |
6ae11b27 LS |
848 | static int free_pool_huge_page(struct hstate *h, nodemask_t *nodes_allowed, |
849 | bool acct_surplus) | |
e8c5c824 | 850 | { |
b2261026 | 851 | int nr_nodes, node; |
e8c5c824 LS |
852 | int ret = 0; |
853 | ||
b2261026 | 854 | for_each_node_mask_to_free(h, nr_nodes, node, nodes_allowed) { |
685f3457 LS |
855 | /* |
856 | * If we're returning unused surplus pages, only examine | |
857 | * nodes with surplus pages. | |
858 | */ | |
b2261026 JK |
859 | if ((!acct_surplus || h->surplus_huge_pages_node[node]) && |
860 | !list_empty(&h->hugepage_freelists[node])) { | |
e8c5c824 | 861 | struct page *page = |
b2261026 | 862 | list_entry(h->hugepage_freelists[node].next, |
e8c5c824 LS |
863 | struct page, lru); |
864 | list_del(&page->lru); | |
865 | h->free_huge_pages--; | |
b2261026 | 866 | h->free_huge_pages_node[node]--; |
685f3457 LS |
867 | if (acct_surplus) { |
868 | h->surplus_huge_pages--; | |
b2261026 | 869 | h->surplus_huge_pages_node[node]--; |
685f3457 | 870 | } |
e8c5c824 LS |
871 | update_and_free_page(h, page); |
872 | ret = 1; | |
9a76db09 | 873 | break; |
e8c5c824 | 874 | } |
b2261026 | 875 | } |
e8c5c824 LS |
876 | |
877 | return ret; | |
878 | } | |
879 | ||
bf50bab2 | 880 | static struct page *alloc_buddy_huge_page(struct hstate *h, int nid) |
7893d1d5 AL |
881 | { |
882 | struct page *page; | |
bf50bab2 | 883 | unsigned int r_nid; |
7893d1d5 | 884 | |
aa888a74 AK |
885 | if (h->order >= MAX_ORDER) |
886 | return NULL; | |
887 | ||
d1c3fb1f NA |
888 | /* |
889 | * Assume we will successfully allocate the surplus page to | |
890 | * prevent racing processes from causing the surplus to exceed | |
891 | * overcommit | |
892 | * | |
893 | * This however introduces a different race, where a process B | |
894 | * tries to grow the static hugepage pool while alloc_pages() is | |
895 | * called by process A. B will only examine the per-node | |
896 | * counters in determining if surplus huge pages can be | |
897 | * converted to normal huge pages in adjust_pool_surplus(). A | |
898 | * won't be able to increment the per-node counter, until the | |
899 | * lock is dropped by B, but B doesn't drop hugetlb_lock until | |
900 | * no more huge pages can be converted from surplus to normal | |
901 | * state (and doesn't try to convert again). Thus, we have a | |
902 | * case where a surplus huge page exists, the pool is grown, and | |
903 | * the surplus huge page still exists after, even though it | |
904 | * should just have been converted to a normal huge page. This | |
905 | * does not leak memory, though, as the hugepage will be freed | |
906 | * once it is out of use. It also does not allow the counters to | |
907 | * go out of whack in adjust_pool_surplus() as we don't modify | |
908 | * the node values until we've gotten the hugepage and only the | |
909 | * per-node value is checked there. | |
910 | */ | |
911 | spin_lock(&hugetlb_lock); | |
a5516438 | 912 | if (h->surplus_huge_pages >= h->nr_overcommit_huge_pages) { |
d1c3fb1f NA |
913 | spin_unlock(&hugetlb_lock); |
914 | return NULL; | |
915 | } else { | |
a5516438 AK |
916 | h->nr_huge_pages++; |
917 | h->surplus_huge_pages++; | |
d1c3fb1f NA |
918 | } |
919 | spin_unlock(&hugetlb_lock); | |
920 | ||
bf50bab2 NH |
921 | if (nid == NUMA_NO_NODE) |
922 | page = alloc_pages(htlb_alloc_mask|__GFP_COMP| | |
923 | __GFP_REPEAT|__GFP_NOWARN, | |
924 | huge_page_order(h)); | |
925 | else | |
926 | page = alloc_pages_exact_node(nid, | |
927 | htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE| | |
928 | __GFP_REPEAT|__GFP_NOWARN, huge_page_order(h)); | |
d1c3fb1f | 929 | |
caff3a2c GS |
930 | if (page && arch_prepare_hugepage(page)) { |
931 | __free_pages(page, huge_page_order(h)); | |
ea5768c7 | 932 | page = NULL; |
caff3a2c GS |
933 | } |
934 | ||
d1c3fb1f | 935 | spin_lock(&hugetlb_lock); |
7893d1d5 | 936 | if (page) { |
0edaecfa | 937 | INIT_LIST_HEAD(&page->lru); |
bf50bab2 | 938 | r_nid = page_to_nid(page); |
7893d1d5 | 939 | set_compound_page_dtor(page, free_huge_page); |
9dd540e2 | 940 | set_hugetlb_cgroup(page, NULL); |
d1c3fb1f NA |
941 | /* |
942 | * We incremented the global counters already | |
943 | */ | |
bf50bab2 NH |
944 | h->nr_huge_pages_node[r_nid]++; |
945 | h->surplus_huge_pages_node[r_nid]++; | |
3b116300 | 946 | __count_vm_event(HTLB_BUDDY_PGALLOC); |
d1c3fb1f | 947 | } else { |
a5516438 AK |
948 | h->nr_huge_pages--; |
949 | h->surplus_huge_pages--; | |
3b116300 | 950 | __count_vm_event(HTLB_BUDDY_PGALLOC_FAIL); |
7893d1d5 | 951 | } |
d1c3fb1f | 952 | spin_unlock(&hugetlb_lock); |
7893d1d5 AL |
953 | |
954 | return page; | |
955 | } | |
956 | ||
bf50bab2 NH |
957 | /* |
958 | * This allocation function is useful in the context where vma is irrelevant. | |
959 | * E.g. soft-offlining uses this function because it only cares physical | |
960 | * address of error page. | |
961 | */ | |
962 | struct page *alloc_huge_page_node(struct hstate *h, int nid) | |
963 | { | |
4ef91848 | 964 | struct page *page = NULL; |
bf50bab2 NH |
965 | |
966 | spin_lock(&hugetlb_lock); | |
4ef91848 JK |
967 | if (h->free_huge_pages - h->resv_huge_pages > 0) |
968 | page = dequeue_huge_page_node(h, nid); | |
bf50bab2 NH |
969 | spin_unlock(&hugetlb_lock); |
970 | ||
94ae8ba7 | 971 | if (!page) |
bf50bab2 NH |
972 | page = alloc_buddy_huge_page(h, nid); |
973 | ||
974 | return page; | |
975 | } | |
976 | ||
e4e574b7 | 977 | /* |
25985edc | 978 | * Increase the hugetlb pool such that it can accommodate a reservation |
e4e574b7 AL |
979 | * of size 'delta'. |
980 | */ | |
a5516438 | 981 | static int gather_surplus_pages(struct hstate *h, int delta) |
e4e574b7 AL |
982 | { |
983 | struct list_head surplus_list; | |
984 | struct page *page, *tmp; | |
985 | int ret, i; | |
986 | int needed, allocated; | |
28073b02 | 987 | bool alloc_ok = true; |
e4e574b7 | 988 | |
a5516438 | 989 | needed = (h->resv_huge_pages + delta) - h->free_huge_pages; |
ac09b3a1 | 990 | if (needed <= 0) { |
a5516438 | 991 | h->resv_huge_pages += delta; |
e4e574b7 | 992 | return 0; |
ac09b3a1 | 993 | } |
e4e574b7 AL |
994 | |
995 | allocated = 0; | |
996 | INIT_LIST_HEAD(&surplus_list); | |
997 | ||
998 | ret = -ENOMEM; | |
999 | retry: | |
1000 | spin_unlock(&hugetlb_lock); | |
1001 | for (i = 0; i < needed; i++) { | |
bf50bab2 | 1002 | page = alloc_buddy_huge_page(h, NUMA_NO_NODE); |
28073b02 HD |
1003 | if (!page) { |
1004 | alloc_ok = false; | |
1005 | break; | |
1006 | } | |
e4e574b7 AL |
1007 | list_add(&page->lru, &surplus_list); |
1008 | } | |
28073b02 | 1009 | allocated += i; |
e4e574b7 AL |
1010 | |
1011 | /* | |
1012 | * After retaking hugetlb_lock, we need to recalculate 'needed' | |
1013 | * because either resv_huge_pages or free_huge_pages may have changed. | |
1014 | */ | |
1015 | spin_lock(&hugetlb_lock); | |
a5516438 AK |
1016 | needed = (h->resv_huge_pages + delta) - |
1017 | (h->free_huge_pages + allocated); | |
28073b02 HD |
1018 | if (needed > 0) { |
1019 | if (alloc_ok) | |
1020 | goto retry; | |
1021 | /* | |
1022 | * We were not able to allocate enough pages to | |
1023 | * satisfy the entire reservation so we free what | |
1024 | * we've allocated so far. | |
1025 | */ | |
1026 | goto free; | |
1027 | } | |
e4e574b7 AL |
1028 | /* |
1029 | * The surplus_list now contains _at_least_ the number of extra pages | |
25985edc | 1030 | * needed to accommodate the reservation. Add the appropriate number |
e4e574b7 | 1031 | * of pages to the hugetlb pool and free the extras back to the buddy |
ac09b3a1 AL |
1032 | * allocator. Commit the entire reservation here to prevent another |
1033 | * process from stealing the pages as they are added to the pool but | |
1034 | * before they are reserved. | |
e4e574b7 AL |
1035 | */ |
1036 | needed += allocated; | |
a5516438 | 1037 | h->resv_huge_pages += delta; |
e4e574b7 | 1038 | ret = 0; |
a9869b83 | 1039 | |
19fc3f0a | 1040 | /* Free the needed pages to the hugetlb pool */ |
e4e574b7 | 1041 | list_for_each_entry_safe(page, tmp, &surplus_list, lru) { |
19fc3f0a AL |
1042 | if ((--needed) < 0) |
1043 | break; | |
a9869b83 NH |
1044 | /* |
1045 | * This page is now managed by the hugetlb allocator and has | |
1046 | * no users -- drop the buddy allocator's reference. | |
1047 | */ | |
1048 | put_page_testzero(page); | |
1049 | VM_BUG_ON(page_count(page)); | |
a5516438 | 1050 | enqueue_huge_page(h, page); |
19fc3f0a | 1051 | } |
28073b02 | 1052 | free: |
b0365c8d | 1053 | spin_unlock(&hugetlb_lock); |
19fc3f0a AL |
1054 | |
1055 | /* Free unnecessary surplus pages to the buddy allocator */ | |
c0d934ba JK |
1056 | list_for_each_entry_safe(page, tmp, &surplus_list, lru) |
1057 | put_page(page); | |
a9869b83 | 1058 | spin_lock(&hugetlb_lock); |
e4e574b7 AL |
1059 | |
1060 | return ret; | |
1061 | } | |
1062 | ||
1063 | /* | |
1064 | * When releasing a hugetlb pool reservation, any surplus pages that were | |
1065 | * allocated to satisfy the reservation must be explicitly freed if they were | |
1066 | * never used. | |
685f3457 | 1067 | * Called with hugetlb_lock held. |
e4e574b7 | 1068 | */ |
a5516438 AK |
1069 | static void return_unused_surplus_pages(struct hstate *h, |
1070 | unsigned long unused_resv_pages) | |
e4e574b7 | 1071 | { |
e4e574b7 AL |
1072 | unsigned long nr_pages; |
1073 | ||
ac09b3a1 | 1074 | /* Uncommit the reservation */ |
a5516438 | 1075 | h->resv_huge_pages -= unused_resv_pages; |
ac09b3a1 | 1076 | |
aa888a74 AK |
1077 | /* Cannot return gigantic pages currently */ |
1078 | if (h->order >= MAX_ORDER) | |
1079 | return; | |
1080 | ||
a5516438 | 1081 | nr_pages = min(unused_resv_pages, h->surplus_huge_pages); |
e4e574b7 | 1082 | |
685f3457 LS |
1083 | /* |
1084 | * We want to release as many surplus pages as possible, spread | |
9b5e5d0f LS |
1085 | * evenly across all nodes with memory. Iterate across these nodes |
1086 | * until we can no longer free unreserved surplus pages. This occurs | |
1087 | * when the nodes with surplus pages have no free pages. | |
1088 | * free_pool_huge_page() will balance the the freed pages across the | |
1089 | * on-line nodes with memory and will handle the hstate accounting. | |
685f3457 LS |
1090 | */ |
1091 | while (nr_pages--) { | |
8cebfcd0 | 1092 | if (!free_pool_huge_page(h, &node_states[N_MEMORY], 1)) |
685f3457 | 1093 | break; |
e4e574b7 AL |
1094 | } |
1095 | } | |
1096 | ||
c37f9fb1 AW |
1097 | /* |
1098 | * Determine if the huge page at addr within the vma has an associated | |
1099 | * reservation. Where it does not we will need to logically increase | |
90481622 DG |
1100 | * reservation and actually increase subpool usage before an allocation |
1101 | * can occur. Where any new reservation would be required the | |
1102 | * reservation change is prepared, but not committed. Once the page | |
1103 | * has been allocated from the subpool and instantiated the change should | |
1104 | * be committed via vma_commit_reservation. No action is required on | |
1105 | * failure. | |
c37f9fb1 | 1106 | */ |
e2f17d94 | 1107 | static long vma_needs_reservation(struct hstate *h, |
a5516438 | 1108 | struct vm_area_struct *vma, unsigned long addr) |
c37f9fb1 AW |
1109 | { |
1110 | struct address_space *mapping = vma->vm_file->f_mapping; | |
1111 | struct inode *inode = mapping->host; | |
1112 | ||
f83a275d | 1113 | if (vma->vm_flags & VM_MAYSHARE) { |
a5516438 | 1114 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
c37f9fb1 AW |
1115 | return region_chg(&inode->i_mapping->private_list, |
1116 | idx, idx + 1); | |
1117 | ||
84afd99b AW |
1118 | } else if (!is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { |
1119 | return 1; | |
c37f9fb1 | 1120 | |
84afd99b | 1121 | } else { |
e2f17d94 | 1122 | long err; |
a5516438 | 1123 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
f522c3ac | 1124 | struct resv_map *resv = vma_resv_map(vma); |
84afd99b | 1125 | |
f522c3ac | 1126 | err = region_chg(&resv->regions, idx, idx + 1); |
84afd99b AW |
1127 | if (err < 0) |
1128 | return err; | |
1129 | return 0; | |
1130 | } | |
c37f9fb1 | 1131 | } |
a5516438 AK |
1132 | static void vma_commit_reservation(struct hstate *h, |
1133 | struct vm_area_struct *vma, unsigned long addr) | |
c37f9fb1 AW |
1134 | { |
1135 | struct address_space *mapping = vma->vm_file->f_mapping; | |
1136 | struct inode *inode = mapping->host; | |
1137 | ||
f83a275d | 1138 | if (vma->vm_flags & VM_MAYSHARE) { |
a5516438 | 1139 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
c37f9fb1 | 1140 | region_add(&inode->i_mapping->private_list, idx, idx + 1); |
84afd99b AW |
1141 | |
1142 | } else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { | |
a5516438 | 1143 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
f522c3ac | 1144 | struct resv_map *resv = vma_resv_map(vma); |
84afd99b AW |
1145 | |
1146 | /* Mark this page used in the map. */ | |
f522c3ac | 1147 | region_add(&resv->regions, idx, idx + 1); |
c37f9fb1 AW |
1148 | } |
1149 | } | |
1150 | ||
a1e78772 | 1151 | static struct page *alloc_huge_page(struct vm_area_struct *vma, |
04f2cbe3 | 1152 | unsigned long addr, int avoid_reserve) |
1da177e4 | 1153 | { |
90481622 | 1154 | struct hugepage_subpool *spool = subpool_vma(vma); |
a5516438 | 1155 | struct hstate *h = hstate_vma(vma); |
348ea204 | 1156 | struct page *page; |
e2f17d94 | 1157 | long chg; |
6d76dcf4 AK |
1158 | int ret, idx; |
1159 | struct hugetlb_cgroup *h_cg; | |
a1e78772 | 1160 | |
6d76dcf4 | 1161 | idx = hstate_index(h); |
a1e78772 | 1162 | /* |
90481622 DG |
1163 | * Processes that did not create the mapping will have no |
1164 | * reserves and will not have accounted against subpool | |
1165 | * limit. Check that the subpool limit can be made before | |
1166 | * satisfying the allocation MAP_NORESERVE mappings may also | |
1167 | * need pages and subpool limit allocated allocated if no reserve | |
1168 | * mapping overlaps. | |
a1e78772 | 1169 | */ |
a5516438 | 1170 | chg = vma_needs_reservation(h, vma, addr); |
c37f9fb1 | 1171 | if (chg < 0) |
76dcee75 | 1172 | return ERR_PTR(-ENOMEM); |
8bb3f12e JK |
1173 | if (chg || avoid_reserve) |
1174 | if (hugepage_subpool_get_pages(spool, 1)) | |
76dcee75 | 1175 | return ERR_PTR(-ENOSPC); |
1da177e4 | 1176 | |
6d76dcf4 AK |
1177 | ret = hugetlb_cgroup_charge_cgroup(idx, pages_per_huge_page(h), &h_cg); |
1178 | if (ret) { | |
8bb3f12e JK |
1179 | if (chg || avoid_reserve) |
1180 | hugepage_subpool_put_pages(spool, 1); | |
6d76dcf4 AK |
1181 | return ERR_PTR(-ENOSPC); |
1182 | } | |
1da177e4 | 1183 | spin_lock(&hugetlb_lock); |
af0ed73e | 1184 | page = dequeue_huge_page_vma(h, vma, addr, avoid_reserve, chg); |
81a6fcae | 1185 | if (!page) { |
94ae8ba7 | 1186 | spin_unlock(&hugetlb_lock); |
bf50bab2 | 1187 | page = alloc_buddy_huge_page(h, NUMA_NO_NODE); |
68842c9b | 1188 | if (!page) { |
6d76dcf4 AK |
1189 | hugetlb_cgroup_uncharge_cgroup(idx, |
1190 | pages_per_huge_page(h), | |
1191 | h_cg); | |
8bb3f12e JK |
1192 | if (chg || avoid_reserve) |
1193 | hugepage_subpool_put_pages(spool, 1); | |
76dcee75 | 1194 | return ERR_PTR(-ENOSPC); |
68842c9b | 1195 | } |
79dbb236 AK |
1196 | spin_lock(&hugetlb_lock); |
1197 | list_move(&page->lru, &h->hugepage_activelist); | |
81a6fcae | 1198 | /* Fall through */ |
68842c9b | 1199 | } |
81a6fcae JK |
1200 | hugetlb_cgroup_commit_charge(idx, pages_per_huge_page(h), h_cg, page); |
1201 | spin_unlock(&hugetlb_lock); | |
348ea204 | 1202 | |
90481622 | 1203 | set_page_private(page, (unsigned long)spool); |
90d8b7e6 | 1204 | |
a5516438 | 1205 | vma_commit_reservation(h, vma, addr); |
90d8b7e6 | 1206 | return page; |
b45b5bd6 DG |
1207 | } |
1208 | ||
91f47662 | 1209 | int __weak alloc_bootmem_huge_page(struct hstate *h) |
aa888a74 AK |
1210 | { |
1211 | struct huge_bootmem_page *m; | |
b2261026 | 1212 | int nr_nodes, node; |
aa888a74 | 1213 | |
b2261026 | 1214 | for_each_node_mask_to_alloc(h, nr_nodes, node, &node_states[N_MEMORY]) { |
aa888a74 AK |
1215 | void *addr; |
1216 | ||
b2261026 | 1217 | addr = __alloc_bootmem_node_nopanic(NODE_DATA(node), |
aa888a74 AK |
1218 | huge_page_size(h), huge_page_size(h), 0); |
1219 | ||
1220 | if (addr) { | |
1221 | /* | |
1222 | * Use the beginning of the huge page to store the | |
1223 | * huge_bootmem_page struct (until gather_bootmem | |
1224 | * puts them into the mem_map). | |
1225 | */ | |
1226 | m = addr; | |
91f47662 | 1227 | goto found; |
aa888a74 | 1228 | } |
aa888a74 AK |
1229 | } |
1230 | return 0; | |
1231 | ||
1232 | found: | |
1233 | BUG_ON((unsigned long)virt_to_phys(m) & (huge_page_size(h) - 1)); | |
1234 | /* Put them into a private list first because mem_map is not up yet */ | |
1235 | list_add(&m->list, &huge_boot_pages); | |
1236 | m->hstate = h; | |
1237 | return 1; | |
1238 | } | |
1239 | ||
18229df5 AW |
1240 | static void prep_compound_huge_page(struct page *page, int order) |
1241 | { | |
1242 | if (unlikely(order > (MAX_ORDER - 1))) | |
1243 | prep_compound_gigantic_page(page, order); | |
1244 | else | |
1245 | prep_compound_page(page, order); | |
1246 | } | |
1247 | ||
aa888a74 AK |
1248 | /* Put bootmem huge pages into the standard lists after mem_map is up */ |
1249 | static void __init gather_bootmem_prealloc(void) | |
1250 | { | |
1251 | struct huge_bootmem_page *m; | |
1252 | ||
1253 | list_for_each_entry(m, &huge_boot_pages, list) { | |
aa888a74 | 1254 | struct hstate *h = m->hstate; |
ee8f248d BB |
1255 | struct page *page; |
1256 | ||
1257 | #ifdef CONFIG_HIGHMEM | |
1258 | page = pfn_to_page(m->phys >> PAGE_SHIFT); | |
1259 | free_bootmem_late((unsigned long)m, | |
1260 | sizeof(struct huge_bootmem_page)); | |
1261 | #else | |
1262 | page = virt_to_page(m); | |
1263 | #endif | |
aa888a74 AK |
1264 | __ClearPageReserved(page); |
1265 | WARN_ON(page_count(page) != 1); | |
18229df5 | 1266 | prep_compound_huge_page(page, h->order); |
aa888a74 | 1267 | prep_new_huge_page(h, page, page_to_nid(page)); |
b0320c7b RA |
1268 | /* |
1269 | * If we had gigantic hugepages allocated at boot time, we need | |
1270 | * to restore the 'stolen' pages to totalram_pages in order to | |
1271 | * fix confusing memory reports from free(1) and another | |
1272 | * side-effects, like CommitLimit going negative. | |
1273 | */ | |
1274 | if (h->order > (MAX_ORDER - 1)) | |
3dcc0571 | 1275 | adjust_managed_page_count(page, 1 << h->order); |
aa888a74 AK |
1276 | } |
1277 | } | |
1278 | ||
8faa8b07 | 1279 | static void __init hugetlb_hstate_alloc_pages(struct hstate *h) |
1da177e4 LT |
1280 | { |
1281 | unsigned long i; | |
a5516438 | 1282 | |
e5ff2159 | 1283 | for (i = 0; i < h->max_huge_pages; ++i) { |
aa888a74 AK |
1284 | if (h->order >= MAX_ORDER) { |
1285 | if (!alloc_bootmem_huge_page(h)) | |
1286 | break; | |
9b5e5d0f | 1287 | } else if (!alloc_fresh_huge_page(h, |
8cebfcd0 | 1288 | &node_states[N_MEMORY])) |
1da177e4 | 1289 | break; |
1da177e4 | 1290 | } |
8faa8b07 | 1291 | h->max_huge_pages = i; |
e5ff2159 AK |
1292 | } |
1293 | ||
1294 | static void __init hugetlb_init_hstates(void) | |
1295 | { | |
1296 | struct hstate *h; | |
1297 | ||
1298 | for_each_hstate(h) { | |
8faa8b07 AK |
1299 | /* oversize hugepages were init'ed in early boot */ |
1300 | if (h->order < MAX_ORDER) | |
1301 | hugetlb_hstate_alloc_pages(h); | |
e5ff2159 AK |
1302 | } |
1303 | } | |
1304 | ||
4abd32db AK |
1305 | static char * __init memfmt(char *buf, unsigned long n) |
1306 | { | |
1307 | if (n >= (1UL << 30)) | |
1308 | sprintf(buf, "%lu GB", n >> 30); | |
1309 | else if (n >= (1UL << 20)) | |
1310 | sprintf(buf, "%lu MB", n >> 20); | |
1311 | else | |
1312 | sprintf(buf, "%lu KB", n >> 10); | |
1313 | return buf; | |
1314 | } | |
1315 | ||
e5ff2159 AK |
1316 | static void __init report_hugepages(void) |
1317 | { | |
1318 | struct hstate *h; | |
1319 | ||
1320 | for_each_hstate(h) { | |
4abd32db | 1321 | char buf[32]; |
ffb22af5 | 1322 | pr_info("HugeTLB registered %s page size, pre-allocated %ld pages\n", |
4abd32db AK |
1323 | memfmt(buf, huge_page_size(h)), |
1324 | h->free_huge_pages); | |
e5ff2159 AK |
1325 | } |
1326 | } | |
1327 | ||
1da177e4 | 1328 | #ifdef CONFIG_HIGHMEM |
6ae11b27 LS |
1329 | static void try_to_free_low(struct hstate *h, unsigned long count, |
1330 | nodemask_t *nodes_allowed) | |
1da177e4 | 1331 | { |
4415cc8d CL |
1332 | int i; |
1333 | ||
aa888a74 AK |
1334 | if (h->order >= MAX_ORDER) |
1335 | return; | |
1336 | ||
6ae11b27 | 1337 | for_each_node_mask(i, *nodes_allowed) { |
1da177e4 | 1338 | struct page *page, *next; |
a5516438 AK |
1339 | struct list_head *freel = &h->hugepage_freelists[i]; |
1340 | list_for_each_entry_safe(page, next, freel, lru) { | |
1341 | if (count >= h->nr_huge_pages) | |
6b0c880d | 1342 | return; |
1da177e4 LT |
1343 | if (PageHighMem(page)) |
1344 | continue; | |
1345 | list_del(&page->lru); | |
e5ff2159 | 1346 | update_and_free_page(h, page); |
a5516438 AK |
1347 | h->free_huge_pages--; |
1348 | h->free_huge_pages_node[page_to_nid(page)]--; | |
1da177e4 LT |
1349 | } |
1350 | } | |
1351 | } | |
1352 | #else | |
6ae11b27 LS |
1353 | static inline void try_to_free_low(struct hstate *h, unsigned long count, |
1354 | nodemask_t *nodes_allowed) | |
1da177e4 LT |
1355 | { |
1356 | } | |
1357 | #endif | |
1358 | ||
20a0307c WF |
1359 | /* |
1360 | * Increment or decrement surplus_huge_pages. Keep node-specific counters | |
1361 | * balanced by operating on them in a round-robin fashion. | |
1362 | * Returns 1 if an adjustment was made. | |
1363 | */ | |
6ae11b27 LS |
1364 | static int adjust_pool_surplus(struct hstate *h, nodemask_t *nodes_allowed, |
1365 | int delta) | |
20a0307c | 1366 | { |
b2261026 | 1367 | int nr_nodes, node; |
20a0307c WF |
1368 | |
1369 | VM_BUG_ON(delta != -1 && delta != 1); | |
20a0307c | 1370 | |
b2261026 JK |
1371 | if (delta < 0) { |
1372 | for_each_node_mask_to_alloc(h, nr_nodes, node, nodes_allowed) { | |
1373 | if (h->surplus_huge_pages_node[node]) | |
1374 | goto found; | |
e8c5c824 | 1375 | } |
b2261026 JK |
1376 | } else { |
1377 | for_each_node_mask_to_free(h, nr_nodes, node, nodes_allowed) { | |
1378 | if (h->surplus_huge_pages_node[node] < | |
1379 | h->nr_huge_pages_node[node]) | |
1380 | goto found; | |
e8c5c824 | 1381 | } |
b2261026 JK |
1382 | } |
1383 | return 0; | |
20a0307c | 1384 | |
b2261026 JK |
1385 | found: |
1386 | h->surplus_huge_pages += delta; | |
1387 | h->surplus_huge_pages_node[node] += delta; | |
1388 | return 1; | |
20a0307c WF |
1389 | } |
1390 | ||
a5516438 | 1391 | #define persistent_huge_pages(h) (h->nr_huge_pages - h->surplus_huge_pages) |
6ae11b27 LS |
1392 | static unsigned long set_max_huge_pages(struct hstate *h, unsigned long count, |
1393 | nodemask_t *nodes_allowed) | |
1da177e4 | 1394 | { |
7893d1d5 | 1395 | unsigned long min_count, ret; |
1da177e4 | 1396 | |
aa888a74 AK |
1397 | if (h->order >= MAX_ORDER) |
1398 | return h->max_huge_pages; | |
1399 | ||
7893d1d5 AL |
1400 | /* |
1401 | * Increase the pool size | |
1402 | * First take pages out of surplus state. Then make up the | |
1403 | * remaining difference by allocating fresh huge pages. | |
d1c3fb1f NA |
1404 | * |
1405 | * We might race with alloc_buddy_huge_page() here and be unable | |
1406 | * to convert a surplus huge page to a normal huge page. That is | |
1407 | * not critical, though, it just means the overall size of the | |
1408 | * pool might be one hugepage larger than it needs to be, but | |
1409 | * within all the constraints specified by the sysctls. | |
7893d1d5 | 1410 | */ |
1da177e4 | 1411 | spin_lock(&hugetlb_lock); |
a5516438 | 1412 | while (h->surplus_huge_pages && count > persistent_huge_pages(h)) { |
6ae11b27 | 1413 | if (!adjust_pool_surplus(h, nodes_allowed, -1)) |
7893d1d5 AL |
1414 | break; |
1415 | } | |
1416 | ||
a5516438 | 1417 | while (count > persistent_huge_pages(h)) { |
7893d1d5 AL |
1418 | /* |
1419 | * If this allocation races such that we no longer need the | |
1420 | * page, free_huge_page will handle it by freeing the page | |
1421 | * and reducing the surplus. | |
1422 | */ | |
1423 | spin_unlock(&hugetlb_lock); | |
6ae11b27 | 1424 | ret = alloc_fresh_huge_page(h, nodes_allowed); |
7893d1d5 AL |
1425 | spin_lock(&hugetlb_lock); |
1426 | if (!ret) | |
1427 | goto out; | |
1428 | ||
536240f2 MG |
1429 | /* Bail for signals. Probably ctrl-c from user */ |
1430 | if (signal_pending(current)) | |
1431 | goto out; | |
7893d1d5 | 1432 | } |
7893d1d5 AL |
1433 | |
1434 | /* | |
1435 | * Decrease the pool size | |
1436 | * First return free pages to the buddy allocator (being careful | |
1437 | * to keep enough around to satisfy reservations). Then place | |
1438 | * pages into surplus state as needed so the pool will shrink | |
1439 | * to the desired size as pages become free. | |
d1c3fb1f NA |
1440 | * |
1441 | * By placing pages into the surplus state independent of the | |
1442 | * overcommit value, we are allowing the surplus pool size to | |
1443 | * exceed overcommit. There are few sane options here. Since | |
1444 | * alloc_buddy_huge_page() is checking the global counter, | |
1445 | * though, we'll note that we're not allowed to exceed surplus | |
1446 | * and won't grow the pool anywhere else. Not until one of the | |
1447 | * sysctls are changed, or the surplus pages go out of use. | |
7893d1d5 | 1448 | */ |
a5516438 | 1449 | min_count = h->resv_huge_pages + h->nr_huge_pages - h->free_huge_pages; |
6b0c880d | 1450 | min_count = max(count, min_count); |
6ae11b27 | 1451 | try_to_free_low(h, min_count, nodes_allowed); |
a5516438 | 1452 | while (min_count < persistent_huge_pages(h)) { |
6ae11b27 | 1453 | if (!free_pool_huge_page(h, nodes_allowed, 0)) |
1da177e4 | 1454 | break; |
1da177e4 | 1455 | } |
a5516438 | 1456 | while (count < persistent_huge_pages(h)) { |
6ae11b27 | 1457 | if (!adjust_pool_surplus(h, nodes_allowed, 1)) |
7893d1d5 AL |
1458 | break; |
1459 | } | |
1460 | out: | |
a5516438 | 1461 | ret = persistent_huge_pages(h); |
1da177e4 | 1462 | spin_unlock(&hugetlb_lock); |
7893d1d5 | 1463 | return ret; |
1da177e4 LT |
1464 | } |
1465 | ||
a3437870 NA |
1466 | #define HSTATE_ATTR_RO(_name) \ |
1467 | static struct kobj_attribute _name##_attr = __ATTR_RO(_name) | |
1468 | ||
1469 | #define HSTATE_ATTR(_name) \ | |
1470 | static struct kobj_attribute _name##_attr = \ | |
1471 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
1472 | ||
1473 | static struct kobject *hugepages_kobj; | |
1474 | static struct kobject *hstate_kobjs[HUGE_MAX_HSTATE]; | |
1475 | ||
9a305230 LS |
1476 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp); |
1477 | ||
1478 | static struct hstate *kobj_to_hstate(struct kobject *kobj, int *nidp) | |
a3437870 NA |
1479 | { |
1480 | int i; | |
9a305230 | 1481 | |
a3437870 | 1482 | for (i = 0; i < HUGE_MAX_HSTATE; i++) |
9a305230 LS |
1483 | if (hstate_kobjs[i] == kobj) { |
1484 | if (nidp) | |
1485 | *nidp = NUMA_NO_NODE; | |
a3437870 | 1486 | return &hstates[i]; |
9a305230 LS |
1487 | } |
1488 | ||
1489 | return kobj_to_node_hstate(kobj, nidp); | |
a3437870 NA |
1490 | } |
1491 | ||
06808b08 | 1492 | static ssize_t nr_hugepages_show_common(struct kobject *kobj, |
a3437870 NA |
1493 | struct kobj_attribute *attr, char *buf) |
1494 | { | |
9a305230 LS |
1495 | struct hstate *h; |
1496 | unsigned long nr_huge_pages; | |
1497 | int nid; | |
1498 | ||
1499 | h = kobj_to_hstate(kobj, &nid); | |
1500 | if (nid == NUMA_NO_NODE) | |
1501 | nr_huge_pages = h->nr_huge_pages; | |
1502 | else | |
1503 | nr_huge_pages = h->nr_huge_pages_node[nid]; | |
1504 | ||
1505 | return sprintf(buf, "%lu\n", nr_huge_pages); | |
a3437870 | 1506 | } |
adbe8726 | 1507 | |
06808b08 LS |
1508 | static ssize_t nr_hugepages_store_common(bool obey_mempolicy, |
1509 | struct kobject *kobj, struct kobj_attribute *attr, | |
1510 | const char *buf, size_t len) | |
a3437870 NA |
1511 | { |
1512 | int err; | |
9a305230 | 1513 | int nid; |
06808b08 | 1514 | unsigned long count; |
9a305230 | 1515 | struct hstate *h; |
bad44b5b | 1516 | NODEMASK_ALLOC(nodemask_t, nodes_allowed, GFP_KERNEL | __GFP_NORETRY); |
a3437870 | 1517 | |
3dbb95f7 | 1518 | err = kstrtoul(buf, 10, &count); |
73ae31e5 | 1519 | if (err) |
adbe8726 | 1520 | goto out; |
a3437870 | 1521 | |
9a305230 | 1522 | h = kobj_to_hstate(kobj, &nid); |
adbe8726 EM |
1523 | if (h->order >= MAX_ORDER) { |
1524 | err = -EINVAL; | |
1525 | goto out; | |
1526 | } | |
1527 | ||
9a305230 LS |
1528 | if (nid == NUMA_NO_NODE) { |
1529 | /* | |
1530 | * global hstate attribute | |
1531 | */ | |
1532 | if (!(obey_mempolicy && | |
1533 | init_nodemask_of_mempolicy(nodes_allowed))) { | |
1534 | NODEMASK_FREE(nodes_allowed); | |
8cebfcd0 | 1535 | nodes_allowed = &node_states[N_MEMORY]; |
9a305230 LS |
1536 | } |
1537 | } else if (nodes_allowed) { | |
1538 | /* | |
1539 | * per node hstate attribute: adjust count to global, | |
1540 | * but restrict alloc/free to the specified node. | |
1541 | */ | |
1542 | count += h->nr_huge_pages - h->nr_huge_pages_node[nid]; | |
1543 | init_nodemask_of_node(nodes_allowed, nid); | |
1544 | } else | |
8cebfcd0 | 1545 | nodes_allowed = &node_states[N_MEMORY]; |
9a305230 | 1546 | |
06808b08 | 1547 | h->max_huge_pages = set_max_huge_pages(h, count, nodes_allowed); |
a3437870 | 1548 | |
8cebfcd0 | 1549 | if (nodes_allowed != &node_states[N_MEMORY]) |
06808b08 LS |
1550 | NODEMASK_FREE(nodes_allowed); |
1551 | ||
1552 | return len; | |
adbe8726 EM |
1553 | out: |
1554 | NODEMASK_FREE(nodes_allowed); | |
1555 | return err; | |
06808b08 LS |
1556 | } |
1557 | ||
1558 | static ssize_t nr_hugepages_show(struct kobject *kobj, | |
1559 | struct kobj_attribute *attr, char *buf) | |
1560 | { | |
1561 | return nr_hugepages_show_common(kobj, attr, buf); | |
1562 | } | |
1563 | ||
1564 | static ssize_t nr_hugepages_store(struct kobject *kobj, | |
1565 | struct kobj_attribute *attr, const char *buf, size_t len) | |
1566 | { | |
1567 | return nr_hugepages_store_common(false, kobj, attr, buf, len); | |
a3437870 NA |
1568 | } |
1569 | HSTATE_ATTR(nr_hugepages); | |
1570 | ||
06808b08 LS |
1571 | #ifdef CONFIG_NUMA |
1572 | ||
1573 | /* | |
1574 | * hstate attribute for optionally mempolicy-based constraint on persistent | |
1575 | * huge page alloc/free. | |
1576 | */ | |
1577 | static ssize_t nr_hugepages_mempolicy_show(struct kobject *kobj, | |
1578 | struct kobj_attribute *attr, char *buf) | |
1579 | { | |
1580 | return nr_hugepages_show_common(kobj, attr, buf); | |
1581 | } | |
1582 | ||
1583 | static ssize_t nr_hugepages_mempolicy_store(struct kobject *kobj, | |
1584 | struct kobj_attribute *attr, const char *buf, size_t len) | |
1585 | { | |
1586 | return nr_hugepages_store_common(true, kobj, attr, buf, len); | |
1587 | } | |
1588 | HSTATE_ATTR(nr_hugepages_mempolicy); | |
1589 | #endif | |
1590 | ||
1591 | ||
a3437870 NA |
1592 | static ssize_t nr_overcommit_hugepages_show(struct kobject *kobj, |
1593 | struct kobj_attribute *attr, char *buf) | |
1594 | { | |
9a305230 | 1595 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
a3437870 NA |
1596 | return sprintf(buf, "%lu\n", h->nr_overcommit_huge_pages); |
1597 | } | |
adbe8726 | 1598 | |
a3437870 NA |
1599 | static ssize_t nr_overcommit_hugepages_store(struct kobject *kobj, |
1600 | struct kobj_attribute *attr, const char *buf, size_t count) | |
1601 | { | |
1602 | int err; | |
1603 | unsigned long input; | |
9a305230 | 1604 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
a3437870 | 1605 | |
adbe8726 EM |
1606 | if (h->order >= MAX_ORDER) |
1607 | return -EINVAL; | |
1608 | ||
3dbb95f7 | 1609 | err = kstrtoul(buf, 10, &input); |
a3437870 | 1610 | if (err) |
73ae31e5 | 1611 | return err; |
a3437870 NA |
1612 | |
1613 | spin_lock(&hugetlb_lock); | |
1614 | h->nr_overcommit_huge_pages = input; | |
1615 | spin_unlock(&hugetlb_lock); | |
1616 | ||
1617 | return count; | |
1618 | } | |
1619 | HSTATE_ATTR(nr_overcommit_hugepages); | |
1620 | ||
1621 | static ssize_t free_hugepages_show(struct kobject *kobj, | |
1622 | struct kobj_attribute *attr, char *buf) | |
1623 | { | |
9a305230 LS |
1624 | struct hstate *h; |
1625 | unsigned long free_huge_pages; | |
1626 | int nid; | |
1627 | ||
1628 | h = kobj_to_hstate(kobj, &nid); | |
1629 | if (nid == NUMA_NO_NODE) | |
1630 | free_huge_pages = h->free_huge_pages; | |
1631 | else | |
1632 | free_huge_pages = h->free_huge_pages_node[nid]; | |
1633 | ||
1634 | return sprintf(buf, "%lu\n", free_huge_pages); | |
a3437870 NA |
1635 | } |
1636 | HSTATE_ATTR_RO(free_hugepages); | |
1637 | ||
1638 | static ssize_t resv_hugepages_show(struct kobject *kobj, | |
1639 | struct kobj_attribute *attr, char *buf) | |
1640 | { | |
9a305230 | 1641 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
a3437870 NA |
1642 | return sprintf(buf, "%lu\n", h->resv_huge_pages); |
1643 | } | |
1644 | HSTATE_ATTR_RO(resv_hugepages); | |
1645 | ||
1646 | static ssize_t surplus_hugepages_show(struct kobject *kobj, | |
1647 | struct kobj_attribute *attr, char *buf) | |
1648 | { | |
9a305230 LS |
1649 | struct hstate *h; |
1650 | unsigned long surplus_huge_pages; | |
1651 | int nid; | |
1652 | ||
1653 | h = kobj_to_hstate(kobj, &nid); | |
1654 | if (nid == NUMA_NO_NODE) | |
1655 | surplus_huge_pages = h->surplus_huge_pages; | |
1656 | else | |
1657 | surplus_huge_pages = h->surplus_huge_pages_node[nid]; | |
1658 | ||
1659 | return sprintf(buf, "%lu\n", surplus_huge_pages); | |
a3437870 NA |
1660 | } |
1661 | HSTATE_ATTR_RO(surplus_hugepages); | |
1662 | ||
1663 | static struct attribute *hstate_attrs[] = { | |
1664 | &nr_hugepages_attr.attr, | |
1665 | &nr_overcommit_hugepages_attr.attr, | |
1666 | &free_hugepages_attr.attr, | |
1667 | &resv_hugepages_attr.attr, | |
1668 | &surplus_hugepages_attr.attr, | |
06808b08 LS |
1669 | #ifdef CONFIG_NUMA |
1670 | &nr_hugepages_mempolicy_attr.attr, | |
1671 | #endif | |
a3437870 NA |
1672 | NULL, |
1673 | }; | |
1674 | ||
1675 | static struct attribute_group hstate_attr_group = { | |
1676 | .attrs = hstate_attrs, | |
1677 | }; | |
1678 | ||
094e9539 JM |
1679 | static int hugetlb_sysfs_add_hstate(struct hstate *h, struct kobject *parent, |
1680 | struct kobject **hstate_kobjs, | |
1681 | struct attribute_group *hstate_attr_group) | |
a3437870 NA |
1682 | { |
1683 | int retval; | |
972dc4de | 1684 | int hi = hstate_index(h); |
a3437870 | 1685 | |
9a305230 LS |
1686 | hstate_kobjs[hi] = kobject_create_and_add(h->name, parent); |
1687 | if (!hstate_kobjs[hi]) | |
a3437870 NA |
1688 | return -ENOMEM; |
1689 | ||
9a305230 | 1690 | retval = sysfs_create_group(hstate_kobjs[hi], hstate_attr_group); |
a3437870 | 1691 | if (retval) |
9a305230 | 1692 | kobject_put(hstate_kobjs[hi]); |
a3437870 NA |
1693 | |
1694 | return retval; | |
1695 | } | |
1696 | ||
1697 | static void __init hugetlb_sysfs_init(void) | |
1698 | { | |
1699 | struct hstate *h; | |
1700 | int err; | |
1701 | ||
1702 | hugepages_kobj = kobject_create_and_add("hugepages", mm_kobj); | |
1703 | if (!hugepages_kobj) | |
1704 | return; | |
1705 | ||
1706 | for_each_hstate(h) { | |
9a305230 LS |
1707 | err = hugetlb_sysfs_add_hstate(h, hugepages_kobj, |
1708 | hstate_kobjs, &hstate_attr_group); | |
a3437870 | 1709 | if (err) |
ffb22af5 | 1710 | pr_err("Hugetlb: Unable to add hstate %s", h->name); |
a3437870 NA |
1711 | } |
1712 | } | |
1713 | ||
9a305230 LS |
1714 | #ifdef CONFIG_NUMA |
1715 | ||
1716 | /* | |
1717 | * node_hstate/s - associate per node hstate attributes, via their kobjects, | |
10fbcf4c KS |
1718 | * with node devices in node_devices[] using a parallel array. The array |
1719 | * index of a node device or _hstate == node id. | |
1720 | * This is here to avoid any static dependency of the node device driver, in | |
9a305230 LS |
1721 | * the base kernel, on the hugetlb module. |
1722 | */ | |
1723 | struct node_hstate { | |
1724 | struct kobject *hugepages_kobj; | |
1725 | struct kobject *hstate_kobjs[HUGE_MAX_HSTATE]; | |
1726 | }; | |
1727 | struct node_hstate node_hstates[MAX_NUMNODES]; | |
1728 | ||
1729 | /* | |
10fbcf4c | 1730 | * A subset of global hstate attributes for node devices |
9a305230 LS |
1731 | */ |
1732 | static struct attribute *per_node_hstate_attrs[] = { | |
1733 | &nr_hugepages_attr.attr, | |
1734 | &free_hugepages_attr.attr, | |
1735 | &surplus_hugepages_attr.attr, | |
1736 | NULL, | |
1737 | }; | |
1738 | ||
1739 | static struct attribute_group per_node_hstate_attr_group = { | |
1740 | .attrs = per_node_hstate_attrs, | |
1741 | }; | |
1742 | ||
1743 | /* | |
10fbcf4c | 1744 | * kobj_to_node_hstate - lookup global hstate for node device hstate attr kobj. |
9a305230 LS |
1745 | * Returns node id via non-NULL nidp. |
1746 | */ | |
1747 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp) | |
1748 | { | |
1749 | int nid; | |
1750 | ||
1751 | for (nid = 0; nid < nr_node_ids; nid++) { | |
1752 | struct node_hstate *nhs = &node_hstates[nid]; | |
1753 | int i; | |
1754 | for (i = 0; i < HUGE_MAX_HSTATE; i++) | |
1755 | if (nhs->hstate_kobjs[i] == kobj) { | |
1756 | if (nidp) | |
1757 | *nidp = nid; | |
1758 | return &hstates[i]; | |
1759 | } | |
1760 | } | |
1761 | ||
1762 | BUG(); | |
1763 | return NULL; | |
1764 | } | |
1765 | ||
1766 | /* | |
10fbcf4c | 1767 | * Unregister hstate attributes from a single node device. |
9a305230 LS |
1768 | * No-op if no hstate attributes attached. |
1769 | */ | |
3cd8b44f | 1770 | static void hugetlb_unregister_node(struct node *node) |
9a305230 LS |
1771 | { |
1772 | struct hstate *h; | |
10fbcf4c | 1773 | struct node_hstate *nhs = &node_hstates[node->dev.id]; |
9a305230 LS |
1774 | |
1775 | if (!nhs->hugepages_kobj) | |
9b5e5d0f | 1776 | return; /* no hstate attributes */ |
9a305230 | 1777 | |
972dc4de AK |
1778 | for_each_hstate(h) { |
1779 | int idx = hstate_index(h); | |
1780 | if (nhs->hstate_kobjs[idx]) { | |
1781 | kobject_put(nhs->hstate_kobjs[idx]); | |
1782 | nhs->hstate_kobjs[idx] = NULL; | |
9a305230 | 1783 | } |
972dc4de | 1784 | } |
9a305230 LS |
1785 | |
1786 | kobject_put(nhs->hugepages_kobj); | |
1787 | nhs->hugepages_kobj = NULL; | |
1788 | } | |
1789 | ||
1790 | /* | |
10fbcf4c | 1791 | * hugetlb module exit: unregister hstate attributes from node devices |
9a305230 LS |
1792 | * that have them. |
1793 | */ | |
1794 | static void hugetlb_unregister_all_nodes(void) | |
1795 | { | |
1796 | int nid; | |
1797 | ||
1798 | /* | |
10fbcf4c | 1799 | * disable node device registrations. |
9a305230 LS |
1800 | */ |
1801 | register_hugetlbfs_with_node(NULL, NULL); | |
1802 | ||
1803 | /* | |
1804 | * remove hstate attributes from any nodes that have them. | |
1805 | */ | |
1806 | for (nid = 0; nid < nr_node_ids; nid++) | |
8732794b | 1807 | hugetlb_unregister_node(node_devices[nid]); |
9a305230 LS |
1808 | } |
1809 | ||
1810 | /* | |
10fbcf4c | 1811 | * Register hstate attributes for a single node device. |
9a305230 LS |
1812 | * No-op if attributes already registered. |
1813 | */ | |
3cd8b44f | 1814 | static void hugetlb_register_node(struct node *node) |
9a305230 LS |
1815 | { |
1816 | struct hstate *h; | |
10fbcf4c | 1817 | struct node_hstate *nhs = &node_hstates[node->dev.id]; |
9a305230 LS |
1818 | int err; |
1819 | ||
1820 | if (nhs->hugepages_kobj) | |
1821 | return; /* already allocated */ | |
1822 | ||
1823 | nhs->hugepages_kobj = kobject_create_and_add("hugepages", | |
10fbcf4c | 1824 | &node->dev.kobj); |
9a305230 LS |
1825 | if (!nhs->hugepages_kobj) |
1826 | return; | |
1827 | ||
1828 | for_each_hstate(h) { | |
1829 | err = hugetlb_sysfs_add_hstate(h, nhs->hugepages_kobj, | |
1830 | nhs->hstate_kobjs, | |
1831 | &per_node_hstate_attr_group); | |
1832 | if (err) { | |
ffb22af5 AM |
1833 | pr_err("Hugetlb: Unable to add hstate %s for node %d\n", |
1834 | h->name, node->dev.id); | |
9a305230 LS |
1835 | hugetlb_unregister_node(node); |
1836 | break; | |
1837 | } | |
1838 | } | |
1839 | } | |
1840 | ||
1841 | /* | |
9b5e5d0f | 1842 | * hugetlb init time: register hstate attributes for all registered node |
10fbcf4c KS |
1843 | * devices of nodes that have memory. All on-line nodes should have |
1844 | * registered their associated device by this time. | |
9a305230 LS |
1845 | */ |
1846 | static void hugetlb_register_all_nodes(void) | |
1847 | { | |
1848 | int nid; | |
1849 | ||
8cebfcd0 | 1850 | for_each_node_state(nid, N_MEMORY) { |
8732794b | 1851 | struct node *node = node_devices[nid]; |
10fbcf4c | 1852 | if (node->dev.id == nid) |
9a305230 LS |
1853 | hugetlb_register_node(node); |
1854 | } | |
1855 | ||
1856 | /* | |
10fbcf4c | 1857 | * Let the node device driver know we're here so it can |
9a305230 LS |
1858 | * [un]register hstate attributes on node hotplug. |
1859 | */ | |
1860 | register_hugetlbfs_with_node(hugetlb_register_node, | |
1861 | hugetlb_unregister_node); | |
1862 | } | |
1863 | #else /* !CONFIG_NUMA */ | |
1864 | ||
1865 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp) | |
1866 | { | |
1867 | BUG(); | |
1868 | if (nidp) | |
1869 | *nidp = -1; | |
1870 | return NULL; | |
1871 | } | |
1872 | ||
1873 | static void hugetlb_unregister_all_nodes(void) { } | |
1874 | ||
1875 | static void hugetlb_register_all_nodes(void) { } | |
1876 | ||
1877 | #endif | |
1878 | ||
a3437870 NA |
1879 | static void __exit hugetlb_exit(void) |
1880 | { | |
1881 | struct hstate *h; | |
1882 | ||
9a305230 LS |
1883 | hugetlb_unregister_all_nodes(); |
1884 | ||
a3437870 | 1885 | for_each_hstate(h) { |
972dc4de | 1886 | kobject_put(hstate_kobjs[hstate_index(h)]); |
a3437870 NA |
1887 | } |
1888 | ||
1889 | kobject_put(hugepages_kobj); | |
1890 | } | |
1891 | module_exit(hugetlb_exit); | |
1892 | ||
1893 | static int __init hugetlb_init(void) | |
1894 | { | |
0ef89d25 BH |
1895 | /* Some platform decide whether they support huge pages at boot |
1896 | * time. On these, such as powerpc, HPAGE_SHIFT is set to 0 when | |
1897 | * there is no such support | |
1898 | */ | |
1899 | if (HPAGE_SHIFT == 0) | |
1900 | return 0; | |
a3437870 | 1901 | |
e11bfbfc NP |
1902 | if (!size_to_hstate(default_hstate_size)) { |
1903 | default_hstate_size = HPAGE_SIZE; | |
1904 | if (!size_to_hstate(default_hstate_size)) | |
1905 | hugetlb_add_hstate(HUGETLB_PAGE_ORDER); | |
a3437870 | 1906 | } |
972dc4de | 1907 | default_hstate_idx = hstate_index(size_to_hstate(default_hstate_size)); |
e11bfbfc NP |
1908 | if (default_hstate_max_huge_pages) |
1909 | default_hstate.max_huge_pages = default_hstate_max_huge_pages; | |
a3437870 NA |
1910 | |
1911 | hugetlb_init_hstates(); | |
aa888a74 | 1912 | gather_bootmem_prealloc(); |
a3437870 NA |
1913 | report_hugepages(); |
1914 | ||
1915 | hugetlb_sysfs_init(); | |
9a305230 | 1916 | hugetlb_register_all_nodes(); |
7179e7bf | 1917 | hugetlb_cgroup_file_init(); |
9a305230 | 1918 | |
a3437870 NA |
1919 | return 0; |
1920 | } | |
1921 | module_init(hugetlb_init); | |
1922 | ||
1923 | /* Should be called on processing a hugepagesz=... option */ | |
1924 | void __init hugetlb_add_hstate(unsigned order) | |
1925 | { | |
1926 | struct hstate *h; | |
8faa8b07 AK |
1927 | unsigned long i; |
1928 | ||
a3437870 | 1929 | if (size_to_hstate(PAGE_SIZE << order)) { |
ffb22af5 | 1930 | pr_warning("hugepagesz= specified twice, ignoring\n"); |
a3437870 NA |
1931 | return; |
1932 | } | |
47d38344 | 1933 | BUG_ON(hugetlb_max_hstate >= HUGE_MAX_HSTATE); |
a3437870 | 1934 | BUG_ON(order == 0); |
47d38344 | 1935 | h = &hstates[hugetlb_max_hstate++]; |
a3437870 NA |
1936 | h->order = order; |
1937 | h->mask = ~((1ULL << (order + PAGE_SHIFT)) - 1); | |
8faa8b07 AK |
1938 | h->nr_huge_pages = 0; |
1939 | h->free_huge_pages = 0; | |
1940 | for (i = 0; i < MAX_NUMNODES; ++i) | |
1941 | INIT_LIST_HEAD(&h->hugepage_freelists[i]); | |
0edaecfa | 1942 | INIT_LIST_HEAD(&h->hugepage_activelist); |
8cebfcd0 LJ |
1943 | h->next_nid_to_alloc = first_node(node_states[N_MEMORY]); |
1944 | h->next_nid_to_free = first_node(node_states[N_MEMORY]); | |
a3437870 NA |
1945 | snprintf(h->name, HSTATE_NAME_LEN, "hugepages-%lukB", |
1946 | huge_page_size(h)/1024); | |
8faa8b07 | 1947 | |
a3437870 NA |
1948 | parsed_hstate = h; |
1949 | } | |
1950 | ||
e11bfbfc | 1951 | static int __init hugetlb_nrpages_setup(char *s) |
a3437870 NA |
1952 | { |
1953 | unsigned long *mhp; | |
8faa8b07 | 1954 | static unsigned long *last_mhp; |
a3437870 NA |
1955 | |
1956 | /* | |
47d38344 | 1957 | * !hugetlb_max_hstate means we haven't parsed a hugepagesz= parameter yet, |
a3437870 NA |
1958 | * so this hugepages= parameter goes to the "default hstate". |
1959 | */ | |
47d38344 | 1960 | if (!hugetlb_max_hstate) |
a3437870 NA |
1961 | mhp = &default_hstate_max_huge_pages; |
1962 | else | |
1963 | mhp = &parsed_hstate->max_huge_pages; | |
1964 | ||
8faa8b07 | 1965 | if (mhp == last_mhp) { |
ffb22af5 AM |
1966 | pr_warning("hugepages= specified twice without " |
1967 | "interleaving hugepagesz=, ignoring\n"); | |
8faa8b07 AK |
1968 | return 1; |
1969 | } | |
1970 | ||
a3437870 NA |
1971 | if (sscanf(s, "%lu", mhp) <= 0) |
1972 | *mhp = 0; | |
1973 | ||
8faa8b07 AK |
1974 | /* |
1975 | * Global state is always initialized later in hugetlb_init. | |
1976 | * But we need to allocate >= MAX_ORDER hstates here early to still | |
1977 | * use the bootmem allocator. | |
1978 | */ | |
47d38344 | 1979 | if (hugetlb_max_hstate && parsed_hstate->order >= MAX_ORDER) |
8faa8b07 AK |
1980 | hugetlb_hstate_alloc_pages(parsed_hstate); |
1981 | ||
1982 | last_mhp = mhp; | |
1983 | ||
a3437870 NA |
1984 | return 1; |
1985 | } | |
e11bfbfc NP |
1986 | __setup("hugepages=", hugetlb_nrpages_setup); |
1987 | ||
1988 | static int __init hugetlb_default_setup(char *s) | |
1989 | { | |
1990 | default_hstate_size = memparse(s, &s); | |
1991 | return 1; | |
1992 | } | |
1993 | __setup("default_hugepagesz=", hugetlb_default_setup); | |
a3437870 | 1994 | |
8a213460 NA |
1995 | static unsigned int cpuset_mems_nr(unsigned int *array) |
1996 | { | |
1997 | int node; | |
1998 | unsigned int nr = 0; | |
1999 | ||
2000 | for_each_node_mask(node, cpuset_current_mems_allowed) | |
2001 | nr += array[node]; | |
2002 | ||
2003 | return nr; | |
2004 | } | |
2005 | ||
2006 | #ifdef CONFIG_SYSCTL | |
06808b08 LS |
2007 | static int hugetlb_sysctl_handler_common(bool obey_mempolicy, |
2008 | struct ctl_table *table, int write, | |
2009 | void __user *buffer, size_t *length, loff_t *ppos) | |
1da177e4 | 2010 | { |
e5ff2159 AK |
2011 | struct hstate *h = &default_hstate; |
2012 | unsigned long tmp; | |
08d4a246 | 2013 | int ret; |
e5ff2159 | 2014 | |
c033a93c | 2015 | tmp = h->max_huge_pages; |
e5ff2159 | 2016 | |
adbe8726 EM |
2017 | if (write && h->order >= MAX_ORDER) |
2018 | return -EINVAL; | |
2019 | ||
e5ff2159 AK |
2020 | table->data = &tmp; |
2021 | table->maxlen = sizeof(unsigned long); | |
08d4a246 MH |
2022 | ret = proc_doulongvec_minmax(table, write, buffer, length, ppos); |
2023 | if (ret) | |
2024 | goto out; | |
e5ff2159 | 2025 | |
06808b08 | 2026 | if (write) { |
bad44b5b DR |
2027 | NODEMASK_ALLOC(nodemask_t, nodes_allowed, |
2028 | GFP_KERNEL | __GFP_NORETRY); | |
06808b08 LS |
2029 | if (!(obey_mempolicy && |
2030 | init_nodemask_of_mempolicy(nodes_allowed))) { | |
2031 | NODEMASK_FREE(nodes_allowed); | |
8cebfcd0 | 2032 | nodes_allowed = &node_states[N_MEMORY]; |
06808b08 LS |
2033 | } |
2034 | h->max_huge_pages = set_max_huge_pages(h, tmp, nodes_allowed); | |
2035 | ||
8cebfcd0 | 2036 | if (nodes_allowed != &node_states[N_MEMORY]) |
06808b08 LS |
2037 | NODEMASK_FREE(nodes_allowed); |
2038 | } | |
08d4a246 MH |
2039 | out: |
2040 | return ret; | |
1da177e4 | 2041 | } |
396faf03 | 2042 | |
06808b08 LS |
2043 | int hugetlb_sysctl_handler(struct ctl_table *table, int write, |
2044 | void __user *buffer, size_t *length, loff_t *ppos) | |
2045 | { | |
2046 | ||
2047 | return hugetlb_sysctl_handler_common(false, table, write, | |
2048 | buffer, length, ppos); | |
2049 | } | |
2050 | ||
2051 | #ifdef CONFIG_NUMA | |
2052 | int hugetlb_mempolicy_sysctl_handler(struct ctl_table *table, int write, | |
2053 | void __user *buffer, size_t *length, loff_t *ppos) | |
2054 | { | |
2055 | return hugetlb_sysctl_handler_common(true, table, write, | |
2056 | buffer, length, ppos); | |
2057 | } | |
2058 | #endif /* CONFIG_NUMA */ | |
2059 | ||
396faf03 | 2060 | int hugetlb_treat_movable_handler(struct ctl_table *table, int write, |
8d65af78 | 2061 | void __user *buffer, |
396faf03 MG |
2062 | size_t *length, loff_t *ppos) |
2063 | { | |
8d65af78 | 2064 | proc_dointvec(table, write, buffer, length, ppos); |
396faf03 MG |
2065 | if (hugepages_treat_as_movable) |
2066 | htlb_alloc_mask = GFP_HIGHUSER_MOVABLE; | |
2067 | else | |
2068 | htlb_alloc_mask = GFP_HIGHUSER; | |
2069 | return 0; | |
2070 | } | |
2071 | ||
a3d0c6aa | 2072 | int hugetlb_overcommit_handler(struct ctl_table *table, int write, |
8d65af78 | 2073 | void __user *buffer, |
a3d0c6aa NA |
2074 | size_t *length, loff_t *ppos) |
2075 | { | |
a5516438 | 2076 | struct hstate *h = &default_hstate; |
e5ff2159 | 2077 | unsigned long tmp; |
08d4a246 | 2078 | int ret; |
e5ff2159 | 2079 | |
c033a93c | 2080 | tmp = h->nr_overcommit_huge_pages; |
e5ff2159 | 2081 | |
adbe8726 EM |
2082 | if (write && h->order >= MAX_ORDER) |
2083 | return -EINVAL; | |
2084 | ||
e5ff2159 AK |
2085 | table->data = &tmp; |
2086 | table->maxlen = sizeof(unsigned long); | |
08d4a246 MH |
2087 | ret = proc_doulongvec_minmax(table, write, buffer, length, ppos); |
2088 | if (ret) | |
2089 | goto out; | |
e5ff2159 AK |
2090 | |
2091 | if (write) { | |
2092 | spin_lock(&hugetlb_lock); | |
2093 | h->nr_overcommit_huge_pages = tmp; | |
2094 | spin_unlock(&hugetlb_lock); | |
2095 | } | |
08d4a246 MH |
2096 | out: |
2097 | return ret; | |
a3d0c6aa NA |
2098 | } |
2099 | ||
1da177e4 LT |
2100 | #endif /* CONFIG_SYSCTL */ |
2101 | ||
e1759c21 | 2102 | void hugetlb_report_meminfo(struct seq_file *m) |
1da177e4 | 2103 | { |
a5516438 | 2104 | struct hstate *h = &default_hstate; |
e1759c21 | 2105 | seq_printf(m, |
4f98a2fe RR |
2106 | "HugePages_Total: %5lu\n" |
2107 | "HugePages_Free: %5lu\n" | |
2108 | "HugePages_Rsvd: %5lu\n" | |
2109 | "HugePages_Surp: %5lu\n" | |
2110 | "Hugepagesize: %8lu kB\n", | |
a5516438 AK |
2111 | h->nr_huge_pages, |
2112 | h->free_huge_pages, | |
2113 | h->resv_huge_pages, | |
2114 | h->surplus_huge_pages, | |
2115 | 1UL << (huge_page_order(h) + PAGE_SHIFT - 10)); | |
1da177e4 LT |
2116 | } |
2117 | ||
2118 | int hugetlb_report_node_meminfo(int nid, char *buf) | |
2119 | { | |
a5516438 | 2120 | struct hstate *h = &default_hstate; |
1da177e4 LT |
2121 | return sprintf(buf, |
2122 | "Node %d HugePages_Total: %5u\n" | |
a1de0919 NA |
2123 | "Node %d HugePages_Free: %5u\n" |
2124 | "Node %d HugePages_Surp: %5u\n", | |
a5516438 AK |
2125 | nid, h->nr_huge_pages_node[nid], |
2126 | nid, h->free_huge_pages_node[nid], | |
2127 | nid, h->surplus_huge_pages_node[nid]); | |
1da177e4 LT |
2128 | } |
2129 | ||
949f7ec5 DR |
2130 | void hugetlb_show_meminfo(void) |
2131 | { | |
2132 | struct hstate *h; | |
2133 | int nid; | |
2134 | ||
2135 | for_each_node_state(nid, N_MEMORY) | |
2136 | for_each_hstate(h) | |
2137 | pr_info("Node %d hugepages_total=%u hugepages_free=%u hugepages_surp=%u hugepages_size=%lukB\n", | |
2138 | nid, | |
2139 | h->nr_huge_pages_node[nid], | |
2140 | h->free_huge_pages_node[nid], | |
2141 | h->surplus_huge_pages_node[nid], | |
2142 | 1UL << (huge_page_order(h) + PAGE_SHIFT - 10)); | |
2143 | } | |
2144 | ||
1da177e4 LT |
2145 | /* Return the number pages of memory we physically have, in PAGE_SIZE units. */ |
2146 | unsigned long hugetlb_total_pages(void) | |
2147 | { | |
d0028588 WL |
2148 | struct hstate *h; |
2149 | unsigned long nr_total_pages = 0; | |
2150 | ||
2151 | for_each_hstate(h) | |
2152 | nr_total_pages += h->nr_huge_pages * pages_per_huge_page(h); | |
2153 | return nr_total_pages; | |
1da177e4 | 2154 | } |
1da177e4 | 2155 | |
a5516438 | 2156 | static int hugetlb_acct_memory(struct hstate *h, long delta) |
fc1b8a73 MG |
2157 | { |
2158 | int ret = -ENOMEM; | |
2159 | ||
2160 | spin_lock(&hugetlb_lock); | |
2161 | /* | |
2162 | * When cpuset is configured, it breaks the strict hugetlb page | |
2163 | * reservation as the accounting is done on a global variable. Such | |
2164 | * reservation is completely rubbish in the presence of cpuset because | |
2165 | * the reservation is not checked against page availability for the | |
2166 | * current cpuset. Application can still potentially OOM'ed by kernel | |
2167 | * with lack of free htlb page in cpuset that the task is in. | |
2168 | * Attempt to enforce strict accounting with cpuset is almost | |
2169 | * impossible (or too ugly) because cpuset is too fluid that | |
2170 | * task or memory node can be dynamically moved between cpusets. | |
2171 | * | |
2172 | * The change of semantics for shared hugetlb mapping with cpuset is | |
2173 | * undesirable. However, in order to preserve some of the semantics, | |
2174 | * we fall back to check against current free page availability as | |
2175 | * a best attempt and hopefully to minimize the impact of changing | |
2176 | * semantics that cpuset has. | |
2177 | */ | |
2178 | if (delta > 0) { | |
a5516438 | 2179 | if (gather_surplus_pages(h, delta) < 0) |
fc1b8a73 MG |
2180 | goto out; |
2181 | ||
a5516438 AK |
2182 | if (delta > cpuset_mems_nr(h->free_huge_pages_node)) { |
2183 | return_unused_surplus_pages(h, delta); | |
fc1b8a73 MG |
2184 | goto out; |
2185 | } | |
2186 | } | |
2187 | ||
2188 | ret = 0; | |
2189 | if (delta < 0) | |
a5516438 | 2190 | return_unused_surplus_pages(h, (unsigned long) -delta); |
fc1b8a73 MG |
2191 | |
2192 | out: | |
2193 | spin_unlock(&hugetlb_lock); | |
2194 | return ret; | |
2195 | } | |
2196 | ||
84afd99b AW |
2197 | static void hugetlb_vm_op_open(struct vm_area_struct *vma) |
2198 | { | |
f522c3ac | 2199 | struct resv_map *resv = vma_resv_map(vma); |
84afd99b AW |
2200 | |
2201 | /* | |
2202 | * This new VMA should share its siblings reservation map if present. | |
2203 | * The VMA will only ever have a valid reservation map pointer where | |
2204 | * it is being copied for another still existing VMA. As that VMA | |
25985edc | 2205 | * has a reference to the reservation map it cannot disappear until |
84afd99b AW |
2206 | * after this open call completes. It is therefore safe to take a |
2207 | * new reference here without additional locking. | |
2208 | */ | |
f522c3ac JK |
2209 | if (resv) |
2210 | kref_get(&resv->refs); | |
84afd99b AW |
2211 | } |
2212 | ||
c50ac050 DH |
2213 | static void resv_map_put(struct vm_area_struct *vma) |
2214 | { | |
f522c3ac | 2215 | struct resv_map *resv = vma_resv_map(vma); |
c50ac050 | 2216 | |
f522c3ac | 2217 | if (!resv) |
c50ac050 | 2218 | return; |
f522c3ac | 2219 | kref_put(&resv->refs, resv_map_release); |
c50ac050 DH |
2220 | } |
2221 | ||
a1e78772 MG |
2222 | static void hugetlb_vm_op_close(struct vm_area_struct *vma) |
2223 | { | |
a5516438 | 2224 | struct hstate *h = hstate_vma(vma); |
f522c3ac | 2225 | struct resv_map *resv = vma_resv_map(vma); |
90481622 | 2226 | struct hugepage_subpool *spool = subpool_vma(vma); |
84afd99b AW |
2227 | unsigned long reserve; |
2228 | unsigned long start; | |
2229 | unsigned long end; | |
2230 | ||
f522c3ac | 2231 | if (resv) { |
a5516438 AK |
2232 | start = vma_hugecache_offset(h, vma, vma->vm_start); |
2233 | end = vma_hugecache_offset(h, vma, vma->vm_end); | |
84afd99b AW |
2234 | |
2235 | reserve = (end - start) - | |
f522c3ac | 2236 | region_count(&resv->regions, start, end); |
84afd99b | 2237 | |
c50ac050 | 2238 | resv_map_put(vma); |
84afd99b | 2239 | |
7251ff78 | 2240 | if (reserve) { |
a5516438 | 2241 | hugetlb_acct_memory(h, -reserve); |
90481622 | 2242 | hugepage_subpool_put_pages(spool, reserve); |
7251ff78 | 2243 | } |
84afd99b | 2244 | } |
a1e78772 MG |
2245 | } |
2246 | ||
1da177e4 LT |
2247 | /* |
2248 | * We cannot handle pagefaults against hugetlb pages at all. They cause | |
2249 | * handle_mm_fault() to try to instantiate regular-sized pages in the | |
2250 | * hugegpage VMA. do_page_fault() is supposed to trap this, so BUG is we get | |
2251 | * this far. | |
2252 | */ | |
d0217ac0 | 2253 | static int hugetlb_vm_op_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
2254 | { |
2255 | BUG(); | |
d0217ac0 | 2256 | return 0; |
1da177e4 LT |
2257 | } |
2258 | ||
f0f37e2f | 2259 | const struct vm_operations_struct hugetlb_vm_ops = { |
d0217ac0 | 2260 | .fault = hugetlb_vm_op_fault, |
84afd99b | 2261 | .open = hugetlb_vm_op_open, |
a1e78772 | 2262 | .close = hugetlb_vm_op_close, |
1da177e4 LT |
2263 | }; |
2264 | ||
1e8f889b DG |
2265 | static pte_t make_huge_pte(struct vm_area_struct *vma, struct page *page, |
2266 | int writable) | |
63551ae0 DG |
2267 | { |
2268 | pte_t entry; | |
2269 | ||
1e8f889b | 2270 | if (writable) { |
106c992a GS |
2271 | entry = huge_pte_mkwrite(huge_pte_mkdirty(mk_huge_pte(page, |
2272 | vma->vm_page_prot))); | |
63551ae0 | 2273 | } else { |
106c992a GS |
2274 | entry = huge_pte_wrprotect(mk_huge_pte(page, |
2275 | vma->vm_page_prot)); | |
63551ae0 DG |
2276 | } |
2277 | entry = pte_mkyoung(entry); | |
2278 | entry = pte_mkhuge(entry); | |
d9ed9faa | 2279 | entry = arch_make_huge_pte(entry, vma, page, writable); |
63551ae0 DG |
2280 | |
2281 | return entry; | |
2282 | } | |
2283 | ||
1e8f889b DG |
2284 | static void set_huge_ptep_writable(struct vm_area_struct *vma, |
2285 | unsigned long address, pte_t *ptep) | |
2286 | { | |
2287 | pte_t entry; | |
2288 | ||
106c992a | 2289 | entry = huge_pte_mkwrite(huge_pte_mkdirty(huge_ptep_get(ptep))); |
32f84528 | 2290 | if (huge_ptep_set_access_flags(vma, address, ptep, entry, 1)) |
4b3073e1 | 2291 | update_mmu_cache(vma, address, ptep); |
1e8f889b DG |
2292 | } |
2293 | ||
2294 | ||
63551ae0 DG |
2295 | int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src, |
2296 | struct vm_area_struct *vma) | |
2297 | { | |
2298 | pte_t *src_pte, *dst_pte, entry; | |
2299 | struct page *ptepage; | |
1c59827d | 2300 | unsigned long addr; |
1e8f889b | 2301 | int cow; |
a5516438 AK |
2302 | struct hstate *h = hstate_vma(vma); |
2303 | unsigned long sz = huge_page_size(h); | |
1e8f889b DG |
2304 | |
2305 | cow = (vma->vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; | |
63551ae0 | 2306 | |
a5516438 | 2307 | for (addr = vma->vm_start; addr < vma->vm_end; addr += sz) { |
c74df32c HD |
2308 | src_pte = huge_pte_offset(src, addr); |
2309 | if (!src_pte) | |
2310 | continue; | |
a5516438 | 2311 | dst_pte = huge_pte_alloc(dst, addr, sz); |
63551ae0 DG |
2312 | if (!dst_pte) |
2313 | goto nomem; | |
c5c99429 LW |
2314 | |
2315 | /* If the pagetables are shared don't copy or take references */ | |
2316 | if (dst_pte == src_pte) | |
2317 | continue; | |
2318 | ||
c74df32c | 2319 | spin_lock(&dst->page_table_lock); |
46478758 | 2320 | spin_lock_nested(&src->page_table_lock, SINGLE_DEPTH_NESTING); |
7f2e9525 | 2321 | if (!huge_pte_none(huge_ptep_get(src_pte))) { |
1e8f889b | 2322 | if (cow) |
7f2e9525 GS |
2323 | huge_ptep_set_wrprotect(src, addr, src_pte); |
2324 | entry = huge_ptep_get(src_pte); | |
1c59827d HD |
2325 | ptepage = pte_page(entry); |
2326 | get_page(ptepage); | |
0fe6e20b | 2327 | page_dup_rmap(ptepage); |
1c59827d HD |
2328 | set_huge_pte_at(dst, addr, dst_pte, entry); |
2329 | } | |
2330 | spin_unlock(&src->page_table_lock); | |
c74df32c | 2331 | spin_unlock(&dst->page_table_lock); |
63551ae0 DG |
2332 | } |
2333 | return 0; | |
2334 | ||
2335 | nomem: | |
2336 | return -ENOMEM; | |
2337 | } | |
2338 | ||
290408d4 NH |
2339 | static int is_hugetlb_entry_migration(pte_t pte) |
2340 | { | |
2341 | swp_entry_t swp; | |
2342 | ||
2343 | if (huge_pte_none(pte) || pte_present(pte)) | |
2344 | return 0; | |
2345 | swp = pte_to_swp_entry(pte); | |
32f84528 | 2346 | if (non_swap_entry(swp) && is_migration_entry(swp)) |
290408d4 | 2347 | return 1; |
32f84528 | 2348 | else |
290408d4 NH |
2349 | return 0; |
2350 | } | |
2351 | ||
fd6a03ed NH |
2352 | static int is_hugetlb_entry_hwpoisoned(pte_t pte) |
2353 | { | |
2354 | swp_entry_t swp; | |
2355 | ||
2356 | if (huge_pte_none(pte) || pte_present(pte)) | |
2357 | return 0; | |
2358 | swp = pte_to_swp_entry(pte); | |
32f84528 | 2359 | if (non_swap_entry(swp) && is_hwpoison_entry(swp)) |
fd6a03ed | 2360 | return 1; |
32f84528 | 2361 | else |
fd6a03ed NH |
2362 | return 0; |
2363 | } | |
2364 | ||
24669e58 AK |
2365 | void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct *vma, |
2366 | unsigned long start, unsigned long end, | |
2367 | struct page *ref_page) | |
63551ae0 | 2368 | { |
24669e58 | 2369 | int force_flush = 0; |
63551ae0 DG |
2370 | struct mm_struct *mm = vma->vm_mm; |
2371 | unsigned long address; | |
c7546f8f | 2372 | pte_t *ptep; |
63551ae0 DG |
2373 | pte_t pte; |
2374 | struct page *page; | |
a5516438 AK |
2375 | struct hstate *h = hstate_vma(vma); |
2376 | unsigned long sz = huge_page_size(h); | |
2ec74c3e SG |
2377 | const unsigned long mmun_start = start; /* For mmu_notifiers */ |
2378 | const unsigned long mmun_end = end; /* For mmu_notifiers */ | |
a5516438 | 2379 | |
63551ae0 | 2380 | WARN_ON(!is_vm_hugetlb_page(vma)); |
a5516438 AK |
2381 | BUG_ON(start & ~huge_page_mask(h)); |
2382 | BUG_ON(end & ~huge_page_mask(h)); | |
63551ae0 | 2383 | |
24669e58 | 2384 | tlb_start_vma(tlb, vma); |
2ec74c3e | 2385 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
24669e58 | 2386 | again: |
508034a3 | 2387 | spin_lock(&mm->page_table_lock); |
a5516438 | 2388 | for (address = start; address < end; address += sz) { |
c7546f8f | 2389 | ptep = huge_pte_offset(mm, address); |
4c887265 | 2390 | if (!ptep) |
c7546f8f DG |
2391 | continue; |
2392 | ||
39dde65c CK |
2393 | if (huge_pmd_unshare(mm, &address, ptep)) |
2394 | continue; | |
2395 | ||
6629326b HD |
2396 | pte = huge_ptep_get(ptep); |
2397 | if (huge_pte_none(pte)) | |
2398 | continue; | |
2399 | ||
2400 | /* | |
2401 | * HWPoisoned hugepage is already unmapped and dropped reference | |
2402 | */ | |
8c4894c6 | 2403 | if (unlikely(is_hugetlb_entry_hwpoisoned(pte))) { |
106c992a | 2404 | huge_pte_clear(mm, address, ptep); |
6629326b | 2405 | continue; |
8c4894c6 | 2406 | } |
6629326b HD |
2407 | |
2408 | page = pte_page(pte); | |
04f2cbe3 MG |
2409 | /* |
2410 | * If a reference page is supplied, it is because a specific | |
2411 | * page is being unmapped, not a range. Ensure the page we | |
2412 | * are about to unmap is the actual page of interest. | |
2413 | */ | |
2414 | if (ref_page) { | |
04f2cbe3 MG |
2415 | if (page != ref_page) |
2416 | continue; | |
2417 | ||
2418 | /* | |
2419 | * Mark the VMA as having unmapped its page so that | |
2420 | * future faults in this VMA will fail rather than | |
2421 | * looking like data was lost | |
2422 | */ | |
2423 | set_vma_resv_flags(vma, HPAGE_RESV_UNMAPPED); | |
2424 | } | |
2425 | ||
c7546f8f | 2426 | pte = huge_ptep_get_and_clear(mm, address, ptep); |
24669e58 | 2427 | tlb_remove_tlb_entry(tlb, ptep, address); |
106c992a | 2428 | if (huge_pte_dirty(pte)) |
6649a386 | 2429 | set_page_dirty(page); |
9e81130b | 2430 | |
24669e58 AK |
2431 | page_remove_rmap(page); |
2432 | force_flush = !__tlb_remove_page(tlb, page); | |
2433 | if (force_flush) | |
2434 | break; | |
9e81130b HD |
2435 | /* Bail out after unmapping reference page if supplied */ |
2436 | if (ref_page) | |
2437 | break; | |
63551ae0 | 2438 | } |
cd2934a3 | 2439 | spin_unlock(&mm->page_table_lock); |
24669e58 AK |
2440 | /* |
2441 | * mmu_gather ran out of room to batch pages, we break out of | |
2442 | * the PTE lock to avoid doing the potential expensive TLB invalidate | |
2443 | * and page-free while holding it. | |
2444 | */ | |
2445 | if (force_flush) { | |
2446 | force_flush = 0; | |
2447 | tlb_flush_mmu(tlb); | |
2448 | if (address < end && !ref_page) | |
2449 | goto again; | |
fe1668ae | 2450 | } |
2ec74c3e | 2451 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
24669e58 | 2452 | tlb_end_vma(tlb, vma); |
1da177e4 | 2453 | } |
63551ae0 | 2454 | |
d833352a MG |
2455 | void __unmap_hugepage_range_final(struct mmu_gather *tlb, |
2456 | struct vm_area_struct *vma, unsigned long start, | |
2457 | unsigned long end, struct page *ref_page) | |
2458 | { | |
2459 | __unmap_hugepage_range(tlb, vma, start, end, ref_page); | |
2460 | ||
2461 | /* | |
2462 | * Clear this flag so that x86's huge_pmd_share page_table_shareable | |
2463 | * test will fail on a vma being torn down, and not grab a page table | |
2464 | * on its way out. We're lucky that the flag has such an appropriate | |
2465 | * name, and can in fact be safely cleared here. We could clear it | |
2466 | * before the __unmap_hugepage_range above, but all that's necessary | |
2467 | * is to clear it before releasing the i_mmap_mutex. This works | |
2468 | * because in the context this is called, the VMA is about to be | |
2469 | * destroyed and the i_mmap_mutex is held. | |
2470 | */ | |
2471 | vma->vm_flags &= ~VM_MAYSHARE; | |
2472 | } | |
2473 | ||
502717f4 | 2474 | void unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start, |
04f2cbe3 | 2475 | unsigned long end, struct page *ref_page) |
502717f4 | 2476 | { |
24669e58 AK |
2477 | struct mm_struct *mm; |
2478 | struct mmu_gather tlb; | |
2479 | ||
2480 | mm = vma->vm_mm; | |
2481 | ||
2b047252 | 2482 | tlb_gather_mmu(&tlb, mm, start, end); |
24669e58 AK |
2483 | __unmap_hugepage_range(&tlb, vma, start, end, ref_page); |
2484 | tlb_finish_mmu(&tlb, start, end); | |
502717f4 CK |
2485 | } |
2486 | ||
04f2cbe3 MG |
2487 | /* |
2488 | * This is called when the original mapper is failing to COW a MAP_PRIVATE | |
2489 | * mappping it owns the reserve page for. The intention is to unmap the page | |
2490 | * from other VMAs and let the children be SIGKILLed if they are faulting the | |
2491 | * same region. | |
2492 | */ | |
2a4b3ded HH |
2493 | static int unmap_ref_private(struct mm_struct *mm, struct vm_area_struct *vma, |
2494 | struct page *page, unsigned long address) | |
04f2cbe3 | 2495 | { |
7526674d | 2496 | struct hstate *h = hstate_vma(vma); |
04f2cbe3 MG |
2497 | struct vm_area_struct *iter_vma; |
2498 | struct address_space *mapping; | |
04f2cbe3 MG |
2499 | pgoff_t pgoff; |
2500 | ||
2501 | /* | |
2502 | * vm_pgoff is in PAGE_SIZE units, hence the different calculation | |
2503 | * from page cache lookup which is in HPAGE_SIZE units. | |
2504 | */ | |
7526674d | 2505 | address = address & huge_page_mask(h); |
36e4f20a MH |
2506 | pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + |
2507 | vma->vm_pgoff; | |
496ad9aa | 2508 | mapping = file_inode(vma->vm_file)->i_mapping; |
04f2cbe3 | 2509 | |
4eb2b1dc MG |
2510 | /* |
2511 | * Take the mapping lock for the duration of the table walk. As | |
2512 | * this mapping should be shared between all the VMAs, | |
2513 | * __unmap_hugepage_range() is called as the lock is already held | |
2514 | */ | |
3d48ae45 | 2515 | mutex_lock(&mapping->i_mmap_mutex); |
6b2dbba8 | 2516 | vma_interval_tree_foreach(iter_vma, &mapping->i_mmap, pgoff, pgoff) { |
04f2cbe3 MG |
2517 | /* Do not unmap the current VMA */ |
2518 | if (iter_vma == vma) | |
2519 | continue; | |
2520 | ||
2521 | /* | |
2522 | * Unmap the page from other VMAs without their own reserves. | |
2523 | * They get marked to be SIGKILLed if they fault in these | |
2524 | * areas. This is because a future no-page fault on this VMA | |
2525 | * could insert a zeroed page instead of the data existing | |
2526 | * from the time of fork. This would look like data corruption | |
2527 | */ | |
2528 | if (!is_vma_resv_set(iter_vma, HPAGE_RESV_OWNER)) | |
24669e58 AK |
2529 | unmap_hugepage_range(iter_vma, address, |
2530 | address + huge_page_size(h), page); | |
04f2cbe3 | 2531 | } |
3d48ae45 | 2532 | mutex_unlock(&mapping->i_mmap_mutex); |
04f2cbe3 MG |
2533 | |
2534 | return 1; | |
2535 | } | |
2536 | ||
0fe6e20b NH |
2537 | /* |
2538 | * Hugetlb_cow() should be called with page lock of the original hugepage held. | |
ef009b25 MH |
2539 | * Called with hugetlb_instantiation_mutex held and pte_page locked so we |
2540 | * cannot race with other handlers or page migration. | |
2541 | * Keep the pte_same checks anyway to make transition from the mutex easier. | |
0fe6e20b | 2542 | */ |
1e8f889b | 2543 | static int hugetlb_cow(struct mm_struct *mm, struct vm_area_struct *vma, |
04f2cbe3 MG |
2544 | unsigned long address, pte_t *ptep, pte_t pte, |
2545 | struct page *pagecache_page) | |
1e8f889b | 2546 | { |
a5516438 | 2547 | struct hstate *h = hstate_vma(vma); |
1e8f889b | 2548 | struct page *old_page, *new_page; |
04f2cbe3 | 2549 | int outside_reserve = 0; |
2ec74c3e SG |
2550 | unsigned long mmun_start; /* For mmu_notifiers */ |
2551 | unsigned long mmun_end; /* For mmu_notifiers */ | |
1e8f889b DG |
2552 | |
2553 | old_page = pte_page(pte); | |
2554 | ||
04f2cbe3 | 2555 | retry_avoidcopy: |
1e8f889b DG |
2556 | /* If no-one else is actually using this page, avoid the copy |
2557 | * and just make the page writable */ | |
37a2140d JK |
2558 | if (page_mapcount(old_page) == 1 && PageAnon(old_page)) { |
2559 | page_move_anon_rmap(old_page, vma, address); | |
1e8f889b | 2560 | set_huge_ptep_writable(vma, address, ptep); |
83c54070 | 2561 | return 0; |
1e8f889b DG |
2562 | } |
2563 | ||
04f2cbe3 MG |
2564 | /* |
2565 | * If the process that created a MAP_PRIVATE mapping is about to | |
2566 | * perform a COW due to a shared page count, attempt to satisfy | |
2567 | * the allocation without using the existing reserves. The pagecache | |
2568 | * page is used to determine if the reserve at this address was | |
2569 | * consumed or not. If reserves were used, a partial faulted mapping | |
2570 | * at the time of fork() could consume its reserves on COW instead | |
2571 | * of the full address range. | |
2572 | */ | |
5944d011 | 2573 | if (is_vma_resv_set(vma, HPAGE_RESV_OWNER) && |
04f2cbe3 MG |
2574 | old_page != pagecache_page) |
2575 | outside_reserve = 1; | |
2576 | ||
1e8f889b | 2577 | page_cache_get(old_page); |
b76c8cfb LW |
2578 | |
2579 | /* Drop page_table_lock as buddy allocator may be called */ | |
2580 | spin_unlock(&mm->page_table_lock); | |
04f2cbe3 | 2581 | new_page = alloc_huge_page(vma, address, outside_reserve); |
1e8f889b | 2582 | |
2fc39cec | 2583 | if (IS_ERR(new_page)) { |
76dcee75 | 2584 | long err = PTR_ERR(new_page); |
1e8f889b | 2585 | page_cache_release(old_page); |
04f2cbe3 MG |
2586 | |
2587 | /* | |
2588 | * If a process owning a MAP_PRIVATE mapping fails to COW, | |
2589 | * it is due to references held by a child and an insufficient | |
2590 | * huge page pool. To guarantee the original mappers | |
2591 | * reliability, unmap the page from child processes. The child | |
2592 | * may get SIGKILLed if it later faults. | |
2593 | */ | |
2594 | if (outside_reserve) { | |
2595 | BUG_ON(huge_pte_none(pte)); | |
2596 | if (unmap_ref_private(mm, vma, old_page, address)) { | |
04f2cbe3 | 2597 | BUG_ON(huge_pte_none(pte)); |
b76c8cfb | 2598 | spin_lock(&mm->page_table_lock); |
a734bcc8 HD |
2599 | ptep = huge_pte_offset(mm, address & huge_page_mask(h)); |
2600 | if (likely(pte_same(huge_ptep_get(ptep), pte))) | |
2601 | goto retry_avoidcopy; | |
2602 | /* | |
2603 | * race occurs while re-acquiring page_table_lock, and | |
2604 | * our job is done. | |
2605 | */ | |
2606 | return 0; | |
04f2cbe3 MG |
2607 | } |
2608 | WARN_ON_ONCE(1); | |
2609 | } | |
2610 | ||
b76c8cfb LW |
2611 | /* Caller expects lock to be held */ |
2612 | spin_lock(&mm->page_table_lock); | |
76dcee75 AK |
2613 | if (err == -ENOMEM) |
2614 | return VM_FAULT_OOM; | |
2615 | else | |
2616 | return VM_FAULT_SIGBUS; | |
1e8f889b DG |
2617 | } |
2618 | ||
0fe6e20b NH |
2619 | /* |
2620 | * When the original hugepage is shared one, it does not have | |
2621 | * anon_vma prepared. | |
2622 | */ | |
44e2aa93 | 2623 | if (unlikely(anon_vma_prepare(vma))) { |
ea4039a3 HD |
2624 | page_cache_release(new_page); |
2625 | page_cache_release(old_page); | |
44e2aa93 DN |
2626 | /* Caller expects lock to be held */ |
2627 | spin_lock(&mm->page_table_lock); | |
0fe6e20b | 2628 | return VM_FAULT_OOM; |
44e2aa93 | 2629 | } |
0fe6e20b | 2630 | |
47ad8475 AA |
2631 | copy_user_huge_page(new_page, old_page, address, vma, |
2632 | pages_per_huge_page(h)); | |
0ed361de | 2633 | __SetPageUptodate(new_page); |
1e8f889b | 2634 | |
2ec74c3e SG |
2635 | mmun_start = address & huge_page_mask(h); |
2636 | mmun_end = mmun_start + huge_page_size(h); | |
2637 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); | |
b76c8cfb LW |
2638 | /* |
2639 | * Retake the page_table_lock to check for racing updates | |
2640 | * before the page tables are altered | |
2641 | */ | |
2642 | spin_lock(&mm->page_table_lock); | |
a5516438 | 2643 | ptep = huge_pte_offset(mm, address & huge_page_mask(h)); |
7f2e9525 | 2644 | if (likely(pte_same(huge_ptep_get(ptep), pte))) { |
07443a85 JK |
2645 | ClearPagePrivate(new_page); |
2646 | ||
1e8f889b | 2647 | /* Break COW */ |
8fe627ec | 2648 | huge_ptep_clear_flush(vma, address, ptep); |
1e8f889b DG |
2649 | set_huge_pte_at(mm, address, ptep, |
2650 | make_huge_pte(vma, new_page, 1)); | |
0fe6e20b | 2651 | page_remove_rmap(old_page); |
cd67f0d2 | 2652 | hugepage_add_new_anon_rmap(new_page, vma, address); |
1e8f889b DG |
2653 | /* Make the old page be freed below */ |
2654 | new_page = old_page; | |
2655 | } | |
2ec74c3e SG |
2656 | spin_unlock(&mm->page_table_lock); |
2657 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); | |
1e8f889b DG |
2658 | page_cache_release(new_page); |
2659 | page_cache_release(old_page); | |
8312034f JK |
2660 | |
2661 | /* Caller expects lock to be held */ | |
2662 | spin_lock(&mm->page_table_lock); | |
83c54070 | 2663 | return 0; |
1e8f889b DG |
2664 | } |
2665 | ||
04f2cbe3 | 2666 | /* Return the pagecache page at a given address within a VMA */ |
a5516438 AK |
2667 | static struct page *hugetlbfs_pagecache_page(struct hstate *h, |
2668 | struct vm_area_struct *vma, unsigned long address) | |
04f2cbe3 MG |
2669 | { |
2670 | struct address_space *mapping; | |
e7c4b0bf | 2671 | pgoff_t idx; |
04f2cbe3 MG |
2672 | |
2673 | mapping = vma->vm_file->f_mapping; | |
a5516438 | 2674 | idx = vma_hugecache_offset(h, vma, address); |
04f2cbe3 MG |
2675 | |
2676 | return find_lock_page(mapping, idx); | |
2677 | } | |
2678 | ||
3ae77f43 HD |
2679 | /* |
2680 | * Return whether there is a pagecache page to back given address within VMA. | |
2681 | * Caller follow_hugetlb_page() holds page_table_lock so we cannot lock_page. | |
2682 | */ | |
2683 | static bool hugetlbfs_pagecache_present(struct hstate *h, | |
2a15efc9 HD |
2684 | struct vm_area_struct *vma, unsigned long address) |
2685 | { | |
2686 | struct address_space *mapping; | |
2687 | pgoff_t idx; | |
2688 | struct page *page; | |
2689 | ||
2690 | mapping = vma->vm_file->f_mapping; | |
2691 | idx = vma_hugecache_offset(h, vma, address); | |
2692 | ||
2693 | page = find_get_page(mapping, idx); | |
2694 | if (page) | |
2695 | put_page(page); | |
2696 | return page != NULL; | |
2697 | } | |
2698 | ||
a1ed3dda | 2699 | static int hugetlb_no_page(struct mm_struct *mm, struct vm_area_struct *vma, |
788c7df4 | 2700 | unsigned long address, pte_t *ptep, unsigned int flags) |
ac9b9c66 | 2701 | { |
a5516438 | 2702 | struct hstate *h = hstate_vma(vma); |
ac9b9c66 | 2703 | int ret = VM_FAULT_SIGBUS; |
409eb8c2 | 2704 | int anon_rmap = 0; |
e7c4b0bf | 2705 | pgoff_t idx; |
4c887265 | 2706 | unsigned long size; |
4c887265 AL |
2707 | struct page *page; |
2708 | struct address_space *mapping; | |
1e8f889b | 2709 | pte_t new_pte; |
4c887265 | 2710 | |
04f2cbe3 MG |
2711 | /* |
2712 | * Currently, we are forced to kill the process in the event the | |
2713 | * original mapper has unmapped pages from the child due to a failed | |
25985edc | 2714 | * COW. Warn that such a situation has occurred as it may not be obvious |
04f2cbe3 MG |
2715 | */ |
2716 | if (is_vma_resv_set(vma, HPAGE_RESV_UNMAPPED)) { | |
ffb22af5 AM |
2717 | pr_warning("PID %d killed due to inadequate hugepage pool\n", |
2718 | current->pid); | |
04f2cbe3 MG |
2719 | return ret; |
2720 | } | |
2721 | ||
4c887265 | 2722 | mapping = vma->vm_file->f_mapping; |
a5516438 | 2723 | idx = vma_hugecache_offset(h, vma, address); |
4c887265 AL |
2724 | |
2725 | /* | |
2726 | * Use page lock to guard against racing truncation | |
2727 | * before we get page_table_lock. | |
2728 | */ | |
6bda666a CL |
2729 | retry: |
2730 | page = find_lock_page(mapping, idx); | |
2731 | if (!page) { | |
a5516438 | 2732 | size = i_size_read(mapping->host) >> huge_page_shift(h); |
ebed4bfc HD |
2733 | if (idx >= size) |
2734 | goto out; | |
04f2cbe3 | 2735 | page = alloc_huge_page(vma, address, 0); |
2fc39cec | 2736 | if (IS_ERR(page)) { |
76dcee75 AK |
2737 | ret = PTR_ERR(page); |
2738 | if (ret == -ENOMEM) | |
2739 | ret = VM_FAULT_OOM; | |
2740 | else | |
2741 | ret = VM_FAULT_SIGBUS; | |
6bda666a CL |
2742 | goto out; |
2743 | } | |
47ad8475 | 2744 | clear_huge_page(page, address, pages_per_huge_page(h)); |
0ed361de | 2745 | __SetPageUptodate(page); |
ac9b9c66 | 2746 | |
f83a275d | 2747 | if (vma->vm_flags & VM_MAYSHARE) { |
6bda666a | 2748 | int err; |
45c682a6 | 2749 | struct inode *inode = mapping->host; |
6bda666a CL |
2750 | |
2751 | err = add_to_page_cache(page, mapping, idx, GFP_KERNEL); | |
2752 | if (err) { | |
2753 | put_page(page); | |
6bda666a CL |
2754 | if (err == -EEXIST) |
2755 | goto retry; | |
2756 | goto out; | |
2757 | } | |
07443a85 | 2758 | ClearPagePrivate(page); |
45c682a6 KC |
2759 | |
2760 | spin_lock(&inode->i_lock); | |
a5516438 | 2761 | inode->i_blocks += blocks_per_huge_page(h); |
45c682a6 | 2762 | spin_unlock(&inode->i_lock); |
23be7468 | 2763 | } else { |
6bda666a | 2764 | lock_page(page); |
0fe6e20b NH |
2765 | if (unlikely(anon_vma_prepare(vma))) { |
2766 | ret = VM_FAULT_OOM; | |
2767 | goto backout_unlocked; | |
2768 | } | |
409eb8c2 | 2769 | anon_rmap = 1; |
23be7468 | 2770 | } |
0fe6e20b | 2771 | } else { |
998b4382 NH |
2772 | /* |
2773 | * If memory error occurs between mmap() and fault, some process | |
2774 | * don't have hwpoisoned swap entry for errored virtual address. | |
2775 | * So we need to block hugepage fault by PG_hwpoison bit check. | |
2776 | */ | |
2777 | if (unlikely(PageHWPoison(page))) { | |
32f84528 | 2778 | ret = VM_FAULT_HWPOISON | |
972dc4de | 2779 | VM_FAULT_SET_HINDEX(hstate_index(h)); |
998b4382 NH |
2780 | goto backout_unlocked; |
2781 | } | |
6bda666a | 2782 | } |
1e8f889b | 2783 | |
57303d80 AW |
2784 | /* |
2785 | * If we are going to COW a private mapping later, we examine the | |
2786 | * pending reservations for this page now. This will ensure that | |
2787 | * any allocations necessary to record that reservation occur outside | |
2788 | * the spinlock. | |
2789 | */ | |
788c7df4 | 2790 | if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) |
2b26736c AW |
2791 | if (vma_needs_reservation(h, vma, address) < 0) { |
2792 | ret = VM_FAULT_OOM; | |
2793 | goto backout_unlocked; | |
2794 | } | |
57303d80 | 2795 | |
ac9b9c66 | 2796 | spin_lock(&mm->page_table_lock); |
a5516438 | 2797 | size = i_size_read(mapping->host) >> huge_page_shift(h); |
4c887265 AL |
2798 | if (idx >= size) |
2799 | goto backout; | |
2800 | ||
83c54070 | 2801 | ret = 0; |
7f2e9525 | 2802 | if (!huge_pte_none(huge_ptep_get(ptep))) |
4c887265 AL |
2803 | goto backout; |
2804 | ||
07443a85 JK |
2805 | if (anon_rmap) { |
2806 | ClearPagePrivate(page); | |
409eb8c2 | 2807 | hugepage_add_new_anon_rmap(page, vma, address); |
07443a85 | 2808 | } |
409eb8c2 HD |
2809 | else |
2810 | page_dup_rmap(page); | |
1e8f889b DG |
2811 | new_pte = make_huge_pte(vma, page, ((vma->vm_flags & VM_WRITE) |
2812 | && (vma->vm_flags & VM_SHARED))); | |
2813 | set_huge_pte_at(mm, address, ptep, new_pte); | |
2814 | ||
788c7df4 | 2815 | if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) { |
1e8f889b | 2816 | /* Optimization, do the COW without a second fault */ |
04f2cbe3 | 2817 | ret = hugetlb_cow(mm, vma, address, ptep, new_pte, page); |
1e8f889b DG |
2818 | } |
2819 | ||
ac9b9c66 | 2820 | spin_unlock(&mm->page_table_lock); |
4c887265 AL |
2821 | unlock_page(page); |
2822 | out: | |
ac9b9c66 | 2823 | return ret; |
4c887265 AL |
2824 | |
2825 | backout: | |
2826 | spin_unlock(&mm->page_table_lock); | |
2b26736c | 2827 | backout_unlocked: |
4c887265 AL |
2828 | unlock_page(page); |
2829 | put_page(page); | |
2830 | goto out; | |
ac9b9c66 HD |
2831 | } |
2832 | ||
86e5216f | 2833 | int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
788c7df4 | 2834 | unsigned long address, unsigned int flags) |
86e5216f AL |
2835 | { |
2836 | pte_t *ptep; | |
2837 | pte_t entry; | |
1e8f889b | 2838 | int ret; |
0fe6e20b | 2839 | struct page *page = NULL; |
57303d80 | 2840 | struct page *pagecache_page = NULL; |
3935baa9 | 2841 | static DEFINE_MUTEX(hugetlb_instantiation_mutex); |
a5516438 | 2842 | struct hstate *h = hstate_vma(vma); |
86e5216f | 2843 | |
1e16a539 KH |
2844 | address &= huge_page_mask(h); |
2845 | ||
fd6a03ed NH |
2846 | ptep = huge_pte_offset(mm, address); |
2847 | if (ptep) { | |
2848 | entry = huge_ptep_get(ptep); | |
290408d4 | 2849 | if (unlikely(is_hugetlb_entry_migration(entry))) { |
30dad309 | 2850 | migration_entry_wait_huge(mm, ptep); |
290408d4 NH |
2851 | return 0; |
2852 | } else if (unlikely(is_hugetlb_entry_hwpoisoned(entry))) | |
32f84528 | 2853 | return VM_FAULT_HWPOISON_LARGE | |
972dc4de | 2854 | VM_FAULT_SET_HINDEX(hstate_index(h)); |
fd6a03ed NH |
2855 | } |
2856 | ||
a5516438 | 2857 | ptep = huge_pte_alloc(mm, address, huge_page_size(h)); |
86e5216f AL |
2858 | if (!ptep) |
2859 | return VM_FAULT_OOM; | |
2860 | ||
3935baa9 DG |
2861 | /* |
2862 | * Serialize hugepage allocation and instantiation, so that we don't | |
2863 | * get spurious allocation failures if two CPUs race to instantiate | |
2864 | * the same page in the page cache. | |
2865 | */ | |
2866 | mutex_lock(&hugetlb_instantiation_mutex); | |
7f2e9525 GS |
2867 | entry = huge_ptep_get(ptep); |
2868 | if (huge_pte_none(entry)) { | |
788c7df4 | 2869 | ret = hugetlb_no_page(mm, vma, address, ptep, flags); |
b4d1d99f | 2870 | goto out_mutex; |
3935baa9 | 2871 | } |
86e5216f | 2872 | |
83c54070 | 2873 | ret = 0; |
1e8f889b | 2874 | |
57303d80 AW |
2875 | /* |
2876 | * If we are going to COW the mapping later, we examine the pending | |
2877 | * reservations for this page now. This will ensure that any | |
2878 | * allocations necessary to record that reservation occur outside the | |
2879 | * spinlock. For private mappings, we also lookup the pagecache | |
2880 | * page now as it is used to determine if a reservation has been | |
2881 | * consumed. | |
2882 | */ | |
106c992a | 2883 | if ((flags & FAULT_FLAG_WRITE) && !huge_pte_write(entry)) { |
2b26736c AW |
2884 | if (vma_needs_reservation(h, vma, address) < 0) { |
2885 | ret = VM_FAULT_OOM; | |
b4d1d99f | 2886 | goto out_mutex; |
2b26736c | 2887 | } |
57303d80 | 2888 | |
f83a275d | 2889 | if (!(vma->vm_flags & VM_MAYSHARE)) |
57303d80 AW |
2890 | pagecache_page = hugetlbfs_pagecache_page(h, |
2891 | vma, address); | |
2892 | } | |
2893 | ||
56c9cfb1 NH |
2894 | /* |
2895 | * hugetlb_cow() requires page locks of pte_page(entry) and | |
2896 | * pagecache_page, so here we need take the former one | |
2897 | * when page != pagecache_page or !pagecache_page. | |
2898 | * Note that locking order is always pagecache_page -> page, | |
2899 | * so no worry about deadlock. | |
2900 | */ | |
2901 | page = pte_page(entry); | |
66aebce7 | 2902 | get_page(page); |
56c9cfb1 | 2903 | if (page != pagecache_page) |
0fe6e20b | 2904 | lock_page(page); |
0fe6e20b | 2905 | |
1e8f889b DG |
2906 | spin_lock(&mm->page_table_lock); |
2907 | /* Check for a racing update before calling hugetlb_cow */ | |
b4d1d99f DG |
2908 | if (unlikely(!pte_same(entry, huge_ptep_get(ptep)))) |
2909 | goto out_page_table_lock; | |
2910 | ||
2911 | ||
788c7df4 | 2912 | if (flags & FAULT_FLAG_WRITE) { |
106c992a | 2913 | if (!huge_pte_write(entry)) { |
57303d80 AW |
2914 | ret = hugetlb_cow(mm, vma, address, ptep, entry, |
2915 | pagecache_page); | |
b4d1d99f DG |
2916 | goto out_page_table_lock; |
2917 | } | |
106c992a | 2918 | entry = huge_pte_mkdirty(entry); |
b4d1d99f DG |
2919 | } |
2920 | entry = pte_mkyoung(entry); | |
788c7df4 HD |
2921 | if (huge_ptep_set_access_flags(vma, address, ptep, entry, |
2922 | flags & FAULT_FLAG_WRITE)) | |
4b3073e1 | 2923 | update_mmu_cache(vma, address, ptep); |
b4d1d99f DG |
2924 | |
2925 | out_page_table_lock: | |
1e8f889b | 2926 | spin_unlock(&mm->page_table_lock); |
57303d80 AW |
2927 | |
2928 | if (pagecache_page) { | |
2929 | unlock_page(pagecache_page); | |
2930 | put_page(pagecache_page); | |
2931 | } | |
1f64d69c DN |
2932 | if (page != pagecache_page) |
2933 | unlock_page(page); | |
66aebce7 | 2934 | put_page(page); |
57303d80 | 2935 | |
b4d1d99f | 2936 | out_mutex: |
3935baa9 | 2937 | mutex_unlock(&hugetlb_instantiation_mutex); |
1e8f889b DG |
2938 | |
2939 | return ret; | |
86e5216f AL |
2940 | } |
2941 | ||
28a35716 ML |
2942 | long follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2943 | struct page **pages, struct vm_area_struct **vmas, | |
2944 | unsigned long *position, unsigned long *nr_pages, | |
2945 | long i, unsigned int flags) | |
63551ae0 | 2946 | { |
d5d4b0aa CK |
2947 | unsigned long pfn_offset; |
2948 | unsigned long vaddr = *position; | |
28a35716 | 2949 | unsigned long remainder = *nr_pages; |
a5516438 | 2950 | struct hstate *h = hstate_vma(vma); |
63551ae0 | 2951 | |
1c59827d | 2952 | spin_lock(&mm->page_table_lock); |
63551ae0 | 2953 | while (vaddr < vma->vm_end && remainder) { |
4c887265 | 2954 | pte_t *pte; |
2a15efc9 | 2955 | int absent; |
4c887265 | 2956 | struct page *page; |
63551ae0 | 2957 | |
4c887265 AL |
2958 | /* |
2959 | * Some archs (sparc64, sh*) have multiple pte_ts to | |
2a15efc9 | 2960 | * each hugepage. We have to make sure we get the |
4c887265 AL |
2961 | * first, for the page indexing below to work. |
2962 | */ | |
a5516438 | 2963 | pte = huge_pte_offset(mm, vaddr & huge_page_mask(h)); |
2a15efc9 HD |
2964 | absent = !pte || huge_pte_none(huge_ptep_get(pte)); |
2965 | ||
2966 | /* | |
2967 | * When coredumping, it suits get_dump_page if we just return | |
3ae77f43 HD |
2968 | * an error where there's an empty slot with no huge pagecache |
2969 | * to back it. This way, we avoid allocating a hugepage, and | |
2970 | * the sparse dumpfile avoids allocating disk blocks, but its | |
2971 | * huge holes still show up with zeroes where they need to be. | |
2a15efc9 | 2972 | */ |
3ae77f43 HD |
2973 | if (absent && (flags & FOLL_DUMP) && |
2974 | !hugetlbfs_pagecache_present(h, vma, vaddr)) { | |
2a15efc9 HD |
2975 | remainder = 0; |
2976 | break; | |
2977 | } | |
63551ae0 | 2978 | |
9cc3a5bd NH |
2979 | /* |
2980 | * We need call hugetlb_fault for both hugepages under migration | |
2981 | * (in which case hugetlb_fault waits for the migration,) and | |
2982 | * hwpoisoned hugepages (in which case we need to prevent the | |
2983 | * caller from accessing to them.) In order to do this, we use | |
2984 | * here is_swap_pte instead of is_hugetlb_entry_migration and | |
2985 | * is_hugetlb_entry_hwpoisoned. This is because it simply covers | |
2986 | * both cases, and because we can't follow correct pages | |
2987 | * directly from any kind of swap entries. | |
2988 | */ | |
2989 | if (absent || is_swap_pte(huge_ptep_get(pte)) || | |
106c992a GS |
2990 | ((flags & FOLL_WRITE) && |
2991 | !huge_pte_write(huge_ptep_get(pte)))) { | |
4c887265 | 2992 | int ret; |
63551ae0 | 2993 | |
4c887265 | 2994 | spin_unlock(&mm->page_table_lock); |
2a15efc9 HD |
2995 | ret = hugetlb_fault(mm, vma, vaddr, |
2996 | (flags & FOLL_WRITE) ? FAULT_FLAG_WRITE : 0); | |
4c887265 | 2997 | spin_lock(&mm->page_table_lock); |
a89182c7 | 2998 | if (!(ret & VM_FAULT_ERROR)) |
4c887265 | 2999 | continue; |
63551ae0 | 3000 | |
4c887265 | 3001 | remainder = 0; |
4c887265 AL |
3002 | break; |
3003 | } | |
3004 | ||
a5516438 | 3005 | pfn_offset = (vaddr & ~huge_page_mask(h)) >> PAGE_SHIFT; |
7f2e9525 | 3006 | page = pte_page(huge_ptep_get(pte)); |
d5d4b0aa | 3007 | same_page: |
d6692183 | 3008 | if (pages) { |
2a15efc9 | 3009 | pages[i] = mem_map_offset(page, pfn_offset); |
4b2e38ad | 3010 | get_page(pages[i]); |
d6692183 | 3011 | } |
63551ae0 DG |
3012 | |
3013 | if (vmas) | |
3014 | vmas[i] = vma; | |
3015 | ||
3016 | vaddr += PAGE_SIZE; | |
d5d4b0aa | 3017 | ++pfn_offset; |
63551ae0 DG |
3018 | --remainder; |
3019 | ++i; | |
d5d4b0aa | 3020 | if (vaddr < vma->vm_end && remainder && |
a5516438 | 3021 | pfn_offset < pages_per_huge_page(h)) { |
d5d4b0aa CK |
3022 | /* |
3023 | * We use pfn_offset to avoid touching the pageframes | |
3024 | * of this compound page. | |
3025 | */ | |
3026 | goto same_page; | |
3027 | } | |
63551ae0 | 3028 | } |
1c59827d | 3029 | spin_unlock(&mm->page_table_lock); |
28a35716 | 3030 | *nr_pages = remainder; |
63551ae0 DG |
3031 | *position = vaddr; |
3032 | ||
2a15efc9 | 3033 | return i ? i : -EFAULT; |
63551ae0 | 3034 | } |
8f860591 | 3035 | |
7da4d641 | 3036 | unsigned long hugetlb_change_protection(struct vm_area_struct *vma, |
8f860591 ZY |
3037 | unsigned long address, unsigned long end, pgprot_t newprot) |
3038 | { | |
3039 | struct mm_struct *mm = vma->vm_mm; | |
3040 | unsigned long start = address; | |
3041 | pte_t *ptep; | |
3042 | pte_t pte; | |
a5516438 | 3043 | struct hstate *h = hstate_vma(vma); |
7da4d641 | 3044 | unsigned long pages = 0; |
8f860591 ZY |
3045 | |
3046 | BUG_ON(address >= end); | |
3047 | flush_cache_range(vma, address, end); | |
3048 | ||
3d48ae45 | 3049 | mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); |
8f860591 | 3050 | spin_lock(&mm->page_table_lock); |
a5516438 | 3051 | for (; address < end; address += huge_page_size(h)) { |
8f860591 ZY |
3052 | ptep = huge_pte_offset(mm, address); |
3053 | if (!ptep) | |
3054 | continue; | |
7da4d641 PZ |
3055 | if (huge_pmd_unshare(mm, &address, ptep)) { |
3056 | pages++; | |
39dde65c | 3057 | continue; |
7da4d641 | 3058 | } |
7f2e9525 | 3059 | if (!huge_pte_none(huge_ptep_get(ptep))) { |
8f860591 | 3060 | pte = huge_ptep_get_and_clear(mm, address, ptep); |
106c992a | 3061 | pte = pte_mkhuge(huge_pte_modify(pte, newprot)); |
be7517d6 | 3062 | pte = arch_make_huge_pte(pte, vma, NULL, 0); |
8f860591 | 3063 | set_huge_pte_at(mm, address, ptep, pte); |
7da4d641 | 3064 | pages++; |
8f860591 ZY |
3065 | } |
3066 | } | |
3067 | spin_unlock(&mm->page_table_lock); | |
d833352a MG |
3068 | /* |
3069 | * Must flush TLB before releasing i_mmap_mutex: x86's huge_pmd_unshare | |
3070 | * may have cleared our pud entry and done put_page on the page table: | |
3071 | * once we release i_mmap_mutex, another task can do the final put_page | |
3072 | * and that page table be reused and filled with junk. | |
3073 | */ | |
8f860591 | 3074 | flush_tlb_range(vma, start, end); |
d833352a | 3075 | mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); |
7da4d641 PZ |
3076 | |
3077 | return pages << h->order; | |
8f860591 ZY |
3078 | } |
3079 | ||
a1e78772 MG |
3080 | int hugetlb_reserve_pages(struct inode *inode, |
3081 | long from, long to, | |
5a6fe125 | 3082 | struct vm_area_struct *vma, |
ca16d140 | 3083 | vm_flags_t vm_flags) |
e4e574b7 | 3084 | { |
17c9d12e | 3085 | long ret, chg; |
a5516438 | 3086 | struct hstate *h = hstate_inode(inode); |
90481622 | 3087 | struct hugepage_subpool *spool = subpool_inode(inode); |
e4e574b7 | 3088 | |
17c9d12e MG |
3089 | /* |
3090 | * Only apply hugepage reservation if asked. At fault time, an | |
3091 | * attempt will be made for VM_NORESERVE to allocate a page | |
90481622 | 3092 | * without using reserves |
17c9d12e | 3093 | */ |
ca16d140 | 3094 | if (vm_flags & VM_NORESERVE) |
17c9d12e MG |
3095 | return 0; |
3096 | ||
a1e78772 MG |
3097 | /* |
3098 | * Shared mappings base their reservation on the number of pages that | |
3099 | * are already allocated on behalf of the file. Private mappings need | |
3100 | * to reserve the full area even if read-only as mprotect() may be | |
3101 | * called to make the mapping read-write. Assume !vma is a shm mapping | |
3102 | */ | |
f83a275d | 3103 | if (!vma || vma->vm_flags & VM_MAYSHARE) |
a1e78772 | 3104 | chg = region_chg(&inode->i_mapping->private_list, from, to); |
17c9d12e MG |
3105 | else { |
3106 | struct resv_map *resv_map = resv_map_alloc(); | |
3107 | if (!resv_map) | |
3108 | return -ENOMEM; | |
3109 | ||
a1e78772 | 3110 | chg = to - from; |
84afd99b | 3111 | |
17c9d12e MG |
3112 | set_vma_resv_map(vma, resv_map); |
3113 | set_vma_resv_flags(vma, HPAGE_RESV_OWNER); | |
3114 | } | |
3115 | ||
c50ac050 DH |
3116 | if (chg < 0) { |
3117 | ret = chg; | |
3118 | goto out_err; | |
3119 | } | |
8a630112 | 3120 | |
90481622 | 3121 | /* There must be enough pages in the subpool for the mapping */ |
c50ac050 DH |
3122 | if (hugepage_subpool_get_pages(spool, chg)) { |
3123 | ret = -ENOSPC; | |
3124 | goto out_err; | |
3125 | } | |
5a6fe125 MG |
3126 | |
3127 | /* | |
17c9d12e | 3128 | * Check enough hugepages are available for the reservation. |
90481622 | 3129 | * Hand the pages back to the subpool if there are not |
5a6fe125 | 3130 | */ |
a5516438 | 3131 | ret = hugetlb_acct_memory(h, chg); |
68842c9b | 3132 | if (ret < 0) { |
90481622 | 3133 | hugepage_subpool_put_pages(spool, chg); |
c50ac050 | 3134 | goto out_err; |
68842c9b | 3135 | } |
17c9d12e MG |
3136 | |
3137 | /* | |
3138 | * Account for the reservations made. Shared mappings record regions | |
3139 | * that have reservations as they are shared by multiple VMAs. | |
3140 | * When the last VMA disappears, the region map says how much | |
3141 | * the reservation was and the page cache tells how much of | |
3142 | * the reservation was consumed. Private mappings are per-VMA and | |
3143 | * only the consumed reservations are tracked. When the VMA | |
3144 | * disappears, the original reservation is the VMA size and the | |
3145 | * consumed reservations are stored in the map. Hence, nothing | |
3146 | * else has to be done for private mappings here | |
3147 | */ | |
f83a275d | 3148 | if (!vma || vma->vm_flags & VM_MAYSHARE) |
a1e78772 | 3149 | region_add(&inode->i_mapping->private_list, from, to); |
a43a8c39 | 3150 | return 0; |
c50ac050 | 3151 | out_err: |
4523e145 DH |
3152 | if (vma) |
3153 | resv_map_put(vma); | |
c50ac050 | 3154 | return ret; |
a43a8c39 CK |
3155 | } |
3156 | ||
3157 | void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed) | |
3158 | { | |
a5516438 | 3159 | struct hstate *h = hstate_inode(inode); |
a43a8c39 | 3160 | long chg = region_truncate(&inode->i_mapping->private_list, offset); |
90481622 | 3161 | struct hugepage_subpool *spool = subpool_inode(inode); |
45c682a6 KC |
3162 | |
3163 | spin_lock(&inode->i_lock); | |
e4c6f8be | 3164 | inode->i_blocks -= (blocks_per_huge_page(h) * freed); |
45c682a6 KC |
3165 | spin_unlock(&inode->i_lock); |
3166 | ||
90481622 | 3167 | hugepage_subpool_put_pages(spool, (chg - freed)); |
a5516438 | 3168 | hugetlb_acct_memory(h, -(chg - freed)); |
a43a8c39 | 3169 | } |
93f70f90 | 3170 | |
3212b535 SC |
3171 | #ifdef CONFIG_ARCH_WANT_HUGE_PMD_SHARE |
3172 | static unsigned long page_table_shareable(struct vm_area_struct *svma, | |
3173 | struct vm_area_struct *vma, | |
3174 | unsigned long addr, pgoff_t idx) | |
3175 | { | |
3176 | unsigned long saddr = ((idx - svma->vm_pgoff) << PAGE_SHIFT) + | |
3177 | svma->vm_start; | |
3178 | unsigned long sbase = saddr & PUD_MASK; | |
3179 | unsigned long s_end = sbase + PUD_SIZE; | |
3180 | ||
3181 | /* Allow segments to share if only one is marked locked */ | |
3182 | unsigned long vm_flags = vma->vm_flags & ~VM_LOCKED; | |
3183 | unsigned long svm_flags = svma->vm_flags & ~VM_LOCKED; | |
3184 | ||
3185 | /* | |
3186 | * match the virtual addresses, permission and the alignment of the | |
3187 | * page table page. | |
3188 | */ | |
3189 | if (pmd_index(addr) != pmd_index(saddr) || | |
3190 | vm_flags != svm_flags || | |
3191 | sbase < svma->vm_start || svma->vm_end < s_end) | |
3192 | return 0; | |
3193 | ||
3194 | return saddr; | |
3195 | } | |
3196 | ||
3197 | static int vma_shareable(struct vm_area_struct *vma, unsigned long addr) | |
3198 | { | |
3199 | unsigned long base = addr & PUD_MASK; | |
3200 | unsigned long end = base + PUD_SIZE; | |
3201 | ||
3202 | /* | |
3203 | * check on proper vm_flags and page table alignment | |
3204 | */ | |
3205 | if (vma->vm_flags & VM_MAYSHARE && | |
3206 | vma->vm_start <= base && end <= vma->vm_end) | |
3207 | return 1; | |
3208 | return 0; | |
3209 | } | |
3210 | ||
3211 | /* | |
3212 | * Search for a shareable pmd page for hugetlb. In any case calls pmd_alloc() | |
3213 | * and returns the corresponding pte. While this is not necessary for the | |
3214 | * !shared pmd case because we can allocate the pmd later as well, it makes the | |
3215 | * code much cleaner. pmd allocation is essential for the shared case because | |
3216 | * pud has to be populated inside the same i_mmap_mutex section - otherwise | |
3217 | * racing tasks could either miss the sharing (see huge_pte_offset) or select a | |
3218 | * bad pmd for sharing. | |
3219 | */ | |
3220 | pte_t *huge_pmd_share(struct mm_struct *mm, unsigned long addr, pud_t *pud) | |
3221 | { | |
3222 | struct vm_area_struct *vma = find_vma(mm, addr); | |
3223 | struct address_space *mapping = vma->vm_file->f_mapping; | |
3224 | pgoff_t idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + | |
3225 | vma->vm_pgoff; | |
3226 | struct vm_area_struct *svma; | |
3227 | unsigned long saddr; | |
3228 | pte_t *spte = NULL; | |
3229 | pte_t *pte; | |
3230 | ||
3231 | if (!vma_shareable(vma, addr)) | |
3232 | return (pte_t *)pmd_alloc(mm, pud, addr); | |
3233 | ||
3234 | mutex_lock(&mapping->i_mmap_mutex); | |
3235 | vma_interval_tree_foreach(svma, &mapping->i_mmap, idx, idx) { | |
3236 | if (svma == vma) | |
3237 | continue; | |
3238 | ||
3239 | saddr = page_table_shareable(svma, vma, addr, idx); | |
3240 | if (saddr) { | |
3241 | spte = huge_pte_offset(svma->vm_mm, saddr); | |
3242 | if (spte) { | |
3243 | get_page(virt_to_page(spte)); | |
3244 | break; | |
3245 | } | |
3246 | } | |
3247 | } | |
3248 | ||
3249 | if (!spte) | |
3250 | goto out; | |
3251 | ||
3252 | spin_lock(&mm->page_table_lock); | |
3253 | if (pud_none(*pud)) | |
3254 | pud_populate(mm, pud, | |
3255 | (pmd_t *)((unsigned long)spte & PAGE_MASK)); | |
3256 | else | |
3257 | put_page(virt_to_page(spte)); | |
3258 | spin_unlock(&mm->page_table_lock); | |
3259 | out: | |
3260 | pte = (pte_t *)pmd_alloc(mm, pud, addr); | |
3261 | mutex_unlock(&mapping->i_mmap_mutex); | |
3262 | return pte; | |
3263 | } | |
3264 | ||
3265 | /* | |
3266 | * unmap huge page backed by shared pte. | |
3267 | * | |
3268 | * Hugetlb pte page is ref counted at the time of mapping. If pte is shared | |
3269 | * indicated by page_count > 1, unmap is achieved by clearing pud and | |
3270 | * decrementing the ref count. If count == 1, the pte page is not shared. | |
3271 | * | |
3272 | * called with vma->vm_mm->page_table_lock held. | |
3273 | * | |
3274 | * returns: 1 successfully unmapped a shared pte page | |
3275 | * 0 the underlying pte page is not shared, or it is the last user | |
3276 | */ | |
3277 | int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep) | |
3278 | { | |
3279 | pgd_t *pgd = pgd_offset(mm, *addr); | |
3280 | pud_t *pud = pud_offset(pgd, *addr); | |
3281 | ||
3282 | BUG_ON(page_count(virt_to_page(ptep)) == 0); | |
3283 | if (page_count(virt_to_page(ptep)) == 1) | |
3284 | return 0; | |
3285 | ||
3286 | pud_clear(pud); | |
3287 | put_page(virt_to_page(ptep)); | |
3288 | *addr = ALIGN(*addr, HPAGE_SIZE * PTRS_PER_PTE) - HPAGE_SIZE; | |
3289 | return 1; | |
3290 | } | |
9e5fc74c SC |
3291 | #define want_pmd_share() (1) |
3292 | #else /* !CONFIG_ARCH_WANT_HUGE_PMD_SHARE */ | |
3293 | pte_t *huge_pmd_share(struct mm_struct *mm, unsigned long addr, pud_t *pud) | |
3294 | { | |
3295 | return NULL; | |
3296 | } | |
3297 | #define want_pmd_share() (0) | |
3212b535 SC |
3298 | #endif /* CONFIG_ARCH_WANT_HUGE_PMD_SHARE */ |
3299 | ||
9e5fc74c SC |
3300 | #ifdef CONFIG_ARCH_WANT_GENERAL_HUGETLB |
3301 | pte_t *huge_pte_alloc(struct mm_struct *mm, | |
3302 | unsigned long addr, unsigned long sz) | |
3303 | { | |
3304 | pgd_t *pgd; | |
3305 | pud_t *pud; | |
3306 | pte_t *pte = NULL; | |
3307 | ||
3308 | pgd = pgd_offset(mm, addr); | |
3309 | pud = pud_alloc(mm, pgd, addr); | |
3310 | if (pud) { | |
3311 | if (sz == PUD_SIZE) { | |
3312 | pte = (pte_t *)pud; | |
3313 | } else { | |
3314 | BUG_ON(sz != PMD_SIZE); | |
3315 | if (want_pmd_share() && pud_none(*pud)) | |
3316 | pte = huge_pmd_share(mm, addr, pud); | |
3317 | else | |
3318 | pte = (pte_t *)pmd_alloc(mm, pud, addr); | |
3319 | } | |
3320 | } | |
3321 | BUG_ON(pte && !pte_none(*pte) && !pte_huge(*pte)); | |
3322 | ||
3323 | return pte; | |
3324 | } | |
3325 | ||
3326 | pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr) | |
3327 | { | |
3328 | pgd_t *pgd; | |
3329 | pud_t *pud; | |
3330 | pmd_t *pmd = NULL; | |
3331 | ||
3332 | pgd = pgd_offset(mm, addr); | |
3333 | if (pgd_present(*pgd)) { | |
3334 | pud = pud_offset(pgd, addr); | |
3335 | if (pud_present(*pud)) { | |
3336 | if (pud_huge(*pud)) | |
3337 | return (pte_t *)pud; | |
3338 | pmd = pmd_offset(pud, addr); | |
3339 | } | |
3340 | } | |
3341 | return (pte_t *) pmd; | |
3342 | } | |
3343 | ||
3344 | struct page * | |
3345 | follow_huge_pmd(struct mm_struct *mm, unsigned long address, | |
3346 | pmd_t *pmd, int write) | |
3347 | { | |
3348 | struct page *page; | |
3349 | ||
3350 | page = pte_page(*(pte_t *)pmd); | |
3351 | if (page) | |
3352 | page += ((address & ~PMD_MASK) >> PAGE_SHIFT); | |
3353 | return page; | |
3354 | } | |
3355 | ||
3356 | struct page * | |
3357 | follow_huge_pud(struct mm_struct *mm, unsigned long address, | |
3358 | pud_t *pud, int write) | |
3359 | { | |
3360 | struct page *page; | |
3361 | ||
3362 | page = pte_page(*(pte_t *)pud); | |
3363 | if (page) | |
3364 | page += ((address & ~PUD_MASK) >> PAGE_SHIFT); | |
3365 | return page; | |
3366 | } | |
3367 | ||
3368 | #else /* !CONFIG_ARCH_WANT_GENERAL_HUGETLB */ | |
3369 | ||
3370 | /* Can be overriden by architectures */ | |
3371 | __attribute__((weak)) struct page * | |
3372 | follow_huge_pud(struct mm_struct *mm, unsigned long address, | |
3373 | pud_t *pud, int write) | |
3374 | { | |
3375 | BUG(); | |
3376 | return NULL; | |
3377 | } | |
3378 | ||
3379 | #endif /* CONFIG_ARCH_WANT_GENERAL_HUGETLB */ | |
3380 | ||
d5bd9106 AK |
3381 | #ifdef CONFIG_MEMORY_FAILURE |
3382 | ||
6de2b1aa NH |
3383 | /* Should be called in hugetlb_lock */ |
3384 | static int is_hugepage_on_freelist(struct page *hpage) | |
3385 | { | |
3386 | struct page *page; | |
3387 | struct page *tmp; | |
3388 | struct hstate *h = page_hstate(hpage); | |
3389 | int nid = page_to_nid(hpage); | |
3390 | ||
3391 | list_for_each_entry_safe(page, tmp, &h->hugepage_freelists[nid], lru) | |
3392 | if (page == hpage) | |
3393 | return 1; | |
3394 | return 0; | |
3395 | } | |
3396 | ||
93f70f90 NH |
3397 | /* |
3398 | * This function is called from memory failure code. | |
3399 | * Assume the caller holds page lock of the head page. | |
3400 | */ | |
6de2b1aa | 3401 | int dequeue_hwpoisoned_huge_page(struct page *hpage) |
93f70f90 NH |
3402 | { |
3403 | struct hstate *h = page_hstate(hpage); | |
3404 | int nid = page_to_nid(hpage); | |
6de2b1aa | 3405 | int ret = -EBUSY; |
93f70f90 NH |
3406 | |
3407 | spin_lock(&hugetlb_lock); | |
6de2b1aa | 3408 | if (is_hugepage_on_freelist(hpage)) { |
56f2fb14 NH |
3409 | /* |
3410 | * Hwpoisoned hugepage isn't linked to activelist or freelist, | |
3411 | * but dangling hpage->lru can trigger list-debug warnings | |
3412 | * (this happens when we call unpoison_memory() on it), | |
3413 | * so let it point to itself with list_del_init(). | |
3414 | */ | |
3415 | list_del_init(&hpage->lru); | |
8c6c2ecb | 3416 | set_page_refcounted(hpage); |
6de2b1aa NH |
3417 | h->free_huge_pages--; |
3418 | h->free_huge_pages_node[nid]--; | |
3419 | ret = 0; | |
3420 | } | |
93f70f90 | 3421 | spin_unlock(&hugetlb_lock); |
6de2b1aa | 3422 | return ret; |
93f70f90 | 3423 | } |
6de2b1aa | 3424 | #endif |