Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Generic hugetlb support. | |
3 | * (C) William Irwin, April 2004 | |
4 | */ | |
1da177e4 LT |
5 | #include <linux/list.h> |
6 | #include <linux/init.h> | |
7 | #include <linux/module.h> | |
8 | #include <linux/mm.h> | |
e1759c21 | 9 | #include <linux/seq_file.h> |
1da177e4 LT |
10 | #include <linux/sysctl.h> |
11 | #include <linux/highmem.h> | |
cddb8a5c | 12 | #include <linux/mmu_notifier.h> |
1da177e4 | 13 | #include <linux/nodemask.h> |
63551ae0 | 14 | #include <linux/pagemap.h> |
5da7ca86 | 15 | #include <linux/mempolicy.h> |
aea47ff3 | 16 | #include <linux/cpuset.h> |
3935baa9 | 17 | #include <linux/mutex.h> |
aa888a74 | 18 | #include <linux/bootmem.h> |
a3437870 | 19 | #include <linux/sysfs.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
0fe6e20b | 21 | #include <linux/rmap.h> |
fd6a03ed NH |
22 | #include <linux/swap.h> |
23 | #include <linux/swapops.h> | |
d6606683 | 24 | |
63551ae0 DG |
25 | #include <asm/page.h> |
26 | #include <asm/pgtable.h> | |
32f84528 | 27 | #include <linux/io.h> |
63551ae0 DG |
28 | |
29 | #include <linux/hugetlb.h> | |
9a305230 | 30 | #include <linux/node.h> |
7835e98b | 31 | #include "internal.h" |
1da177e4 LT |
32 | |
33 | const unsigned long hugetlb_zero = 0, hugetlb_infinity = ~0UL; | |
396faf03 MG |
34 | static gfp_t htlb_alloc_mask = GFP_HIGHUSER; |
35 | unsigned long hugepages_treat_as_movable; | |
a5516438 | 36 | |
e5ff2159 AK |
37 | static int max_hstate; |
38 | unsigned int default_hstate_idx; | |
39 | struct hstate hstates[HUGE_MAX_HSTATE]; | |
40 | ||
53ba51d2 JT |
41 | __initdata LIST_HEAD(huge_boot_pages); |
42 | ||
e5ff2159 AK |
43 | /* for command line parsing */ |
44 | static struct hstate * __initdata parsed_hstate; | |
45 | static unsigned long __initdata default_hstate_max_huge_pages; | |
e11bfbfc | 46 | static unsigned long __initdata default_hstate_size; |
e5ff2159 AK |
47 | |
48 | #define for_each_hstate(h) \ | |
49 | for ((h) = hstates; (h) < &hstates[max_hstate]; (h)++) | |
396faf03 | 50 | |
3935baa9 DG |
51 | /* |
52 | * Protects updates to hugepage_freelists, nr_huge_pages, and free_huge_pages | |
53 | */ | |
54 | static DEFINE_SPINLOCK(hugetlb_lock); | |
0bd0f9fb | 55 | |
96822904 AW |
56 | /* |
57 | * Region tracking -- allows tracking of reservations and instantiated pages | |
58 | * across the pages in a mapping. | |
84afd99b AW |
59 | * |
60 | * The region data structures are protected by a combination of the mmap_sem | |
61 | * and the hugetlb_instantion_mutex. To access or modify a region the caller | |
62 | * must either hold the mmap_sem for write, or the mmap_sem for read and | |
63 | * the hugetlb_instantiation mutex: | |
64 | * | |
32f84528 | 65 | * down_write(&mm->mmap_sem); |
84afd99b | 66 | * or |
32f84528 CF |
67 | * down_read(&mm->mmap_sem); |
68 | * mutex_lock(&hugetlb_instantiation_mutex); | |
96822904 AW |
69 | */ |
70 | struct file_region { | |
71 | struct list_head link; | |
72 | long from; | |
73 | long to; | |
74 | }; | |
75 | ||
76 | static long region_add(struct list_head *head, long f, long t) | |
77 | { | |
78 | struct file_region *rg, *nrg, *trg; | |
79 | ||
80 | /* Locate the region we are either in or before. */ | |
81 | list_for_each_entry(rg, head, link) | |
82 | if (f <= rg->to) | |
83 | break; | |
84 | ||
85 | /* Round our left edge to the current segment if it encloses us. */ | |
86 | if (f > rg->from) | |
87 | f = rg->from; | |
88 | ||
89 | /* Check for and consume any regions we now overlap with. */ | |
90 | nrg = rg; | |
91 | list_for_each_entry_safe(rg, trg, rg->link.prev, link) { | |
92 | if (&rg->link == head) | |
93 | break; | |
94 | if (rg->from > t) | |
95 | break; | |
96 | ||
97 | /* If this area reaches higher then extend our area to | |
98 | * include it completely. If this is not the first area | |
99 | * which we intend to reuse, free it. */ | |
100 | if (rg->to > t) | |
101 | t = rg->to; | |
102 | if (rg != nrg) { | |
103 | list_del(&rg->link); | |
104 | kfree(rg); | |
105 | } | |
106 | } | |
107 | nrg->from = f; | |
108 | nrg->to = t; | |
109 | return 0; | |
110 | } | |
111 | ||
112 | static long region_chg(struct list_head *head, long f, long t) | |
113 | { | |
114 | struct file_region *rg, *nrg; | |
115 | long chg = 0; | |
116 | ||
117 | /* Locate the region we are before or in. */ | |
118 | list_for_each_entry(rg, head, link) | |
119 | if (f <= rg->to) | |
120 | break; | |
121 | ||
122 | /* If we are below the current region then a new region is required. | |
123 | * Subtle, allocate a new region at the position but make it zero | |
124 | * size such that we can guarantee to record the reservation. */ | |
125 | if (&rg->link == head || t < rg->from) { | |
126 | nrg = kmalloc(sizeof(*nrg), GFP_KERNEL); | |
127 | if (!nrg) | |
128 | return -ENOMEM; | |
129 | nrg->from = f; | |
130 | nrg->to = f; | |
131 | INIT_LIST_HEAD(&nrg->link); | |
132 | list_add(&nrg->link, rg->link.prev); | |
133 | ||
134 | return t - f; | |
135 | } | |
136 | ||
137 | /* Round our left edge to the current segment if it encloses us. */ | |
138 | if (f > rg->from) | |
139 | f = rg->from; | |
140 | chg = t - f; | |
141 | ||
142 | /* Check for and consume any regions we now overlap with. */ | |
143 | list_for_each_entry(rg, rg->link.prev, link) { | |
144 | if (&rg->link == head) | |
145 | break; | |
146 | if (rg->from > t) | |
147 | return chg; | |
148 | ||
25985edc | 149 | /* We overlap with this area, if it extends further than |
96822904 AW |
150 | * us then we must extend ourselves. Account for its |
151 | * existing reservation. */ | |
152 | if (rg->to > t) { | |
153 | chg += rg->to - t; | |
154 | t = rg->to; | |
155 | } | |
156 | chg -= rg->to - rg->from; | |
157 | } | |
158 | return chg; | |
159 | } | |
160 | ||
161 | static long region_truncate(struct list_head *head, long end) | |
162 | { | |
163 | struct file_region *rg, *trg; | |
164 | long chg = 0; | |
165 | ||
166 | /* Locate the region we are either in or before. */ | |
167 | list_for_each_entry(rg, head, link) | |
168 | if (end <= rg->to) | |
169 | break; | |
170 | if (&rg->link == head) | |
171 | return 0; | |
172 | ||
173 | /* If we are in the middle of a region then adjust it. */ | |
174 | if (end > rg->from) { | |
175 | chg = rg->to - end; | |
176 | rg->to = end; | |
177 | rg = list_entry(rg->link.next, typeof(*rg), link); | |
178 | } | |
179 | ||
180 | /* Drop any remaining regions. */ | |
181 | list_for_each_entry_safe(rg, trg, rg->link.prev, link) { | |
182 | if (&rg->link == head) | |
183 | break; | |
184 | chg += rg->to - rg->from; | |
185 | list_del(&rg->link); | |
186 | kfree(rg); | |
187 | } | |
188 | return chg; | |
189 | } | |
190 | ||
84afd99b AW |
191 | static long region_count(struct list_head *head, long f, long t) |
192 | { | |
193 | struct file_region *rg; | |
194 | long chg = 0; | |
195 | ||
196 | /* Locate each segment we overlap with, and count that overlap. */ | |
197 | list_for_each_entry(rg, head, link) { | |
198 | int seg_from; | |
199 | int seg_to; | |
200 | ||
201 | if (rg->to <= f) | |
202 | continue; | |
203 | if (rg->from >= t) | |
204 | break; | |
205 | ||
206 | seg_from = max(rg->from, f); | |
207 | seg_to = min(rg->to, t); | |
208 | ||
209 | chg += seg_to - seg_from; | |
210 | } | |
211 | ||
212 | return chg; | |
213 | } | |
214 | ||
e7c4b0bf AW |
215 | /* |
216 | * Convert the address within this vma to the page offset within | |
217 | * the mapping, in pagecache page units; huge pages here. | |
218 | */ | |
a5516438 AK |
219 | static pgoff_t vma_hugecache_offset(struct hstate *h, |
220 | struct vm_area_struct *vma, unsigned long address) | |
e7c4b0bf | 221 | { |
a5516438 AK |
222 | return ((address - vma->vm_start) >> huge_page_shift(h)) + |
223 | (vma->vm_pgoff >> huge_page_order(h)); | |
e7c4b0bf AW |
224 | } |
225 | ||
0fe6e20b NH |
226 | pgoff_t linear_hugepage_index(struct vm_area_struct *vma, |
227 | unsigned long address) | |
228 | { | |
229 | return vma_hugecache_offset(hstate_vma(vma), vma, address); | |
230 | } | |
231 | ||
08fba699 MG |
232 | /* |
233 | * Return the size of the pages allocated when backing a VMA. In the majority | |
234 | * cases this will be same size as used by the page table entries. | |
235 | */ | |
236 | unsigned long vma_kernel_pagesize(struct vm_area_struct *vma) | |
237 | { | |
238 | struct hstate *hstate; | |
239 | ||
240 | if (!is_vm_hugetlb_page(vma)) | |
241 | return PAGE_SIZE; | |
242 | ||
243 | hstate = hstate_vma(vma); | |
244 | ||
245 | return 1UL << (hstate->order + PAGE_SHIFT); | |
246 | } | |
f340ca0f | 247 | EXPORT_SYMBOL_GPL(vma_kernel_pagesize); |
08fba699 | 248 | |
3340289d MG |
249 | /* |
250 | * Return the page size being used by the MMU to back a VMA. In the majority | |
251 | * of cases, the page size used by the kernel matches the MMU size. On | |
252 | * architectures where it differs, an architecture-specific version of this | |
253 | * function is required. | |
254 | */ | |
255 | #ifndef vma_mmu_pagesize | |
256 | unsigned long vma_mmu_pagesize(struct vm_area_struct *vma) | |
257 | { | |
258 | return vma_kernel_pagesize(vma); | |
259 | } | |
260 | #endif | |
261 | ||
84afd99b AW |
262 | /* |
263 | * Flags for MAP_PRIVATE reservations. These are stored in the bottom | |
264 | * bits of the reservation map pointer, which are always clear due to | |
265 | * alignment. | |
266 | */ | |
267 | #define HPAGE_RESV_OWNER (1UL << 0) | |
268 | #define HPAGE_RESV_UNMAPPED (1UL << 1) | |
04f2cbe3 | 269 | #define HPAGE_RESV_MASK (HPAGE_RESV_OWNER | HPAGE_RESV_UNMAPPED) |
84afd99b | 270 | |
a1e78772 MG |
271 | /* |
272 | * These helpers are used to track how many pages are reserved for | |
273 | * faults in a MAP_PRIVATE mapping. Only the process that called mmap() | |
274 | * is guaranteed to have their future faults succeed. | |
275 | * | |
276 | * With the exception of reset_vma_resv_huge_pages() which is called at fork(), | |
277 | * the reserve counters are updated with the hugetlb_lock held. It is safe | |
278 | * to reset the VMA at fork() time as it is not in use yet and there is no | |
279 | * chance of the global counters getting corrupted as a result of the values. | |
84afd99b AW |
280 | * |
281 | * The private mapping reservation is represented in a subtly different | |
282 | * manner to a shared mapping. A shared mapping has a region map associated | |
283 | * with the underlying file, this region map represents the backing file | |
284 | * pages which have ever had a reservation assigned which this persists even | |
285 | * after the page is instantiated. A private mapping has a region map | |
286 | * associated with the original mmap which is attached to all VMAs which | |
287 | * reference it, this region map represents those offsets which have consumed | |
288 | * reservation ie. where pages have been instantiated. | |
a1e78772 | 289 | */ |
e7c4b0bf AW |
290 | static unsigned long get_vma_private_data(struct vm_area_struct *vma) |
291 | { | |
292 | return (unsigned long)vma->vm_private_data; | |
293 | } | |
294 | ||
295 | static void set_vma_private_data(struct vm_area_struct *vma, | |
296 | unsigned long value) | |
297 | { | |
298 | vma->vm_private_data = (void *)value; | |
299 | } | |
300 | ||
84afd99b AW |
301 | struct resv_map { |
302 | struct kref refs; | |
303 | struct list_head regions; | |
304 | }; | |
305 | ||
2a4b3ded | 306 | static struct resv_map *resv_map_alloc(void) |
84afd99b AW |
307 | { |
308 | struct resv_map *resv_map = kmalloc(sizeof(*resv_map), GFP_KERNEL); | |
309 | if (!resv_map) | |
310 | return NULL; | |
311 | ||
312 | kref_init(&resv_map->refs); | |
313 | INIT_LIST_HEAD(&resv_map->regions); | |
314 | ||
315 | return resv_map; | |
316 | } | |
317 | ||
2a4b3ded | 318 | static void resv_map_release(struct kref *ref) |
84afd99b AW |
319 | { |
320 | struct resv_map *resv_map = container_of(ref, struct resv_map, refs); | |
321 | ||
322 | /* Clear out any active regions before we release the map. */ | |
323 | region_truncate(&resv_map->regions, 0); | |
324 | kfree(resv_map); | |
325 | } | |
326 | ||
327 | static struct resv_map *vma_resv_map(struct vm_area_struct *vma) | |
a1e78772 MG |
328 | { |
329 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); | |
f83a275d | 330 | if (!(vma->vm_flags & VM_MAYSHARE)) |
84afd99b AW |
331 | return (struct resv_map *)(get_vma_private_data(vma) & |
332 | ~HPAGE_RESV_MASK); | |
2a4b3ded | 333 | return NULL; |
a1e78772 MG |
334 | } |
335 | ||
84afd99b | 336 | static void set_vma_resv_map(struct vm_area_struct *vma, struct resv_map *map) |
a1e78772 MG |
337 | { |
338 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); | |
f83a275d | 339 | VM_BUG_ON(vma->vm_flags & VM_MAYSHARE); |
a1e78772 | 340 | |
84afd99b AW |
341 | set_vma_private_data(vma, (get_vma_private_data(vma) & |
342 | HPAGE_RESV_MASK) | (unsigned long)map); | |
04f2cbe3 MG |
343 | } |
344 | ||
345 | static void set_vma_resv_flags(struct vm_area_struct *vma, unsigned long flags) | |
346 | { | |
04f2cbe3 | 347 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); |
f83a275d | 348 | VM_BUG_ON(vma->vm_flags & VM_MAYSHARE); |
e7c4b0bf AW |
349 | |
350 | set_vma_private_data(vma, get_vma_private_data(vma) | flags); | |
04f2cbe3 MG |
351 | } |
352 | ||
353 | static int is_vma_resv_set(struct vm_area_struct *vma, unsigned long flag) | |
354 | { | |
355 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); | |
e7c4b0bf AW |
356 | |
357 | return (get_vma_private_data(vma) & flag) != 0; | |
a1e78772 MG |
358 | } |
359 | ||
360 | /* Decrement the reserved pages in the hugepage pool by one */ | |
a5516438 AK |
361 | static void decrement_hugepage_resv_vma(struct hstate *h, |
362 | struct vm_area_struct *vma) | |
a1e78772 | 363 | { |
c37f9fb1 AW |
364 | if (vma->vm_flags & VM_NORESERVE) |
365 | return; | |
366 | ||
f83a275d | 367 | if (vma->vm_flags & VM_MAYSHARE) { |
a1e78772 | 368 | /* Shared mappings always use reserves */ |
a5516438 | 369 | h->resv_huge_pages--; |
84afd99b | 370 | } else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { |
a1e78772 MG |
371 | /* |
372 | * Only the process that called mmap() has reserves for | |
373 | * private mappings. | |
374 | */ | |
a5516438 | 375 | h->resv_huge_pages--; |
a1e78772 MG |
376 | } |
377 | } | |
378 | ||
04f2cbe3 | 379 | /* Reset counters to 0 and clear all HPAGE_RESV_* flags */ |
a1e78772 MG |
380 | void reset_vma_resv_huge_pages(struct vm_area_struct *vma) |
381 | { | |
382 | VM_BUG_ON(!is_vm_hugetlb_page(vma)); | |
f83a275d | 383 | if (!(vma->vm_flags & VM_MAYSHARE)) |
a1e78772 MG |
384 | vma->vm_private_data = (void *)0; |
385 | } | |
386 | ||
387 | /* Returns true if the VMA has associated reserve pages */ | |
7f09ca51 | 388 | static int vma_has_reserves(struct vm_area_struct *vma) |
a1e78772 | 389 | { |
f83a275d | 390 | if (vma->vm_flags & VM_MAYSHARE) |
7f09ca51 MG |
391 | return 1; |
392 | if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) | |
393 | return 1; | |
394 | return 0; | |
a1e78772 MG |
395 | } |
396 | ||
0ebabb41 NH |
397 | static void copy_gigantic_page(struct page *dst, struct page *src) |
398 | { | |
399 | int i; | |
400 | struct hstate *h = page_hstate(src); | |
401 | struct page *dst_base = dst; | |
402 | struct page *src_base = src; | |
403 | ||
404 | for (i = 0; i < pages_per_huge_page(h); ) { | |
405 | cond_resched(); | |
406 | copy_highpage(dst, src); | |
407 | ||
408 | i++; | |
409 | dst = mem_map_next(dst, dst_base, i); | |
410 | src = mem_map_next(src, src_base, i); | |
411 | } | |
412 | } | |
413 | ||
414 | void copy_huge_page(struct page *dst, struct page *src) | |
415 | { | |
416 | int i; | |
417 | struct hstate *h = page_hstate(src); | |
418 | ||
419 | if (unlikely(pages_per_huge_page(h) > MAX_ORDER_NR_PAGES)) { | |
420 | copy_gigantic_page(dst, src); | |
421 | return; | |
422 | } | |
423 | ||
424 | might_sleep(); | |
425 | for (i = 0; i < pages_per_huge_page(h); i++) { | |
426 | cond_resched(); | |
427 | copy_highpage(dst + i, src + i); | |
428 | } | |
429 | } | |
430 | ||
a5516438 | 431 | static void enqueue_huge_page(struct hstate *h, struct page *page) |
1da177e4 LT |
432 | { |
433 | int nid = page_to_nid(page); | |
a5516438 AK |
434 | list_add(&page->lru, &h->hugepage_freelists[nid]); |
435 | h->free_huge_pages++; | |
436 | h->free_huge_pages_node[nid]++; | |
1da177e4 LT |
437 | } |
438 | ||
bf50bab2 NH |
439 | static struct page *dequeue_huge_page_node(struct hstate *h, int nid) |
440 | { | |
441 | struct page *page; | |
442 | ||
443 | if (list_empty(&h->hugepage_freelists[nid])) | |
444 | return NULL; | |
445 | page = list_entry(h->hugepage_freelists[nid].next, struct page, lru); | |
446 | list_del(&page->lru); | |
a9869b83 | 447 | set_page_refcounted(page); |
bf50bab2 NH |
448 | h->free_huge_pages--; |
449 | h->free_huge_pages_node[nid]--; | |
450 | return page; | |
451 | } | |
452 | ||
a5516438 AK |
453 | static struct page *dequeue_huge_page_vma(struct hstate *h, |
454 | struct vm_area_struct *vma, | |
04f2cbe3 | 455 | unsigned long address, int avoid_reserve) |
1da177e4 | 456 | { |
cc9a6c87 | 457 | struct page *page; |
480eccf9 | 458 | struct mempolicy *mpol; |
19770b32 | 459 | nodemask_t *nodemask; |
c0ff7453 | 460 | struct zonelist *zonelist; |
dd1a239f MG |
461 | struct zone *zone; |
462 | struct zoneref *z; | |
cc9a6c87 | 463 | unsigned int cpuset_mems_cookie; |
1da177e4 | 464 | |
cc9a6c87 MG |
465 | retry_cpuset: |
466 | cpuset_mems_cookie = get_mems_allowed(); | |
c0ff7453 MX |
467 | zonelist = huge_zonelist(vma, address, |
468 | htlb_alloc_mask, &mpol, &nodemask); | |
a1e78772 MG |
469 | /* |
470 | * A child process with MAP_PRIVATE mappings created by their parent | |
471 | * have no page reserves. This check ensures that reservations are | |
472 | * not "stolen". The child may still get SIGKILLed | |
473 | */ | |
7f09ca51 | 474 | if (!vma_has_reserves(vma) && |
a5516438 | 475 | h->free_huge_pages - h->resv_huge_pages == 0) |
c0ff7453 | 476 | goto err; |
a1e78772 | 477 | |
04f2cbe3 | 478 | /* If reserves cannot be used, ensure enough pages are in the pool */ |
a5516438 | 479 | if (avoid_reserve && h->free_huge_pages - h->resv_huge_pages == 0) |
6eab04a8 | 480 | goto err; |
04f2cbe3 | 481 | |
19770b32 MG |
482 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
483 | MAX_NR_ZONES - 1, nodemask) { | |
bf50bab2 NH |
484 | if (cpuset_zone_allowed_softwall(zone, htlb_alloc_mask)) { |
485 | page = dequeue_huge_page_node(h, zone_to_nid(zone)); | |
486 | if (page) { | |
487 | if (!avoid_reserve) | |
488 | decrement_hugepage_resv_vma(h, vma); | |
489 | break; | |
490 | } | |
3abf7afd | 491 | } |
1da177e4 | 492 | } |
cc9a6c87 | 493 | |
52cd3b07 | 494 | mpol_cond_put(mpol); |
cc9a6c87 MG |
495 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
496 | goto retry_cpuset; | |
1da177e4 | 497 | return page; |
cc9a6c87 MG |
498 | |
499 | err: | |
500 | mpol_cond_put(mpol); | |
501 | return NULL; | |
1da177e4 LT |
502 | } |
503 | ||
a5516438 | 504 | static void update_and_free_page(struct hstate *h, struct page *page) |
6af2acb6 AL |
505 | { |
506 | int i; | |
a5516438 | 507 | |
18229df5 AW |
508 | VM_BUG_ON(h->order >= MAX_ORDER); |
509 | ||
a5516438 AK |
510 | h->nr_huge_pages--; |
511 | h->nr_huge_pages_node[page_to_nid(page)]--; | |
512 | for (i = 0; i < pages_per_huge_page(h); i++) { | |
32f84528 CF |
513 | page[i].flags &= ~(1 << PG_locked | 1 << PG_error | |
514 | 1 << PG_referenced | 1 << PG_dirty | | |
515 | 1 << PG_active | 1 << PG_reserved | | |
516 | 1 << PG_private | 1 << PG_writeback); | |
6af2acb6 AL |
517 | } |
518 | set_compound_page_dtor(page, NULL); | |
519 | set_page_refcounted(page); | |
7f2e9525 | 520 | arch_release_hugepage(page); |
a5516438 | 521 | __free_pages(page, huge_page_order(h)); |
6af2acb6 AL |
522 | } |
523 | ||
e5ff2159 AK |
524 | struct hstate *size_to_hstate(unsigned long size) |
525 | { | |
526 | struct hstate *h; | |
527 | ||
528 | for_each_hstate(h) { | |
529 | if (huge_page_size(h) == size) | |
530 | return h; | |
531 | } | |
532 | return NULL; | |
533 | } | |
534 | ||
27a85ef1 DG |
535 | static void free_huge_page(struct page *page) |
536 | { | |
a5516438 AK |
537 | /* |
538 | * Can't pass hstate in here because it is called from the | |
539 | * compound page destructor. | |
540 | */ | |
e5ff2159 | 541 | struct hstate *h = page_hstate(page); |
7893d1d5 | 542 | int nid = page_to_nid(page); |
c79fb75e | 543 | struct address_space *mapping; |
27a85ef1 | 544 | |
c79fb75e | 545 | mapping = (struct address_space *) page_private(page); |
e5df70ab | 546 | set_page_private(page, 0); |
23be7468 | 547 | page->mapping = NULL; |
7893d1d5 | 548 | BUG_ON(page_count(page)); |
0fe6e20b | 549 | BUG_ON(page_mapcount(page)); |
27a85ef1 DG |
550 | INIT_LIST_HEAD(&page->lru); |
551 | ||
552 | spin_lock(&hugetlb_lock); | |
aa888a74 | 553 | if (h->surplus_huge_pages_node[nid] && huge_page_order(h) < MAX_ORDER) { |
a5516438 AK |
554 | update_and_free_page(h, page); |
555 | h->surplus_huge_pages--; | |
556 | h->surplus_huge_pages_node[nid]--; | |
7893d1d5 | 557 | } else { |
a5516438 | 558 | enqueue_huge_page(h, page); |
7893d1d5 | 559 | } |
27a85ef1 | 560 | spin_unlock(&hugetlb_lock); |
c79fb75e | 561 | if (mapping) |
9a119c05 | 562 | hugetlb_put_quota(mapping, 1); |
27a85ef1 DG |
563 | } |
564 | ||
a5516438 | 565 | static void prep_new_huge_page(struct hstate *h, struct page *page, int nid) |
b7ba30c6 AK |
566 | { |
567 | set_compound_page_dtor(page, free_huge_page); | |
568 | spin_lock(&hugetlb_lock); | |
a5516438 AK |
569 | h->nr_huge_pages++; |
570 | h->nr_huge_pages_node[nid]++; | |
b7ba30c6 AK |
571 | spin_unlock(&hugetlb_lock); |
572 | put_page(page); /* free it into the hugepage allocator */ | |
573 | } | |
574 | ||
20a0307c WF |
575 | static void prep_compound_gigantic_page(struct page *page, unsigned long order) |
576 | { | |
577 | int i; | |
578 | int nr_pages = 1 << order; | |
579 | struct page *p = page + 1; | |
580 | ||
581 | /* we rely on prep_new_huge_page to set the destructor */ | |
582 | set_compound_order(page, order); | |
583 | __SetPageHead(page); | |
584 | for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) { | |
585 | __SetPageTail(p); | |
58a84aa9 | 586 | set_page_count(p, 0); |
20a0307c WF |
587 | p->first_page = page; |
588 | } | |
589 | } | |
590 | ||
591 | int PageHuge(struct page *page) | |
592 | { | |
593 | compound_page_dtor *dtor; | |
594 | ||
595 | if (!PageCompound(page)) | |
596 | return 0; | |
597 | ||
598 | page = compound_head(page); | |
599 | dtor = get_compound_page_dtor(page); | |
600 | ||
601 | return dtor == free_huge_page; | |
602 | } | |
43131e14 NH |
603 | EXPORT_SYMBOL_GPL(PageHuge); |
604 | ||
a5516438 | 605 | static struct page *alloc_fresh_huge_page_node(struct hstate *h, int nid) |
1da177e4 | 606 | { |
1da177e4 | 607 | struct page *page; |
f96efd58 | 608 | |
aa888a74 AK |
609 | if (h->order >= MAX_ORDER) |
610 | return NULL; | |
611 | ||
6484eb3e | 612 | page = alloc_pages_exact_node(nid, |
551883ae NA |
613 | htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE| |
614 | __GFP_REPEAT|__GFP_NOWARN, | |
a5516438 | 615 | huge_page_order(h)); |
1da177e4 | 616 | if (page) { |
7f2e9525 | 617 | if (arch_prepare_hugepage(page)) { |
caff3a2c | 618 | __free_pages(page, huge_page_order(h)); |
7b8ee84d | 619 | return NULL; |
7f2e9525 | 620 | } |
a5516438 | 621 | prep_new_huge_page(h, page, nid); |
1da177e4 | 622 | } |
63b4613c NA |
623 | |
624 | return page; | |
625 | } | |
626 | ||
9a76db09 | 627 | /* |
6ae11b27 LS |
628 | * common helper functions for hstate_next_node_to_{alloc|free}. |
629 | * We may have allocated or freed a huge page based on a different | |
630 | * nodes_allowed previously, so h->next_node_to_{alloc|free} might | |
631 | * be outside of *nodes_allowed. Ensure that we use an allowed | |
632 | * node for alloc or free. | |
9a76db09 | 633 | */ |
6ae11b27 | 634 | static int next_node_allowed(int nid, nodemask_t *nodes_allowed) |
9a76db09 | 635 | { |
6ae11b27 | 636 | nid = next_node(nid, *nodes_allowed); |
9a76db09 | 637 | if (nid == MAX_NUMNODES) |
6ae11b27 | 638 | nid = first_node(*nodes_allowed); |
9a76db09 LS |
639 | VM_BUG_ON(nid >= MAX_NUMNODES); |
640 | ||
641 | return nid; | |
642 | } | |
643 | ||
6ae11b27 LS |
644 | static int get_valid_node_allowed(int nid, nodemask_t *nodes_allowed) |
645 | { | |
646 | if (!node_isset(nid, *nodes_allowed)) | |
647 | nid = next_node_allowed(nid, nodes_allowed); | |
648 | return nid; | |
649 | } | |
650 | ||
5ced66c9 | 651 | /* |
6ae11b27 LS |
652 | * returns the previously saved node ["this node"] from which to |
653 | * allocate a persistent huge page for the pool and advance the | |
654 | * next node from which to allocate, handling wrap at end of node | |
655 | * mask. | |
5ced66c9 | 656 | */ |
6ae11b27 LS |
657 | static int hstate_next_node_to_alloc(struct hstate *h, |
658 | nodemask_t *nodes_allowed) | |
5ced66c9 | 659 | { |
6ae11b27 LS |
660 | int nid; |
661 | ||
662 | VM_BUG_ON(!nodes_allowed); | |
663 | ||
664 | nid = get_valid_node_allowed(h->next_nid_to_alloc, nodes_allowed); | |
665 | h->next_nid_to_alloc = next_node_allowed(nid, nodes_allowed); | |
9a76db09 | 666 | |
9a76db09 | 667 | return nid; |
5ced66c9 AK |
668 | } |
669 | ||
6ae11b27 | 670 | static int alloc_fresh_huge_page(struct hstate *h, nodemask_t *nodes_allowed) |
63b4613c NA |
671 | { |
672 | struct page *page; | |
673 | int start_nid; | |
674 | int next_nid; | |
675 | int ret = 0; | |
676 | ||
6ae11b27 | 677 | start_nid = hstate_next_node_to_alloc(h, nodes_allowed); |
e8c5c824 | 678 | next_nid = start_nid; |
63b4613c NA |
679 | |
680 | do { | |
e8c5c824 | 681 | page = alloc_fresh_huge_page_node(h, next_nid); |
9a76db09 | 682 | if (page) { |
63b4613c | 683 | ret = 1; |
9a76db09 LS |
684 | break; |
685 | } | |
6ae11b27 | 686 | next_nid = hstate_next_node_to_alloc(h, nodes_allowed); |
9a76db09 | 687 | } while (next_nid != start_nid); |
63b4613c | 688 | |
3b116300 AL |
689 | if (ret) |
690 | count_vm_event(HTLB_BUDDY_PGALLOC); | |
691 | else | |
692 | count_vm_event(HTLB_BUDDY_PGALLOC_FAIL); | |
693 | ||
63b4613c | 694 | return ret; |
1da177e4 LT |
695 | } |
696 | ||
e8c5c824 | 697 | /* |
6ae11b27 LS |
698 | * helper for free_pool_huge_page() - return the previously saved |
699 | * node ["this node"] from which to free a huge page. Advance the | |
700 | * next node id whether or not we find a free huge page to free so | |
701 | * that the next attempt to free addresses the next node. | |
e8c5c824 | 702 | */ |
6ae11b27 | 703 | static int hstate_next_node_to_free(struct hstate *h, nodemask_t *nodes_allowed) |
e8c5c824 | 704 | { |
6ae11b27 LS |
705 | int nid; |
706 | ||
707 | VM_BUG_ON(!nodes_allowed); | |
708 | ||
709 | nid = get_valid_node_allowed(h->next_nid_to_free, nodes_allowed); | |
710 | h->next_nid_to_free = next_node_allowed(nid, nodes_allowed); | |
9a76db09 | 711 | |
9a76db09 | 712 | return nid; |
e8c5c824 LS |
713 | } |
714 | ||
715 | /* | |
716 | * Free huge page from pool from next node to free. | |
717 | * Attempt to keep persistent huge pages more or less | |
718 | * balanced over allowed nodes. | |
719 | * Called with hugetlb_lock locked. | |
720 | */ | |
6ae11b27 LS |
721 | static int free_pool_huge_page(struct hstate *h, nodemask_t *nodes_allowed, |
722 | bool acct_surplus) | |
e8c5c824 LS |
723 | { |
724 | int start_nid; | |
725 | int next_nid; | |
726 | int ret = 0; | |
727 | ||
6ae11b27 | 728 | start_nid = hstate_next_node_to_free(h, nodes_allowed); |
e8c5c824 LS |
729 | next_nid = start_nid; |
730 | ||
731 | do { | |
685f3457 LS |
732 | /* |
733 | * If we're returning unused surplus pages, only examine | |
734 | * nodes with surplus pages. | |
735 | */ | |
736 | if ((!acct_surplus || h->surplus_huge_pages_node[next_nid]) && | |
737 | !list_empty(&h->hugepage_freelists[next_nid])) { | |
e8c5c824 LS |
738 | struct page *page = |
739 | list_entry(h->hugepage_freelists[next_nid].next, | |
740 | struct page, lru); | |
741 | list_del(&page->lru); | |
742 | h->free_huge_pages--; | |
743 | h->free_huge_pages_node[next_nid]--; | |
685f3457 LS |
744 | if (acct_surplus) { |
745 | h->surplus_huge_pages--; | |
746 | h->surplus_huge_pages_node[next_nid]--; | |
747 | } | |
e8c5c824 LS |
748 | update_and_free_page(h, page); |
749 | ret = 1; | |
9a76db09 | 750 | break; |
e8c5c824 | 751 | } |
6ae11b27 | 752 | next_nid = hstate_next_node_to_free(h, nodes_allowed); |
9a76db09 | 753 | } while (next_nid != start_nid); |
e8c5c824 LS |
754 | |
755 | return ret; | |
756 | } | |
757 | ||
bf50bab2 | 758 | static struct page *alloc_buddy_huge_page(struct hstate *h, int nid) |
7893d1d5 AL |
759 | { |
760 | struct page *page; | |
bf50bab2 | 761 | unsigned int r_nid; |
7893d1d5 | 762 | |
aa888a74 AK |
763 | if (h->order >= MAX_ORDER) |
764 | return NULL; | |
765 | ||
d1c3fb1f NA |
766 | /* |
767 | * Assume we will successfully allocate the surplus page to | |
768 | * prevent racing processes from causing the surplus to exceed | |
769 | * overcommit | |
770 | * | |
771 | * This however introduces a different race, where a process B | |
772 | * tries to grow the static hugepage pool while alloc_pages() is | |
773 | * called by process A. B will only examine the per-node | |
774 | * counters in determining if surplus huge pages can be | |
775 | * converted to normal huge pages in adjust_pool_surplus(). A | |
776 | * won't be able to increment the per-node counter, until the | |
777 | * lock is dropped by B, but B doesn't drop hugetlb_lock until | |
778 | * no more huge pages can be converted from surplus to normal | |
779 | * state (and doesn't try to convert again). Thus, we have a | |
780 | * case where a surplus huge page exists, the pool is grown, and | |
781 | * the surplus huge page still exists after, even though it | |
782 | * should just have been converted to a normal huge page. This | |
783 | * does not leak memory, though, as the hugepage will be freed | |
784 | * once it is out of use. It also does not allow the counters to | |
785 | * go out of whack in adjust_pool_surplus() as we don't modify | |
786 | * the node values until we've gotten the hugepage and only the | |
787 | * per-node value is checked there. | |
788 | */ | |
789 | spin_lock(&hugetlb_lock); | |
a5516438 | 790 | if (h->surplus_huge_pages >= h->nr_overcommit_huge_pages) { |
d1c3fb1f NA |
791 | spin_unlock(&hugetlb_lock); |
792 | return NULL; | |
793 | } else { | |
a5516438 AK |
794 | h->nr_huge_pages++; |
795 | h->surplus_huge_pages++; | |
d1c3fb1f NA |
796 | } |
797 | spin_unlock(&hugetlb_lock); | |
798 | ||
bf50bab2 NH |
799 | if (nid == NUMA_NO_NODE) |
800 | page = alloc_pages(htlb_alloc_mask|__GFP_COMP| | |
801 | __GFP_REPEAT|__GFP_NOWARN, | |
802 | huge_page_order(h)); | |
803 | else | |
804 | page = alloc_pages_exact_node(nid, | |
805 | htlb_alloc_mask|__GFP_COMP|__GFP_THISNODE| | |
806 | __GFP_REPEAT|__GFP_NOWARN, huge_page_order(h)); | |
d1c3fb1f | 807 | |
caff3a2c GS |
808 | if (page && arch_prepare_hugepage(page)) { |
809 | __free_pages(page, huge_page_order(h)); | |
ea5768c7 | 810 | page = NULL; |
caff3a2c GS |
811 | } |
812 | ||
d1c3fb1f | 813 | spin_lock(&hugetlb_lock); |
7893d1d5 | 814 | if (page) { |
bf50bab2 | 815 | r_nid = page_to_nid(page); |
7893d1d5 | 816 | set_compound_page_dtor(page, free_huge_page); |
d1c3fb1f NA |
817 | /* |
818 | * We incremented the global counters already | |
819 | */ | |
bf50bab2 NH |
820 | h->nr_huge_pages_node[r_nid]++; |
821 | h->surplus_huge_pages_node[r_nid]++; | |
3b116300 | 822 | __count_vm_event(HTLB_BUDDY_PGALLOC); |
d1c3fb1f | 823 | } else { |
a5516438 AK |
824 | h->nr_huge_pages--; |
825 | h->surplus_huge_pages--; | |
3b116300 | 826 | __count_vm_event(HTLB_BUDDY_PGALLOC_FAIL); |
7893d1d5 | 827 | } |
d1c3fb1f | 828 | spin_unlock(&hugetlb_lock); |
7893d1d5 AL |
829 | |
830 | return page; | |
831 | } | |
832 | ||
bf50bab2 NH |
833 | /* |
834 | * This allocation function is useful in the context where vma is irrelevant. | |
835 | * E.g. soft-offlining uses this function because it only cares physical | |
836 | * address of error page. | |
837 | */ | |
838 | struct page *alloc_huge_page_node(struct hstate *h, int nid) | |
839 | { | |
840 | struct page *page; | |
841 | ||
842 | spin_lock(&hugetlb_lock); | |
843 | page = dequeue_huge_page_node(h, nid); | |
844 | spin_unlock(&hugetlb_lock); | |
845 | ||
846 | if (!page) | |
847 | page = alloc_buddy_huge_page(h, nid); | |
848 | ||
849 | return page; | |
850 | } | |
851 | ||
e4e574b7 | 852 | /* |
25985edc | 853 | * Increase the hugetlb pool such that it can accommodate a reservation |
e4e574b7 AL |
854 | * of size 'delta'. |
855 | */ | |
a5516438 | 856 | static int gather_surplus_pages(struct hstate *h, int delta) |
e4e574b7 AL |
857 | { |
858 | struct list_head surplus_list; | |
859 | struct page *page, *tmp; | |
860 | int ret, i; | |
861 | int needed, allocated; | |
28073b02 | 862 | bool alloc_ok = true; |
e4e574b7 | 863 | |
a5516438 | 864 | needed = (h->resv_huge_pages + delta) - h->free_huge_pages; |
ac09b3a1 | 865 | if (needed <= 0) { |
a5516438 | 866 | h->resv_huge_pages += delta; |
e4e574b7 | 867 | return 0; |
ac09b3a1 | 868 | } |
e4e574b7 AL |
869 | |
870 | allocated = 0; | |
871 | INIT_LIST_HEAD(&surplus_list); | |
872 | ||
873 | ret = -ENOMEM; | |
874 | retry: | |
875 | spin_unlock(&hugetlb_lock); | |
876 | for (i = 0; i < needed; i++) { | |
bf50bab2 | 877 | page = alloc_buddy_huge_page(h, NUMA_NO_NODE); |
28073b02 HD |
878 | if (!page) { |
879 | alloc_ok = false; | |
880 | break; | |
881 | } | |
e4e574b7 AL |
882 | list_add(&page->lru, &surplus_list); |
883 | } | |
28073b02 | 884 | allocated += i; |
e4e574b7 AL |
885 | |
886 | /* | |
887 | * After retaking hugetlb_lock, we need to recalculate 'needed' | |
888 | * because either resv_huge_pages or free_huge_pages may have changed. | |
889 | */ | |
890 | spin_lock(&hugetlb_lock); | |
a5516438 AK |
891 | needed = (h->resv_huge_pages + delta) - |
892 | (h->free_huge_pages + allocated); | |
28073b02 HD |
893 | if (needed > 0) { |
894 | if (alloc_ok) | |
895 | goto retry; | |
896 | /* | |
897 | * We were not able to allocate enough pages to | |
898 | * satisfy the entire reservation so we free what | |
899 | * we've allocated so far. | |
900 | */ | |
901 | goto free; | |
902 | } | |
e4e574b7 AL |
903 | /* |
904 | * The surplus_list now contains _at_least_ the number of extra pages | |
25985edc | 905 | * needed to accommodate the reservation. Add the appropriate number |
e4e574b7 | 906 | * of pages to the hugetlb pool and free the extras back to the buddy |
ac09b3a1 AL |
907 | * allocator. Commit the entire reservation here to prevent another |
908 | * process from stealing the pages as they are added to the pool but | |
909 | * before they are reserved. | |
e4e574b7 AL |
910 | */ |
911 | needed += allocated; | |
a5516438 | 912 | h->resv_huge_pages += delta; |
e4e574b7 | 913 | ret = 0; |
a9869b83 | 914 | |
19fc3f0a | 915 | /* Free the needed pages to the hugetlb pool */ |
e4e574b7 | 916 | list_for_each_entry_safe(page, tmp, &surplus_list, lru) { |
19fc3f0a AL |
917 | if ((--needed) < 0) |
918 | break; | |
e4e574b7 | 919 | list_del(&page->lru); |
a9869b83 NH |
920 | /* |
921 | * This page is now managed by the hugetlb allocator and has | |
922 | * no users -- drop the buddy allocator's reference. | |
923 | */ | |
924 | put_page_testzero(page); | |
925 | VM_BUG_ON(page_count(page)); | |
a5516438 | 926 | enqueue_huge_page(h, page); |
19fc3f0a | 927 | } |
28073b02 | 928 | free: |
b0365c8d | 929 | spin_unlock(&hugetlb_lock); |
19fc3f0a AL |
930 | |
931 | /* Free unnecessary surplus pages to the buddy allocator */ | |
932 | if (!list_empty(&surplus_list)) { | |
19fc3f0a AL |
933 | list_for_each_entry_safe(page, tmp, &surplus_list, lru) { |
934 | list_del(&page->lru); | |
a9869b83 | 935 | put_page(page); |
af767cbd | 936 | } |
e4e574b7 | 937 | } |
a9869b83 | 938 | spin_lock(&hugetlb_lock); |
e4e574b7 AL |
939 | |
940 | return ret; | |
941 | } | |
942 | ||
943 | /* | |
944 | * When releasing a hugetlb pool reservation, any surplus pages that were | |
945 | * allocated to satisfy the reservation must be explicitly freed if they were | |
946 | * never used. | |
685f3457 | 947 | * Called with hugetlb_lock held. |
e4e574b7 | 948 | */ |
a5516438 AK |
949 | static void return_unused_surplus_pages(struct hstate *h, |
950 | unsigned long unused_resv_pages) | |
e4e574b7 | 951 | { |
e4e574b7 AL |
952 | unsigned long nr_pages; |
953 | ||
ac09b3a1 | 954 | /* Uncommit the reservation */ |
a5516438 | 955 | h->resv_huge_pages -= unused_resv_pages; |
ac09b3a1 | 956 | |
aa888a74 AK |
957 | /* Cannot return gigantic pages currently */ |
958 | if (h->order >= MAX_ORDER) | |
959 | return; | |
960 | ||
a5516438 | 961 | nr_pages = min(unused_resv_pages, h->surplus_huge_pages); |
e4e574b7 | 962 | |
685f3457 LS |
963 | /* |
964 | * We want to release as many surplus pages as possible, spread | |
9b5e5d0f LS |
965 | * evenly across all nodes with memory. Iterate across these nodes |
966 | * until we can no longer free unreserved surplus pages. This occurs | |
967 | * when the nodes with surplus pages have no free pages. | |
968 | * free_pool_huge_page() will balance the the freed pages across the | |
969 | * on-line nodes with memory and will handle the hstate accounting. | |
685f3457 LS |
970 | */ |
971 | while (nr_pages--) { | |
9b5e5d0f | 972 | if (!free_pool_huge_page(h, &node_states[N_HIGH_MEMORY], 1)) |
685f3457 | 973 | break; |
e4e574b7 AL |
974 | } |
975 | } | |
976 | ||
c37f9fb1 AW |
977 | /* |
978 | * Determine if the huge page at addr within the vma has an associated | |
979 | * reservation. Where it does not we will need to logically increase | |
980 | * reservation and actually increase quota before an allocation can occur. | |
981 | * Where any new reservation would be required the reservation change is | |
982 | * prepared, but not committed. Once the page has been quota'd allocated | |
983 | * an instantiated the change should be committed via vma_commit_reservation. | |
984 | * No action is required on failure. | |
985 | */ | |
e2f17d94 | 986 | static long vma_needs_reservation(struct hstate *h, |
a5516438 | 987 | struct vm_area_struct *vma, unsigned long addr) |
c37f9fb1 AW |
988 | { |
989 | struct address_space *mapping = vma->vm_file->f_mapping; | |
990 | struct inode *inode = mapping->host; | |
991 | ||
f83a275d | 992 | if (vma->vm_flags & VM_MAYSHARE) { |
a5516438 | 993 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
c37f9fb1 AW |
994 | return region_chg(&inode->i_mapping->private_list, |
995 | idx, idx + 1); | |
996 | ||
84afd99b AW |
997 | } else if (!is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { |
998 | return 1; | |
c37f9fb1 | 999 | |
84afd99b | 1000 | } else { |
e2f17d94 | 1001 | long err; |
a5516438 | 1002 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
84afd99b AW |
1003 | struct resv_map *reservations = vma_resv_map(vma); |
1004 | ||
1005 | err = region_chg(&reservations->regions, idx, idx + 1); | |
1006 | if (err < 0) | |
1007 | return err; | |
1008 | return 0; | |
1009 | } | |
c37f9fb1 | 1010 | } |
a5516438 AK |
1011 | static void vma_commit_reservation(struct hstate *h, |
1012 | struct vm_area_struct *vma, unsigned long addr) | |
c37f9fb1 AW |
1013 | { |
1014 | struct address_space *mapping = vma->vm_file->f_mapping; | |
1015 | struct inode *inode = mapping->host; | |
1016 | ||
f83a275d | 1017 | if (vma->vm_flags & VM_MAYSHARE) { |
a5516438 | 1018 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
c37f9fb1 | 1019 | region_add(&inode->i_mapping->private_list, idx, idx + 1); |
84afd99b AW |
1020 | |
1021 | } else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) { | |
a5516438 | 1022 | pgoff_t idx = vma_hugecache_offset(h, vma, addr); |
84afd99b AW |
1023 | struct resv_map *reservations = vma_resv_map(vma); |
1024 | ||
1025 | /* Mark this page used in the map. */ | |
1026 | region_add(&reservations->regions, idx, idx + 1); | |
c37f9fb1 AW |
1027 | } |
1028 | } | |
1029 | ||
a1e78772 | 1030 | static struct page *alloc_huge_page(struct vm_area_struct *vma, |
04f2cbe3 | 1031 | unsigned long addr, int avoid_reserve) |
1da177e4 | 1032 | { |
a5516438 | 1033 | struct hstate *h = hstate_vma(vma); |
348ea204 | 1034 | struct page *page; |
a1e78772 MG |
1035 | struct address_space *mapping = vma->vm_file->f_mapping; |
1036 | struct inode *inode = mapping->host; | |
e2f17d94 | 1037 | long chg; |
a1e78772 MG |
1038 | |
1039 | /* | |
1040 | * Processes that did not create the mapping will have no reserves and | |
1041 | * will not have accounted against quota. Check that the quota can be | |
1042 | * made before satisfying the allocation | |
c37f9fb1 AW |
1043 | * MAP_NORESERVE mappings may also need pages and quota allocated |
1044 | * if no reserve mapping overlaps. | |
a1e78772 | 1045 | */ |
a5516438 | 1046 | chg = vma_needs_reservation(h, vma, addr); |
c37f9fb1 | 1047 | if (chg < 0) |
e0dcd8a0 | 1048 | return ERR_PTR(-VM_FAULT_OOM); |
c37f9fb1 | 1049 | if (chg) |
a1e78772 | 1050 | if (hugetlb_get_quota(inode->i_mapping, chg)) |
e0dcd8a0 | 1051 | return ERR_PTR(-VM_FAULT_SIGBUS); |
1da177e4 LT |
1052 | |
1053 | spin_lock(&hugetlb_lock); | |
a5516438 | 1054 | page = dequeue_huge_page_vma(h, vma, addr, avoid_reserve); |
1da177e4 | 1055 | spin_unlock(&hugetlb_lock); |
b45b5bd6 | 1056 | |
68842c9b | 1057 | if (!page) { |
bf50bab2 | 1058 | page = alloc_buddy_huge_page(h, NUMA_NO_NODE); |
68842c9b | 1059 | if (!page) { |
a1e78772 | 1060 | hugetlb_put_quota(inode->i_mapping, chg); |
4a6018f7 | 1061 | return ERR_PTR(-VM_FAULT_SIGBUS); |
68842c9b KC |
1062 | } |
1063 | } | |
348ea204 | 1064 | |
a1e78772 | 1065 | set_page_private(page, (unsigned long) mapping); |
90d8b7e6 | 1066 | |
a5516438 | 1067 | vma_commit_reservation(h, vma, addr); |
c37f9fb1 | 1068 | |
90d8b7e6 | 1069 | return page; |
b45b5bd6 DG |
1070 | } |
1071 | ||
91f47662 | 1072 | int __weak alloc_bootmem_huge_page(struct hstate *h) |
aa888a74 AK |
1073 | { |
1074 | struct huge_bootmem_page *m; | |
9b5e5d0f | 1075 | int nr_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); |
aa888a74 AK |
1076 | |
1077 | while (nr_nodes) { | |
1078 | void *addr; | |
1079 | ||
1080 | addr = __alloc_bootmem_node_nopanic( | |
6ae11b27 | 1081 | NODE_DATA(hstate_next_node_to_alloc(h, |
9b5e5d0f | 1082 | &node_states[N_HIGH_MEMORY])), |
aa888a74 AK |
1083 | huge_page_size(h), huge_page_size(h), 0); |
1084 | ||
1085 | if (addr) { | |
1086 | /* | |
1087 | * Use the beginning of the huge page to store the | |
1088 | * huge_bootmem_page struct (until gather_bootmem | |
1089 | * puts them into the mem_map). | |
1090 | */ | |
1091 | m = addr; | |
91f47662 | 1092 | goto found; |
aa888a74 | 1093 | } |
aa888a74 AK |
1094 | nr_nodes--; |
1095 | } | |
1096 | return 0; | |
1097 | ||
1098 | found: | |
1099 | BUG_ON((unsigned long)virt_to_phys(m) & (huge_page_size(h) - 1)); | |
1100 | /* Put them into a private list first because mem_map is not up yet */ | |
1101 | list_add(&m->list, &huge_boot_pages); | |
1102 | m->hstate = h; | |
1103 | return 1; | |
1104 | } | |
1105 | ||
18229df5 AW |
1106 | static void prep_compound_huge_page(struct page *page, int order) |
1107 | { | |
1108 | if (unlikely(order > (MAX_ORDER - 1))) | |
1109 | prep_compound_gigantic_page(page, order); | |
1110 | else | |
1111 | prep_compound_page(page, order); | |
1112 | } | |
1113 | ||
aa888a74 AK |
1114 | /* Put bootmem huge pages into the standard lists after mem_map is up */ |
1115 | static void __init gather_bootmem_prealloc(void) | |
1116 | { | |
1117 | struct huge_bootmem_page *m; | |
1118 | ||
1119 | list_for_each_entry(m, &huge_boot_pages, list) { | |
aa888a74 | 1120 | struct hstate *h = m->hstate; |
ee8f248d BB |
1121 | struct page *page; |
1122 | ||
1123 | #ifdef CONFIG_HIGHMEM | |
1124 | page = pfn_to_page(m->phys >> PAGE_SHIFT); | |
1125 | free_bootmem_late((unsigned long)m, | |
1126 | sizeof(struct huge_bootmem_page)); | |
1127 | #else | |
1128 | page = virt_to_page(m); | |
1129 | #endif | |
aa888a74 AK |
1130 | __ClearPageReserved(page); |
1131 | WARN_ON(page_count(page) != 1); | |
18229df5 | 1132 | prep_compound_huge_page(page, h->order); |
aa888a74 | 1133 | prep_new_huge_page(h, page, page_to_nid(page)); |
b0320c7b RA |
1134 | /* |
1135 | * If we had gigantic hugepages allocated at boot time, we need | |
1136 | * to restore the 'stolen' pages to totalram_pages in order to | |
1137 | * fix confusing memory reports from free(1) and another | |
1138 | * side-effects, like CommitLimit going negative. | |
1139 | */ | |
1140 | if (h->order > (MAX_ORDER - 1)) | |
1141 | totalram_pages += 1 << h->order; | |
aa888a74 AK |
1142 | } |
1143 | } | |
1144 | ||
8faa8b07 | 1145 | static void __init hugetlb_hstate_alloc_pages(struct hstate *h) |
1da177e4 LT |
1146 | { |
1147 | unsigned long i; | |
a5516438 | 1148 | |
e5ff2159 | 1149 | for (i = 0; i < h->max_huge_pages; ++i) { |
aa888a74 AK |
1150 | if (h->order >= MAX_ORDER) { |
1151 | if (!alloc_bootmem_huge_page(h)) | |
1152 | break; | |
9b5e5d0f LS |
1153 | } else if (!alloc_fresh_huge_page(h, |
1154 | &node_states[N_HIGH_MEMORY])) | |
1da177e4 | 1155 | break; |
1da177e4 | 1156 | } |
8faa8b07 | 1157 | h->max_huge_pages = i; |
e5ff2159 AK |
1158 | } |
1159 | ||
1160 | static void __init hugetlb_init_hstates(void) | |
1161 | { | |
1162 | struct hstate *h; | |
1163 | ||
1164 | for_each_hstate(h) { | |
8faa8b07 AK |
1165 | /* oversize hugepages were init'ed in early boot */ |
1166 | if (h->order < MAX_ORDER) | |
1167 | hugetlb_hstate_alloc_pages(h); | |
e5ff2159 AK |
1168 | } |
1169 | } | |
1170 | ||
4abd32db AK |
1171 | static char * __init memfmt(char *buf, unsigned long n) |
1172 | { | |
1173 | if (n >= (1UL << 30)) | |
1174 | sprintf(buf, "%lu GB", n >> 30); | |
1175 | else if (n >= (1UL << 20)) | |
1176 | sprintf(buf, "%lu MB", n >> 20); | |
1177 | else | |
1178 | sprintf(buf, "%lu KB", n >> 10); | |
1179 | return buf; | |
1180 | } | |
1181 | ||
e5ff2159 AK |
1182 | static void __init report_hugepages(void) |
1183 | { | |
1184 | struct hstate *h; | |
1185 | ||
1186 | for_each_hstate(h) { | |
4abd32db AK |
1187 | char buf[32]; |
1188 | printk(KERN_INFO "HugeTLB registered %s page size, " | |
1189 | "pre-allocated %ld pages\n", | |
1190 | memfmt(buf, huge_page_size(h)), | |
1191 | h->free_huge_pages); | |
e5ff2159 AK |
1192 | } |
1193 | } | |
1194 | ||
1da177e4 | 1195 | #ifdef CONFIG_HIGHMEM |
6ae11b27 LS |
1196 | static void try_to_free_low(struct hstate *h, unsigned long count, |
1197 | nodemask_t *nodes_allowed) | |
1da177e4 | 1198 | { |
4415cc8d CL |
1199 | int i; |
1200 | ||
aa888a74 AK |
1201 | if (h->order >= MAX_ORDER) |
1202 | return; | |
1203 | ||
6ae11b27 | 1204 | for_each_node_mask(i, *nodes_allowed) { |
1da177e4 | 1205 | struct page *page, *next; |
a5516438 AK |
1206 | struct list_head *freel = &h->hugepage_freelists[i]; |
1207 | list_for_each_entry_safe(page, next, freel, lru) { | |
1208 | if (count >= h->nr_huge_pages) | |
6b0c880d | 1209 | return; |
1da177e4 LT |
1210 | if (PageHighMem(page)) |
1211 | continue; | |
1212 | list_del(&page->lru); | |
e5ff2159 | 1213 | update_and_free_page(h, page); |
a5516438 AK |
1214 | h->free_huge_pages--; |
1215 | h->free_huge_pages_node[page_to_nid(page)]--; | |
1da177e4 LT |
1216 | } |
1217 | } | |
1218 | } | |
1219 | #else | |
6ae11b27 LS |
1220 | static inline void try_to_free_low(struct hstate *h, unsigned long count, |
1221 | nodemask_t *nodes_allowed) | |
1da177e4 LT |
1222 | { |
1223 | } | |
1224 | #endif | |
1225 | ||
20a0307c WF |
1226 | /* |
1227 | * Increment or decrement surplus_huge_pages. Keep node-specific counters | |
1228 | * balanced by operating on them in a round-robin fashion. | |
1229 | * Returns 1 if an adjustment was made. | |
1230 | */ | |
6ae11b27 LS |
1231 | static int adjust_pool_surplus(struct hstate *h, nodemask_t *nodes_allowed, |
1232 | int delta) | |
20a0307c | 1233 | { |
e8c5c824 | 1234 | int start_nid, next_nid; |
20a0307c WF |
1235 | int ret = 0; |
1236 | ||
1237 | VM_BUG_ON(delta != -1 && delta != 1); | |
20a0307c | 1238 | |
e8c5c824 | 1239 | if (delta < 0) |
6ae11b27 | 1240 | start_nid = hstate_next_node_to_alloc(h, nodes_allowed); |
e8c5c824 | 1241 | else |
6ae11b27 | 1242 | start_nid = hstate_next_node_to_free(h, nodes_allowed); |
e8c5c824 LS |
1243 | next_nid = start_nid; |
1244 | ||
1245 | do { | |
1246 | int nid = next_nid; | |
1247 | if (delta < 0) { | |
e8c5c824 LS |
1248 | /* |
1249 | * To shrink on this node, there must be a surplus page | |
1250 | */ | |
9a76db09 | 1251 | if (!h->surplus_huge_pages_node[nid]) { |
6ae11b27 LS |
1252 | next_nid = hstate_next_node_to_alloc(h, |
1253 | nodes_allowed); | |
e8c5c824 | 1254 | continue; |
9a76db09 | 1255 | } |
e8c5c824 LS |
1256 | } |
1257 | if (delta > 0) { | |
e8c5c824 LS |
1258 | /* |
1259 | * Surplus cannot exceed the total number of pages | |
1260 | */ | |
1261 | if (h->surplus_huge_pages_node[nid] >= | |
9a76db09 | 1262 | h->nr_huge_pages_node[nid]) { |
6ae11b27 LS |
1263 | next_nid = hstate_next_node_to_free(h, |
1264 | nodes_allowed); | |
e8c5c824 | 1265 | continue; |
9a76db09 | 1266 | } |
e8c5c824 | 1267 | } |
20a0307c WF |
1268 | |
1269 | h->surplus_huge_pages += delta; | |
1270 | h->surplus_huge_pages_node[nid] += delta; | |
1271 | ret = 1; | |
1272 | break; | |
e8c5c824 | 1273 | } while (next_nid != start_nid); |
20a0307c | 1274 | |
20a0307c WF |
1275 | return ret; |
1276 | } | |
1277 | ||
a5516438 | 1278 | #define persistent_huge_pages(h) (h->nr_huge_pages - h->surplus_huge_pages) |
6ae11b27 LS |
1279 | static unsigned long set_max_huge_pages(struct hstate *h, unsigned long count, |
1280 | nodemask_t *nodes_allowed) | |
1da177e4 | 1281 | { |
7893d1d5 | 1282 | unsigned long min_count, ret; |
1da177e4 | 1283 | |
aa888a74 AK |
1284 | if (h->order >= MAX_ORDER) |
1285 | return h->max_huge_pages; | |
1286 | ||
7893d1d5 AL |
1287 | /* |
1288 | * Increase the pool size | |
1289 | * First take pages out of surplus state. Then make up the | |
1290 | * remaining difference by allocating fresh huge pages. | |
d1c3fb1f NA |
1291 | * |
1292 | * We might race with alloc_buddy_huge_page() here and be unable | |
1293 | * to convert a surplus huge page to a normal huge page. That is | |
1294 | * not critical, though, it just means the overall size of the | |
1295 | * pool might be one hugepage larger than it needs to be, but | |
1296 | * within all the constraints specified by the sysctls. | |
7893d1d5 | 1297 | */ |
1da177e4 | 1298 | spin_lock(&hugetlb_lock); |
a5516438 | 1299 | while (h->surplus_huge_pages && count > persistent_huge_pages(h)) { |
6ae11b27 | 1300 | if (!adjust_pool_surplus(h, nodes_allowed, -1)) |
7893d1d5 AL |
1301 | break; |
1302 | } | |
1303 | ||
a5516438 | 1304 | while (count > persistent_huge_pages(h)) { |
7893d1d5 AL |
1305 | /* |
1306 | * If this allocation races such that we no longer need the | |
1307 | * page, free_huge_page will handle it by freeing the page | |
1308 | * and reducing the surplus. | |
1309 | */ | |
1310 | spin_unlock(&hugetlb_lock); | |
6ae11b27 | 1311 | ret = alloc_fresh_huge_page(h, nodes_allowed); |
7893d1d5 AL |
1312 | spin_lock(&hugetlb_lock); |
1313 | if (!ret) | |
1314 | goto out; | |
1315 | ||
536240f2 MG |
1316 | /* Bail for signals. Probably ctrl-c from user */ |
1317 | if (signal_pending(current)) | |
1318 | goto out; | |
7893d1d5 | 1319 | } |
7893d1d5 AL |
1320 | |
1321 | /* | |
1322 | * Decrease the pool size | |
1323 | * First return free pages to the buddy allocator (being careful | |
1324 | * to keep enough around to satisfy reservations). Then place | |
1325 | * pages into surplus state as needed so the pool will shrink | |
1326 | * to the desired size as pages become free. | |
d1c3fb1f NA |
1327 | * |
1328 | * By placing pages into the surplus state independent of the | |
1329 | * overcommit value, we are allowing the surplus pool size to | |
1330 | * exceed overcommit. There are few sane options here. Since | |
1331 | * alloc_buddy_huge_page() is checking the global counter, | |
1332 | * though, we'll note that we're not allowed to exceed surplus | |
1333 | * and won't grow the pool anywhere else. Not until one of the | |
1334 | * sysctls are changed, or the surplus pages go out of use. | |
7893d1d5 | 1335 | */ |
a5516438 | 1336 | min_count = h->resv_huge_pages + h->nr_huge_pages - h->free_huge_pages; |
6b0c880d | 1337 | min_count = max(count, min_count); |
6ae11b27 | 1338 | try_to_free_low(h, min_count, nodes_allowed); |
a5516438 | 1339 | while (min_count < persistent_huge_pages(h)) { |
6ae11b27 | 1340 | if (!free_pool_huge_page(h, nodes_allowed, 0)) |
1da177e4 | 1341 | break; |
1da177e4 | 1342 | } |
a5516438 | 1343 | while (count < persistent_huge_pages(h)) { |
6ae11b27 | 1344 | if (!adjust_pool_surplus(h, nodes_allowed, 1)) |
7893d1d5 AL |
1345 | break; |
1346 | } | |
1347 | out: | |
a5516438 | 1348 | ret = persistent_huge_pages(h); |
1da177e4 | 1349 | spin_unlock(&hugetlb_lock); |
7893d1d5 | 1350 | return ret; |
1da177e4 LT |
1351 | } |
1352 | ||
a3437870 NA |
1353 | #define HSTATE_ATTR_RO(_name) \ |
1354 | static struct kobj_attribute _name##_attr = __ATTR_RO(_name) | |
1355 | ||
1356 | #define HSTATE_ATTR(_name) \ | |
1357 | static struct kobj_attribute _name##_attr = \ | |
1358 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
1359 | ||
1360 | static struct kobject *hugepages_kobj; | |
1361 | static struct kobject *hstate_kobjs[HUGE_MAX_HSTATE]; | |
1362 | ||
9a305230 LS |
1363 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp); |
1364 | ||
1365 | static struct hstate *kobj_to_hstate(struct kobject *kobj, int *nidp) | |
a3437870 NA |
1366 | { |
1367 | int i; | |
9a305230 | 1368 | |
a3437870 | 1369 | for (i = 0; i < HUGE_MAX_HSTATE; i++) |
9a305230 LS |
1370 | if (hstate_kobjs[i] == kobj) { |
1371 | if (nidp) | |
1372 | *nidp = NUMA_NO_NODE; | |
a3437870 | 1373 | return &hstates[i]; |
9a305230 LS |
1374 | } |
1375 | ||
1376 | return kobj_to_node_hstate(kobj, nidp); | |
a3437870 NA |
1377 | } |
1378 | ||
06808b08 | 1379 | static ssize_t nr_hugepages_show_common(struct kobject *kobj, |
a3437870 NA |
1380 | struct kobj_attribute *attr, char *buf) |
1381 | { | |
9a305230 LS |
1382 | struct hstate *h; |
1383 | unsigned long nr_huge_pages; | |
1384 | int nid; | |
1385 | ||
1386 | h = kobj_to_hstate(kobj, &nid); | |
1387 | if (nid == NUMA_NO_NODE) | |
1388 | nr_huge_pages = h->nr_huge_pages; | |
1389 | else | |
1390 | nr_huge_pages = h->nr_huge_pages_node[nid]; | |
1391 | ||
1392 | return sprintf(buf, "%lu\n", nr_huge_pages); | |
a3437870 | 1393 | } |
adbe8726 | 1394 | |
06808b08 LS |
1395 | static ssize_t nr_hugepages_store_common(bool obey_mempolicy, |
1396 | struct kobject *kobj, struct kobj_attribute *attr, | |
1397 | const char *buf, size_t len) | |
a3437870 NA |
1398 | { |
1399 | int err; | |
9a305230 | 1400 | int nid; |
06808b08 | 1401 | unsigned long count; |
9a305230 | 1402 | struct hstate *h; |
bad44b5b | 1403 | NODEMASK_ALLOC(nodemask_t, nodes_allowed, GFP_KERNEL | __GFP_NORETRY); |
a3437870 | 1404 | |
06808b08 | 1405 | err = strict_strtoul(buf, 10, &count); |
73ae31e5 | 1406 | if (err) |
adbe8726 | 1407 | goto out; |
a3437870 | 1408 | |
9a305230 | 1409 | h = kobj_to_hstate(kobj, &nid); |
adbe8726 EM |
1410 | if (h->order >= MAX_ORDER) { |
1411 | err = -EINVAL; | |
1412 | goto out; | |
1413 | } | |
1414 | ||
9a305230 LS |
1415 | if (nid == NUMA_NO_NODE) { |
1416 | /* | |
1417 | * global hstate attribute | |
1418 | */ | |
1419 | if (!(obey_mempolicy && | |
1420 | init_nodemask_of_mempolicy(nodes_allowed))) { | |
1421 | NODEMASK_FREE(nodes_allowed); | |
1422 | nodes_allowed = &node_states[N_HIGH_MEMORY]; | |
1423 | } | |
1424 | } else if (nodes_allowed) { | |
1425 | /* | |
1426 | * per node hstate attribute: adjust count to global, | |
1427 | * but restrict alloc/free to the specified node. | |
1428 | */ | |
1429 | count += h->nr_huge_pages - h->nr_huge_pages_node[nid]; | |
1430 | init_nodemask_of_node(nodes_allowed, nid); | |
1431 | } else | |
1432 | nodes_allowed = &node_states[N_HIGH_MEMORY]; | |
1433 | ||
06808b08 | 1434 | h->max_huge_pages = set_max_huge_pages(h, count, nodes_allowed); |
a3437870 | 1435 | |
9b5e5d0f | 1436 | if (nodes_allowed != &node_states[N_HIGH_MEMORY]) |
06808b08 LS |
1437 | NODEMASK_FREE(nodes_allowed); |
1438 | ||
1439 | return len; | |
adbe8726 EM |
1440 | out: |
1441 | NODEMASK_FREE(nodes_allowed); | |
1442 | return err; | |
06808b08 LS |
1443 | } |
1444 | ||
1445 | static ssize_t nr_hugepages_show(struct kobject *kobj, | |
1446 | struct kobj_attribute *attr, char *buf) | |
1447 | { | |
1448 | return nr_hugepages_show_common(kobj, attr, buf); | |
1449 | } | |
1450 | ||
1451 | static ssize_t nr_hugepages_store(struct kobject *kobj, | |
1452 | struct kobj_attribute *attr, const char *buf, size_t len) | |
1453 | { | |
1454 | return nr_hugepages_store_common(false, kobj, attr, buf, len); | |
a3437870 NA |
1455 | } |
1456 | HSTATE_ATTR(nr_hugepages); | |
1457 | ||
06808b08 LS |
1458 | #ifdef CONFIG_NUMA |
1459 | ||
1460 | /* | |
1461 | * hstate attribute for optionally mempolicy-based constraint on persistent | |
1462 | * huge page alloc/free. | |
1463 | */ | |
1464 | static ssize_t nr_hugepages_mempolicy_show(struct kobject *kobj, | |
1465 | struct kobj_attribute *attr, char *buf) | |
1466 | { | |
1467 | return nr_hugepages_show_common(kobj, attr, buf); | |
1468 | } | |
1469 | ||
1470 | static ssize_t nr_hugepages_mempolicy_store(struct kobject *kobj, | |
1471 | struct kobj_attribute *attr, const char *buf, size_t len) | |
1472 | { | |
1473 | return nr_hugepages_store_common(true, kobj, attr, buf, len); | |
1474 | } | |
1475 | HSTATE_ATTR(nr_hugepages_mempolicy); | |
1476 | #endif | |
1477 | ||
1478 | ||
a3437870 NA |
1479 | static ssize_t nr_overcommit_hugepages_show(struct kobject *kobj, |
1480 | struct kobj_attribute *attr, char *buf) | |
1481 | { | |
9a305230 | 1482 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
a3437870 NA |
1483 | return sprintf(buf, "%lu\n", h->nr_overcommit_huge_pages); |
1484 | } | |
adbe8726 | 1485 | |
a3437870 NA |
1486 | static ssize_t nr_overcommit_hugepages_store(struct kobject *kobj, |
1487 | struct kobj_attribute *attr, const char *buf, size_t count) | |
1488 | { | |
1489 | int err; | |
1490 | unsigned long input; | |
9a305230 | 1491 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
a3437870 | 1492 | |
adbe8726 EM |
1493 | if (h->order >= MAX_ORDER) |
1494 | return -EINVAL; | |
1495 | ||
a3437870 NA |
1496 | err = strict_strtoul(buf, 10, &input); |
1497 | if (err) | |
73ae31e5 | 1498 | return err; |
a3437870 NA |
1499 | |
1500 | spin_lock(&hugetlb_lock); | |
1501 | h->nr_overcommit_huge_pages = input; | |
1502 | spin_unlock(&hugetlb_lock); | |
1503 | ||
1504 | return count; | |
1505 | } | |
1506 | HSTATE_ATTR(nr_overcommit_hugepages); | |
1507 | ||
1508 | static ssize_t free_hugepages_show(struct kobject *kobj, | |
1509 | struct kobj_attribute *attr, char *buf) | |
1510 | { | |
9a305230 LS |
1511 | struct hstate *h; |
1512 | unsigned long free_huge_pages; | |
1513 | int nid; | |
1514 | ||
1515 | h = kobj_to_hstate(kobj, &nid); | |
1516 | if (nid == NUMA_NO_NODE) | |
1517 | free_huge_pages = h->free_huge_pages; | |
1518 | else | |
1519 | free_huge_pages = h->free_huge_pages_node[nid]; | |
1520 | ||
1521 | return sprintf(buf, "%lu\n", free_huge_pages); | |
a3437870 NA |
1522 | } |
1523 | HSTATE_ATTR_RO(free_hugepages); | |
1524 | ||
1525 | static ssize_t resv_hugepages_show(struct kobject *kobj, | |
1526 | struct kobj_attribute *attr, char *buf) | |
1527 | { | |
9a305230 | 1528 | struct hstate *h = kobj_to_hstate(kobj, NULL); |
a3437870 NA |
1529 | return sprintf(buf, "%lu\n", h->resv_huge_pages); |
1530 | } | |
1531 | HSTATE_ATTR_RO(resv_hugepages); | |
1532 | ||
1533 | static ssize_t surplus_hugepages_show(struct kobject *kobj, | |
1534 | struct kobj_attribute *attr, char *buf) | |
1535 | { | |
9a305230 LS |
1536 | struct hstate *h; |
1537 | unsigned long surplus_huge_pages; | |
1538 | int nid; | |
1539 | ||
1540 | h = kobj_to_hstate(kobj, &nid); | |
1541 | if (nid == NUMA_NO_NODE) | |
1542 | surplus_huge_pages = h->surplus_huge_pages; | |
1543 | else | |
1544 | surplus_huge_pages = h->surplus_huge_pages_node[nid]; | |
1545 | ||
1546 | return sprintf(buf, "%lu\n", surplus_huge_pages); | |
a3437870 NA |
1547 | } |
1548 | HSTATE_ATTR_RO(surplus_hugepages); | |
1549 | ||
1550 | static struct attribute *hstate_attrs[] = { | |
1551 | &nr_hugepages_attr.attr, | |
1552 | &nr_overcommit_hugepages_attr.attr, | |
1553 | &free_hugepages_attr.attr, | |
1554 | &resv_hugepages_attr.attr, | |
1555 | &surplus_hugepages_attr.attr, | |
06808b08 LS |
1556 | #ifdef CONFIG_NUMA |
1557 | &nr_hugepages_mempolicy_attr.attr, | |
1558 | #endif | |
a3437870 NA |
1559 | NULL, |
1560 | }; | |
1561 | ||
1562 | static struct attribute_group hstate_attr_group = { | |
1563 | .attrs = hstate_attrs, | |
1564 | }; | |
1565 | ||
094e9539 JM |
1566 | static int hugetlb_sysfs_add_hstate(struct hstate *h, struct kobject *parent, |
1567 | struct kobject **hstate_kobjs, | |
1568 | struct attribute_group *hstate_attr_group) | |
a3437870 NA |
1569 | { |
1570 | int retval; | |
9a305230 | 1571 | int hi = h - hstates; |
a3437870 | 1572 | |
9a305230 LS |
1573 | hstate_kobjs[hi] = kobject_create_and_add(h->name, parent); |
1574 | if (!hstate_kobjs[hi]) | |
a3437870 NA |
1575 | return -ENOMEM; |
1576 | ||
9a305230 | 1577 | retval = sysfs_create_group(hstate_kobjs[hi], hstate_attr_group); |
a3437870 | 1578 | if (retval) |
9a305230 | 1579 | kobject_put(hstate_kobjs[hi]); |
a3437870 NA |
1580 | |
1581 | return retval; | |
1582 | } | |
1583 | ||
1584 | static void __init hugetlb_sysfs_init(void) | |
1585 | { | |
1586 | struct hstate *h; | |
1587 | int err; | |
1588 | ||
1589 | hugepages_kobj = kobject_create_and_add("hugepages", mm_kobj); | |
1590 | if (!hugepages_kobj) | |
1591 | return; | |
1592 | ||
1593 | for_each_hstate(h) { | |
9a305230 LS |
1594 | err = hugetlb_sysfs_add_hstate(h, hugepages_kobj, |
1595 | hstate_kobjs, &hstate_attr_group); | |
a3437870 NA |
1596 | if (err) |
1597 | printk(KERN_ERR "Hugetlb: Unable to add hstate %s", | |
1598 | h->name); | |
1599 | } | |
1600 | } | |
1601 | ||
9a305230 LS |
1602 | #ifdef CONFIG_NUMA |
1603 | ||
1604 | /* | |
1605 | * node_hstate/s - associate per node hstate attributes, via their kobjects, | |
10fbcf4c KS |
1606 | * with node devices in node_devices[] using a parallel array. The array |
1607 | * index of a node device or _hstate == node id. | |
1608 | * This is here to avoid any static dependency of the node device driver, in | |
9a305230 LS |
1609 | * the base kernel, on the hugetlb module. |
1610 | */ | |
1611 | struct node_hstate { | |
1612 | struct kobject *hugepages_kobj; | |
1613 | struct kobject *hstate_kobjs[HUGE_MAX_HSTATE]; | |
1614 | }; | |
1615 | struct node_hstate node_hstates[MAX_NUMNODES]; | |
1616 | ||
1617 | /* | |
10fbcf4c | 1618 | * A subset of global hstate attributes for node devices |
9a305230 LS |
1619 | */ |
1620 | static struct attribute *per_node_hstate_attrs[] = { | |
1621 | &nr_hugepages_attr.attr, | |
1622 | &free_hugepages_attr.attr, | |
1623 | &surplus_hugepages_attr.attr, | |
1624 | NULL, | |
1625 | }; | |
1626 | ||
1627 | static struct attribute_group per_node_hstate_attr_group = { | |
1628 | .attrs = per_node_hstate_attrs, | |
1629 | }; | |
1630 | ||
1631 | /* | |
10fbcf4c | 1632 | * kobj_to_node_hstate - lookup global hstate for node device hstate attr kobj. |
9a305230 LS |
1633 | * Returns node id via non-NULL nidp. |
1634 | */ | |
1635 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp) | |
1636 | { | |
1637 | int nid; | |
1638 | ||
1639 | for (nid = 0; nid < nr_node_ids; nid++) { | |
1640 | struct node_hstate *nhs = &node_hstates[nid]; | |
1641 | int i; | |
1642 | for (i = 0; i < HUGE_MAX_HSTATE; i++) | |
1643 | if (nhs->hstate_kobjs[i] == kobj) { | |
1644 | if (nidp) | |
1645 | *nidp = nid; | |
1646 | return &hstates[i]; | |
1647 | } | |
1648 | } | |
1649 | ||
1650 | BUG(); | |
1651 | return NULL; | |
1652 | } | |
1653 | ||
1654 | /* | |
10fbcf4c | 1655 | * Unregister hstate attributes from a single node device. |
9a305230 LS |
1656 | * No-op if no hstate attributes attached. |
1657 | */ | |
1658 | void hugetlb_unregister_node(struct node *node) | |
1659 | { | |
1660 | struct hstate *h; | |
10fbcf4c | 1661 | struct node_hstate *nhs = &node_hstates[node->dev.id]; |
9a305230 LS |
1662 | |
1663 | if (!nhs->hugepages_kobj) | |
9b5e5d0f | 1664 | return; /* no hstate attributes */ |
9a305230 LS |
1665 | |
1666 | for_each_hstate(h) | |
1667 | if (nhs->hstate_kobjs[h - hstates]) { | |
1668 | kobject_put(nhs->hstate_kobjs[h - hstates]); | |
1669 | nhs->hstate_kobjs[h - hstates] = NULL; | |
1670 | } | |
1671 | ||
1672 | kobject_put(nhs->hugepages_kobj); | |
1673 | nhs->hugepages_kobj = NULL; | |
1674 | } | |
1675 | ||
1676 | /* | |
10fbcf4c | 1677 | * hugetlb module exit: unregister hstate attributes from node devices |
9a305230 LS |
1678 | * that have them. |
1679 | */ | |
1680 | static void hugetlb_unregister_all_nodes(void) | |
1681 | { | |
1682 | int nid; | |
1683 | ||
1684 | /* | |
10fbcf4c | 1685 | * disable node device registrations. |
9a305230 LS |
1686 | */ |
1687 | register_hugetlbfs_with_node(NULL, NULL); | |
1688 | ||
1689 | /* | |
1690 | * remove hstate attributes from any nodes that have them. | |
1691 | */ | |
1692 | for (nid = 0; nid < nr_node_ids; nid++) | |
1693 | hugetlb_unregister_node(&node_devices[nid]); | |
1694 | } | |
1695 | ||
1696 | /* | |
10fbcf4c | 1697 | * Register hstate attributes for a single node device. |
9a305230 LS |
1698 | * No-op if attributes already registered. |
1699 | */ | |
1700 | void hugetlb_register_node(struct node *node) | |
1701 | { | |
1702 | struct hstate *h; | |
10fbcf4c | 1703 | struct node_hstate *nhs = &node_hstates[node->dev.id]; |
9a305230 LS |
1704 | int err; |
1705 | ||
1706 | if (nhs->hugepages_kobj) | |
1707 | return; /* already allocated */ | |
1708 | ||
1709 | nhs->hugepages_kobj = kobject_create_and_add("hugepages", | |
10fbcf4c | 1710 | &node->dev.kobj); |
9a305230 LS |
1711 | if (!nhs->hugepages_kobj) |
1712 | return; | |
1713 | ||
1714 | for_each_hstate(h) { | |
1715 | err = hugetlb_sysfs_add_hstate(h, nhs->hugepages_kobj, | |
1716 | nhs->hstate_kobjs, | |
1717 | &per_node_hstate_attr_group); | |
1718 | if (err) { | |
1719 | printk(KERN_ERR "Hugetlb: Unable to add hstate %s" | |
1720 | " for node %d\n", | |
10fbcf4c | 1721 | h->name, node->dev.id); |
9a305230 LS |
1722 | hugetlb_unregister_node(node); |
1723 | break; | |
1724 | } | |
1725 | } | |
1726 | } | |
1727 | ||
1728 | /* | |
9b5e5d0f | 1729 | * hugetlb init time: register hstate attributes for all registered node |
10fbcf4c KS |
1730 | * devices of nodes that have memory. All on-line nodes should have |
1731 | * registered their associated device by this time. | |
9a305230 LS |
1732 | */ |
1733 | static void hugetlb_register_all_nodes(void) | |
1734 | { | |
1735 | int nid; | |
1736 | ||
9b5e5d0f | 1737 | for_each_node_state(nid, N_HIGH_MEMORY) { |
9a305230 | 1738 | struct node *node = &node_devices[nid]; |
10fbcf4c | 1739 | if (node->dev.id == nid) |
9a305230 LS |
1740 | hugetlb_register_node(node); |
1741 | } | |
1742 | ||
1743 | /* | |
10fbcf4c | 1744 | * Let the node device driver know we're here so it can |
9a305230 LS |
1745 | * [un]register hstate attributes on node hotplug. |
1746 | */ | |
1747 | register_hugetlbfs_with_node(hugetlb_register_node, | |
1748 | hugetlb_unregister_node); | |
1749 | } | |
1750 | #else /* !CONFIG_NUMA */ | |
1751 | ||
1752 | static struct hstate *kobj_to_node_hstate(struct kobject *kobj, int *nidp) | |
1753 | { | |
1754 | BUG(); | |
1755 | if (nidp) | |
1756 | *nidp = -1; | |
1757 | return NULL; | |
1758 | } | |
1759 | ||
1760 | static void hugetlb_unregister_all_nodes(void) { } | |
1761 | ||
1762 | static void hugetlb_register_all_nodes(void) { } | |
1763 | ||
1764 | #endif | |
1765 | ||
a3437870 NA |
1766 | static void __exit hugetlb_exit(void) |
1767 | { | |
1768 | struct hstate *h; | |
1769 | ||
9a305230 LS |
1770 | hugetlb_unregister_all_nodes(); |
1771 | ||
a3437870 NA |
1772 | for_each_hstate(h) { |
1773 | kobject_put(hstate_kobjs[h - hstates]); | |
1774 | } | |
1775 | ||
1776 | kobject_put(hugepages_kobj); | |
1777 | } | |
1778 | module_exit(hugetlb_exit); | |
1779 | ||
1780 | static int __init hugetlb_init(void) | |
1781 | { | |
0ef89d25 BH |
1782 | /* Some platform decide whether they support huge pages at boot |
1783 | * time. On these, such as powerpc, HPAGE_SHIFT is set to 0 when | |
1784 | * there is no such support | |
1785 | */ | |
1786 | if (HPAGE_SHIFT == 0) | |
1787 | return 0; | |
a3437870 | 1788 | |
e11bfbfc NP |
1789 | if (!size_to_hstate(default_hstate_size)) { |
1790 | default_hstate_size = HPAGE_SIZE; | |
1791 | if (!size_to_hstate(default_hstate_size)) | |
1792 | hugetlb_add_hstate(HUGETLB_PAGE_ORDER); | |
a3437870 | 1793 | } |
e11bfbfc NP |
1794 | default_hstate_idx = size_to_hstate(default_hstate_size) - hstates; |
1795 | if (default_hstate_max_huge_pages) | |
1796 | default_hstate.max_huge_pages = default_hstate_max_huge_pages; | |
a3437870 NA |
1797 | |
1798 | hugetlb_init_hstates(); | |
1799 | ||
aa888a74 AK |
1800 | gather_bootmem_prealloc(); |
1801 | ||
a3437870 NA |
1802 | report_hugepages(); |
1803 | ||
1804 | hugetlb_sysfs_init(); | |
1805 | ||
9a305230 LS |
1806 | hugetlb_register_all_nodes(); |
1807 | ||
a3437870 NA |
1808 | return 0; |
1809 | } | |
1810 | module_init(hugetlb_init); | |
1811 | ||
1812 | /* Should be called on processing a hugepagesz=... option */ | |
1813 | void __init hugetlb_add_hstate(unsigned order) | |
1814 | { | |
1815 | struct hstate *h; | |
8faa8b07 AK |
1816 | unsigned long i; |
1817 | ||
a3437870 NA |
1818 | if (size_to_hstate(PAGE_SIZE << order)) { |
1819 | printk(KERN_WARNING "hugepagesz= specified twice, ignoring\n"); | |
1820 | return; | |
1821 | } | |
1822 | BUG_ON(max_hstate >= HUGE_MAX_HSTATE); | |
1823 | BUG_ON(order == 0); | |
1824 | h = &hstates[max_hstate++]; | |
1825 | h->order = order; | |
1826 | h->mask = ~((1ULL << (order + PAGE_SHIFT)) - 1); | |
8faa8b07 AK |
1827 | h->nr_huge_pages = 0; |
1828 | h->free_huge_pages = 0; | |
1829 | for (i = 0; i < MAX_NUMNODES; ++i) | |
1830 | INIT_LIST_HEAD(&h->hugepage_freelists[i]); | |
9b5e5d0f LS |
1831 | h->next_nid_to_alloc = first_node(node_states[N_HIGH_MEMORY]); |
1832 | h->next_nid_to_free = first_node(node_states[N_HIGH_MEMORY]); | |
a3437870 NA |
1833 | snprintf(h->name, HSTATE_NAME_LEN, "hugepages-%lukB", |
1834 | huge_page_size(h)/1024); | |
8faa8b07 | 1835 | |
a3437870 NA |
1836 | parsed_hstate = h; |
1837 | } | |
1838 | ||
e11bfbfc | 1839 | static int __init hugetlb_nrpages_setup(char *s) |
a3437870 NA |
1840 | { |
1841 | unsigned long *mhp; | |
8faa8b07 | 1842 | static unsigned long *last_mhp; |
a3437870 NA |
1843 | |
1844 | /* | |
1845 | * !max_hstate means we haven't parsed a hugepagesz= parameter yet, | |
1846 | * so this hugepages= parameter goes to the "default hstate". | |
1847 | */ | |
1848 | if (!max_hstate) | |
1849 | mhp = &default_hstate_max_huge_pages; | |
1850 | else | |
1851 | mhp = &parsed_hstate->max_huge_pages; | |
1852 | ||
8faa8b07 AK |
1853 | if (mhp == last_mhp) { |
1854 | printk(KERN_WARNING "hugepages= specified twice without " | |
1855 | "interleaving hugepagesz=, ignoring\n"); | |
1856 | return 1; | |
1857 | } | |
1858 | ||
a3437870 NA |
1859 | if (sscanf(s, "%lu", mhp) <= 0) |
1860 | *mhp = 0; | |
1861 | ||
8faa8b07 AK |
1862 | /* |
1863 | * Global state is always initialized later in hugetlb_init. | |
1864 | * But we need to allocate >= MAX_ORDER hstates here early to still | |
1865 | * use the bootmem allocator. | |
1866 | */ | |
1867 | if (max_hstate && parsed_hstate->order >= MAX_ORDER) | |
1868 | hugetlb_hstate_alloc_pages(parsed_hstate); | |
1869 | ||
1870 | last_mhp = mhp; | |
1871 | ||
a3437870 NA |
1872 | return 1; |
1873 | } | |
e11bfbfc NP |
1874 | __setup("hugepages=", hugetlb_nrpages_setup); |
1875 | ||
1876 | static int __init hugetlb_default_setup(char *s) | |
1877 | { | |
1878 | default_hstate_size = memparse(s, &s); | |
1879 | return 1; | |
1880 | } | |
1881 | __setup("default_hugepagesz=", hugetlb_default_setup); | |
a3437870 | 1882 | |
8a213460 NA |
1883 | static unsigned int cpuset_mems_nr(unsigned int *array) |
1884 | { | |
1885 | int node; | |
1886 | unsigned int nr = 0; | |
1887 | ||
1888 | for_each_node_mask(node, cpuset_current_mems_allowed) | |
1889 | nr += array[node]; | |
1890 | ||
1891 | return nr; | |
1892 | } | |
1893 | ||
1894 | #ifdef CONFIG_SYSCTL | |
06808b08 LS |
1895 | static int hugetlb_sysctl_handler_common(bool obey_mempolicy, |
1896 | struct ctl_table *table, int write, | |
1897 | void __user *buffer, size_t *length, loff_t *ppos) | |
1da177e4 | 1898 | { |
e5ff2159 AK |
1899 | struct hstate *h = &default_hstate; |
1900 | unsigned long tmp; | |
08d4a246 | 1901 | int ret; |
e5ff2159 | 1902 | |
c033a93c | 1903 | tmp = h->max_huge_pages; |
e5ff2159 | 1904 | |
adbe8726 EM |
1905 | if (write && h->order >= MAX_ORDER) |
1906 | return -EINVAL; | |
1907 | ||
e5ff2159 AK |
1908 | table->data = &tmp; |
1909 | table->maxlen = sizeof(unsigned long); | |
08d4a246 MH |
1910 | ret = proc_doulongvec_minmax(table, write, buffer, length, ppos); |
1911 | if (ret) | |
1912 | goto out; | |
e5ff2159 | 1913 | |
06808b08 | 1914 | if (write) { |
bad44b5b DR |
1915 | NODEMASK_ALLOC(nodemask_t, nodes_allowed, |
1916 | GFP_KERNEL | __GFP_NORETRY); | |
06808b08 LS |
1917 | if (!(obey_mempolicy && |
1918 | init_nodemask_of_mempolicy(nodes_allowed))) { | |
1919 | NODEMASK_FREE(nodes_allowed); | |
1920 | nodes_allowed = &node_states[N_HIGH_MEMORY]; | |
1921 | } | |
1922 | h->max_huge_pages = set_max_huge_pages(h, tmp, nodes_allowed); | |
1923 | ||
1924 | if (nodes_allowed != &node_states[N_HIGH_MEMORY]) | |
1925 | NODEMASK_FREE(nodes_allowed); | |
1926 | } | |
08d4a246 MH |
1927 | out: |
1928 | return ret; | |
1da177e4 | 1929 | } |
396faf03 | 1930 | |
06808b08 LS |
1931 | int hugetlb_sysctl_handler(struct ctl_table *table, int write, |
1932 | void __user *buffer, size_t *length, loff_t *ppos) | |
1933 | { | |
1934 | ||
1935 | return hugetlb_sysctl_handler_common(false, table, write, | |
1936 | buffer, length, ppos); | |
1937 | } | |
1938 | ||
1939 | #ifdef CONFIG_NUMA | |
1940 | int hugetlb_mempolicy_sysctl_handler(struct ctl_table *table, int write, | |
1941 | void __user *buffer, size_t *length, loff_t *ppos) | |
1942 | { | |
1943 | return hugetlb_sysctl_handler_common(true, table, write, | |
1944 | buffer, length, ppos); | |
1945 | } | |
1946 | #endif /* CONFIG_NUMA */ | |
1947 | ||
396faf03 | 1948 | int hugetlb_treat_movable_handler(struct ctl_table *table, int write, |
8d65af78 | 1949 | void __user *buffer, |
396faf03 MG |
1950 | size_t *length, loff_t *ppos) |
1951 | { | |
8d65af78 | 1952 | proc_dointvec(table, write, buffer, length, ppos); |
396faf03 MG |
1953 | if (hugepages_treat_as_movable) |
1954 | htlb_alloc_mask = GFP_HIGHUSER_MOVABLE; | |
1955 | else | |
1956 | htlb_alloc_mask = GFP_HIGHUSER; | |
1957 | return 0; | |
1958 | } | |
1959 | ||
a3d0c6aa | 1960 | int hugetlb_overcommit_handler(struct ctl_table *table, int write, |
8d65af78 | 1961 | void __user *buffer, |
a3d0c6aa NA |
1962 | size_t *length, loff_t *ppos) |
1963 | { | |
a5516438 | 1964 | struct hstate *h = &default_hstate; |
e5ff2159 | 1965 | unsigned long tmp; |
08d4a246 | 1966 | int ret; |
e5ff2159 | 1967 | |
c033a93c | 1968 | tmp = h->nr_overcommit_huge_pages; |
e5ff2159 | 1969 | |
adbe8726 EM |
1970 | if (write && h->order >= MAX_ORDER) |
1971 | return -EINVAL; | |
1972 | ||
e5ff2159 AK |
1973 | table->data = &tmp; |
1974 | table->maxlen = sizeof(unsigned long); | |
08d4a246 MH |
1975 | ret = proc_doulongvec_minmax(table, write, buffer, length, ppos); |
1976 | if (ret) | |
1977 | goto out; | |
e5ff2159 AK |
1978 | |
1979 | if (write) { | |
1980 | spin_lock(&hugetlb_lock); | |
1981 | h->nr_overcommit_huge_pages = tmp; | |
1982 | spin_unlock(&hugetlb_lock); | |
1983 | } | |
08d4a246 MH |
1984 | out: |
1985 | return ret; | |
a3d0c6aa NA |
1986 | } |
1987 | ||
1da177e4 LT |
1988 | #endif /* CONFIG_SYSCTL */ |
1989 | ||
e1759c21 | 1990 | void hugetlb_report_meminfo(struct seq_file *m) |
1da177e4 | 1991 | { |
a5516438 | 1992 | struct hstate *h = &default_hstate; |
e1759c21 | 1993 | seq_printf(m, |
4f98a2fe RR |
1994 | "HugePages_Total: %5lu\n" |
1995 | "HugePages_Free: %5lu\n" | |
1996 | "HugePages_Rsvd: %5lu\n" | |
1997 | "HugePages_Surp: %5lu\n" | |
1998 | "Hugepagesize: %8lu kB\n", | |
a5516438 AK |
1999 | h->nr_huge_pages, |
2000 | h->free_huge_pages, | |
2001 | h->resv_huge_pages, | |
2002 | h->surplus_huge_pages, | |
2003 | 1UL << (huge_page_order(h) + PAGE_SHIFT - 10)); | |
1da177e4 LT |
2004 | } |
2005 | ||
2006 | int hugetlb_report_node_meminfo(int nid, char *buf) | |
2007 | { | |
a5516438 | 2008 | struct hstate *h = &default_hstate; |
1da177e4 LT |
2009 | return sprintf(buf, |
2010 | "Node %d HugePages_Total: %5u\n" | |
a1de0919 NA |
2011 | "Node %d HugePages_Free: %5u\n" |
2012 | "Node %d HugePages_Surp: %5u\n", | |
a5516438 AK |
2013 | nid, h->nr_huge_pages_node[nid], |
2014 | nid, h->free_huge_pages_node[nid], | |
2015 | nid, h->surplus_huge_pages_node[nid]); | |
1da177e4 LT |
2016 | } |
2017 | ||
1da177e4 LT |
2018 | /* Return the number pages of memory we physically have, in PAGE_SIZE units. */ |
2019 | unsigned long hugetlb_total_pages(void) | |
2020 | { | |
a5516438 AK |
2021 | struct hstate *h = &default_hstate; |
2022 | return h->nr_huge_pages * pages_per_huge_page(h); | |
1da177e4 | 2023 | } |
1da177e4 | 2024 | |
a5516438 | 2025 | static int hugetlb_acct_memory(struct hstate *h, long delta) |
fc1b8a73 MG |
2026 | { |
2027 | int ret = -ENOMEM; | |
2028 | ||
2029 | spin_lock(&hugetlb_lock); | |
2030 | /* | |
2031 | * When cpuset is configured, it breaks the strict hugetlb page | |
2032 | * reservation as the accounting is done on a global variable. Such | |
2033 | * reservation is completely rubbish in the presence of cpuset because | |
2034 | * the reservation is not checked against page availability for the | |
2035 | * current cpuset. Application can still potentially OOM'ed by kernel | |
2036 | * with lack of free htlb page in cpuset that the task is in. | |
2037 | * Attempt to enforce strict accounting with cpuset is almost | |
2038 | * impossible (or too ugly) because cpuset is too fluid that | |
2039 | * task or memory node can be dynamically moved between cpusets. | |
2040 | * | |
2041 | * The change of semantics for shared hugetlb mapping with cpuset is | |
2042 | * undesirable. However, in order to preserve some of the semantics, | |
2043 | * we fall back to check against current free page availability as | |
2044 | * a best attempt and hopefully to minimize the impact of changing | |
2045 | * semantics that cpuset has. | |
2046 | */ | |
2047 | if (delta > 0) { | |
a5516438 | 2048 | if (gather_surplus_pages(h, delta) < 0) |
fc1b8a73 MG |
2049 | goto out; |
2050 | ||
a5516438 AK |
2051 | if (delta > cpuset_mems_nr(h->free_huge_pages_node)) { |
2052 | return_unused_surplus_pages(h, delta); | |
fc1b8a73 MG |
2053 | goto out; |
2054 | } | |
2055 | } | |
2056 | ||
2057 | ret = 0; | |
2058 | if (delta < 0) | |
a5516438 | 2059 | return_unused_surplus_pages(h, (unsigned long) -delta); |
fc1b8a73 MG |
2060 | |
2061 | out: | |
2062 | spin_unlock(&hugetlb_lock); | |
2063 | return ret; | |
2064 | } | |
2065 | ||
84afd99b AW |
2066 | static void hugetlb_vm_op_open(struct vm_area_struct *vma) |
2067 | { | |
2068 | struct resv_map *reservations = vma_resv_map(vma); | |
2069 | ||
2070 | /* | |
2071 | * This new VMA should share its siblings reservation map if present. | |
2072 | * The VMA will only ever have a valid reservation map pointer where | |
2073 | * it is being copied for another still existing VMA. As that VMA | |
25985edc | 2074 | * has a reference to the reservation map it cannot disappear until |
84afd99b AW |
2075 | * after this open call completes. It is therefore safe to take a |
2076 | * new reference here without additional locking. | |
2077 | */ | |
2078 | if (reservations) | |
2079 | kref_get(&reservations->refs); | |
2080 | } | |
2081 | ||
a1e78772 MG |
2082 | static void hugetlb_vm_op_close(struct vm_area_struct *vma) |
2083 | { | |
a5516438 | 2084 | struct hstate *h = hstate_vma(vma); |
84afd99b AW |
2085 | struct resv_map *reservations = vma_resv_map(vma); |
2086 | unsigned long reserve; | |
2087 | unsigned long start; | |
2088 | unsigned long end; | |
2089 | ||
2090 | if (reservations) { | |
a5516438 AK |
2091 | start = vma_hugecache_offset(h, vma, vma->vm_start); |
2092 | end = vma_hugecache_offset(h, vma, vma->vm_end); | |
84afd99b AW |
2093 | |
2094 | reserve = (end - start) - | |
2095 | region_count(&reservations->regions, start, end); | |
2096 | ||
2097 | kref_put(&reservations->refs, resv_map_release); | |
2098 | ||
7251ff78 | 2099 | if (reserve) { |
a5516438 | 2100 | hugetlb_acct_memory(h, -reserve); |
7251ff78 AL |
2101 | hugetlb_put_quota(vma->vm_file->f_mapping, reserve); |
2102 | } | |
84afd99b | 2103 | } |
a1e78772 MG |
2104 | } |
2105 | ||
1da177e4 LT |
2106 | /* |
2107 | * We cannot handle pagefaults against hugetlb pages at all. They cause | |
2108 | * handle_mm_fault() to try to instantiate regular-sized pages in the | |
2109 | * hugegpage VMA. do_page_fault() is supposed to trap this, so BUG is we get | |
2110 | * this far. | |
2111 | */ | |
d0217ac0 | 2112 | static int hugetlb_vm_op_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
2113 | { |
2114 | BUG(); | |
d0217ac0 | 2115 | return 0; |
1da177e4 LT |
2116 | } |
2117 | ||
f0f37e2f | 2118 | const struct vm_operations_struct hugetlb_vm_ops = { |
d0217ac0 | 2119 | .fault = hugetlb_vm_op_fault, |
84afd99b | 2120 | .open = hugetlb_vm_op_open, |
a1e78772 | 2121 | .close = hugetlb_vm_op_close, |
1da177e4 LT |
2122 | }; |
2123 | ||
1e8f889b DG |
2124 | static pte_t make_huge_pte(struct vm_area_struct *vma, struct page *page, |
2125 | int writable) | |
63551ae0 DG |
2126 | { |
2127 | pte_t entry; | |
2128 | ||
1e8f889b | 2129 | if (writable) { |
63551ae0 DG |
2130 | entry = |
2131 | pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot))); | |
2132 | } else { | |
7f2e9525 | 2133 | entry = huge_pte_wrprotect(mk_pte(page, vma->vm_page_prot)); |
63551ae0 DG |
2134 | } |
2135 | entry = pte_mkyoung(entry); | |
2136 | entry = pte_mkhuge(entry); | |
2137 | ||
2138 | return entry; | |
2139 | } | |
2140 | ||
1e8f889b DG |
2141 | static void set_huge_ptep_writable(struct vm_area_struct *vma, |
2142 | unsigned long address, pte_t *ptep) | |
2143 | { | |
2144 | pte_t entry; | |
2145 | ||
7f2e9525 | 2146 | entry = pte_mkwrite(pte_mkdirty(huge_ptep_get(ptep))); |
32f84528 | 2147 | if (huge_ptep_set_access_flags(vma, address, ptep, entry, 1)) |
4b3073e1 | 2148 | update_mmu_cache(vma, address, ptep); |
1e8f889b DG |
2149 | } |
2150 | ||
2151 | ||
63551ae0 DG |
2152 | int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src, |
2153 | struct vm_area_struct *vma) | |
2154 | { | |
2155 | pte_t *src_pte, *dst_pte, entry; | |
2156 | struct page *ptepage; | |
1c59827d | 2157 | unsigned long addr; |
1e8f889b | 2158 | int cow; |
a5516438 AK |
2159 | struct hstate *h = hstate_vma(vma); |
2160 | unsigned long sz = huge_page_size(h); | |
1e8f889b DG |
2161 | |
2162 | cow = (vma->vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; | |
63551ae0 | 2163 | |
a5516438 | 2164 | for (addr = vma->vm_start; addr < vma->vm_end; addr += sz) { |
c74df32c HD |
2165 | src_pte = huge_pte_offset(src, addr); |
2166 | if (!src_pte) | |
2167 | continue; | |
a5516438 | 2168 | dst_pte = huge_pte_alloc(dst, addr, sz); |
63551ae0 DG |
2169 | if (!dst_pte) |
2170 | goto nomem; | |
c5c99429 LW |
2171 | |
2172 | /* If the pagetables are shared don't copy or take references */ | |
2173 | if (dst_pte == src_pte) | |
2174 | continue; | |
2175 | ||
c74df32c | 2176 | spin_lock(&dst->page_table_lock); |
46478758 | 2177 | spin_lock_nested(&src->page_table_lock, SINGLE_DEPTH_NESTING); |
7f2e9525 | 2178 | if (!huge_pte_none(huge_ptep_get(src_pte))) { |
1e8f889b | 2179 | if (cow) |
7f2e9525 GS |
2180 | huge_ptep_set_wrprotect(src, addr, src_pte); |
2181 | entry = huge_ptep_get(src_pte); | |
1c59827d HD |
2182 | ptepage = pte_page(entry); |
2183 | get_page(ptepage); | |
0fe6e20b | 2184 | page_dup_rmap(ptepage); |
1c59827d HD |
2185 | set_huge_pte_at(dst, addr, dst_pte, entry); |
2186 | } | |
2187 | spin_unlock(&src->page_table_lock); | |
c74df32c | 2188 | spin_unlock(&dst->page_table_lock); |
63551ae0 DG |
2189 | } |
2190 | return 0; | |
2191 | ||
2192 | nomem: | |
2193 | return -ENOMEM; | |
2194 | } | |
2195 | ||
290408d4 NH |
2196 | static int is_hugetlb_entry_migration(pte_t pte) |
2197 | { | |
2198 | swp_entry_t swp; | |
2199 | ||
2200 | if (huge_pte_none(pte) || pte_present(pte)) | |
2201 | return 0; | |
2202 | swp = pte_to_swp_entry(pte); | |
32f84528 | 2203 | if (non_swap_entry(swp) && is_migration_entry(swp)) |
290408d4 | 2204 | return 1; |
32f84528 | 2205 | else |
290408d4 NH |
2206 | return 0; |
2207 | } | |
2208 | ||
fd6a03ed NH |
2209 | static int is_hugetlb_entry_hwpoisoned(pte_t pte) |
2210 | { | |
2211 | swp_entry_t swp; | |
2212 | ||
2213 | if (huge_pte_none(pte) || pte_present(pte)) | |
2214 | return 0; | |
2215 | swp = pte_to_swp_entry(pte); | |
32f84528 | 2216 | if (non_swap_entry(swp) && is_hwpoison_entry(swp)) |
fd6a03ed | 2217 | return 1; |
32f84528 | 2218 | else |
fd6a03ed NH |
2219 | return 0; |
2220 | } | |
2221 | ||
502717f4 | 2222 | void __unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start, |
04f2cbe3 | 2223 | unsigned long end, struct page *ref_page) |
63551ae0 DG |
2224 | { |
2225 | struct mm_struct *mm = vma->vm_mm; | |
2226 | unsigned long address; | |
c7546f8f | 2227 | pte_t *ptep; |
63551ae0 DG |
2228 | pte_t pte; |
2229 | struct page *page; | |
fe1668ae | 2230 | struct page *tmp; |
a5516438 AK |
2231 | struct hstate *h = hstate_vma(vma); |
2232 | unsigned long sz = huge_page_size(h); | |
2233 | ||
c0a499c2 | 2234 | /* |
3d48ae45 | 2235 | * A page gathering list, protected by per file i_mmap_mutex. The |
c0a499c2 CK |
2236 | * lock is used to avoid list corruption from multiple unmapping |
2237 | * of the same page since we are using page->lru. | |
2238 | */ | |
fe1668ae | 2239 | LIST_HEAD(page_list); |
63551ae0 DG |
2240 | |
2241 | WARN_ON(!is_vm_hugetlb_page(vma)); | |
a5516438 AK |
2242 | BUG_ON(start & ~huge_page_mask(h)); |
2243 | BUG_ON(end & ~huge_page_mask(h)); | |
63551ae0 | 2244 | |
cddb8a5c | 2245 | mmu_notifier_invalidate_range_start(mm, start, end); |
508034a3 | 2246 | spin_lock(&mm->page_table_lock); |
a5516438 | 2247 | for (address = start; address < end; address += sz) { |
c7546f8f | 2248 | ptep = huge_pte_offset(mm, address); |
4c887265 | 2249 | if (!ptep) |
c7546f8f DG |
2250 | continue; |
2251 | ||
39dde65c CK |
2252 | if (huge_pmd_unshare(mm, &address, ptep)) |
2253 | continue; | |
2254 | ||
04f2cbe3 MG |
2255 | /* |
2256 | * If a reference page is supplied, it is because a specific | |
2257 | * page is being unmapped, not a range. Ensure the page we | |
2258 | * are about to unmap is the actual page of interest. | |
2259 | */ | |
2260 | if (ref_page) { | |
2261 | pte = huge_ptep_get(ptep); | |
2262 | if (huge_pte_none(pte)) | |
2263 | continue; | |
2264 | page = pte_page(pte); | |
2265 | if (page != ref_page) | |
2266 | continue; | |
2267 | ||
2268 | /* | |
2269 | * Mark the VMA as having unmapped its page so that | |
2270 | * future faults in this VMA will fail rather than | |
2271 | * looking like data was lost | |
2272 | */ | |
2273 | set_vma_resv_flags(vma, HPAGE_RESV_UNMAPPED); | |
2274 | } | |
2275 | ||
c7546f8f | 2276 | pte = huge_ptep_get_and_clear(mm, address, ptep); |
7f2e9525 | 2277 | if (huge_pte_none(pte)) |
63551ae0 | 2278 | continue; |
c7546f8f | 2279 | |
fd6a03ed NH |
2280 | /* |
2281 | * HWPoisoned hugepage is already unmapped and dropped reference | |
2282 | */ | |
2283 | if (unlikely(is_hugetlb_entry_hwpoisoned(pte))) | |
2284 | continue; | |
2285 | ||
63551ae0 | 2286 | page = pte_page(pte); |
6649a386 KC |
2287 | if (pte_dirty(pte)) |
2288 | set_page_dirty(page); | |
fe1668ae | 2289 | list_add(&page->lru, &page_list); |
9e81130b HD |
2290 | |
2291 | /* Bail out after unmapping reference page if supplied */ | |
2292 | if (ref_page) | |
2293 | break; | |
63551ae0 | 2294 | } |
508034a3 | 2295 | flush_tlb_range(vma, start, end); |
cd2934a3 | 2296 | spin_unlock(&mm->page_table_lock); |
cddb8a5c | 2297 | mmu_notifier_invalidate_range_end(mm, start, end); |
fe1668ae | 2298 | list_for_each_entry_safe(page, tmp, &page_list, lru) { |
0fe6e20b | 2299 | page_remove_rmap(page); |
fe1668ae CK |
2300 | list_del(&page->lru); |
2301 | put_page(page); | |
2302 | } | |
1da177e4 | 2303 | } |
63551ae0 | 2304 | |
502717f4 | 2305 | void unmap_hugepage_range(struct vm_area_struct *vma, unsigned long start, |
04f2cbe3 | 2306 | unsigned long end, struct page *ref_page) |
502717f4 | 2307 | { |
3d48ae45 | 2308 | mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); |
a137e1cc | 2309 | __unmap_hugepage_range(vma, start, end, ref_page); |
3d48ae45 | 2310 | mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); |
502717f4 CK |
2311 | } |
2312 | ||
04f2cbe3 MG |
2313 | /* |
2314 | * This is called when the original mapper is failing to COW a MAP_PRIVATE | |
2315 | * mappping it owns the reserve page for. The intention is to unmap the page | |
2316 | * from other VMAs and let the children be SIGKILLed if they are faulting the | |
2317 | * same region. | |
2318 | */ | |
2a4b3ded HH |
2319 | static int unmap_ref_private(struct mm_struct *mm, struct vm_area_struct *vma, |
2320 | struct page *page, unsigned long address) | |
04f2cbe3 | 2321 | { |
7526674d | 2322 | struct hstate *h = hstate_vma(vma); |
04f2cbe3 MG |
2323 | struct vm_area_struct *iter_vma; |
2324 | struct address_space *mapping; | |
2325 | struct prio_tree_iter iter; | |
2326 | pgoff_t pgoff; | |
2327 | ||
2328 | /* | |
2329 | * vm_pgoff is in PAGE_SIZE units, hence the different calculation | |
2330 | * from page cache lookup which is in HPAGE_SIZE units. | |
2331 | */ | |
7526674d | 2332 | address = address & huge_page_mask(h); |
0c176d52 | 2333 | pgoff = vma_hugecache_offset(h, vma, address); |
04f2cbe3 MG |
2334 | mapping = (struct address_space *)page_private(page); |
2335 | ||
4eb2b1dc MG |
2336 | /* |
2337 | * Take the mapping lock for the duration of the table walk. As | |
2338 | * this mapping should be shared between all the VMAs, | |
2339 | * __unmap_hugepage_range() is called as the lock is already held | |
2340 | */ | |
3d48ae45 | 2341 | mutex_lock(&mapping->i_mmap_mutex); |
04f2cbe3 MG |
2342 | vma_prio_tree_foreach(iter_vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
2343 | /* Do not unmap the current VMA */ | |
2344 | if (iter_vma == vma) | |
2345 | continue; | |
2346 | ||
2347 | /* | |
2348 | * Unmap the page from other VMAs without their own reserves. | |
2349 | * They get marked to be SIGKILLed if they fault in these | |
2350 | * areas. This is because a future no-page fault on this VMA | |
2351 | * could insert a zeroed page instead of the data existing | |
2352 | * from the time of fork. This would look like data corruption | |
2353 | */ | |
2354 | if (!is_vma_resv_set(iter_vma, HPAGE_RESV_OWNER)) | |
4eb2b1dc | 2355 | __unmap_hugepage_range(iter_vma, |
7526674d | 2356 | address, address + huge_page_size(h), |
04f2cbe3 MG |
2357 | page); |
2358 | } | |
3d48ae45 | 2359 | mutex_unlock(&mapping->i_mmap_mutex); |
04f2cbe3 MG |
2360 | |
2361 | return 1; | |
2362 | } | |
2363 | ||
0fe6e20b NH |
2364 | /* |
2365 | * Hugetlb_cow() should be called with page lock of the original hugepage held. | |
ef009b25 MH |
2366 | * Called with hugetlb_instantiation_mutex held and pte_page locked so we |
2367 | * cannot race with other handlers or page migration. | |
2368 | * Keep the pte_same checks anyway to make transition from the mutex easier. | |
0fe6e20b | 2369 | */ |
1e8f889b | 2370 | static int hugetlb_cow(struct mm_struct *mm, struct vm_area_struct *vma, |
04f2cbe3 MG |
2371 | unsigned long address, pte_t *ptep, pte_t pte, |
2372 | struct page *pagecache_page) | |
1e8f889b | 2373 | { |
a5516438 | 2374 | struct hstate *h = hstate_vma(vma); |
1e8f889b | 2375 | struct page *old_page, *new_page; |
79ac6ba4 | 2376 | int avoidcopy; |
04f2cbe3 | 2377 | int outside_reserve = 0; |
1e8f889b DG |
2378 | |
2379 | old_page = pte_page(pte); | |
2380 | ||
04f2cbe3 | 2381 | retry_avoidcopy: |
1e8f889b DG |
2382 | /* If no-one else is actually using this page, avoid the copy |
2383 | * and just make the page writable */ | |
0fe6e20b | 2384 | avoidcopy = (page_mapcount(old_page) == 1); |
1e8f889b | 2385 | if (avoidcopy) { |
56c9cfb1 NH |
2386 | if (PageAnon(old_page)) |
2387 | page_move_anon_rmap(old_page, vma, address); | |
1e8f889b | 2388 | set_huge_ptep_writable(vma, address, ptep); |
83c54070 | 2389 | return 0; |
1e8f889b DG |
2390 | } |
2391 | ||
04f2cbe3 MG |
2392 | /* |
2393 | * If the process that created a MAP_PRIVATE mapping is about to | |
2394 | * perform a COW due to a shared page count, attempt to satisfy | |
2395 | * the allocation without using the existing reserves. The pagecache | |
2396 | * page is used to determine if the reserve at this address was | |
2397 | * consumed or not. If reserves were used, a partial faulted mapping | |
2398 | * at the time of fork() could consume its reserves on COW instead | |
2399 | * of the full address range. | |
2400 | */ | |
f83a275d | 2401 | if (!(vma->vm_flags & VM_MAYSHARE) && |
04f2cbe3 MG |
2402 | is_vma_resv_set(vma, HPAGE_RESV_OWNER) && |
2403 | old_page != pagecache_page) | |
2404 | outside_reserve = 1; | |
2405 | ||
1e8f889b | 2406 | page_cache_get(old_page); |
b76c8cfb LW |
2407 | |
2408 | /* Drop page_table_lock as buddy allocator may be called */ | |
2409 | spin_unlock(&mm->page_table_lock); | |
04f2cbe3 | 2410 | new_page = alloc_huge_page(vma, address, outside_reserve); |
1e8f889b | 2411 | |
2fc39cec | 2412 | if (IS_ERR(new_page)) { |
1e8f889b | 2413 | page_cache_release(old_page); |
04f2cbe3 MG |
2414 | |
2415 | /* | |
2416 | * If a process owning a MAP_PRIVATE mapping fails to COW, | |
2417 | * it is due to references held by a child and an insufficient | |
2418 | * huge page pool. To guarantee the original mappers | |
2419 | * reliability, unmap the page from child processes. The child | |
2420 | * may get SIGKILLed if it later faults. | |
2421 | */ | |
2422 | if (outside_reserve) { | |
2423 | BUG_ON(huge_pte_none(pte)); | |
2424 | if (unmap_ref_private(mm, vma, old_page, address)) { | |
2425 | BUG_ON(page_count(old_page) != 1); | |
2426 | BUG_ON(huge_pte_none(pte)); | |
b76c8cfb | 2427 | spin_lock(&mm->page_table_lock); |
a734bcc8 HD |
2428 | ptep = huge_pte_offset(mm, address & huge_page_mask(h)); |
2429 | if (likely(pte_same(huge_ptep_get(ptep), pte))) | |
2430 | goto retry_avoidcopy; | |
2431 | /* | |
2432 | * race occurs while re-acquiring page_table_lock, and | |
2433 | * our job is done. | |
2434 | */ | |
2435 | return 0; | |
04f2cbe3 MG |
2436 | } |
2437 | WARN_ON_ONCE(1); | |
2438 | } | |
2439 | ||
b76c8cfb LW |
2440 | /* Caller expects lock to be held */ |
2441 | spin_lock(&mm->page_table_lock); | |
2fc39cec | 2442 | return -PTR_ERR(new_page); |
1e8f889b DG |
2443 | } |
2444 | ||
0fe6e20b NH |
2445 | /* |
2446 | * When the original hugepage is shared one, it does not have | |
2447 | * anon_vma prepared. | |
2448 | */ | |
44e2aa93 | 2449 | if (unlikely(anon_vma_prepare(vma))) { |
ea4039a3 HD |
2450 | page_cache_release(new_page); |
2451 | page_cache_release(old_page); | |
44e2aa93 DN |
2452 | /* Caller expects lock to be held */ |
2453 | spin_lock(&mm->page_table_lock); | |
0fe6e20b | 2454 | return VM_FAULT_OOM; |
44e2aa93 | 2455 | } |
0fe6e20b | 2456 | |
47ad8475 AA |
2457 | copy_user_huge_page(new_page, old_page, address, vma, |
2458 | pages_per_huge_page(h)); | |
0ed361de | 2459 | __SetPageUptodate(new_page); |
1e8f889b | 2460 | |
b76c8cfb LW |
2461 | /* |
2462 | * Retake the page_table_lock to check for racing updates | |
2463 | * before the page tables are altered | |
2464 | */ | |
2465 | spin_lock(&mm->page_table_lock); | |
a5516438 | 2466 | ptep = huge_pte_offset(mm, address & huge_page_mask(h)); |
7f2e9525 | 2467 | if (likely(pte_same(huge_ptep_get(ptep), pte))) { |
1e8f889b | 2468 | /* Break COW */ |
3edd4fc9 DD |
2469 | mmu_notifier_invalidate_range_start(mm, |
2470 | address & huge_page_mask(h), | |
2471 | (address & huge_page_mask(h)) + huge_page_size(h)); | |
8fe627ec | 2472 | huge_ptep_clear_flush(vma, address, ptep); |
1e8f889b DG |
2473 | set_huge_pte_at(mm, address, ptep, |
2474 | make_huge_pte(vma, new_page, 1)); | |
0fe6e20b | 2475 | page_remove_rmap(old_page); |
cd67f0d2 | 2476 | hugepage_add_new_anon_rmap(new_page, vma, address); |
1e8f889b DG |
2477 | /* Make the old page be freed below */ |
2478 | new_page = old_page; | |
3edd4fc9 DD |
2479 | mmu_notifier_invalidate_range_end(mm, |
2480 | address & huge_page_mask(h), | |
2481 | (address & huge_page_mask(h)) + huge_page_size(h)); | |
1e8f889b DG |
2482 | } |
2483 | page_cache_release(new_page); | |
2484 | page_cache_release(old_page); | |
83c54070 | 2485 | return 0; |
1e8f889b DG |
2486 | } |
2487 | ||
04f2cbe3 | 2488 | /* Return the pagecache page at a given address within a VMA */ |
a5516438 AK |
2489 | static struct page *hugetlbfs_pagecache_page(struct hstate *h, |
2490 | struct vm_area_struct *vma, unsigned long address) | |
04f2cbe3 MG |
2491 | { |
2492 | struct address_space *mapping; | |
e7c4b0bf | 2493 | pgoff_t idx; |
04f2cbe3 MG |
2494 | |
2495 | mapping = vma->vm_file->f_mapping; | |
a5516438 | 2496 | idx = vma_hugecache_offset(h, vma, address); |
04f2cbe3 MG |
2497 | |
2498 | return find_lock_page(mapping, idx); | |
2499 | } | |
2500 | ||
3ae77f43 HD |
2501 | /* |
2502 | * Return whether there is a pagecache page to back given address within VMA. | |
2503 | * Caller follow_hugetlb_page() holds page_table_lock so we cannot lock_page. | |
2504 | */ | |
2505 | static bool hugetlbfs_pagecache_present(struct hstate *h, | |
2a15efc9 HD |
2506 | struct vm_area_struct *vma, unsigned long address) |
2507 | { | |
2508 | struct address_space *mapping; | |
2509 | pgoff_t idx; | |
2510 | struct page *page; | |
2511 | ||
2512 | mapping = vma->vm_file->f_mapping; | |
2513 | idx = vma_hugecache_offset(h, vma, address); | |
2514 | ||
2515 | page = find_get_page(mapping, idx); | |
2516 | if (page) | |
2517 | put_page(page); | |
2518 | return page != NULL; | |
2519 | } | |
2520 | ||
a1ed3dda | 2521 | static int hugetlb_no_page(struct mm_struct *mm, struct vm_area_struct *vma, |
788c7df4 | 2522 | unsigned long address, pte_t *ptep, unsigned int flags) |
ac9b9c66 | 2523 | { |
a5516438 | 2524 | struct hstate *h = hstate_vma(vma); |
ac9b9c66 | 2525 | int ret = VM_FAULT_SIGBUS; |
409eb8c2 | 2526 | int anon_rmap = 0; |
e7c4b0bf | 2527 | pgoff_t idx; |
4c887265 | 2528 | unsigned long size; |
4c887265 AL |
2529 | struct page *page; |
2530 | struct address_space *mapping; | |
1e8f889b | 2531 | pte_t new_pte; |
4c887265 | 2532 | |
04f2cbe3 MG |
2533 | /* |
2534 | * Currently, we are forced to kill the process in the event the | |
2535 | * original mapper has unmapped pages from the child due to a failed | |
25985edc | 2536 | * COW. Warn that such a situation has occurred as it may not be obvious |
04f2cbe3 MG |
2537 | */ |
2538 | if (is_vma_resv_set(vma, HPAGE_RESV_UNMAPPED)) { | |
2539 | printk(KERN_WARNING | |
2540 | "PID %d killed due to inadequate hugepage pool\n", | |
2541 | current->pid); | |
2542 | return ret; | |
2543 | } | |
2544 | ||
4c887265 | 2545 | mapping = vma->vm_file->f_mapping; |
a5516438 | 2546 | idx = vma_hugecache_offset(h, vma, address); |
4c887265 AL |
2547 | |
2548 | /* | |
2549 | * Use page lock to guard against racing truncation | |
2550 | * before we get page_table_lock. | |
2551 | */ | |
6bda666a CL |
2552 | retry: |
2553 | page = find_lock_page(mapping, idx); | |
2554 | if (!page) { | |
a5516438 | 2555 | size = i_size_read(mapping->host) >> huge_page_shift(h); |
ebed4bfc HD |
2556 | if (idx >= size) |
2557 | goto out; | |
04f2cbe3 | 2558 | page = alloc_huge_page(vma, address, 0); |
2fc39cec AL |
2559 | if (IS_ERR(page)) { |
2560 | ret = -PTR_ERR(page); | |
6bda666a CL |
2561 | goto out; |
2562 | } | |
47ad8475 | 2563 | clear_huge_page(page, address, pages_per_huge_page(h)); |
0ed361de | 2564 | __SetPageUptodate(page); |
ac9b9c66 | 2565 | |
f83a275d | 2566 | if (vma->vm_flags & VM_MAYSHARE) { |
6bda666a | 2567 | int err; |
45c682a6 | 2568 | struct inode *inode = mapping->host; |
6bda666a CL |
2569 | |
2570 | err = add_to_page_cache(page, mapping, idx, GFP_KERNEL); | |
2571 | if (err) { | |
2572 | put_page(page); | |
6bda666a CL |
2573 | if (err == -EEXIST) |
2574 | goto retry; | |
2575 | goto out; | |
2576 | } | |
45c682a6 KC |
2577 | |
2578 | spin_lock(&inode->i_lock); | |
a5516438 | 2579 | inode->i_blocks += blocks_per_huge_page(h); |
45c682a6 | 2580 | spin_unlock(&inode->i_lock); |
23be7468 | 2581 | } else { |
6bda666a | 2582 | lock_page(page); |
0fe6e20b NH |
2583 | if (unlikely(anon_vma_prepare(vma))) { |
2584 | ret = VM_FAULT_OOM; | |
2585 | goto backout_unlocked; | |
2586 | } | |
409eb8c2 | 2587 | anon_rmap = 1; |
23be7468 | 2588 | } |
0fe6e20b | 2589 | } else { |
998b4382 NH |
2590 | /* |
2591 | * If memory error occurs between mmap() and fault, some process | |
2592 | * don't have hwpoisoned swap entry for errored virtual address. | |
2593 | * So we need to block hugepage fault by PG_hwpoison bit check. | |
2594 | */ | |
2595 | if (unlikely(PageHWPoison(page))) { | |
32f84528 | 2596 | ret = VM_FAULT_HWPOISON | |
aa50d3a7 | 2597 | VM_FAULT_SET_HINDEX(h - hstates); |
998b4382 NH |
2598 | goto backout_unlocked; |
2599 | } | |
6bda666a | 2600 | } |
1e8f889b | 2601 | |
57303d80 AW |
2602 | /* |
2603 | * If we are going to COW a private mapping later, we examine the | |
2604 | * pending reservations for this page now. This will ensure that | |
2605 | * any allocations necessary to record that reservation occur outside | |
2606 | * the spinlock. | |
2607 | */ | |
788c7df4 | 2608 | if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) |
2b26736c AW |
2609 | if (vma_needs_reservation(h, vma, address) < 0) { |
2610 | ret = VM_FAULT_OOM; | |
2611 | goto backout_unlocked; | |
2612 | } | |
57303d80 | 2613 | |
ac9b9c66 | 2614 | spin_lock(&mm->page_table_lock); |
a5516438 | 2615 | size = i_size_read(mapping->host) >> huge_page_shift(h); |
4c887265 AL |
2616 | if (idx >= size) |
2617 | goto backout; | |
2618 | ||
83c54070 | 2619 | ret = 0; |
7f2e9525 | 2620 | if (!huge_pte_none(huge_ptep_get(ptep))) |
4c887265 AL |
2621 | goto backout; |
2622 | ||
409eb8c2 HD |
2623 | if (anon_rmap) |
2624 | hugepage_add_new_anon_rmap(page, vma, address); | |
2625 | else | |
2626 | page_dup_rmap(page); | |
1e8f889b DG |
2627 | new_pte = make_huge_pte(vma, page, ((vma->vm_flags & VM_WRITE) |
2628 | && (vma->vm_flags & VM_SHARED))); | |
2629 | set_huge_pte_at(mm, address, ptep, new_pte); | |
2630 | ||
788c7df4 | 2631 | if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) { |
1e8f889b | 2632 | /* Optimization, do the COW without a second fault */ |
04f2cbe3 | 2633 | ret = hugetlb_cow(mm, vma, address, ptep, new_pte, page); |
1e8f889b DG |
2634 | } |
2635 | ||
ac9b9c66 | 2636 | spin_unlock(&mm->page_table_lock); |
4c887265 AL |
2637 | unlock_page(page); |
2638 | out: | |
ac9b9c66 | 2639 | return ret; |
4c887265 AL |
2640 | |
2641 | backout: | |
2642 | spin_unlock(&mm->page_table_lock); | |
2b26736c | 2643 | backout_unlocked: |
4c887265 AL |
2644 | unlock_page(page); |
2645 | put_page(page); | |
2646 | goto out; | |
ac9b9c66 HD |
2647 | } |
2648 | ||
86e5216f | 2649 | int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
788c7df4 | 2650 | unsigned long address, unsigned int flags) |
86e5216f AL |
2651 | { |
2652 | pte_t *ptep; | |
2653 | pte_t entry; | |
1e8f889b | 2654 | int ret; |
0fe6e20b | 2655 | struct page *page = NULL; |
57303d80 | 2656 | struct page *pagecache_page = NULL; |
3935baa9 | 2657 | static DEFINE_MUTEX(hugetlb_instantiation_mutex); |
a5516438 | 2658 | struct hstate *h = hstate_vma(vma); |
86e5216f | 2659 | |
1e16a539 KH |
2660 | address &= huge_page_mask(h); |
2661 | ||
fd6a03ed NH |
2662 | ptep = huge_pte_offset(mm, address); |
2663 | if (ptep) { | |
2664 | entry = huge_ptep_get(ptep); | |
290408d4 NH |
2665 | if (unlikely(is_hugetlb_entry_migration(entry))) { |
2666 | migration_entry_wait(mm, (pmd_t *)ptep, address); | |
2667 | return 0; | |
2668 | } else if (unlikely(is_hugetlb_entry_hwpoisoned(entry))) | |
32f84528 | 2669 | return VM_FAULT_HWPOISON_LARGE | |
aa50d3a7 | 2670 | VM_FAULT_SET_HINDEX(h - hstates); |
fd6a03ed NH |
2671 | } |
2672 | ||
a5516438 | 2673 | ptep = huge_pte_alloc(mm, address, huge_page_size(h)); |
86e5216f AL |
2674 | if (!ptep) |
2675 | return VM_FAULT_OOM; | |
2676 | ||
3935baa9 DG |
2677 | /* |
2678 | * Serialize hugepage allocation and instantiation, so that we don't | |
2679 | * get spurious allocation failures if two CPUs race to instantiate | |
2680 | * the same page in the page cache. | |
2681 | */ | |
2682 | mutex_lock(&hugetlb_instantiation_mutex); | |
7f2e9525 GS |
2683 | entry = huge_ptep_get(ptep); |
2684 | if (huge_pte_none(entry)) { | |
788c7df4 | 2685 | ret = hugetlb_no_page(mm, vma, address, ptep, flags); |
b4d1d99f | 2686 | goto out_mutex; |
3935baa9 | 2687 | } |
86e5216f | 2688 | |
83c54070 | 2689 | ret = 0; |
1e8f889b | 2690 | |
57303d80 AW |
2691 | /* |
2692 | * If we are going to COW the mapping later, we examine the pending | |
2693 | * reservations for this page now. This will ensure that any | |
2694 | * allocations necessary to record that reservation occur outside the | |
2695 | * spinlock. For private mappings, we also lookup the pagecache | |
2696 | * page now as it is used to determine if a reservation has been | |
2697 | * consumed. | |
2698 | */ | |
788c7df4 | 2699 | if ((flags & FAULT_FLAG_WRITE) && !pte_write(entry)) { |
2b26736c AW |
2700 | if (vma_needs_reservation(h, vma, address) < 0) { |
2701 | ret = VM_FAULT_OOM; | |
b4d1d99f | 2702 | goto out_mutex; |
2b26736c | 2703 | } |
57303d80 | 2704 | |
f83a275d | 2705 | if (!(vma->vm_flags & VM_MAYSHARE)) |
57303d80 AW |
2706 | pagecache_page = hugetlbfs_pagecache_page(h, |
2707 | vma, address); | |
2708 | } | |
2709 | ||
56c9cfb1 NH |
2710 | /* |
2711 | * hugetlb_cow() requires page locks of pte_page(entry) and | |
2712 | * pagecache_page, so here we need take the former one | |
2713 | * when page != pagecache_page or !pagecache_page. | |
2714 | * Note that locking order is always pagecache_page -> page, | |
2715 | * so no worry about deadlock. | |
2716 | */ | |
2717 | page = pte_page(entry); | |
2718 | if (page != pagecache_page) | |
0fe6e20b | 2719 | lock_page(page); |
0fe6e20b | 2720 | |
1e8f889b DG |
2721 | spin_lock(&mm->page_table_lock); |
2722 | /* Check for a racing update before calling hugetlb_cow */ | |
b4d1d99f DG |
2723 | if (unlikely(!pte_same(entry, huge_ptep_get(ptep)))) |
2724 | goto out_page_table_lock; | |
2725 | ||
2726 | ||
788c7df4 | 2727 | if (flags & FAULT_FLAG_WRITE) { |
b4d1d99f | 2728 | if (!pte_write(entry)) { |
57303d80 AW |
2729 | ret = hugetlb_cow(mm, vma, address, ptep, entry, |
2730 | pagecache_page); | |
b4d1d99f DG |
2731 | goto out_page_table_lock; |
2732 | } | |
2733 | entry = pte_mkdirty(entry); | |
2734 | } | |
2735 | entry = pte_mkyoung(entry); | |
788c7df4 HD |
2736 | if (huge_ptep_set_access_flags(vma, address, ptep, entry, |
2737 | flags & FAULT_FLAG_WRITE)) | |
4b3073e1 | 2738 | update_mmu_cache(vma, address, ptep); |
b4d1d99f DG |
2739 | |
2740 | out_page_table_lock: | |
1e8f889b | 2741 | spin_unlock(&mm->page_table_lock); |
57303d80 AW |
2742 | |
2743 | if (pagecache_page) { | |
2744 | unlock_page(pagecache_page); | |
2745 | put_page(pagecache_page); | |
2746 | } | |
1f64d69c DN |
2747 | if (page != pagecache_page) |
2748 | unlock_page(page); | |
57303d80 | 2749 | |
b4d1d99f | 2750 | out_mutex: |
3935baa9 | 2751 | mutex_unlock(&hugetlb_instantiation_mutex); |
1e8f889b DG |
2752 | |
2753 | return ret; | |
86e5216f AL |
2754 | } |
2755 | ||
ceb86879 AK |
2756 | /* Can be overriden by architectures */ |
2757 | __attribute__((weak)) struct page * | |
2758 | follow_huge_pud(struct mm_struct *mm, unsigned long address, | |
2759 | pud_t *pud, int write) | |
2760 | { | |
2761 | BUG(); | |
2762 | return NULL; | |
2763 | } | |
2764 | ||
63551ae0 DG |
2765 | int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2766 | struct page **pages, struct vm_area_struct **vmas, | |
5b23dbe8 | 2767 | unsigned long *position, int *length, int i, |
2a15efc9 | 2768 | unsigned int flags) |
63551ae0 | 2769 | { |
d5d4b0aa CK |
2770 | unsigned long pfn_offset; |
2771 | unsigned long vaddr = *position; | |
63551ae0 | 2772 | int remainder = *length; |
a5516438 | 2773 | struct hstate *h = hstate_vma(vma); |
63551ae0 | 2774 | |
1c59827d | 2775 | spin_lock(&mm->page_table_lock); |
63551ae0 | 2776 | while (vaddr < vma->vm_end && remainder) { |
4c887265 | 2777 | pte_t *pte; |
2a15efc9 | 2778 | int absent; |
4c887265 | 2779 | struct page *page; |
63551ae0 | 2780 | |
4c887265 AL |
2781 | /* |
2782 | * Some archs (sparc64, sh*) have multiple pte_ts to | |
2a15efc9 | 2783 | * each hugepage. We have to make sure we get the |
4c887265 AL |
2784 | * first, for the page indexing below to work. |
2785 | */ | |
a5516438 | 2786 | pte = huge_pte_offset(mm, vaddr & huge_page_mask(h)); |
2a15efc9 HD |
2787 | absent = !pte || huge_pte_none(huge_ptep_get(pte)); |
2788 | ||
2789 | /* | |
2790 | * When coredumping, it suits get_dump_page if we just return | |
3ae77f43 HD |
2791 | * an error where there's an empty slot with no huge pagecache |
2792 | * to back it. This way, we avoid allocating a hugepage, and | |
2793 | * the sparse dumpfile avoids allocating disk blocks, but its | |
2794 | * huge holes still show up with zeroes where they need to be. | |
2a15efc9 | 2795 | */ |
3ae77f43 HD |
2796 | if (absent && (flags & FOLL_DUMP) && |
2797 | !hugetlbfs_pagecache_present(h, vma, vaddr)) { | |
2a15efc9 HD |
2798 | remainder = 0; |
2799 | break; | |
2800 | } | |
63551ae0 | 2801 | |
2a15efc9 HD |
2802 | if (absent || |
2803 | ((flags & FOLL_WRITE) && !pte_write(huge_ptep_get(pte)))) { | |
4c887265 | 2804 | int ret; |
63551ae0 | 2805 | |
4c887265 | 2806 | spin_unlock(&mm->page_table_lock); |
2a15efc9 HD |
2807 | ret = hugetlb_fault(mm, vma, vaddr, |
2808 | (flags & FOLL_WRITE) ? FAULT_FLAG_WRITE : 0); | |
4c887265 | 2809 | spin_lock(&mm->page_table_lock); |
a89182c7 | 2810 | if (!(ret & VM_FAULT_ERROR)) |
4c887265 | 2811 | continue; |
63551ae0 | 2812 | |
4c887265 | 2813 | remainder = 0; |
4c887265 AL |
2814 | break; |
2815 | } | |
2816 | ||
a5516438 | 2817 | pfn_offset = (vaddr & ~huge_page_mask(h)) >> PAGE_SHIFT; |
7f2e9525 | 2818 | page = pte_page(huge_ptep_get(pte)); |
d5d4b0aa | 2819 | same_page: |
d6692183 | 2820 | if (pages) { |
2a15efc9 | 2821 | pages[i] = mem_map_offset(page, pfn_offset); |
4b2e38ad | 2822 | get_page(pages[i]); |
d6692183 | 2823 | } |
63551ae0 DG |
2824 | |
2825 | if (vmas) | |
2826 | vmas[i] = vma; | |
2827 | ||
2828 | vaddr += PAGE_SIZE; | |
d5d4b0aa | 2829 | ++pfn_offset; |
63551ae0 DG |
2830 | --remainder; |
2831 | ++i; | |
d5d4b0aa | 2832 | if (vaddr < vma->vm_end && remainder && |
a5516438 | 2833 | pfn_offset < pages_per_huge_page(h)) { |
d5d4b0aa CK |
2834 | /* |
2835 | * We use pfn_offset to avoid touching the pageframes | |
2836 | * of this compound page. | |
2837 | */ | |
2838 | goto same_page; | |
2839 | } | |
63551ae0 | 2840 | } |
1c59827d | 2841 | spin_unlock(&mm->page_table_lock); |
63551ae0 DG |
2842 | *length = remainder; |
2843 | *position = vaddr; | |
2844 | ||
2a15efc9 | 2845 | return i ? i : -EFAULT; |
63551ae0 | 2846 | } |
8f860591 ZY |
2847 | |
2848 | void hugetlb_change_protection(struct vm_area_struct *vma, | |
2849 | unsigned long address, unsigned long end, pgprot_t newprot) | |
2850 | { | |
2851 | struct mm_struct *mm = vma->vm_mm; | |
2852 | unsigned long start = address; | |
2853 | pte_t *ptep; | |
2854 | pte_t pte; | |
a5516438 | 2855 | struct hstate *h = hstate_vma(vma); |
8f860591 ZY |
2856 | |
2857 | BUG_ON(address >= end); | |
2858 | flush_cache_range(vma, address, end); | |
2859 | ||
3d48ae45 | 2860 | mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); |
8f860591 | 2861 | spin_lock(&mm->page_table_lock); |
a5516438 | 2862 | for (; address < end; address += huge_page_size(h)) { |
8f860591 ZY |
2863 | ptep = huge_pte_offset(mm, address); |
2864 | if (!ptep) | |
2865 | continue; | |
39dde65c CK |
2866 | if (huge_pmd_unshare(mm, &address, ptep)) |
2867 | continue; | |
7f2e9525 | 2868 | if (!huge_pte_none(huge_ptep_get(ptep))) { |
8f860591 ZY |
2869 | pte = huge_ptep_get_and_clear(mm, address, ptep); |
2870 | pte = pte_mkhuge(pte_modify(pte, newprot)); | |
2871 | set_huge_pte_at(mm, address, ptep, pte); | |
8f860591 ZY |
2872 | } |
2873 | } | |
2874 | spin_unlock(&mm->page_table_lock); | |
3d48ae45 | 2875 | mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); |
8f860591 ZY |
2876 | |
2877 | flush_tlb_range(vma, start, end); | |
2878 | } | |
2879 | ||
a1e78772 MG |
2880 | int hugetlb_reserve_pages(struct inode *inode, |
2881 | long from, long to, | |
5a6fe125 | 2882 | struct vm_area_struct *vma, |
ca16d140 | 2883 | vm_flags_t vm_flags) |
e4e574b7 | 2884 | { |
17c9d12e | 2885 | long ret, chg; |
a5516438 | 2886 | struct hstate *h = hstate_inode(inode); |
e4e574b7 | 2887 | |
17c9d12e MG |
2888 | /* |
2889 | * Only apply hugepage reservation if asked. At fault time, an | |
2890 | * attempt will be made for VM_NORESERVE to allocate a page | |
2891 | * and filesystem quota without using reserves | |
2892 | */ | |
ca16d140 | 2893 | if (vm_flags & VM_NORESERVE) |
17c9d12e MG |
2894 | return 0; |
2895 | ||
a1e78772 MG |
2896 | /* |
2897 | * Shared mappings base their reservation on the number of pages that | |
2898 | * are already allocated on behalf of the file. Private mappings need | |
2899 | * to reserve the full area even if read-only as mprotect() may be | |
2900 | * called to make the mapping read-write. Assume !vma is a shm mapping | |
2901 | */ | |
f83a275d | 2902 | if (!vma || vma->vm_flags & VM_MAYSHARE) |
a1e78772 | 2903 | chg = region_chg(&inode->i_mapping->private_list, from, to); |
17c9d12e MG |
2904 | else { |
2905 | struct resv_map *resv_map = resv_map_alloc(); | |
2906 | if (!resv_map) | |
2907 | return -ENOMEM; | |
2908 | ||
a1e78772 | 2909 | chg = to - from; |
84afd99b | 2910 | |
17c9d12e MG |
2911 | set_vma_resv_map(vma, resv_map); |
2912 | set_vma_resv_flags(vma, HPAGE_RESV_OWNER); | |
2913 | } | |
2914 | ||
e4e574b7 AL |
2915 | if (chg < 0) |
2916 | return chg; | |
8a630112 | 2917 | |
17c9d12e | 2918 | /* There must be enough filesystem quota for the mapping */ |
90d8b7e6 AL |
2919 | if (hugetlb_get_quota(inode->i_mapping, chg)) |
2920 | return -ENOSPC; | |
5a6fe125 MG |
2921 | |
2922 | /* | |
17c9d12e MG |
2923 | * Check enough hugepages are available for the reservation. |
2924 | * Hand back the quota if there are not | |
5a6fe125 | 2925 | */ |
a5516438 | 2926 | ret = hugetlb_acct_memory(h, chg); |
68842c9b KC |
2927 | if (ret < 0) { |
2928 | hugetlb_put_quota(inode->i_mapping, chg); | |
a43a8c39 | 2929 | return ret; |
68842c9b | 2930 | } |
17c9d12e MG |
2931 | |
2932 | /* | |
2933 | * Account for the reservations made. Shared mappings record regions | |
2934 | * that have reservations as they are shared by multiple VMAs. | |
2935 | * When the last VMA disappears, the region map says how much | |
2936 | * the reservation was and the page cache tells how much of | |
2937 | * the reservation was consumed. Private mappings are per-VMA and | |
2938 | * only the consumed reservations are tracked. When the VMA | |
2939 | * disappears, the original reservation is the VMA size and the | |
2940 | * consumed reservations are stored in the map. Hence, nothing | |
2941 | * else has to be done for private mappings here | |
2942 | */ | |
f83a275d | 2943 | if (!vma || vma->vm_flags & VM_MAYSHARE) |
a1e78772 | 2944 | region_add(&inode->i_mapping->private_list, from, to); |
a43a8c39 CK |
2945 | return 0; |
2946 | } | |
2947 | ||
2948 | void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed) | |
2949 | { | |
a5516438 | 2950 | struct hstate *h = hstate_inode(inode); |
a43a8c39 | 2951 | long chg = region_truncate(&inode->i_mapping->private_list, offset); |
45c682a6 KC |
2952 | |
2953 | spin_lock(&inode->i_lock); | |
e4c6f8be | 2954 | inode->i_blocks -= (blocks_per_huge_page(h) * freed); |
45c682a6 KC |
2955 | spin_unlock(&inode->i_lock); |
2956 | ||
90d8b7e6 | 2957 | hugetlb_put_quota(inode->i_mapping, (chg - freed)); |
a5516438 | 2958 | hugetlb_acct_memory(h, -(chg - freed)); |
a43a8c39 | 2959 | } |
93f70f90 | 2960 | |
d5bd9106 AK |
2961 | #ifdef CONFIG_MEMORY_FAILURE |
2962 | ||
6de2b1aa NH |
2963 | /* Should be called in hugetlb_lock */ |
2964 | static int is_hugepage_on_freelist(struct page *hpage) | |
2965 | { | |
2966 | struct page *page; | |
2967 | struct page *tmp; | |
2968 | struct hstate *h = page_hstate(hpage); | |
2969 | int nid = page_to_nid(hpage); | |
2970 | ||
2971 | list_for_each_entry_safe(page, tmp, &h->hugepage_freelists[nid], lru) | |
2972 | if (page == hpage) | |
2973 | return 1; | |
2974 | return 0; | |
2975 | } | |
2976 | ||
93f70f90 NH |
2977 | /* |
2978 | * This function is called from memory failure code. | |
2979 | * Assume the caller holds page lock of the head page. | |
2980 | */ | |
6de2b1aa | 2981 | int dequeue_hwpoisoned_huge_page(struct page *hpage) |
93f70f90 NH |
2982 | { |
2983 | struct hstate *h = page_hstate(hpage); | |
2984 | int nid = page_to_nid(hpage); | |
6de2b1aa | 2985 | int ret = -EBUSY; |
93f70f90 NH |
2986 | |
2987 | spin_lock(&hugetlb_lock); | |
6de2b1aa NH |
2988 | if (is_hugepage_on_freelist(hpage)) { |
2989 | list_del(&hpage->lru); | |
8c6c2ecb | 2990 | set_page_refcounted(hpage); |
6de2b1aa NH |
2991 | h->free_huge_pages--; |
2992 | h->free_huge_pages_node[nid]--; | |
2993 | ret = 0; | |
2994 | } | |
93f70f90 | 2995 | spin_unlock(&hugetlb_lock); |
6de2b1aa | 2996 | return ret; |
93f70f90 | 2997 | } |
6de2b1aa | 2998 | #endif |