Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
41151e77 | 2 | * PPC Huge TLB Page Support for Kernel. |
1da177e4 LT |
3 | * |
4 | * Copyright (C) 2003 David Gibson, IBM Corporation. | |
41151e77 | 5 | * Copyright (C) 2011 Becky Bruce, Freescale Semiconductor |
1da177e4 LT |
6 | * |
7 | * Based on the IA-32 version: | |
8 | * Copyright (C) 2002, Rohit Seth <rohit.seth@intel.com> | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/mm.h> |
883a3e52 | 12 | #include <linux/io.h> |
5a0e3ad6 | 13 | #include <linux/slab.h> |
1da177e4 | 14 | #include <linux/hugetlb.h> |
342d3db7 | 15 | #include <linux/export.h> |
41151e77 BB |
16 | #include <linux/of_fdt.h> |
17 | #include <linux/memblock.h> | |
18 | #include <linux/bootmem.h> | |
13020be8 | 19 | #include <linux/moduleparam.h> |
883a3e52 | 20 | #include <asm/pgtable.h> |
1da177e4 LT |
21 | #include <asm/pgalloc.h> |
22 | #include <asm/tlb.h> | |
41151e77 | 23 | #include <asm/setup.h> |
1da177e4 | 24 | |
91224346 JT |
25 | #define PAGE_SHIFT_64K 16 |
26 | #define PAGE_SHIFT_16M 24 | |
27 | #define PAGE_SHIFT_16G 34 | |
4ec161cf | 28 | |
41151e77 | 29 | unsigned int HPAGE_SHIFT; |
ec4b2c0c | 30 | |
41151e77 BB |
31 | /* |
32 | * Tracks gpages after the device tree is scanned and before the | |
a6146888 BB |
33 | * huge_boot_pages list is ready. On non-Freescale implementations, this is |
34 | * just used to track 16G pages and so is a single array. FSL-based | |
35 | * implementations may have more than one gpage size, so we need multiple | |
36 | * arrays | |
41151e77 | 37 | */ |
881fde1d | 38 | #ifdef CONFIG_PPC_FSL_BOOK3E |
41151e77 BB |
39 | #define MAX_NUMBER_GPAGES 128 |
40 | struct psize_gpages { | |
41 | u64 gpage_list[MAX_NUMBER_GPAGES]; | |
42 | unsigned int nr_gpages; | |
43 | }; | |
44 | static struct psize_gpages gpage_freearray[MMU_PAGE_COUNT]; | |
881fde1d BB |
45 | #else |
46 | #define MAX_NUMBER_GPAGES 1024 | |
47 | static u64 gpage_freearray[MAX_NUMBER_GPAGES]; | |
48 | static unsigned nr_gpages; | |
41151e77 | 49 | #endif |
f10a04c0 | 50 | |
0d9ea754 JT |
51 | static inline int shift_to_mmu_psize(unsigned int shift) |
52 | { | |
d1837cba DG |
53 | int psize; |
54 | ||
55 | for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) | |
56 | if (mmu_psize_defs[psize].shift == shift) | |
57 | return psize; | |
0d9ea754 JT |
58 | return -1; |
59 | } | |
60 | ||
61 | static inline unsigned int mmu_psize_to_shift(unsigned int mmu_psize) | |
62 | { | |
63 | if (mmu_psize_defs[mmu_psize].shift) | |
64 | return mmu_psize_defs[mmu_psize].shift; | |
65 | BUG(); | |
66 | } | |
67 | ||
a4fe3ce7 DG |
68 | #define hugepd_none(hpd) ((hpd).pd == 0) |
69 | ||
a4fe3ce7 DG |
70 | pte_t *find_linux_pte_or_hugepte(pgd_t *pgdir, unsigned long ea, unsigned *shift) |
71 | { | |
72 | pgd_t *pg; | |
73 | pud_t *pu; | |
74 | pmd_t *pm; | |
75 | hugepd_t *hpdp = NULL; | |
76 | unsigned pdshift = PGDIR_SHIFT; | |
77 | ||
78 | if (shift) | |
79 | *shift = 0; | |
80 | ||
81 | pg = pgdir + pgd_index(ea); | |
82 | if (is_hugepd(pg)) { | |
83 | hpdp = (hugepd_t *)pg; | |
84 | } else if (!pgd_none(*pg)) { | |
85 | pdshift = PUD_SHIFT; | |
86 | pu = pud_offset(pg, ea); | |
87 | if (is_hugepd(pu)) | |
88 | hpdp = (hugepd_t *)pu; | |
89 | else if (!pud_none(*pu)) { | |
90 | pdshift = PMD_SHIFT; | |
91 | pm = pmd_offset(pu, ea); | |
92 | if (is_hugepd(pm)) | |
93 | hpdp = (hugepd_t *)pm; | |
94 | else if (!pmd_none(*pm)) { | |
41151e77 | 95 | return pte_offset_kernel(pm, ea); |
a4fe3ce7 DG |
96 | } |
97 | } | |
98 | } | |
99 | ||
100 | if (!hpdp) | |
101 | return NULL; | |
102 | ||
103 | if (shift) | |
104 | *shift = hugepd_shift(*hpdp); | |
105 | return hugepte_offset(hpdp, ea, pdshift); | |
106 | } | |
342d3db7 | 107 | EXPORT_SYMBOL_GPL(find_linux_pte_or_hugepte); |
a4fe3ce7 DG |
108 | |
109 | pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr) | |
110 | { | |
111 | return find_linux_pte_or_hugepte(mm->pgd, addr, NULL); | |
112 | } | |
113 | ||
f10a04c0 | 114 | static int __hugepte_alloc(struct mm_struct *mm, hugepd_t *hpdp, |
a4fe3ce7 | 115 | unsigned long address, unsigned pdshift, unsigned pshift) |
f10a04c0 | 116 | { |
41151e77 BB |
117 | struct kmem_cache *cachep; |
118 | pte_t *new; | |
119 | ||
881fde1d | 120 | #ifdef CONFIG_PPC_FSL_BOOK3E |
41151e77 BB |
121 | int i; |
122 | int num_hugepd = 1 << (pshift - pdshift); | |
123 | cachep = hugepte_cache; | |
881fde1d BB |
124 | #else |
125 | cachep = PGT_CACHE(pdshift - pshift); | |
41151e77 BB |
126 | #endif |
127 | ||
128 | new = kmem_cache_zalloc(cachep, GFP_KERNEL|__GFP_REPEAT); | |
f10a04c0 | 129 | |
a4fe3ce7 DG |
130 | BUG_ON(pshift > HUGEPD_SHIFT_MASK); |
131 | BUG_ON((unsigned long)new & HUGEPD_SHIFT_MASK); | |
132 | ||
f10a04c0 DG |
133 | if (! new) |
134 | return -ENOMEM; | |
135 | ||
136 | spin_lock(&mm->page_table_lock); | |
881fde1d | 137 | #ifdef CONFIG_PPC_FSL_BOOK3E |
41151e77 BB |
138 | /* |
139 | * We have multiple higher-level entries that point to the same | |
140 | * actual pte location. Fill in each as we go and backtrack on error. | |
141 | * We need all of these so the DTLB pgtable walk code can find the | |
142 | * right higher-level entry without knowing if it's a hugepage or not. | |
143 | */ | |
144 | for (i = 0; i < num_hugepd; i++, hpdp++) { | |
145 | if (unlikely(!hugepd_none(*hpdp))) | |
146 | break; | |
147 | else | |
148 | hpdp->pd = ((unsigned long)new & ~PD_HUGE) | pshift; | |
149 | } | |
150 | /* If we bailed from the for loop early, an error occurred, clean up */ | |
151 | if (i < num_hugepd) { | |
152 | for (i = i - 1 ; i >= 0; i--, hpdp--) | |
153 | hpdp->pd = 0; | |
154 | kmem_cache_free(cachep, new); | |
155 | } | |
a1cd5419 BB |
156 | #else |
157 | if (!hugepd_none(*hpdp)) | |
158 | kmem_cache_free(cachep, new); | |
159 | else | |
160 | hpdp->pd = ((unsigned long)new & ~PD_HUGE) | pshift; | |
41151e77 | 161 | #endif |
f10a04c0 DG |
162 | spin_unlock(&mm->page_table_lock); |
163 | return 0; | |
164 | } | |
165 | ||
a1cd5419 BB |
166 | /* |
167 | * These macros define how to determine which level of the page table holds | |
168 | * the hpdp. | |
169 | */ | |
170 | #ifdef CONFIG_PPC_FSL_BOOK3E | |
171 | #define HUGEPD_PGD_SHIFT PGDIR_SHIFT | |
172 | #define HUGEPD_PUD_SHIFT PUD_SHIFT | |
173 | #else | |
174 | #define HUGEPD_PGD_SHIFT PUD_SHIFT | |
175 | #define HUGEPD_PUD_SHIFT PMD_SHIFT | |
176 | #endif | |
177 | ||
a4fe3ce7 | 178 | pte_t *huge_pte_alloc(struct mm_struct *mm, unsigned long addr, unsigned long sz) |
0b26425c | 179 | { |
a4fe3ce7 DG |
180 | pgd_t *pg; |
181 | pud_t *pu; | |
182 | pmd_t *pm; | |
183 | hugepd_t *hpdp = NULL; | |
184 | unsigned pshift = __ffs(sz); | |
185 | unsigned pdshift = PGDIR_SHIFT; | |
186 | ||
187 | addr &= ~(sz-1); | |
188 | ||
189 | pg = pgd_offset(mm, addr); | |
a1cd5419 BB |
190 | |
191 | if (pshift >= HUGEPD_PGD_SHIFT) { | |
a4fe3ce7 DG |
192 | hpdp = (hugepd_t *)pg; |
193 | } else { | |
194 | pdshift = PUD_SHIFT; | |
195 | pu = pud_alloc(mm, pg, addr); | |
a1cd5419 | 196 | if (pshift >= HUGEPD_PUD_SHIFT) { |
a4fe3ce7 DG |
197 | hpdp = (hugepd_t *)pu; |
198 | } else { | |
199 | pdshift = PMD_SHIFT; | |
200 | pm = pmd_alloc(mm, pu, addr); | |
201 | hpdp = (hugepd_t *)pm; | |
202 | } | |
203 | } | |
204 | ||
205 | if (!hpdp) | |
206 | return NULL; | |
207 | ||
208 | BUG_ON(!hugepd_none(*hpdp) && !hugepd_ok(*hpdp)); | |
209 | ||
210 | if (hugepd_none(*hpdp) && __hugepte_alloc(mm, hpdp, addr, pdshift, pshift)) | |
211 | return NULL; | |
212 | ||
213 | return hugepte_offset(hpdp, addr, pdshift); | |
4ec161cf | 214 | } |
4ec161cf | 215 | |
881fde1d | 216 | #ifdef CONFIG_PPC_FSL_BOOK3E |
658013e9 JT |
217 | /* Build list of addresses of gigantic pages. This function is used in early |
218 | * boot before the buddy or bootmem allocator is setup. | |
219 | */ | |
41151e77 BB |
220 | void add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages) |
221 | { | |
222 | unsigned int idx = shift_to_mmu_psize(__ffs(page_size)); | |
223 | int i; | |
224 | ||
225 | if (addr == 0) | |
226 | return; | |
227 | ||
228 | gpage_freearray[idx].nr_gpages = number_of_pages; | |
229 | ||
230 | for (i = 0; i < number_of_pages; i++) { | |
231 | gpage_freearray[idx].gpage_list[i] = addr; | |
232 | addr += page_size; | |
233 | } | |
234 | } | |
235 | ||
236 | /* | |
237 | * Moves the gigantic page addresses from the temporary list to the | |
238 | * huge_boot_pages list. | |
239 | */ | |
240 | int alloc_bootmem_huge_page(struct hstate *hstate) | |
241 | { | |
242 | struct huge_bootmem_page *m; | |
243 | int idx = shift_to_mmu_psize(hstate->order + PAGE_SHIFT); | |
244 | int nr_gpages = gpage_freearray[idx].nr_gpages; | |
245 | ||
246 | if (nr_gpages == 0) | |
247 | return 0; | |
248 | ||
249 | #ifdef CONFIG_HIGHMEM | |
250 | /* | |
251 | * If gpages can be in highmem we can't use the trick of storing the | |
252 | * data structure in the page; allocate space for this | |
253 | */ | |
254 | m = alloc_bootmem(sizeof(struct huge_bootmem_page)); | |
255 | m->phys = gpage_freearray[idx].gpage_list[--nr_gpages]; | |
256 | #else | |
257 | m = phys_to_virt(gpage_freearray[idx].gpage_list[--nr_gpages]); | |
258 | #endif | |
259 | ||
260 | list_add(&m->list, &huge_boot_pages); | |
261 | gpage_freearray[idx].nr_gpages = nr_gpages; | |
262 | gpage_freearray[idx].gpage_list[nr_gpages] = 0; | |
263 | m->hstate = hstate; | |
264 | ||
265 | return 1; | |
266 | } | |
267 | /* | |
268 | * Scan the command line hugepagesz= options for gigantic pages; store those in | |
269 | * a list that we use to allocate the memory once all options are parsed. | |
270 | */ | |
271 | ||
272 | unsigned long gpage_npages[MMU_PAGE_COUNT]; | |
273 | ||
89528127 PG |
274 | static int __init do_gpage_early_setup(char *param, char *val, |
275 | const char *unused) | |
41151e77 BB |
276 | { |
277 | static phys_addr_t size; | |
278 | unsigned long npages; | |
279 | ||
280 | /* | |
281 | * The hugepagesz and hugepages cmdline options are interleaved. We | |
282 | * use the size variable to keep track of whether or not this was done | |
283 | * properly and skip over instances where it is incorrect. Other | |
284 | * command-line parsing code will issue warnings, so we don't need to. | |
285 | * | |
286 | */ | |
287 | if ((strcmp(param, "default_hugepagesz") == 0) || | |
288 | (strcmp(param, "hugepagesz") == 0)) { | |
289 | size = memparse(val, NULL); | |
290 | } else if (strcmp(param, "hugepages") == 0) { | |
291 | if (size != 0) { | |
292 | if (sscanf(val, "%lu", &npages) <= 0) | |
293 | npages = 0; | |
294 | gpage_npages[shift_to_mmu_psize(__ffs(size))] = npages; | |
295 | size = 0; | |
296 | } | |
297 | } | |
298 | return 0; | |
299 | } | |
300 | ||
301 | ||
302 | /* | |
303 | * This function allocates physical space for pages that are larger than the | |
304 | * buddy allocator can handle. We want to allocate these in highmem because | |
305 | * the amount of lowmem is limited. This means that this function MUST be | |
306 | * called before lowmem_end_addr is set up in MMU_init() in order for the lmb | |
307 | * allocate to grab highmem. | |
308 | */ | |
309 | void __init reserve_hugetlb_gpages(void) | |
310 | { | |
311 | static __initdata char cmdline[COMMAND_LINE_SIZE]; | |
312 | phys_addr_t size, base; | |
313 | int i; | |
314 | ||
315 | strlcpy(cmdline, boot_command_line, COMMAND_LINE_SIZE); | |
026cee00 PM |
316 | parse_args("hugetlb gpages", cmdline, NULL, 0, 0, 0, |
317 | &do_gpage_early_setup); | |
41151e77 BB |
318 | |
319 | /* | |
320 | * Walk gpage list in reverse, allocating larger page sizes first. | |
321 | * Skip over unsupported sizes, or sizes that have 0 gpages allocated. | |
322 | * When we reach the point in the list where pages are no longer | |
323 | * considered gpages, we're done. | |
324 | */ | |
325 | for (i = MMU_PAGE_COUNT-1; i >= 0; i--) { | |
326 | if (mmu_psize_defs[i].shift == 0 || gpage_npages[i] == 0) | |
327 | continue; | |
328 | else if (mmu_psize_to_shift(i) < (MAX_ORDER + PAGE_SHIFT)) | |
329 | break; | |
330 | ||
331 | size = (phys_addr_t)(1ULL << mmu_psize_to_shift(i)); | |
332 | base = memblock_alloc_base(size * gpage_npages[i], size, | |
333 | MEMBLOCK_ALLOC_ANYWHERE); | |
334 | add_gpage(base, size, gpage_npages[i]); | |
335 | } | |
336 | } | |
337 | ||
881fde1d | 338 | #else /* !PPC_FSL_BOOK3E */ |
41151e77 BB |
339 | |
340 | /* Build list of addresses of gigantic pages. This function is used in early | |
341 | * boot before the buddy or bootmem allocator is setup. | |
342 | */ | |
343 | void add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages) | |
658013e9 JT |
344 | { |
345 | if (!addr) | |
346 | return; | |
347 | while (number_of_pages > 0) { | |
348 | gpage_freearray[nr_gpages] = addr; | |
349 | nr_gpages++; | |
350 | number_of_pages--; | |
351 | addr += page_size; | |
352 | } | |
353 | } | |
354 | ||
ec4b2c0c | 355 | /* Moves the gigantic page addresses from the temporary list to the |
0d9ea754 JT |
356 | * huge_boot_pages list. |
357 | */ | |
358 | int alloc_bootmem_huge_page(struct hstate *hstate) | |
ec4b2c0c JT |
359 | { |
360 | struct huge_bootmem_page *m; | |
361 | if (nr_gpages == 0) | |
362 | return 0; | |
363 | m = phys_to_virt(gpage_freearray[--nr_gpages]); | |
364 | gpage_freearray[nr_gpages] = 0; | |
365 | list_add(&m->list, &huge_boot_pages); | |
0d9ea754 | 366 | m->hstate = hstate; |
ec4b2c0c JT |
367 | return 1; |
368 | } | |
41151e77 | 369 | #endif |
ec4b2c0c | 370 | |
39dde65c CK |
371 | int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep) |
372 | { | |
373 | return 0; | |
374 | } | |
375 | ||
881fde1d | 376 | #ifdef CONFIG_PPC_FSL_BOOK3E |
41151e77 BB |
377 | #define HUGEPD_FREELIST_SIZE \ |
378 | ((PAGE_SIZE - sizeof(struct hugepd_freelist)) / sizeof(pte_t)) | |
379 | ||
380 | struct hugepd_freelist { | |
381 | struct rcu_head rcu; | |
382 | unsigned int index; | |
383 | void *ptes[0]; | |
384 | }; | |
385 | ||
386 | static DEFINE_PER_CPU(struct hugepd_freelist *, hugepd_freelist_cur); | |
387 | ||
388 | static void hugepd_free_rcu_callback(struct rcu_head *head) | |
389 | { | |
390 | struct hugepd_freelist *batch = | |
391 | container_of(head, struct hugepd_freelist, rcu); | |
392 | unsigned int i; | |
393 | ||
394 | for (i = 0; i < batch->index; i++) | |
395 | kmem_cache_free(hugepte_cache, batch->ptes[i]); | |
396 | ||
397 | free_page((unsigned long)batch); | |
398 | } | |
399 | ||
400 | static void hugepd_free(struct mmu_gather *tlb, void *hugepte) | |
401 | { | |
402 | struct hugepd_freelist **batchp; | |
403 | ||
404 | batchp = &__get_cpu_var(hugepd_freelist_cur); | |
405 | ||
406 | if (atomic_read(&tlb->mm->mm_users) < 2 || | |
407 | cpumask_equal(mm_cpumask(tlb->mm), | |
408 | cpumask_of(smp_processor_id()))) { | |
409 | kmem_cache_free(hugepte_cache, hugepte); | |
410 | return; | |
411 | } | |
412 | ||
413 | if (*batchp == NULL) { | |
414 | *batchp = (struct hugepd_freelist *)__get_free_page(GFP_ATOMIC); | |
415 | (*batchp)->index = 0; | |
416 | } | |
417 | ||
418 | (*batchp)->ptes[(*batchp)->index++] = hugepte; | |
419 | if ((*batchp)->index == HUGEPD_FREELIST_SIZE) { | |
420 | call_rcu_sched(&(*batchp)->rcu, hugepd_free_rcu_callback); | |
421 | *batchp = NULL; | |
422 | } | |
423 | } | |
424 | #endif | |
425 | ||
a4fe3ce7 DG |
426 | static void free_hugepd_range(struct mmu_gather *tlb, hugepd_t *hpdp, int pdshift, |
427 | unsigned long start, unsigned long end, | |
428 | unsigned long floor, unsigned long ceiling) | |
f10a04c0 DG |
429 | { |
430 | pte_t *hugepte = hugepd_page(*hpdp); | |
41151e77 BB |
431 | int i; |
432 | ||
a4fe3ce7 | 433 | unsigned long pdmask = ~((1UL << pdshift) - 1); |
41151e77 BB |
434 | unsigned int num_hugepd = 1; |
435 | ||
881fde1d BB |
436 | #ifdef CONFIG_PPC_FSL_BOOK3E |
437 | /* Note: On fsl the hpdp may be the first of several */ | |
41151e77 | 438 | num_hugepd = (1 << (hugepd_shift(*hpdp) - pdshift)); |
881fde1d BB |
439 | #else |
440 | unsigned int shift = hugepd_shift(*hpdp); | |
41151e77 | 441 | #endif |
a4fe3ce7 DG |
442 | |
443 | start &= pdmask; | |
444 | if (start < floor) | |
445 | return; | |
446 | if (ceiling) { | |
447 | ceiling &= pdmask; | |
448 | if (! ceiling) | |
449 | return; | |
450 | } | |
451 | if (end - 1 > ceiling - 1) | |
452 | return; | |
f10a04c0 | 453 | |
41151e77 BB |
454 | for (i = 0; i < num_hugepd; i++, hpdp++) |
455 | hpdp->pd = 0; | |
456 | ||
f10a04c0 | 457 | tlb->need_flush = 1; |
881fde1d BB |
458 | |
459 | #ifdef CONFIG_PPC_FSL_BOOK3E | |
41151e77 | 460 | hugepd_free(tlb, hugepte); |
881fde1d BB |
461 | #else |
462 | pgtable_free_tlb(tlb, hugepte, pdshift - shift); | |
41151e77 | 463 | #endif |
f10a04c0 DG |
464 | } |
465 | ||
f10a04c0 DG |
466 | static void hugetlb_free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
467 | unsigned long addr, unsigned long end, | |
a4fe3ce7 | 468 | unsigned long floor, unsigned long ceiling) |
f10a04c0 DG |
469 | { |
470 | pmd_t *pmd; | |
471 | unsigned long next; | |
472 | unsigned long start; | |
473 | ||
474 | start = addr; | |
f10a04c0 | 475 | do { |
a1cd5419 | 476 | pmd = pmd_offset(pud, addr); |
f10a04c0 DG |
477 | next = pmd_addr_end(addr, end); |
478 | if (pmd_none(*pmd)) | |
479 | continue; | |
a1cd5419 BB |
480 | #ifdef CONFIG_PPC_FSL_BOOK3E |
481 | /* | |
482 | * Increment next by the size of the huge mapping since | |
483 | * there may be more than one entry at this level for a | |
484 | * single hugepage, but all of them point to | |
485 | * the same kmem cache that holds the hugepte. | |
486 | */ | |
487 | next = addr + (1 << hugepd_shift(*(hugepd_t *)pmd)); | |
488 | #endif | |
a4fe3ce7 DG |
489 | free_hugepd_range(tlb, (hugepd_t *)pmd, PMD_SHIFT, |
490 | addr, next, floor, ceiling); | |
a1cd5419 | 491 | } while (addr = next, addr != end); |
f10a04c0 DG |
492 | |
493 | start &= PUD_MASK; | |
494 | if (start < floor) | |
495 | return; | |
496 | if (ceiling) { | |
497 | ceiling &= PUD_MASK; | |
498 | if (!ceiling) | |
499 | return; | |
1da177e4 | 500 | } |
f10a04c0 DG |
501 | if (end - 1 > ceiling - 1) |
502 | return; | |
1da177e4 | 503 | |
f10a04c0 DG |
504 | pmd = pmd_offset(pud, start); |
505 | pud_clear(pud); | |
9e1b32ca | 506 | pmd_free_tlb(tlb, pmd, start); |
f10a04c0 | 507 | } |
f10a04c0 DG |
508 | |
509 | static void hugetlb_free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, | |
510 | unsigned long addr, unsigned long end, | |
511 | unsigned long floor, unsigned long ceiling) | |
512 | { | |
513 | pud_t *pud; | |
514 | unsigned long next; | |
515 | unsigned long start; | |
516 | ||
517 | start = addr; | |
f10a04c0 | 518 | do { |
a1cd5419 | 519 | pud = pud_offset(pgd, addr); |
f10a04c0 | 520 | next = pud_addr_end(addr, end); |
a4fe3ce7 | 521 | if (!is_hugepd(pud)) { |
4ec161cf JT |
522 | if (pud_none_or_clear_bad(pud)) |
523 | continue; | |
0d9ea754 | 524 | hugetlb_free_pmd_range(tlb, pud, addr, next, floor, |
a4fe3ce7 | 525 | ceiling); |
4ec161cf | 526 | } else { |
a1cd5419 BB |
527 | #ifdef CONFIG_PPC_FSL_BOOK3E |
528 | /* | |
529 | * Increment next by the size of the huge mapping since | |
530 | * there may be more than one entry at this level for a | |
531 | * single hugepage, but all of them point to | |
532 | * the same kmem cache that holds the hugepte. | |
533 | */ | |
534 | next = addr + (1 << hugepd_shift(*(hugepd_t *)pud)); | |
535 | #endif | |
a4fe3ce7 DG |
536 | free_hugepd_range(tlb, (hugepd_t *)pud, PUD_SHIFT, |
537 | addr, next, floor, ceiling); | |
4ec161cf | 538 | } |
a1cd5419 | 539 | } while (addr = next, addr != end); |
f10a04c0 DG |
540 | |
541 | start &= PGDIR_MASK; | |
542 | if (start < floor) | |
543 | return; | |
544 | if (ceiling) { | |
545 | ceiling &= PGDIR_MASK; | |
546 | if (!ceiling) | |
547 | return; | |
548 | } | |
549 | if (end - 1 > ceiling - 1) | |
550 | return; | |
551 | ||
552 | pud = pud_offset(pgd, start); | |
553 | pgd_clear(pgd); | |
9e1b32ca | 554 | pud_free_tlb(tlb, pud, start); |
f10a04c0 DG |
555 | } |
556 | ||
557 | /* | |
558 | * This function frees user-level page tables of a process. | |
559 | * | |
560 | * Must be called with pagetable lock held. | |
561 | */ | |
42b77728 | 562 | void hugetlb_free_pgd_range(struct mmu_gather *tlb, |
f10a04c0 DG |
563 | unsigned long addr, unsigned long end, |
564 | unsigned long floor, unsigned long ceiling) | |
565 | { | |
566 | pgd_t *pgd; | |
567 | unsigned long next; | |
f10a04c0 DG |
568 | |
569 | /* | |
a4fe3ce7 DG |
570 | * Because there are a number of different possible pagetable |
571 | * layouts for hugepage ranges, we limit knowledge of how | |
572 | * things should be laid out to the allocation path | |
573 | * (huge_pte_alloc(), above). Everything else works out the | |
574 | * structure as it goes from information in the hugepd | |
575 | * pointers. That means that we can't here use the | |
576 | * optimization used in the normal page free_pgd_range(), of | |
577 | * checking whether we're actually covering a large enough | |
578 | * range to have to do anything at the top level of the walk | |
579 | * instead of at the bottom. | |
f10a04c0 | 580 | * |
a4fe3ce7 DG |
581 | * To make sense of this, you should probably go read the big |
582 | * block comment at the top of the normal free_pgd_range(), | |
583 | * too. | |
f10a04c0 | 584 | */ |
f10a04c0 | 585 | |
f10a04c0 | 586 | do { |
f10a04c0 | 587 | next = pgd_addr_end(addr, end); |
41151e77 | 588 | pgd = pgd_offset(tlb->mm, addr); |
a4fe3ce7 | 589 | if (!is_hugepd(pgd)) { |
0b26425c DG |
590 | if (pgd_none_or_clear_bad(pgd)) |
591 | continue; | |
592 | hugetlb_free_pud_range(tlb, pgd, addr, next, floor, ceiling); | |
593 | } else { | |
881fde1d | 594 | #ifdef CONFIG_PPC_FSL_BOOK3E |
41151e77 BB |
595 | /* |
596 | * Increment next by the size of the huge mapping since | |
881fde1d BB |
597 | * there may be more than one entry at the pgd level |
598 | * for a single hugepage, but all of them point to the | |
599 | * same kmem cache that holds the hugepte. | |
41151e77 BB |
600 | */ |
601 | next = addr + (1 << hugepd_shift(*(hugepd_t *)pgd)); | |
602 | #endif | |
a4fe3ce7 DG |
603 | free_hugepd_range(tlb, (hugepd_t *)pgd, PGDIR_SHIFT, |
604 | addr, next, floor, ceiling); | |
0b26425c | 605 | } |
41151e77 | 606 | } while (addr = next, addr != end); |
1da177e4 LT |
607 | } |
608 | ||
1da177e4 LT |
609 | struct page * |
610 | follow_huge_addr(struct mm_struct *mm, unsigned long address, int write) | |
611 | { | |
612 | pte_t *ptep; | |
613 | struct page *page; | |
a4fe3ce7 DG |
614 | unsigned shift; |
615 | unsigned long mask; | |
616 | ||
617 | ptep = find_linux_pte_or_hugepte(mm->pgd, address, &shift); | |
1da177e4 | 618 | |
0d9ea754 | 619 | /* Verify it is a huge page else bail. */ |
a4fe3ce7 | 620 | if (!ptep || !shift) |
1da177e4 LT |
621 | return ERR_PTR(-EINVAL); |
622 | ||
a4fe3ce7 | 623 | mask = (1UL << shift) - 1; |
1da177e4 | 624 | page = pte_page(*ptep); |
a4fe3ce7 DG |
625 | if (page) |
626 | page += (address & mask) / PAGE_SIZE; | |
1da177e4 LT |
627 | |
628 | return page; | |
629 | } | |
630 | ||
631 | int pmd_huge(pmd_t pmd) | |
632 | { | |
633 | return 0; | |
634 | } | |
635 | ||
ceb86879 AK |
636 | int pud_huge(pud_t pud) |
637 | { | |
638 | return 0; | |
639 | } | |
640 | ||
1da177e4 LT |
641 | struct page * |
642 | follow_huge_pmd(struct mm_struct *mm, unsigned long address, | |
643 | pmd_t *pmd, int write) | |
644 | { | |
645 | BUG(); | |
646 | return NULL; | |
647 | } | |
648 | ||
a4fe3ce7 DG |
649 | static noinline int gup_hugepte(pte_t *ptep, unsigned long sz, unsigned long addr, |
650 | unsigned long end, int write, struct page **pages, int *nr) | |
651 | { | |
652 | unsigned long mask; | |
653 | unsigned long pte_end; | |
3526741f | 654 | struct page *head, *page, *tail; |
a4fe3ce7 DG |
655 | pte_t pte; |
656 | int refs; | |
657 | ||
658 | pte_end = (addr + sz) & ~(sz-1); | |
659 | if (pte_end < end) | |
660 | end = pte_end; | |
661 | ||
662 | pte = *ptep; | |
663 | mask = _PAGE_PRESENT | _PAGE_USER; | |
664 | if (write) | |
665 | mask |= _PAGE_RW; | |
666 | ||
667 | if ((pte_val(pte) & mask) != mask) | |
668 | return 0; | |
669 | ||
670 | /* hugepages are never "special" */ | |
671 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
672 | ||
673 | refs = 0; | |
674 | head = pte_page(pte); | |
675 | ||
676 | page = head + ((addr & (sz-1)) >> PAGE_SHIFT); | |
3526741f | 677 | tail = page; |
a4fe3ce7 DG |
678 | do { |
679 | VM_BUG_ON(compound_head(page) != head); | |
680 | pages[*nr] = page; | |
681 | (*nr)++; | |
682 | page++; | |
683 | refs++; | |
684 | } while (addr += PAGE_SIZE, addr != end); | |
685 | ||
686 | if (!page_cache_add_speculative(head, refs)) { | |
687 | *nr -= refs; | |
688 | return 0; | |
689 | } | |
690 | ||
691 | if (unlikely(pte_val(pte) != pte_val(*ptep))) { | |
692 | /* Could be optimized better */ | |
85964684 AA |
693 | *nr -= refs; |
694 | while (refs--) | |
405e44f2 | 695 | put_page(head); |
cf592bf7 AA |
696 | return 0; |
697 | } | |
698 | ||
699 | /* | |
700 | * Any tail page need their mapcount reference taken before we | |
701 | * return. | |
702 | */ | |
703 | while (refs--) { | |
704 | if (PageTail(tail)) | |
705 | get_huge_page_tail(tail); | |
706 | tail++; | |
a4fe3ce7 DG |
707 | } |
708 | ||
709 | return 1; | |
710 | } | |
711 | ||
39adfa54 DG |
712 | static unsigned long hugepte_addr_end(unsigned long addr, unsigned long end, |
713 | unsigned long sz) | |
714 | { | |
715 | unsigned long __boundary = (addr + sz) & ~(sz-1); | |
716 | return (__boundary - 1 < end - 1) ? __boundary : end; | |
717 | } | |
718 | ||
a4fe3ce7 DG |
719 | int gup_hugepd(hugepd_t *hugepd, unsigned pdshift, |
720 | unsigned long addr, unsigned long end, | |
721 | int write, struct page **pages, int *nr) | |
722 | { | |
723 | pte_t *ptep; | |
724 | unsigned long sz = 1UL << hugepd_shift(*hugepd); | |
39adfa54 | 725 | unsigned long next; |
a4fe3ce7 DG |
726 | |
727 | ptep = hugepte_offset(hugepd, addr, pdshift); | |
728 | do { | |
39adfa54 | 729 | next = hugepte_addr_end(addr, end, sz); |
a4fe3ce7 DG |
730 | if (!gup_hugepte(ptep, sz, addr, end, write, pages, nr)) |
731 | return 0; | |
39adfa54 | 732 | } while (ptep++, addr = next, addr != end); |
a4fe3ce7 DG |
733 | |
734 | return 1; | |
735 | } | |
1da177e4 | 736 | |
76512959 | 737 | #ifdef CONFIG_PPC_MM_SLICES |
1da177e4 LT |
738 | unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr, |
739 | unsigned long len, unsigned long pgoff, | |
740 | unsigned long flags) | |
741 | { | |
0d9ea754 JT |
742 | struct hstate *hstate = hstate_file(file); |
743 | int mmu_psize = shift_to_mmu_psize(huge_page_shift(hstate)); | |
48f797de | 744 | |
0d9ea754 | 745 | return slice_get_unmapped_area(addr, len, flags, mmu_psize, 1, 0); |
1da177e4 | 746 | } |
76512959 | 747 | #endif |
1da177e4 | 748 | |
3340289d MG |
749 | unsigned long vma_mmu_pagesize(struct vm_area_struct *vma) |
750 | { | |
25c29f9e | 751 | #ifdef CONFIG_PPC_MM_SLICES |
3340289d MG |
752 | unsigned int psize = get_slice_psize(vma->vm_mm, vma->vm_start); |
753 | ||
754 | return 1UL << mmu_psize_to_shift(psize); | |
41151e77 BB |
755 | #else |
756 | if (!is_vm_hugetlb_page(vma)) | |
757 | return PAGE_SIZE; | |
758 | ||
759 | return huge_page_size(hstate_vma(vma)); | |
760 | #endif | |
761 | } | |
762 | ||
763 | static inline bool is_power_of_4(unsigned long x) | |
764 | { | |
765 | if (is_power_of_2(x)) | |
766 | return (__ilog2(x) % 2) ? false : true; | |
767 | return false; | |
3340289d MG |
768 | } |
769 | ||
d1837cba | 770 | static int __init add_huge_page_size(unsigned long long size) |
4ec161cf | 771 | { |
d1837cba DG |
772 | int shift = __ffs(size); |
773 | int mmu_psize; | |
a4fe3ce7 | 774 | |
4ec161cf | 775 | /* Check that it is a page size supported by the hardware and |
d1837cba | 776 | * that it fits within pagetable and slice limits. */ |
41151e77 BB |
777 | #ifdef CONFIG_PPC_FSL_BOOK3E |
778 | if ((size < PAGE_SIZE) || !is_power_of_4(size)) | |
779 | return -EINVAL; | |
780 | #else | |
d1837cba DG |
781 | if (!is_power_of_2(size) |
782 | || (shift > SLICE_HIGH_SHIFT) || (shift <= PAGE_SHIFT)) | |
783 | return -EINVAL; | |
41151e77 | 784 | #endif |
91224346 | 785 | |
d1837cba DG |
786 | if ((mmu_psize = shift_to_mmu_psize(shift)) < 0) |
787 | return -EINVAL; | |
788 | ||
789 | #ifdef CONFIG_SPU_FS_64K_LS | |
790 | /* Disable support for 64K huge pages when 64K SPU local store | |
791 | * support is enabled as the current implementation conflicts. | |
792 | */ | |
793 | if (shift == PAGE_SHIFT_64K) | |
794 | return -EINVAL; | |
795 | #endif /* CONFIG_SPU_FS_64K_LS */ | |
796 | ||
797 | BUG_ON(mmu_psize_defs[mmu_psize].shift != shift); | |
798 | ||
799 | /* Return if huge page size has already been setup */ | |
800 | if (size_to_hstate(size)) | |
801 | return 0; | |
802 | ||
803 | hugetlb_add_hstate(shift - PAGE_SHIFT); | |
804 | ||
805 | return 0; | |
4ec161cf JT |
806 | } |
807 | ||
808 | static int __init hugepage_setup_sz(char *str) | |
809 | { | |
810 | unsigned long long size; | |
4ec161cf JT |
811 | |
812 | size = memparse(str, &str); | |
813 | ||
d1837cba | 814 | if (add_huge_page_size(size) != 0) |
4ec161cf JT |
815 | printk(KERN_WARNING "Invalid huge page size specified(%llu)\n", size); |
816 | ||
817 | return 1; | |
818 | } | |
819 | __setup("hugepagesz=", hugepage_setup_sz); | |
820 | ||
881fde1d | 821 | #ifdef CONFIG_PPC_FSL_BOOK3E |
41151e77 BB |
822 | struct kmem_cache *hugepte_cache; |
823 | static int __init hugetlbpage_init(void) | |
824 | { | |
825 | int psize; | |
826 | ||
827 | for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) { | |
828 | unsigned shift; | |
829 | ||
830 | if (!mmu_psize_defs[psize].shift) | |
831 | continue; | |
832 | ||
833 | shift = mmu_psize_to_shift(psize); | |
834 | ||
835 | /* Don't treat normal page sizes as huge... */ | |
836 | if (shift != PAGE_SHIFT) | |
837 | if (add_huge_page_size(1ULL << shift) < 0) | |
838 | continue; | |
839 | } | |
840 | ||
841 | /* | |
842 | * Create a kmem cache for hugeptes. The bottom bits in the pte have | |
843 | * size information encoded in them, so align them to allow this | |
844 | */ | |
845 | hugepte_cache = kmem_cache_create("hugepte-cache", sizeof(pte_t), | |
846 | HUGEPD_SHIFT_MASK + 1, 0, NULL); | |
847 | if (hugepte_cache == NULL) | |
848 | panic("%s: Unable to create kmem cache for hugeptes\n", | |
849 | __func__); | |
850 | ||
851 | /* Default hpage size = 4M */ | |
852 | if (mmu_psize_defs[MMU_PAGE_4M].shift) | |
853 | HPAGE_SHIFT = mmu_psize_defs[MMU_PAGE_4M].shift; | |
854 | else | |
855 | panic("%s: Unable to set default huge page size\n", __func__); | |
856 | ||
857 | ||
858 | return 0; | |
859 | } | |
860 | #else | |
f10a04c0 DG |
861 | static int __init hugetlbpage_init(void) |
862 | { | |
a4fe3ce7 | 863 | int psize; |
0d9ea754 | 864 | |
44ae3ab3 | 865 | if (!mmu_has_feature(MMU_FTR_16M_PAGE)) |
f10a04c0 | 866 | return -ENODEV; |
00df438e | 867 | |
d1837cba DG |
868 | for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) { |
869 | unsigned shift; | |
870 | unsigned pdshift; | |
0d9ea754 | 871 | |
d1837cba DG |
872 | if (!mmu_psize_defs[psize].shift) |
873 | continue; | |
00df438e | 874 | |
d1837cba DG |
875 | shift = mmu_psize_to_shift(psize); |
876 | ||
877 | if (add_huge_page_size(1ULL << shift) < 0) | |
878 | continue; | |
879 | ||
880 | if (shift < PMD_SHIFT) | |
881 | pdshift = PMD_SHIFT; | |
882 | else if (shift < PUD_SHIFT) | |
883 | pdshift = PUD_SHIFT; | |
884 | else | |
885 | pdshift = PGDIR_SHIFT; | |
886 | ||
887 | pgtable_cache_add(pdshift - shift, NULL); | |
888 | if (!PGT_CACHE(pdshift - shift)) | |
889 | panic("hugetlbpage_init(): could not create " | |
890 | "pgtable cache for %d bit pagesize\n", shift); | |
0d9ea754 | 891 | } |
f10a04c0 | 892 | |
d1837cba DG |
893 | /* Set default large page size. Currently, we pick 16M or 1M |
894 | * depending on what is available | |
895 | */ | |
896 | if (mmu_psize_defs[MMU_PAGE_16M].shift) | |
897 | HPAGE_SHIFT = mmu_psize_defs[MMU_PAGE_16M].shift; | |
898 | else if (mmu_psize_defs[MMU_PAGE_1M].shift) | |
899 | HPAGE_SHIFT = mmu_psize_defs[MMU_PAGE_1M].shift; | |
900 | ||
f10a04c0 DG |
901 | return 0; |
902 | } | |
41151e77 | 903 | #endif |
f10a04c0 | 904 | module_init(hugetlbpage_init); |
0895ecda DG |
905 | |
906 | void flush_dcache_icache_hugepage(struct page *page) | |
907 | { | |
908 | int i; | |
41151e77 | 909 | void *start; |
0895ecda DG |
910 | |
911 | BUG_ON(!PageCompound(page)); | |
912 | ||
41151e77 BB |
913 | for (i = 0; i < (1UL << compound_order(page)); i++) { |
914 | if (!PageHighMem(page)) { | |
915 | __flush_dcache_icache(page_address(page+i)); | |
916 | } else { | |
2480b208 | 917 | start = kmap_atomic(page+i); |
41151e77 | 918 | __flush_dcache_icache(start); |
2480b208 | 919 | kunmap_atomic(start); |
41151e77 BB |
920 | } |
921 | } | |
0895ecda | 922 | } |