Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * PPC64 (POWER4) Huge TLB Page Support for Kernel. | |
3 | * | |
4 | * Copyright (C) 2003 David Gibson, IBM Corporation. | |
5 | * | |
6 | * Based on the IA-32 version: | |
7 | * Copyright (C) 2002, Rohit Seth <rohit.seth@intel.com> | |
8 | */ | |
9 | ||
10 | #include <linux/init.h> | |
11 | #include <linux/fs.h> | |
12 | #include <linux/mm.h> | |
13 | #include <linux/hugetlb.h> | |
14 | #include <linux/pagemap.h> | |
1da177e4 LT |
15 | #include <linux/slab.h> |
16 | #include <linux/err.h> | |
17 | #include <linux/sysctl.h> | |
18 | #include <asm/mman.h> | |
19 | #include <asm/pgalloc.h> | |
20 | #include <asm/tlb.h> | |
21 | #include <asm/tlbflush.h> | |
22 | #include <asm/mmu_context.h> | |
23 | #include <asm/machdep.h> | |
24 | #include <asm/cputable.h> | |
94b2a439 | 25 | #include <asm/spu.h> |
1da177e4 | 26 | |
91224346 JT |
27 | #define PAGE_SHIFT_64K 16 |
28 | #define PAGE_SHIFT_16M 24 | |
29 | #define PAGE_SHIFT_16G 34 | |
4ec161cf | 30 | |
c594adad DG |
31 | #define NUM_LOW_AREAS (0x100000000UL >> SID_SHIFT) |
32 | #define NUM_HIGH_AREAS (PGTABLE_RANGE >> HTLB_AREA_SHIFT) | |
ec4b2c0c JT |
33 | #define MAX_NUMBER_GPAGES 1024 |
34 | ||
35 | /* Tracks the 16G pages after the device tree is scanned and before the | |
36 | * huge_boot_pages list is ready. */ | |
37 | static unsigned long gpage_freearray[MAX_NUMBER_GPAGES]; | |
38 | static unsigned nr_gpages; | |
c594adad | 39 | |
4ec161cf JT |
40 | unsigned int hugepte_shift; |
41 | #define PTRS_PER_HUGEPTE (1 << hugepte_shift) | |
42 | #define HUGEPTE_TABLE_SIZE (sizeof(pte_t) << hugepte_shift) | |
f10a04c0 | 43 | |
4ec161cf | 44 | #define HUGEPD_SHIFT (HPAGE_SHIFT + hugepte_shift) |
f10a04c0 DG |
45 | #define HUGEPD_SIZE (1UL << HUGEPD_SHIFT) |
46 | #define HUGEPD_MASK (~(HUGEPD_SIZE-1)) | |
47 | ||
48 | #define huge_pgtable_cache (pgtable_cache[HUGEPTE_CACHE_NUM]) | |
49 | ||
50 | /* Flag to mark huge PD pointers. This means pmd_bad() and pud_bad() | |
51 | * will choke on pointers to hugepte tables, which is handy for | |
52 | * catching screwups early. */ | |
53 | #define HUGEPD_OK 0x1 | |
54 | ||
55 | typedef struct { unsigned long pd; } hugepd_t; | |
56 | ||
57 | #define hugepd_none(hpd) ((hpd).pd == 0) | |
58 | ||
59 | static inline pte_t *hugepd_page(hugepd_t hpd) | |
60 | { | |
61 | BUG_ON(!(hpd.pd & HUGEPD_OK)); | |
62 | return (pte_t *)(hpd.pd & ~HUGEPD_OK); | |
63 | } | |
64 | ||
65 | static inline pte_t *hugepte_offset(hugepd_t *hpdp, unsigned long addr) | |
66 | { | |
67 | unsigned long idx = ((addr >> HPAGE_SHIFT) & (PTRS_PER_HUGEPTE-1)); | |
68 | pte_t *dir = hugepd_page(*hpdp); | |
69 | ||
70 | return dir + idx; | |
71 | } | |
72 | ||
73 | static int __hugepte_alloc(struct mm_struct *mm, hugepd_t *hpdp, | |
74 | unsigned long address) | |
75 | { | |
76 | pte_t *new = kmem_cache_alloc(huge_pgtable_cache, | |
77 | GFP_KERNEL|__GFP_REPEAT); | |
78 | ||
79 | if (! new) | |
80 | return -ENOMEM; | |
81 | ||
82 | spin_lock(&mm->page_table_lock); | |
83 | if (!hugepd_none(*hpdp)) | |
84 | kmem_cache_free(huge_pgtable_cache, new); | |
85 | else | |
86 | hpdp->pd = (unsigned long)new | HUGEPD_OK; | |
87 | spin_unlock(&mm->page_table_lock); | |
88 | return 0; | |
89 | } | |
90 | ||
4ec161cf JT |
91 | /* Base page size affects how we walk hugetlb page tables */ |
92 | #ifdef CONFIG_PPC_64K_PAGES | |
93 | #define hpmd_offset(pud, addr) pmd_offset(pud, addr) | |
94 | #define hpmd_alloc(mm, pud, addr) pmd_alloc(mm, pud, addr) | |
95 | #else | |
96 | static inline | |
97 | pmd_t *hpmd_offset(pud_t *pud, unsigned long addr) | |
98 | { | |
91224346 | 99 | if (HPAGE_SHIFT == PAGE_SHIFT_64K) |
4ec161cf JT |
100 | return pmd_offset(pud, addr); |
101 | else | |
102 | return (pmd_t *) pud; | |
103 | } | |
104 | static inline | |
105 | pmd_t *hpmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long addr) | |
106 | { | |
91224346 | 107 | if (HPAGE_SHIFT == PAGE_SHIFT_64K) |
4ec161cf JT |
108 | return pmd_alloc(mm, pud, addr); |
109 | else | |
110 | return (pmd_t *) pud; | |
111 | } | |
112 | #endif | |
113 | ||
658013e9 JT |
114 | /* Build list of addresses of gigantic pages. This function is used in early |
115 | * boot before the buddy or bootmem allocator is setup. | |
116 | */ | |
117 | void add_gpage(unsigned long addr, unsigned long page_size, | |
118 | unsigned long number_of_pages) | |
119 | { | |
120 | if (!addr) | |
121 | return; | |
122 | while (number_of_pages > 0) { | |
123 | gpage_freearray[nr_gpages] = addr; | |
124 | nr_gpages++; | |
125 | number_of_pages--; | |
126 | addr += page_size; | |
127 | } | |
128 | } | |
129 | ||
ec4b2c0c JT |
130 | /* Moves the gigantic page addresses from the temporary list to the |
131 | * huge_boot_pages list. */ | |
132 | int alloc_bootmem_huge_page(struct hstate *h) | |
133 | { | |
134 | struct huge_bootmem_page *m; | |
135 | if (nr_gpages == 0) | |
136 | return 0; | |
137 | m = phys_to_virt(gpage_freearray[--nr_gpages]); | |
138 | gpage_freearray[nr_gpages] = 0; | |
139 | list_add(&m->list, &huge_boot_pages); | |
140 | m->hstate = h; | |
141 | return 1; | |
142 | } | |
143 | ||
144 | ||
e28f7faf DG |
145 | /* Modelled after find_linux_pte() */ |
146 | pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr) | |
1da177e4 | 147 | { |
e28f7faf DG |
148 | pgd_t *pg; |
149 | pud_t *pu; | |
4ec161cf | 150 | pmd_t *pm; |
1da177e4 | 151 | |
d0f13e3c | 152 | BUG_ON(get_slice_psize(mm, addr) != mmu_huge_psize); |
1da177e4 | 153 | |
e28f7faf DG |
154 | addr &= HPAGE_MASK; |
155 | ||
156 | pg = pgd_offset(mm, addr); | |
157 | if (!pgd_none(*pg)) { | |
158 | pu = pud_offset(pg, addr); | |
159 | if (!pud_none(*pu)) { | |
4ec161cf | 160 | pm = hpmd_offset(pu, addr); |
f10a04c0 DG |
161 | if (!pmd_none(*pm)) |
162 | return hugepte_offset((hugepd_t *)pm, addr); | |
e28f7faf DG |
163 | } |
164 | } | |
1da177e4 | 165 | |
e28f7faf | 166 | return NULL; |
1da177e4 LT |
167 | } |
168 | ||
a5516438 AK |
169 | pte_t *huge_pte_alloc(struct mm_struct *mm, |
170 | unsigned long addr, unsigned long sz) | |
1da177e4 | 171 | { |
e28f7faf DG |
172 | pgd_t *pg; |
173 | pud_t *pu; | |
4ec161cf | 174 | pmd_t *pm; |
f10a04c0 | 175 | hugepd_t *hpdp = NULL; |
1da177e4 | 176 | |
d0f13e3c | 177 | BUG_ON(get_slice_psize(mm, addr) != mmu_huge_psize); |
1da177e4 | 178 | |
e28f7faf | 179 | addr &= HPAGE_MASK; |
1da177e4 | 180 | |
e28f7faf DG |
181 | pg = pgd_offset(mm, addr); |
182 | pu = pud_alloc(mm, pg, addr); | |
1da177e4 | 183 | |
e28f7faf | 184 | if (pu) { |
4ec161cf | 185 | pm = hpmd_alloc(mm, pu, addr); |
f10a04c0 DG |
186 | if (pm) |
187 | hpdp = (hugepd_t *)pm; | |
f10a04c0 DG |
188 | } |
189 | ||
190 | if (! hpdp) | |
191 | return NULL; | |
192 | ||
193 | if (hugepd_none(*hpdp) && __hugepte_alloc(mm, hpdp, addr)) | |
194 | return NULL; | |
195 | ||
196 | return hugepte_offset(hpdp, addr); | |
197 | } | |
198 | ||
39dde65c CK |
199 | int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep) |
200 | { | |
201 | return 0; | |
202 | } | |
203 | ||
f10a04c0 DG |
204 | static void free_hugepte_range(struct mmu_gather *tlb, hugepd_t *hpdp) |
205 | { | |
206 | pte_t *hugepte = hugepd_page(*hpdp); | |
207 | ||
208 | hpdp->pd = 0; | |
209 | tlb->need_flush = 1; | |
210 | pgtable_free_tlb(tlb, pgtable_free_cache(hugepte, HUGEPTE_CACHE_NUM, | |
c9169f87 | 211 | PGF_CACHENUM_MASK)); |
f10a04c0 DG |
212 | } |
213 | ||
f10a04c0 DG |
214 | static void hugetlb_free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
215 | unsigned long addr, unsigned long end, | |
216 | unsigned long floor, unsigned long ceiling) | |
217 | { | |
218 | pmd_t *pmd; | |
219 | unsigned long next; | |
220 | unsigned long start; | |
221 | ||
222 | start = addr; | |
223 | pmd = pmd_offset(pud, addr); | |
224 | do { | |
225 | next = pmd_addr_end(addr, end); | |
226 | if (pmd_none(*pmd)) | |
227 | continue; | |
228 | free_hugepte_range(tlb, (hugepd_t *)pmd); | |
229 | } while (pmd++, addr = next, addr != end); | |
230 | ||
231 | start &= PUD_MASK; | |
232 | if (start < floor) | |
233 | return; | |
234 | if (ceiling) { | |
235 | ceiling &= PUD_MASK; | |
236 | if (!ceiling) | |
237 | return; | |
1da177e4 | 238 | } |
f10a04c0 DG |
239 | if (end - 1 > ceiling - 1) |
240 | return; | |
1da177e4 | 241 | |
f10a04c0 DG |
242 | pmd = pmd_offset(pud, start); |
243 | pud_clear(pud); | |
244 | pmd_free_tlb(tlb, pmd); | |
245 | } | |
f10a04c0 DG |
246 | |
247 | static void hugetlb_free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, | |
248 | unsigned long addr, unsigned long end, | |
249 | unsigned long floor, unsigned long ceiling) | |
250 | { | |
251 | pud_t *pud; | |
252 | unsigned long next; | |
253 | unsigned long start; | |
254 | ||
255 | start = addr; | |
256 | pud = pud_offset(pgd, addr); | |
257 | do { | |
258 | next = pud_addr_end(addr, end); | |
259 | #ifdef CONFIG_PPC_64K_PAGES | |
260 | if (pud_none_or_clear_bad(pud)) | |
261 | continue; | |
262 | hugetlb_free_pmd_range(tlb, pud, addr, next, floor, ceiling); | |
263 | #else | |
91224346 | 264 | if (HPAGE_SHIFT == PAGE_SHIFT_64K) { |
4ec161cf JT |
265 | if (pud_none_or_clear_bad(pud)) |
266 | continue; | |
267 | hugetlb_free_pmd_range(tlb, pud, addr, next, floor, ceiling); | |
268 | } else { | |
269 | if (pud_none(*pud)) | |
270 | continue; | |
271 | free_hugepte_range(tlb, (hugepd_t *)pud); | |
272 | } | |
f10a04c0 DG |
273 | #endif |
274 | } while (pud++, addr = next, addr != end); | |
275 | ||
276 | start &= PGDIR_MASK; | |
277 | if (start < floor) | |
278 | return; | |
279 | if (ceiling) { | |
280 | ceiling &= PGDIR_MASK; | |
281 | if (!ceiling) | |
282 | return; | |
283 | } | |
284 | if (end - 1 > ceiling - 1) | |
285 | return; | |
286 | ||
287 | pud = pud_offset(pgd, start); | |
288 | pgd_clear(pgd); | |
289 | pud_free_tlb(tlb, pud); | |
290 | } | |
291 | ||
292 | /* | |
293 | * This function frees user-level page tables of a process. | |
294 | * | |
295 | * Must be called with pagetable lock held. | |
296 | */ | |
42b77728 | 297 | void hugetlb_free_pgd_range(struct mmu_gather *tlb, |
f10a04c0 DG |
298 | unsigned long addr, unsigned long end, |
299 | unsigned long floor, unsigned long ceiling) | |
300 | { | |
301 | pgd_t *pgd; | |
302 | unsigned long next; | |
303 | unsigned long start; | |
304 | ||
305 | /* | |
306 | * Comments below take from the normal free_pgd_range(). They | |
307 | * apply here too. The tests against HUGEPD_MASK below are | |
308 | * essential, because we *don't* test for this at the bottom | |
309 | * level. Without them we'll attempt to free a hugepte table | |
310 | * when we unmap just part of it, even if there are other | |
311 | * active mappings using it. | |
312 | * | |
313 | * The next few lines have given us lots of grief... | |
314 | * | |
315 | * Why are we testing HUGEPD* at this top level? Because | |
316 | * often there will be no work to do at all, and we'd prefer | |
317 | * not to go all the way down to the bottom just to discover | |
318 | * that. | |
319 | * | |
320 | * Why all these "- 1"s? Because 0 represents both the bottom | |
321 | * of the address space and the top of it (using -1 for the | |
322 | * top wouldn't help much: the masks would do the wrong thing). | |
323 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
324 | * the address space, but end 0 and ceiling 0 refer to the top | |
325 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
326 | * that end 0 case should be mythical). | |
327 | * | |
328 | * Wherever addr is brought up or ceiling brought down, we | |
329 | * must be careful to reject "the opposite 0" before it | |
330 | * confuses the subsequent tests. But what about where end is | |
331 | * brought down by HUGEPD_SIZE below? no, end can't go down to | |
332 | * 0 there. | |
333 | * | |
334 | * Whereas we round start (addr) and ceiling down, by different | |
335 | * masks at different levels, in order to test whether a table | |
336 | * now has no other vmas using it, so can be freed, we don't | |
337 | * bother to round floor or end up - the tests don't need that. | |
338 | */ | |
339 | ||
340 | addr &= HUGEPD_MASK; | |
341 | if (addr < floor) { | |
342 | addr += HUGEPD_SIZE; | |
343 | if (!addr) | |
344 | return; | |
345 | } | |
346 | if (ceiling) { | |
347 | ceiling &= HUGEPD_MASK; | |
348 | if (!ceiling) | |
349 | return; | |
350 | } | |
351 | if (end - 1 > ceiling - 1) | |
352 | end -= HUGEPD_SIZE; | |
353 | if (addr > end - 1) | |
354 | return; | |
355 | ||
356 | start = addr; | |
42b77728 | 357 | pgd = pgd_offset(tlb->mm, addr); |
f10a04c0 | 358 | do { |
42b77728 | 359 | BUG_ON(get_slice_psize(tlb->mm, addr) != mmu_huge_psize); |
f10a04c0 DG |
360 | next = pgd_addr_end(addr, end); |
361 | if (pgd_none_or_clear_bad(pgd)) | |
362 | continue; | |
42b77728 | 363 | hugetlb_free_pud_range(tlb, pgd, addr, next, floor, ceiling); |
f10a04c0 | 364 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
365 | } |
366 | ||
e28f7faf DG |
367 | void set_huge_pte_at(struct mm_struct *mm, unsigned long addr, |
368 | pte_t *ptep, pte_t pte) | |
369 | { | |
e28f7faf | 370 | if (pte_present(*ptep)) { |
3c726f8d | 371 | /* We open-code pte_clear because we need to pass the right |
a741e679 BH |
372 | * argument to hpte_need_flush (huge / !huge). Might not be |
373 | * necessary anymore if we make hpte_need_flush() get the | |
374 | * page size from the slices | |
3c726f8d | 375 | */ |
a741e679 | 376 | pte_update(mm, addr & HPAGE_MASK, ptep, ~0UL, 1); |
e28f7faf | 377 | } |
3c726f8d | 378 | *ptep = __pte(pte_val(pte) & ~_PAGE_HPTEFLAGS); |
1da177e4 LT |
379 | } |
380 | ||
e28f7faf DG |
381 | pte_t huge_ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, |
382 | pte_t *ptep) | |
1da177e4 | 383 | { |
a741e679 | 384 | unsigned long old = pte_update(mm, addr, ptep, ~0UL, 1); |
e28f7faf | 385 | return __pte(old); |
1da177e4 LT |
386 | } |
387 | ||
1da177e4 LT |
388 | struct page * |
389 | follow_huge_addr(struct mm_struct *mm, unsigned long address, int write) | |
390 | { | |
391 | pte_t *ptep; | |
392 | struct page *page; | |
393 | ||
d0f13e3c | 394 | if (get_slice_psize(mm, address) != mmu_huge_psize) |
1da177e4 LT |
395 | return ERR_PTR(-EINVAL); |
396 | ||
397 | ptep = huge_pte_offset(mm, address); | |
398 | page = pte_page(*ptep); | |
399 | if (page) | |
400 | page += (address % HPAGE_SIZE) / PAGE_SIZE; | |
401 | ||
402 | return page; | |
403 | } | |
404 | ||
405 | int pmd_huge(pmd_t pmd) | |
406 | { | |
407 | return 0; | |
408 | } | |
409 | ||
ceb86879 AK |
410 | int pud_huge(pud_t pud) |
411 | { | |
412 | return 0; | |
413 | } | |
414 | ||
1da177e4 LT |
415 | struct page * |
416 | follow_huge_pmd(struct mm_struct *mm, unsigned long address, | |
417 | pmd_t *pmd, int write) | |
418 | { | |
419 | BUG(); | |
420 | return NULL; | |
421 | } | |
422 | ||
1da177e4 LT |
423 | |
424 | unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr, | |
425 | unsigned long len, unsigned long pgoff, | |
426 | unsigned long flags) | |
427 | { | |
d0f13e3c BH |
428 | return slice_get_unmapped_area(addr, len, flags, |
429 | mmu_huge_psize, 1, 0); | |
1da177e4 LT |
430 | } |
431 | ||
cbf52afd DG |
432 | /* |
433 | * Called by asm hashtable.S for doing lazy icache flush | |
434 | */ | |
435 | static unsigned int hash_huge_page_do_lazy_icache(unsigned long rflags, | |
436 | pte_t pte, int trap) | |
437 | { | |
438 | struct page *page; | |
439 | int i; | |
440 | ||
441 | if (!pfn_valid(pte_pfn(pte))) | |
442 | return rflags; | |
443 | ||
444 | page = pte_page(pte); | |
445 | ||
446 | /* page is dirty */ | |
447 | if (!test_bit(PG_arch_1, &page->flags) && !PageReserved(page)) { | |
448 | if (trap == 0x400) { | |
449 | for (i = 0; i < (HPAGE_SIZE / PAGE_SIZE); i++) | |
450 | __flush_dcache_icache(page_address(page+i)); | |
451 | set_bit(PG_arch_1, &page->flags); | |
452 | } else { | |
453 | rflags |= HPTE_R_N; | |
454 | } | |
455 | } | |
456 | return rflags; | |
457 | } | |
458 | ||
1da177e4 | 459 | int hash_huge_page(struct mm_struct *mm, unsigned long access, |
cbf52afd DG |
460 | unsigned long ea, unsigned long vsid, int local, |
461 | unsigned long trap) | |
1da177e4 LT |
462 | { |
463 | pte_t *ptep; | |
3c726f8d BH |
464 | unsigned long old_pte, new_pte; |
465 | unsigned long va, rflags, pa; | |
1da177e4 LT |
466 | long slot; |
467 | int err = 1; | |
1189be65 | 468 | int ssize = user_segment_size(ea); |
1da177e4 | 469 | |
1da177e4 LT |
470 | ptep = huge_pte_offset(mm, ea); |
471 | ||
472 | /* Search the Linux page table for a match with va */ | |
1189be65 | 473 | va = hpt_va(ea, vsid, ssize); |
1da177e4 LT |
474 | |
475 | /* | |
476 | * If no pte found or not present, send the problem up to | |
477 | * do_page_fault | |
478 | */ | |
479 | if (unlikely(!ptep || pte_none(*ptep))) | |
480 | goto out; | |
481 | ||
1da177e4 LT |
482 | /* |
483 | * Check the user's access rights to the page. If access should be | |
484 | * prevented then send the problem up to do_page_fault. | |
485 | */ | |
486 | if (unlikely(access & ~pte_val(*ptep))) | |
487 | goto out; | |
488 | /* | |
489 | * At this point, we have a pte (old_pte) which can be used to build | |
490 | * or update an HPTE. There are 2 cases: | |
491 | * | |
492 | * 1. There is a valid (present) pte with no associated HPTE (this is | |
493 | * the most common case) | |
494 | * 2. There is a valid (present) pte with an associated HPTE. The | |
495 | * current values of the pp bits in the HPTE prevent access | |
496 | * because we are doing software DIRTY bit management and the | |
497 | * page is currently not DIRTY. | |
498 | */ | |
499 | ||
500 | ||
3c726f8d BH |
501 | do { |
502 | old_pte = pte_val(*ptep); | |
503 | if (old_pte & _PAGE_BUSY) | |
504 | goto out; | |
41743a4e | 505 | new_pte = old_pte | _PAGE_BUSY | _PAGE_ACCESSED; |
3c726f8d BH |
506 | } while(old_pte != __cmpxchg_u64((unsigned long *)ptep, |
507 | old_pte, new_pte)); | |
508 | ||
509 | rflags = 0x2 | (!(new_pte & _PAGE_RW)); | |
1da177e4 | 510 | /* _PAGE_EXEC -> HW_NO_EXEC since it's inverted */ |
3c726f8d | 511 | rflags |= ((new_pte & _PAGE_EXEC) ? 0 : HPTE_R_N); |
cbf52afd DG |
512 | if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE)) |
513 | /* No CPU has hugepages but lacks no execute, so we | |
514 | * don't need to worry about that case */ | |
515 | rflags = hash_huge_page_do_lazy_icache(rflags, __pte(old_pte), | |
516 | trap); | |
1da177e4 LT |
517 | |
518 | /* Check if pte already has an hpte (case 2) */ | |
3c726f8d | 519 | if (unlikely(old_pte & _PAGE_HASHPTE)) { |
1da177e4 LT |
520 | /* There MIGHT be an HPTE for this pte */ |
521 | unsigned long hash, slot; | |
522 | ||
1189be65 | 523 | hash = hpt_hash(va, HPAGE_SHIFT, ssize); |
3c726f8d | 524 | if (old_pte & _PAGE_F_SECOND) |
1da177e4 LT |
525 | hash = ~hash; |
526 | slot = (hash & htab_hash_mask) * HPTES_PER_GROUP; | |
3c726f8d | 527 | slot += (old_pte & _PAGE_F_GIX) >> 12; |
1da177e4 | 528 | |
325c82a0 | 529 | if (ppc_md.hpte_updatepp(slot, rflags, va, mmu_huge_psize, |
1189be65 | 530 | ssize, local) == -1) |
3c726f8d | 531 | old_pte &= ~_PAGE_HPTEFLAGS; |
1da177e4 LT |
532 | } |
533 | ||
3c726f8d | 534 | if (likely(!(old_pte & _PAGE_HASHPTE))) { |
1189be65 | 535 | unsigned long hash = hpt_hash(va, HPAGE_SHIFT, ssize); |
1da177e4 LT |
536 | unsigned long hpte_group; |
537 | ||
3c726f8d | 538 | pa = pte_pfn(__pte(old_pte)) << PAGE_SHIFT; |
1da177e4 LT |
539 | |
540 | repeat: | |
541 | hpte_group = ((hash & htab_hash_mask) * | |
542 | HPTES_PER_GROUP) & ~0x7UL; | |
543 | ||
3c726f8d | 544 | /* clear HPTE slot informations in new PTE */ |
41743a4e BH |
545 | #ifdef CONFIG_PPC_64K_PAGES |
546 | new_pte = (new_pte & ~_PAGE_HPTEFLAGS) | _PAGE_HPTE_SUB0; | |
547 | #else | |
3c726f8d | 548 | new_pte = (new_pte & ~_PAGE_HPTEFLAGS) | _PAGE_HASHPTE; |
41743a4e | 549 | #endif |
1da177e4 | 550 | /* Add in WIMG bits */ |
87e9ab13 DK |
551 | rflags |= (new_pte & (_PAGE_WRITETHRU | _PAGE_NO_CACHE | |
552 | _PAGE_COHERENT | _PAGE_GUARDED)); | |
1da177e4 | 553 | |
3c726f8d BH |
554 | /* Insert into the hash table, primary slot */ |
555 | slot = ppc_md.hpte_insert(hpte_group, va, pa, rflags, 0, | |
1189be65 | 556 | mmu_huge_psize, ssize); |
1da177e4 LT |
557 | |
558 | /* Primary is full, try the secondary */ | |
559 | if (unlikely(slot == -1)) { | |
1da177e4 LT |
560 | hpte_group = ((~hash & htab_hash_mask) * |
561 | HPTES_PER_GROUP) & ~0x7UL; | |
3c726f8d | 562 | slot = ppc_md.hpte_insert(hpte_group, va, pa, rflags, |
67b10813 | 563 | HPTE_V_SECONDARY, |
1189be65 | 564 | mmu_huge_psize, ssize); |
1da177e4 LT |
565 | if (slot == -1) { |
566 | if (mftb() & 0x1) | |
67b10813 BH |
567 | hpte_group = ((hash & htab_hash_mask) * |
568 | HPTES_PER_GROUP)&~0x7UL; | |
1da177e4 LT |
569 | |
570 | ppc_md.hpte_remove(hpte_group); | |
571 | goto repeat; | |
572 | } | |
573 | } | |
574 | ||
575 | if (unlikely(slot == -2)) | |
576 | panic("hash_huge_page: pte_insert failed\n"); | |
577 | ||
d649bd7b | 578 | new_pte |= (slot << 12) & (_PAGE_F_SECOND | _PAGE_F_GIX); |
1da177e4 LT |
579 | } |
580 | ||
3c726f8d | 581 | /* |
01edcd89 | 582 | * No need to use ldarx/stdcx here |
3c726f8d BH |
583 | */ |
584 | *ptep = __pte(new_pte & ~_PAGE_BUSY); | |
585 | ||
1da177e4 LT |
586 | err = 0; |
587 | ||
588 | out: | |
1da177e4 LT |
589 | return err; |
590 | } | |
f10a04c0 | 591 | |
4ec161cf JT |
592 | void set_huge_psize(int psize) |
593 | { | |
594 | /* Check that it is a page size supported by the hardware and | |
595 | * that it fits within pagetable limits. */ | |
91224346 JT |
596 | if (mmu_psize_defs[psize].shift && |
597 | mmu_psize_defs[psize].shift < SID_SHIFT_1T && | |
4ec161cf | 598 | (mmu_psize_defs[psize].shift > MIN_HUGEPTE_SHIFT || |
91224346 JT |
599 | mmu_psize_defs[psize].shift == PAGE_SHIFT_64K || |
600 | mmu_psize_defs[psize].shift == PAGE_SHIFT_16G)) { | |
601 | /* Return if huge page size is the same as the | |
602 | * base page size. */ | |
603 | if (mmu_psize_defs[psize].shift == PAGE_SHIFT) | |
604 | return; | |
605 | ||
4ec161cf JT |
606 | HPAGE_SHIFT = mmu_psize_defs[psize].shift; |
607 | mmu_huge_psize = psize; | |
4ec161cf | 608 | |
91224346 JT |
609 | switch (HPAGE_SHIFT) { |
610 | case PAGE_SHIFT_64K: | |
611 | /* We only allow 64k hpages with 4k base page, | |
612 | * which was checked above, and always put them | |
613 | * at the PMD */ | |
614 | hugepte_shift = PMD_SHIFT; | |
615 | break; | |
616 | case PAGE_SHIFT_16M: | |
617 | /* 16M pages can be at two different levels | |
618 | * of pagestables based on base page size */ | |
619 | if (PAGE_SHIFT == PAGE_SHIFT_64K) | |
620 | hugepte_shift = PMD_SHIFT; | |
621 | else /* 4k base page */ | |
622 | hugepte_shift = PUD_SHIFT; | |
623 | break; | |
624 | case PAGE_SHIFT_16G: | |
625 | /* 16G pages are always at PGD level */ | |
626 | hugepte_shift = PGDIR_SHIFT; | |
627 | break; | |
628 | } | |
629 | hugepte_shift -= HPAGE_SHIFT; | |
4ec161cf JT |
630 | } else |
631 | HPAGE_SHIFT = 0; | |
632 | } | |
633 | ||
634 | static int __init hugepage_setup_sz(char *str) | |
635 | { | |
636 | unsigned long long size; | |
637 | int mmu_psize = -1; | |
638 | int shift; | |
639 | ||
640 | size = memparse(str, &str); | |
641 | ||
642 | shift = __ffs(size); | |
643 | switch (shift) { | |
644 | #ifndef CONFIG_PPC_64K_PAGES | |
91224346 | 645 | case PAGE_SHIFT_64K: |
4ec161cf JT |
646 | mmu_psize = MMU_PAGE_64K; |
647 | break; | |
648 | #endif | |
91224346 | 649 | case PAGE_SHIFT_16M: |
4ec161cf JT |
650 | mmu_psize = MMU_PAGE_16M; |
651 | break; | |
91224346 JT |
652 | case PAGE_SHIFT_16G: |
653 | mmu_psize = MMU_PAGE_16G; | |
654 | break; | |
4ec161cf JT |
655 | } |
656 | ||
91224346 | 657 | if (mmu_psize >= 0 && mmu_psize_defs[mmu_psize].shift) { |
4ec161cf | 658 | set_huge_psize(mmu_psize); |
91224346 JT |
659 | hugetlb_add_hstate(shift - PAGE_SHIFT); |
660 | } | |
4ec161cf JT |
661 | else |
662 | printk(KERN_WARNING "Invalid huge page size specified(%llu)\n", size); | |
663 | ||
664 | return 1; | |
665 | } | |
666 | __setup("hugepagesz=", hugepage_setup_sz); | |
667 | ||
4ba9b9d0 | 668 | static void zero_ctor(struct kmem_cache *cache, void *addr) |
f10a04c0 DG |
669 | { |
670 | memset(addr, 0, kmem_cache_size(cache)); | |
671 | } | |
672 | ||
673 | static int __init hugetlbpage_init(void) | |
674 | { | |
675 | if (!cpu_has_feature(CPU_FTR_16M_PAGE)) | |
676 | return -ENODEV; | |
677 | ||
678 | huge_pgtable_cache = kmem_cache_create("hugepte_cache", | |
679 | HUGEPTE_TABLE_SIZE, | |
680 | HUGEPTE_TABLE_SIZE, | |
f0f3980b | 681 | 0, |
20c2df83 | 682 | zero_ctor); |
f10a04c0 DG |
683 | if (! huge_pgtable_cache) |
684 | panic("hugetlbpage_init(): could not create hugepte cache\n"); | |
685 | ||
686 | return 0; | |
687 | } | |
688 | ||
689 | module_init(hugetlbpage_init); |