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 | |
a4fe3ce7 DG |
51 | #define hugepd_none(hpd) ((hpd).pd == 0) |
52 | ||
a4fe3ce7 DG |
53 | pte_t *find_linux_pte_or_hugepte(pgd_t *pgdir, unsigned long ea, unsigned *shift) |
54 | { | |
55 | pgd_t *pg; | |
56 | pud_t *pu; | |
57 | pmd_t *pm; | |
58 | hugepd_t *hpdp = NULL; | |
59 | unsigned pdshift = PGDIR_SHIFT; | |
60 | ||
61 | if (shift) | |
62 | *shift = 0; | |
63 | ||
64 | pg = pgdir + pgd_index(ea); | |
65 | if (is_hugepd(pg)) { | |
66 | hpdp = (hugepd_t *)pg; | |
67 | } else if (!pgd_none(*pg)) { | |
68 | pdshift = PUD_SHIFT; | |
69 | pu = pud_offset(pg, ea); | |
70 | if (is_hugepd(pu)) | |
71 | hpdp = (hugepd_t *)pu; | |
72 | else if (!pud_none(*pu)) { | |
73 | pdshift = PMD_SHIFT; | |
74 | pm = pmd_offset(pu, ea); | |
75 | if (is_hugepd(pm)) | |
76 | hpdp = (hugepd_t *)pm; | |
77 | else if (!pmd_none(*pm)) { | |
41151e77 | 78 | return pte_offset_kernel(pm, ea); |
a4fe3ce7 DG |
79 | } |
80 | } | |
81 | } | |
82 | ||
83 | if (!hpdp) | |
84 | return NULL; | |
85 | ||
86 | if (shift) | |
87 | *shift = hugepd_shift(*hpdp); | |
88 | return hugepte_offset(hpdp, ea, pdshift); | |
89 | } | |
342d3db7 | 90 | EXPORT_SYMBOL_GPL(find_linux_pte_or_hugepte); |
a4fe3ce7 DG |
91 | |
92 | pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr) | |
93 | { | |
94 | return find_linux_pte_or_hugepte(mm->pgd, addr, NULL); | |
95 | } | |
96 | ||
f10a04c0 | 97 | static int __hugepte_alloc(struct mm_struct *mm, hugepd_t *hpdp, |
a4fe3ce7 | 98 | unsigned long address, unsigned pdshift, unsigned pshift) |
f10a04c0 | 99 | { |
41151e77 BB |
100 | struct kmem_cache *cachep; |
101 | pte_t *new; | |
102 | ||
881fde1d | 103 | #ifdef CONFIG_PPC_FSL_BOOK3E |
41151e77 BB |
104 | int i; |
105 | int num_hugepd = 1 << (pshift - pdshift); | |
106 | cachep = hugepte_cache; | |
881fde1d BB |
107 | #else |
108 | cachep = PGT_CACHE(pdshift - pshift); | |
41151e77 BB |
109 | #endif |
110 | ||
111 | new = kmem_cache_zalloc(cachep, GFP_KERNEL|__GFP_REPEAT); | |
f10a04c0 | 112 | |
a4fe3ce7 DG |
113 | BUG_ON(pshift > HUGEPD_SHIFT_MASK); |
114 | BUG_ON((unsigned long)new & HUGEPD_SHIFT_MASK); | |
115 | ||
f10a04c0 DG |
116 | if (! new) |
117 | return -ENOMEM; | |
118 | ||
119 | spin_lock(&mm->page_table_lock); | |
881fde1d | 120 | #ifdef CONFIG_PPC_FSL_BOOK3E |
41151e77 BB |
121 | /* |
122 | * We have multiple higher-level entries that point to the same | |
123 | * actual pte location. Fill in each as we go and backtrack on error. | |
124 | * We need all of these so the DTLB pgtable walk code can find the | |
125 | * right higher-level entry without knowing if it's a hugepage or not. | |
126 | */ | |
127 | for (i = 0; i < num_hugepd; i++, hpdp++) { | |
128 | if (unlikely(!hugepd_none(*hpdp))) | |
129 | break; | |
130 | else | |
cf9427b8 | 131 | /* We use the old format for PPC_FSL_BOOK3E */ |
41151e77 BB |
132 | hpdp->pd = ((unsigned long)new & ~PD_HUGE) | pshift; |
133 | } | |
134 | /* If we bailed from the for loop early, an error occurred, clean up */ | |
135 | if (i < num_hugepd) { | |
136 | for (i = i - 1 ; i >= 0; i--, hpdp--) | |
137 | hpdp->pd = 0; | |
138 | kmem_cache_free(cachep, new); | |
139 | } | |
a1cd5419 BB |
140 | #else |
141 | if (!hugepd_none(*hpdp)) | |
142 | kmem_cache_free(cachep, new); | |
cf9427b8 AK |
143 | else { |
144 | #ifdef CONFIG_PPC_BOOK3S_64 | |
145 | hpdp->pd = (unsigned long)new | | |
146 | (shift_to_mmu_psize(pshift) << 2); | |
147 | #else | |
a1cd5419 | 148 | hpdp->pd = ((unsigned long)new & ~PD_HUGE) | pshift; |
cf9427b8 AK |
149 | #endif |
150 | } | |
41151e77 | 151 | #endif |
f10a04c0 DG |
152 | spin_unlock(&mm->page_table_lock); |
153 | return 0; | |
154 | } | |
155 | ||
a1cd5419 BB |
156 | /* |
157 | * These macros define how to determine which level of the page table holds | |
158 | * the hpdp. | |
159 | */ | |
160 | #ifdef CONFIG_PPC_FSL_BOOK3E | |
161 | #define HUGEPD_PGD_SHIFT PGDIR_SHIFT | |
162 | #define HUGEPD_PUD_SHIFT PUD_SHIFT | |
163 | #else | |
164 | #define HUGEPD_PGD_SHIFT PUD_SHIFT | |
165 | #define HUGEPD_PUD_SHIFT PMD_SHIFT | |
166 | #endif | |
167 | ||
a4fe3ce7 | 168 | pte_t *huge_pte_alloc(struct mm_struct *mm, unsigned long addr, unsigned long sz) |
0b26425c | 169 | { |
a4fe3ce7 DG |
170 | pgd_t *pg; |
171 | pud_t *pu; | |
172 | pmd_t *pm; | |
173 | hugepd_t *hpdp = NULL; | |
174 | unsigned pshift = __ffs(sz); | |
175 | unsigned pdshift = PGDIR_SHIFT; | |
176 | ||
177 | addr &= ~(sz-1); | |
178 | ||
179 | pg = pgd_offset(mm, addr); | |
a1cd5419 BB |
180 | |
181 | if (pshift >= HUGEPD_PGD_SHIFT) { | |
a4fe3ce7 DG |
182 | hpdp = (hugepd_t *)pg; |
183 | } else { | |
184 | pdshift = PUD_SHIFT; | |
185 | pu = pud_alloc(mm, pg, addr); | |
a1cd5419 | 186 | if (pshift >= HUGEPD_PUD_SHIFT) { |
a4fe3ce7 DG |
187 | hpdp = (hugepd_t *)pu; |
188 | } else { | |
189 | pdshift = PMD_SHIFT; | |
190 | pm = pmd_alloc(mm, pu, addr); | |
191 | hpdp = (hugepd_t *)pm; | |
192 | } | |
193 | } | |
194 | ||
195 | if (!hpdp) | |
196 | return NULL; | |
197 | ||
198 | BUG_ON(!hugepd_none(*hpdp) && !hugepd_ok(*hpdp)); | |
199 | ||
200 | if (hugepd_none(*hpdp) && __hugepte_alloc(mm, hpdp, addr, pdshift, pshift)) | |
201 | return NULL; | |
202 | ||
203 | return hugepte_offset(hpdp, addr, pdshift); | |
4ec161cf | 204 | } |
4ec161cf | 205 | |
881fde1d | 206 | #ifdef CONFIG_PPC_FSL_BOOK3E |
658013e9 JT |
207 | /* Build list of addresses of gigantic pages. This function is used in early |
208 | * boot before the buddy or bootmem allocator is setup. | |
209 | */ | |
41151e77 BB |
210 | void add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages) |
211 | { | |
212 | unsigned int idx = shift_to_mmu_psize(__ffs(page_size)); | |
213 | int i; | |
214 | ||
215 | if (addr == 0) | |
216 | return; | |
217 | ||
218 | gpage_freearray[idx].nr_gpages = number_of_pages; | |
219 | ||
220 | for (i = 0; i < number_of_pages; i++) { | |
221 | gpage_freearray[idx].gpage_list[i] = addr; | |
222 | addr += page_size; | |
223 | } | |
224 | } | |
225 | ||
226 | /* | |
227 | * Moves the gigantic page addresses from the temporary list to the | |
228 | * huge_boot_pages list. | |
229 | */ | |
230 | int alloc_bootmem_huge_page(struct hstate *hstate) | |
231 | { | |
232 | struct huge_bootmem_page *m; | |
233 | int idx = shift_to_mmu_psize(hstate->order + PAGE_SHIFT); | |
234 | int nr_gpages = gpage_freearray[idx].nr_gpages; | |
235 | ||
236 | if (nr_gpages == 0) | |
237 | return 0; | |
238 | ||
239 | #ifdef CONFIG_HIGHMEM | |
240 | /* | |
241 | * If gpages can be in highmem we can't use the trick of storing the | |
242 | * data structure in the page; allocate space for this | |
243 | */ | |
244 | m = alloc_bootmem(sizeof(struct huge_bootmem_page)); | |
245 | m->phys = gpage_freearray[idx].gpage_list[--nr_gpages]; | |
246 | #else | |
247 | m = phys_to_virt(gpage_freearray[idx].gpage_list[--nr_gpages]); | |
248 | #endif | |
249 | ||
250 | list_add(&m->list, &huge_boot_pages); | |
251 | gpage_freearray[idx].nr_gpages = nr_gpages; | |
252 | gpage_freearray[idx].gpage_list[nr_gpages] = 0; | |
253 | m->hstate = hstate; | |
254 | ||
255 | return 1; | |
256 | } | |
257 | /* | |
258 | * Scan the command line hugepagesz= options for gigantic pages; store those in | |
259 | * a list that we use to allocate the memory once all options are parsed. | |
260 | */ | |
261 | ||
262 | unsigned long gpage_npages[MMU_PAGE_COUNT]; | |
263 | ||
89528127 PG |
264 | static int __init do_gpage_early_setup(char *param, char *val, |
265 | const char *unused) | |
41151e77 BB |
266 | { |
267 | static phys_addr_t size; | |
268 | unsigned long npages; | |
269 | ||
270 | /* | |
271 | * The hugepagesz and hugepages cmdline options are interleaved. We | |
272 | * use the size variable to keep track of whether or not this was done | |
273 | * properly and skip over instances where it is incorrect. Other | |
274 | * command-line parsing code will issue warnings, so we don't need to. | |
275 | * | |
276 | */ | |
277 | if ((strcmp(param, "default_hugepagesz") == 0) || | |
278 | (strcmp(param, "hugepagesz") == 0)) { | |
279 | size = memparse(val, NULL); | |
280 | } else if (strcmp(param, "hugepages") == 0) { | |
281 | if (size != 0) { | |
282 | if (sscanf(val, "%lu", &npages) <= 0) | |
283 | npages = 0; | |
284 | gpage_npages[shift_to_mmu_psize(__ffs(size))] = npages; | |
285 | size = 0; | |
286 | } | |
287 | } | |
288 | return 0; | |
289 | } | |
290 | ||
291 | ||
292 | /* | |
293 | * This function allocates physical space for pages that are larger than the | |
294 | * buddy allocator can handle. We want to allocate these in highmem because | |
295 | * the amount of lowmem is limited. This means that this function MUST be | |
296 | * called before lowmem_end_addr is set up in MMU_init() in order for the lmb | |
297 | * allocate to grab highmem. | |
298 | */ | |
299 | void __init reserve_hugetlb_gpages(void) | |
300 | { | |
301 | static __initdata char cmdline[COMMAND_LINE_SIZE]; | |
302 | phys_addr_t size, base; | |
303 | int i; | |
304 | ||
305 | strlcpy(cmdline, boot_command_line, COMMAND_LINE_SIZE); | |
026cee00 PM |
306 | parse_args("hugetlb gpages", cmdline, NULL, 0, 0, 0, |
307 | &do_gpage_early_setup); | |
41151e77 BB |
308 | |
309 | /* | |
310 | * Walk gpage list in reverse, allocating larger page sizes first. | |
311 | * Skip over unsupported sizes, or sizes that have 0 gpages allocated. | |
312 | * When we reach the point in the list where pages are no longer | |
313 | * considered gpages, we're done. | |
314 | */ | |
315 | for (i = MMU_PAGE_COUNT-1; i >= 0; i--) { | |
316 | if (mmu_psize_defs[i].shift == 0 || gpage_npages[i] == 0) | |
317 | continue; | |
318 | else if (mmu_psize_to_shift(i) < (MAX_ORDER + PAGE_SHIFT)) | |
319 | break; | |
320 | ||
321 | size = (phys_addr_t)(1ULL << mmu_psize_to_shift(i)); | |
322 | base = memblock_alloc_base(size * gpage_npages[i], size, | |
323 | MEMBLOCK_ALLOC_ANYWHERE); | |
324 | add_gpage(base, size, gpage_npages[i]); | |
325 | } | |
326 | } | |
327 | ||
881fde1d | 328 | #else /* !PPC_FSL_BOOK3E */ |
41151e77 BB |
329 | |
330 | /* Build list of addresses of gigantic pages. This function is used in early | |
331 | * boot before the buddy or bootmem allocator is setup. | |
332 | */ | |
333 | void add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages) | |
658013e9 JT |
334 | { |
335 | if (!addr) | |
336 | return; | |
337 | while (number_of_pages > 0) { | |
338 | gpage_freearray[nr_gpages] = addr; | |
339 | nr_gpages++; | |
340 | number_of_pages--; | |
341 | addr += page_size; | |
342 | } | |
343 | } | |
344 | ||
ec4b2c0c | 345 | /* Moves the gigantic page addresses from the temporary list to the |
0d9ea754 JT |
346 | * huge_boot_pages list. |
347 | */ | |
348 | int alloc_bootmem_huge_page(struct hstate *hstate) | |
ec4b2c0c JT |
349 | { |
350 | struct huge_bootmem_page *m; | |
351 | if (nr_gpages == 0) | |
352 | return 0; | |
353 | m = phys_to_virt(gpage_freearray[--nr_gpages]); | |
354 | gpage_freearray[nr_gpages] = 0; | |
355 | list_add(&m->list, &huge_boot_pages); | |
0d9ea754 | 356 | m->hstate = hstate; |
ec4b2c0c JT |
357 | return 1; |
358 | } | |
41151e77 | 359 | #endif |
ec4b2c0c | 360 | |
39dde65c CK |
361 | int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep) |
362 | { | |
363 | return 0; | |
364 | } | |
365 | ||
881fde1d | 366 | #ifdef CONFIG_PPC_FSL_BOOK3E |
41151e77 BB |
367 | #define HUGEPD_FREELIST_SIZE \ |
368 | ((PAGE_SIZE - sizeof(struct hugepd_freelist)) / sizeof(pte_t)) | |
369 | ||
370 | struct hugepd_freelist { | |
371 | struct rcu_head rcu; | |
372 | unsigned int index; | |
373 | void *ptes[0]; | |
374 | }; | |
375 | ||
376 | static DEFINE_PER_CPU(struct hugepd_freelist *, hugepd_freelist_cur); | |
377 | ||
378 | static void hugepd_free_rcu_callback(struct rcu_head *head) | |
379 | { | |
380 | struct hugepd_freelist *batch = | |
381 | container_of(head, struct hugepd_freelist, rcu); | |
382 | unsigned int i; | |
383 | ||
384 | for (i = 0; i < batch->index; i++) | |
385 | kmem_cache_free(hugepte_cache, batch->ptes[i]); | |
386 | ||
387 | free_page((unsigned long)batch); | |
388 | } | |
389 | ||
390 | static void hugepd_free(struct mmu_gather *tlb, void *hugepte) | |
391 | { | |
392 | struct hugepd_freelist **batchp; | |
393 | ||
394 | batchp = &__get_cpu_var(hugepd_freelist_cur); | |
395 | ||
396 | if (atomic_read(&tlb->mm->mm_users) < 2 || | |
397 | cpumask_equal(mm_cpumask(tlb->mm), | |
398 | cpumask_of(smp_processor_id()))) { | |
399 | kmem_cache_free(hugepte_cache, hugepte); | |
400 | return; | |
401 | } | |
402 | ||
403 | if (*batchp == NULL) { | |
404 | *batchp = (struct hugepd_freelist *)__get_free_page(GFP_ATOMIC); | |
405 | (*batchp)->index = 0; | |
406 | } | |
407 | ||
408 | (*batchp)->ptes[(*batchp)->index++] = hugepte; | |
409 | if ((*batchp)->index == HUGEPD_FREELIST_SIZE) { | |
410 | call_rcu_sched(&(*batchp)->rcu, hugepd_free_rcu_callback); | |
411 | *batchp = NULL; | |
412 | } | |
413 | } | |
414 | #endif | |
415 | ||
a4fe3ce7 DG |
416 | static void free_hugepd_range(struct mmu_gather *tlb, hugepd_t *hpdp, int pdshift, |
417 | unsigned long start, unsigned long end, | |
418 | unsigned long floor, unsigned long ceiling) | |
f10a04c0 DG |
419 | { |
420 | pte_t *hugepte = hugepd_page(*hpdp); | |
41151e77 BB |
421 | int i; |
422 | ||
a4fe3ce7 | 423 | unsigned long pdmask = ~((1UL << pdshift) - 1); |
41151e77 BB |
424 | unsigned int num_hugepd = 1; |
425 | ||
881fde1d BB |
426 | #ifdef CONFIG_PPC_FSL_BOOK3E |
427 | /* Note: On fsl the hpdp may be the first of several */ | |
41151e77 | 428 | num_hugepd = (1 << (hugepd_shift(*hpdp) - pdshift)); |
881fde1d BB |
429 | #else |
430 | unsigned int shift = hugepd_shift(*hpdp); | |
41151e77 | 431 | #endif |
a4fe3ce7 DG |
432 | |
433 | start &= pdmask; | |
434 | if (start < floor) | |
435 | return; | |
436 | if (ceiling) { | |
437 | ceiling &= pdmask; | |
438 | if (! ceiling) | |
439 | return; | |
440 | } | |
441 | if (end - 1 > ceiling - 1) | |
442 | return; | |
f10a04c0 | 443 | |
41151e77 BB |
444 | for (i = 0; i < num_hugepd; i++, hpdp++) |
445 | hpdp->pd = 0; | |
446 | ||
f10a04c0 | 447 | tlb->need_flush = 1; |
881fde1d BB |
448 | |
449 | #ifdef CONFIG_PPC_FSL_BOOK3E | |
41151e77 | 450 | hugepd_free(tlb, hugepte); |
881fde1d BB |
451 | #else |
452 | pgtable_free_tlb(tlb, hugepte, pdshift - shift); | |
41151e77 | 453 | #endif |
f10a04c0 DG |
454 | } |
455 | ||
f10a04c0 DG |
456 | static void hugetlb_free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
457 | unsigned long addr, unsigned long end, | |
a4fe3ce7 | 458 | unsigned long floor, unsigned long ceiling) |
f10a04c0 DG |
459 | { |
460 | pmd_t *pmd; | |
461 | unsigned long next; | |
462 | unsigned long start; | |
463 | ||
464 | start = addr; | |
f10a04c0 | 465 | do { |
a1cd5419 | 466 | pmd = pmd_offset(pud, addr); |
f10a04c0 DG |
467 | next = pmd_addr_end(addr, end); |
468 | if (pmd_none(*pmd)) | |
469 | continue; | |
a1cd5419 BB |
470 | #ifdef CONFIG_PPC_FSL_BOOK3E |
471 | /* | |
472 | * Increment next by the size of the huge mapping since | |
473 | * there may be more than one entry at this level for a | |
474 | * single hugepage, but all of them point to | |
475 | * the same kmem cache that holds the hugepte. | |
476 | */ | |
477 | next = addr + (1 << hugepd_shift(*(hugepd_t *)pmd)); | |
478 | #endif | |
a4fe3ce7 DG |
479 | free_hugepd_range(tlb, (hugepd_t *)pmd, PMD_SHIFT, |
480 | addr, next, floor, ceiling); | |
a1cd5419 | 481 | } while (addr = next, addr != end); |
f10a04c0 DG |
482 | |
483 | start &= PUD_MASK; | |
484 | if (start < floor) | |
485 | return; | |
486 | if (ceiling) { | |
487 | ceiling &= PUD_MASK; | |
488 | if (!ceiling) | |
489 | return; | |
1da177e4 | 490 | } |
f10a04c0 DG |
491 | if (end - 1 > ceiling - 1) |
492 | return; | |
1da177e4 | 493 | |
f10a04c0 DG |
494 | pmd = pmd_offset(pud, start); |
495 | pud_clear(pud); | |
9e1b32ca | 496 | pmd_free_tlb(tlb, pmd, start); |
f10a04c0 | 497 | } |
f10a04c0 DG |
498 | |
499 | static void hugetlb_free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, | |
500 | unsigned long addr, unsigned long end, | |
501 | unsigned long floor, unsigned long ceiling) | |
502 | { | |
503 | pud_t *pud; | |
504 | unsigned long next; | |
505 | unsigned long start; | |
506 | ||
507 | start = addr; | |
f10a04c0 | 508 | do { |
a1cd5419 | 509 | pud = pud_offset(pgd, addr); |
f10a04c0 | 510 | next = pud_addr_end(addr, end); |
a4fe3ce7 | 511 | if (!is_hugepd(pud)) { |
4ec161cf JT |
512 | if (pud_none_or_clear_bad(pud)) |
513 | continue; | |
0d9ea754 | 514 | hugetlb_free_pmd_range(tlb, pud, addr, next, floor, |
a4fe3ce7 | 515 | ceiling); |
4ec161cf | 516 | } else { |
a1cd5419 BB |
517 | #ifdef CONFIG_PPC_FSL_BOOK3E |
518 | /* | |
519 | * Increment next by the size of the huge mapping since | |
520 | * there may be more than one entry at this level for a | |
521 | * single hugepage, but all of them point to | |
522 | * the same kmem cache that holds the hugepte. | |
523 | */ | |
524 | next = addr + (1 << hugepd_shift(*(hugepd_t *)pud)); | |
525 | #endif | |
a4fe3ce7 DG |
526 | free_hugepd_range(tlb, (hugepd_t *)pud, PUD_SHIFT, |
527 | addr, next, floor, ceiling); | |
4ec161cf | 528 | } |
a1cd5419 | 529 | } while (addr = next, addr != end); |
f10a04c0 DG |
530 | |
531 | start &= PGDIR_MASK; | |
532 | if (start < floor) | |
533 | return; | |
534 | if (ceiling) { | |
535 | ceiling &= PGDIR_MASK; | |
536 | if (!ceiling) | |
537 | return; | |
538 | } | |
539 | if (end - 1 > ceiling - 1) | |
540 | return; | |
541 | ||
542 | pud = pud_offset(pgd, start); | |
543 | pgd_clear(pgd); | |
9e1b32ca | 544 | pud_free_tlb(tlb, pud, start); |
f10a04c0 DG |
545 | } |
546 | ||
547 | /* | |
548 | * This function frees user-level page tables of a process. | |
549 | * | |
550 | * Must be called with pagetable lock held. | |
551 | */ | |
42b77728 | 552 | void hugetlb_free_pgd_range(struct mmu_gather *tlb, |
f10a04c0 DG |
553 | unsigned long addr, unsigned long end, |
554 | unsigned long floor, unsigned long ceiling) | |
555 | { | |
556 | pgd_t *pgd; | |
557 | unsigned long next; | |
f10a04c0 DG |
558 | |
559 | /* | |
a4fe3ce7 DG |
560 | * Because there are a number of different possible pagetable |
561 | * layouts for hugepage ranges, we limit knowledge of how | |
562 | * things should be laid out to the allocation path | |
563 | * (huge_pte_alloc(), above). Everything else works out the | |
564 | * structure as it goes from information in the hugepd | |
565 | * pointers. That means that we can't here use the | |
566 | * optimization used in the normal page free_pgd_range(), of | |
567 | * checking whether we're actually covering a large enough | |
568 | * range to have to do anything at the top level of the walk | |
569 | * instead of at the bottom. | |
f10a04c0 | 570 | * |
a4fe3ce7 DG |
571 | * To make sense of this, you should probably go read the big |
572 | * block comment at the top of the normal free_pgd_range(), | |
573 | * too. | |
f10a04c0 | 574 | */ |
f10a04c0 | 575 | |
f10a04c0 | 576 | do { |
f10a04c0 | 577 | next = pgd_addr_end(addr, end); |
41151e77 | 578 | pgd = pgd_offset(tlb->mm, addr); |
a4fe3ce7 | 579 | if (!is_hugepd(pgd)) { |
0b26425c DG |
580 | if (pgd_none_or_clear_bad(pgd)) |
581 | continue; | |
582 | hugetlb_free_pud_range(tlb, pgd, addr, next, floor, ceiling); | |
583 | } else { | |
881fde1d | 584 | #ifdef CONFIG_PPC_FSL_BOOK3E |
41151e77 BB |
585 | /* |
586 | * Increment next by the size of the huge mapping since | |
881fde1d BB |
587 | * there may be more than one entry at the pgd level |
588 | * for a single hugepage, but all of them point to the | |
589 | * same kmem cache that holds the hugepte. | |
41151e77 BB |
590 | */ |
591 | next = addr + (1 << hugepd_shift(*(hugepd_t *)pgd)); | |
592 | #endif | |
a4fe3ce7 DG |
593 | free_hugepd_range(tlb, (hugepd_t *)pgd, PGDIR_SHIFT, |
594 | addr, next, floor, ceiling); | |
0b26425c | 595 | } |
41151e77 | 596 | } while (addr = next, addr != end); |
1da177e4 LT |
597 | } |
598 | ||
1da177e4 LT |
599 | struct page * |
600 | follow_huge_addr(struct mm_struct *mm, unsigned long address, int write) | |
601 | { | |
602 | pte_t *ptep; | |
603 | struct page *page; | |
a4fe3ce7 DG |
604 | unsigned shift; |
605 | unsigned long mask; | |
606 | ||
607 | ptep = find_linux_pte_or_hugepte(mm->pgd, address, &shift); | |
1da177e4 | 608 | |
0d9ea754 | 609 | /* Verify it is a huge page else bail. */ |
a4fe3ce7 | 610 | if (!ptep || !shift) |
1da177e4 LT |
611 | return ERR_PTR(-EINVAL); |
612 | ||
a4fe3ce7 | 613 | mask = (1UL << shift) - 1; |
1da177e4 | 614 | page = pte_page(*ptep); |
a4fe3ce7 DG |
615 | if (page) |
616 | page += (address & mask) / PAGE_SIZE; | |
1da177e4 LT |
617 | |
618 | return page; | |
619 | } | |
620 | ||
621 | int pmd_huge(pmd_t pmd) | |
622 | { | |
623 | return 0; | |
624 | } | |
625 | ||
ceb86879 AK |
626 | int pud_huge(pud_t pud) |
627 | { | |
628 | return 0; | |
629 | } | |
630 | ||
1da177e4 LT |
631 | struct page * |
632 | follow_huge_pmd(struct mm_struct *mm, unsigned long address, | |
633 | pmd_t *pmd, int write) | |
634 | { | |
635 | BUG(); | |
636 | return NULL; | |
637 | } | |
638 | ||
a4fe3ce7 DG |
639 | static noinline int gup_hugepte(pte_t *ptep, unsigned long sz, unsigned long addr, |
640 | unsigned long end, int write, struct page **pages, int *nr) | |
641 | { | |
642 | unsigned long mask; | |
643 | unsigned long pte_end; | |
3526741f | 644 | struct page *head, *page, *tail; |
a4fe3ce7 DG |
645 | pte_t pte; |
646 | int refs; | |
647 | ||
648 | pte_end = (addr + sz) & ~(sz-1); | |
649 | if (pte_end < end) | |
650 | end = pte_end; | |
651 | ||
652 | pte = *ptep; | |
653 | mask = _PAGE_PRESENT | _PAGE_USER; | |
654 | if (write) | |
655 | mask |= _PAGE_RW; | |
656 | ||
657 | if ((pte_val(pte) & mask) != mask) | |
658 | return 0; | |
659 | ||
660 | /* hugepages are never "special" */ | |
661 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
662 | ||
663 | refs = 0; | |
664 | head = pte_page(pte); | |
665 | ||
666 | page = head + ((addr & (sz-1)) >> PAGE_SHIFT); | |
3526741f | 667 | tail = page; |
a4fe3ce7 DG |
668 | do { |
669 | VM_BUG_ON(compound_head(page) != head); | |
670 | pages[*nr] = page; | |
671 | (*nr)++; | |
672 | page++; | |
673 | refs++; | |
674 | } while (addr += PAGE_SIZE, addr != end); | |
675 | ||
676 | if (!page_cache_add_speculative(head, refs)) { | |
677 | *nr -= refs; | |
678 | return 0; | |
679 | } | |
680 | ||
681 | if (unlikely(pte_val(pte) != pte_val(*ptep))) { | |
682 | /* Could be optimized better */ | |
85964684 AA |
683 | *nr -= refs; |
684 | while (refs--) | |
405e44f2 | 685 | put_page(head); |
cf592bf7 AA |
686 | return 0; |
687 | } | |
688 | ||
689 | /* | |
690 | * Any tail page need their mapcount reference taken before we | |
691 | * return. | |
692 | */ | |
693 | while (refs--) { | |
694 | if (PageTail(tail)) | |
695 | get_huge_page_tail(tail); | |
696 | tail++; | |
a4fe3ce7 DG |
697 | } |
698 | ||
699 | return 1; | |
700 | } | |
701 | ||
39adfa54 DG |
702 | static unsigned long hugepte_addr_end(unsigned long addr, unsigned long end, |
703 | unsigned long sz) | |
704 | { | |
705 | unsigned long __boundary = (addr + sz) & ~(sz-1); | |
706 | return (__boundary - 1 < end - 1) ? __boundary : end; | |
707 | } | |
708 | ||
a4fe3ce7 DG |
709 | int gup_hugepd(hugepd_t *hugepd, unsigned pdshift, |
710 | unsigned long addr, unsigned long end, | |
711 | int write, struct page **pages, int *nr) | |
712 | { | |
713 | pte_t *ptep; | |
714 | unsigned long sz = 1UL << hugepd_shift(*hugepd); | |
39adfa54 | 715 | unsigned long next; |
a4fe3ce7 DG |
716 | |
717 | ptep = hugepte_offset(hugepd, addr, pdshift); | |
718 | do { | |
39adfa54 | 719 | next = hugepte_addr_end(addr, end, sz); |
a4fe3ce7 DG |
720 | if (!gup_hugepte(ptep, sz, addr, end, write, pages, nr)) |
721 | return 0; | |
39adfa54 | 722 | } while (ptep++, addr = next, addr != end); |
a4fe3ce7 DG |
723 | |
724 | return 1; | |
725 | } | |
1da177e4 | 726 | |
76512959 | 727 | #ifdef CONFIG_PPC_MM_SLICES |
1da177e4 LT |
728 | unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr, |
729 | unsigned long len, unsigned long pgoff, | |
730 | unsigned long flags) | |
731 | { | |
0d9ea754 JT |
732 | struct hstate *hstate = hstate_file(file); |
733 | int mmu_psize = shift_to_mmu_psize(huge_page_shift(hstate)); | |
48f797de | 734 | |
34d07177 | 735 | return slice_get_unmapped_area(addr, len, flags, mmu_psize, 1); |
1da177e4 | 736 | } |
76512959 | 737 | #endif |
1da177e4 | 738 | |
3340289d MG |
739 | unsigned long vma_mmu_pagesize(struct vm_area_struct *vma) |
740 | { | |
25c29f9e | 741 | #ifdef CONFIG_PPC_MM_SLICES |
3340289d MG |
742 | unsigned int psize = get_slice_psize(vma->vm_mm, vma->vm_start); |
743 | ||
744 | return 1UL << mmu_psize_to_shift(psize); | |
41151e77 BB |
745 | #else |
746 | if (!is_vm_hugetlb_page(vma)) | |
747 | return PAGE_SIZE; | |
748 | ||
749 | return huge_page_size(hstate_vma(vma)); | |
750 | #endif | |
751 | } | |
752 | ||
753 | static inline bool is_power_of_4(unsigned long x) | |
754 | { | |
755 | if (is_power_of_2(x)) | |
756 | return (__ilog2(x) % 2) ? false : true; | |
757 | return false; | |
3340289d MG |
758 | } |
759 | ||
d1837cba | 760 | static int __init add_huge_page_size(unsigned long long size) |
4ec161cf | 761 | { |
d1837cba DG |
762 | int shift = __ffs(size); |
763 | int mmu_psize; | |
a4fe3ce7 | 764 | |
4ec161cf | 765 | /* Check that it is a page size supported by the hardware and |
d1837cba | 766 | * that it fits within pagetable and slice limits. */ |
41151e77 BB |
767 | #ifdef CONFIG_PPC_FSL_BOOK3E |
768 | if ((size < PAGE_SIZE) || !is_power_of_4(size)) | |
769 | return -EINVAL; | |
770 | #else | |
d1837cba DG |
771 | if (!is_power_of_2(size) |
772 | || (shift > SLICE_HIGH_SHIFT) || (shift <= PAGE_SHIFT)) | |
773 | return -EINVAL; | |
41151e77 | 774 | #endif |
91224346 | 775 | |
d1837cba DG |
776 | if ((mmu_psize = shift_to_mmu_psize(shift)) < 0) |
777 | return -EINVAL; | |
778 | ||
779 | #ifdef CONFIG_SPU_FS_64K_LS | |
780 | /* Disable support for 64K huge pages when 64K SPU local store | |
781 | * support is enabled as the current implementation conflicts. | |
782 | */ | |
783 | if (shift == PAGE_SHIFT_64K) | |
784 | return -EINVAL; | |
785 | #endif /* CONFIG_SPU_FS_64K_LS */ | |
786 | ||
787 | BUG_ON(mmu_psize_defs[mmu_psize].shift != shift); | |
788 | ||
789 | /* Return if huge page size has already been setup */ | |
790 | if (size_to_hstate(size)) | |
791 | return 0; | |
792 | ||
793 | hugetlb_add_hstate(shift - PAGE_SHIFT); | |
794 | ||
795 | return 0; | |
4ec161cf JT |
796 | } |
797 | ||
798 | static int __init hugepage_setup_sz(char *str) | |
799 | { | |
800 | unsigned long long size; | |
4ec161cf JT |
801 | |
802 | size = memparse(str, &str); | |
803 | ||
d1837cba | 804 | if (add_huge_page_size(size) != 0) |
4ec161cf JT |
805 | printk(KERN_WARNING "Invalid huge page size specified(%llu)\n", size); |
806 | ||
807 | return 1; | |
808 | } | |
809 | __setup("hugepagesz=", hugepage_setup_sz); | |
810 | ||
881fde1d | 811 | #ifdef CONFIG_PPC_FSL_BOOK3E |
41151e77 BB |
812 | struct kmem_cache *hugepte_cache; |
813 | static int __init hugetlbpage_init(void) | |
814 | { | |
815 | int psize; | |
816 | ||
817 | for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) { | |
818 | unsigned shift; | |
819 | ||
820 | if (!mmu_psize_defs[psize].shift) | |
821 | continue; | |
822 | ||
823 | shift = mmu_psize_to_shift(psize); | |
824 | ||
825 | /* Don't treat normal page sizes as huge... */ | |
826 | if (shift != PAGE_SHIFT) | |
827 | if (add_huge_page_size(1ULL << shift) < 0) | |
828 | continue; | |
829 | } | |
830 | ||
831 | /* | |
832 | * Create a kmem cache for hugeptes. The bottom bits in the pte have | |
833 | * size information encoded in them, so align them to allow this | |
834 | */ | |
835 | hugepte_cache = kmem_cache_create("hugepte-cache", sizeof(pte_t), | |
836 | HUGEPD_SHIFT_MASK + 1, 0, NULL); | |
837 | if (hugepte_cache == NULL) | |
838 | panic("%s: Unable to create kmem cache for hugeptes\n", | |
839 | __func__); | |
840 | ||
841 | /* Default hpage size = 4M */ | |
842 | if (mmu_psize_defs[MMU_PAGE_4M].shift) | |
843 | HPAGE_SHIFT = mmu_psize_defs[MMU_PAGE_4M].shift; | |
844 | else | |
845 | panic("%s: Unable to set default huge page size\n", __func__); | |
846 | ||
847 | ||
848 | return 0; | |
849 | } | |
850 | #else | |
f10a04c0 DG |
851 | static int __init hugetlbpage_init(void) |
852 | { | |
a4fe3ce7 | 853 | int psize; |
0d9ea754 | 854 | |
44ae3ab3 | 855 | if (!mmu_has_feature(MMU_FTR_16M_PAGE)) |
f10a04c0 | 856 | return -ENODEV; |
00df438e | 857 | |
d1837cba DG |
858 | for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) { |
859 | unsigned shift; | |
860 | unsigned pdshift; | |
0d9ea754 | 861 | |
d1837cba DG |
862 | if (!mmu_psize_defs[psize].shift) |
863 | continue; | |
00df438e | 864 | |
d1837cba DG |
865 | shift = mmu_psize_to_shift(psize); |
866 | ||
867 | if (add_huge_page_size(1ULL << shift) < 0) | |
868 | continue; | |
869 | ||
870 | if (shift < PMD_SHIFT) | |
871 | pdshift = PMD_SHIFT; | |
872 | else if (shift < PUD_SHIFT) | |
873 | pdshift = PUD_SHIFT; | |
874 | else | |
875 | pdshift = PGDIR_SHIFT; | |
876 | ||
877 | pgtable_cache_add(pdshift - shift, NULL); | |
878 | if (!PGT_CACHE(pdshift - shift)) | |
879 | panic("hugetlbpage_init(): could not create " | |
880 | "pgtable cache for %d bit pagesize\n", shift); | |
0d9ea754 | 881 | } |
f10a04c0 | 882 | |
d1837cba DG |
883 | /* Set default large page size. Currently, we pick 16M or 1M |
884 | * depending on what is available | |
885 | */ | |
886 | if (mmu_psize_defs[MMU_PAGE_16M].shift) | |
887 | HPAGE_SHIFT = mmu_psize_defs[MMU_PAGE_16M].shift; | |
888 | else if (mmu_psize_defs[MMU_PAGE_1M].shift) | |
889 | HPAGE_SHIFT = mmu_psize_defs[MMU_PAGE_1M].shift; | |
890 | ||
f10a04c0 DG |
891 | return 0; |
892 | } | |
41151e77 | 893 | #endif |
f10a04c0 | 894 | module_init(hugetlbpage_init); |
0895ecda DG |
895 | |
896 | void flush_dcache_icache_hugepage(struct page *page) | |
897 | { | |
898 | int i; | |
41151e77 | 899 | void *start; |
0895ecda DG |
900 | |
901 | BUG_ON(!PageCompound(page)); | |
902 | ||
41151e77 BB |
903 | for (i = 0; i < (1UL << compound_order(page)); i++) { |
904 | if (!PageHighMem(page)) { | |
905 | __flush_dcache_icache(page_address(page+i)); | |
906 | } else { | |
2480b208 | 907 | start = kmap_atomic(page+i); |
41151e77 | 908 | __flush_dcache_icache(start); |
2480b208 | 909 | kunmap_atomic(start); |
41151e77 BB |
910 | } |
911 | } | |
0895ecda | 912 | } |