1 /*
2 * PPC Huge TLB Page Support for Kernel.
3 *
4 * Copyright (C) 2003 David Gibson, IBM Corporation.
5 * Copyright (C) 2011 Becky Bruce, Freescale Semiconductor
6 *
7 * Based on the IA-32 version:
8 * Copyright (C) 2002, Rohit Seth <rohit.seth@intel.com>
9 */
10
11 #include <linux/mm.h>
12 #include <linux/io.h>
13 #include <linux/slab.h>
14 #include <linux/hugetlb.h>
15 #include <linux/export.h>
16 #include <linux/of_fdt.h>
17 #include <linux/memblock.h>
18 #include <linux/moduleparam.h>
19 #include <linux/swap.h>
20 #include <linux/swapops.h>
21 #include <linux/kmemleak.h>
22 #include <asm/pgalloc.h>
23 #include <asm/tlb.h>
24 #include <asm/setup.h>
25 #include <asm/hugetlb.h>
26 #include <asm/pte-walk.h>
27
28 bool hugetlb_disabled = false;
29
30 #define hugepd_none(hpd) (hpd_val(hpd) == 0)
31
32 #define PTE_T_ORDER (__builtin_ffs(sizeof(pte_basic_t)) - \
33 __builtin_ffs(sizeof(void *)))
34
huge_pte_offset(struct mm_struct * mm,unsigned long addr,unsigned long sz)35 pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr, unsigned long sz)
36 {
37 /*
38 * Only called for hugetlbfs pages, hence can ignore THP and the
39 * irq disabled walk.
40 */
41 return __find_linux_pte(mm->pgd, addr, NULL, NULL);
42 }
43
__hugepte_alloc(struct mm_struct * mm,hugepd_t * hpdp,unsigned long address,unsigned int pdshift,unsigned int pshift,spinlock_t * ptl)44 static int __hugepte_alloc(struct mm_struct *mm, hugepd_t *hpdp,
45 unsigned long address, unsigned int pdshift,
46 unsigned int pshift, spinlock_t *ptl)
47 {
48 struct kmem_cache *cachep;
49 pte_t *new;
50 int i;
51 int num_hugepd;
52
53 if (pshift >= pdshift) {
54 cachep = PGT_CACHE(PTE_T_ORDER);
55 num_hugepd = 1 << (pshift - pdshift);
56 } else {
57 cachep = PGT_CACHE(pdshift - pshift);
58 num_hugepd = 1;
59 }
60
61 if (!cachep) {
62 WARN_ONCE(1, "No page table cache created for hugetlb tables");
63 return -ENOMEM;
64 }
65
66 new = kmem_cache_alloc(cachep, pgtable_gfp_flags(mm, GFP_KERNEL));
67
68 BUG_ON(pshift > HUGEPD_SHIFT_MASK);
69 BUG_ON((unsigned long)new & HUGEPD_SHIFT_MASK);
70
71 if (!new)
72 return -ENOMEM;
73
74 /*
75 * Make sure other cpus find the hugepd set only after a
76 * properly initialized page table is visible to them.
77 * For more details look for comment in __pte_alloc().
78 */
79 smp_wmb();
80
81 spin_lock(ptl);
82 /*
83 * We have multiple higher-level entries that point to the same
84 * actual pte location. Fill in each as we go and backtrack on error.
85 * We need all of these so the DTLB pgtable walk code can find the
86 * right higher-level entry without knowing if it's a hugepage or not.
87 */
88 for (i = 0; i < num_hugepd; i++, hpdp++) {
89 if (unlikely(!hugepd_none(*hpdp)))
90 break;
91 hugepd_populate(hpdp, new, pshift);
92 }
93 /* If we bailed from the for loop early, an error occurred, clean up */
94 if (i < num_hugepd) {
95 for (i = i - 1 ; i >= 0; i--, hpdp--)
96 *hpdp = __hugepd(0);
97 kmem_cache_free(cachep, new);
98 } else {
99 kmemleak_ignore(new);
100 }
101 spin_unlock(ptl);
102 return 0;
103 }
104
105 /*
106 * At this point we do the placement change only for BOOK3S 64. This would
107 * possibly work on other subarchs.
108 */
huge_pte_alloc(struct mm_struct * mm,struct vm_area_struct * vma,unsigned long addr,unsigned long sz)109 pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
110 unsigned long addr, unsigned long sz)
111 {
112 pgd_t *pg;
113 p4d_t *p4;
114 pud_t *pu;
115 pmd_t *pm;
116 hugepd_t *hpdp = NULL;
117 unsigned pshift = __ffs(sz);
118 unsigned pdshift = PGDIR_SHIFT;
119 spinlock_t *ptl;
120
121 addr &= ~(sz-1);
122 pg = pgd_offset(mm, addr);
123 p4 = p4d_offset(pg, addr);
124
125 #ifdef CONFIG_PPC_BOOK3S_64
126 if (pshift == PGDIR_SHIFT)
127 /* 16GB huge page */
128 return (pte_t *) p4;
129 else if (pshift > PUD_SHIFT) {
130 /*
131 * We need to use hugepd table
132 */
133 ptl = &mm->page_table_lock;
134 hpdp = (hugepd_t *)p4;
135 } else {
136 pdshift = PUD_SHIFT;
137 pu = pud_alloc(mm, p4, addr);
138 if (!pu)
139 return NULL;
140 if (pshift == PUD_SHIFT)
141 return (pte_t *)pu;
142 else if (pshift > PMD_SHIFT) {
143 ptl = pud_lockptr(mm, pu);
144 hpdp = (hugepd_t *)pu;
145 } else {
146 pdshift = PMD_SHIFT;
147 pm = pmd_alloc(mm, pu, addr);
148 if (!pm)
149 return NULL;
150 if (pshift == PMD_SHIFT)
151 /* 16MB hugepage */
152 return (pte_t *)pm;
153 else {
154 ptl = pmd_lockptr(mm, pm);
155 hpdp = (hugepd_t *)pm;
156 }
157 }
158 }
159 #else
160 if (pshift >= PGDIR_SHIFT) {
161 ptl = &mm->page_table_lock;
162 hpdp = (hugepd_t *)p4;
163 } else {
164 pdshift = PUD_SHIFT;
165 pu = pud_alloc(mm, p4, addr);
166 if (!pu)
167 return NULL;
168 if (pshift >= PUD_SHIFT) {
169 ptl = pud_lockptr(mm, pu);
170 hpdp = (hugepd_t *)pu;
171 } else {
172 pdshift = PMD_SHIFT;
173 pm = pmd_alloc(mm, pu, addr);
174 if (!pm)
175 return NULL;
176 ptl = pmd_lockptr(mm, pm);
177 hpdp = (hugepd_t *)pm;
178 }
179 }
180 #endif
181 if (!hpdp)
182 return NULL;
183
184 if (IS_ENABLED(CONFIG_PPC_8xx) && pshift < PMD_SHIFT)
185 return pte_alloc_map(mm, (pmd_t *)hpdp, addr);
186
187 BUG_ON(!hugepd_none(*hpdp) && !hugepd_ok(*hpdp));
188
189 if (hugepd_none(*hpdp) && __hugepte_alloc(mm, hpdp, addr,
190 pdshift, pshift, ptl))
191 return NULL;
192
193 return hugepte_offset(*hpdp, addr, pdshift);
194 }
195
196 #ifdef CONFIG_PPC_BOOK3S_64
197 /*
198 * Tracks gpages after the device tree is scanned and before the
199 * huge_boot_pages list is ready on pseries.
200 */
201 #define MAX_NUMBER_GPAGES 1024
202 __initdata static u64 gpage_freearray[MAX_NUMBER_GPAGES];
203 __initdata static unsigned nr_gpages;
204
205 /*
206 * Build list of addresses of gigantic pages. This function is used in early
207 * boot before the buddy allocator is setup.
208 */
pseries_add_gpage(u64 addr,u64 page_size,unsigned long number_of_pages)209 void __init pseries_add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages)
210 {
211 if (!addr)
212 return;
213 while (number_of_pages > 0) {
214 gpage_freearray[nr_gpages] = addr;
215 nr_gpages++;
216 number_of_pages--;
217 addr += page_size;
218 }
219 }
220
pseries_alloc_bootmem_huge_page(struct hstate * hstate)221 int __init pseries_alloc_bootmem_huge_page(struct hstate *hstate)
222 {
223 struct huge_bootmem_page *m;
224 if (nr_gpages == 0)
225 return 0;
226 m = phys_to_virt(gpage_freearray[--nr_gpages]);
227 gpage_freearray[nr_gpages] = 0;
228 list_add(&m->list, &huge_boot_pages);
229 m->hstate = hstate;
230 return 1;
231 }
232 #endif
233
234
alloc_bootmem_huge_page(struct hstate * h)235 int __init alloc_bootmem_huge_page(struct hstate *h)
236 {
237
238 #ifdef CONFIG_PPC_BOOK3S_64
239 if (firmware_has_feature(FW_FEATURE_LPAR) && !radix_enabled())
240 return pseries_alloc_bootmem_huge_page(h);
241 #endif
242 return __alloc_bootmem_huge_page(h);
243 }
244
245 #ifndef CONFIG_PPC_BOOK3S_64
246 #define HUGEPD_FREELIST_SIZE \
247 ((PAGE_SIZE - sizeof(struct hugepd_freelist)) / sizeof(pte_t))
248
249 struct hugepd_freelist {
250 struct rcu_head rcu;
251 unsigned int index;
252 void *ptes[];
253 };
254
255 static DEFINE_PER_CPU(struct hugepd_freelist *, hugepd_freelist_cur);
256
hugepd_free_rcu_callback(struct rcu_head * head)257 static void hugepd_free_rcu_callback(struct rcu_head *head)
258 {
259 struct hugepd_freelist *batch =
260 container_of(head, struct hugepd_freelist, rcu);
261 unsigned int i;
262
263 for (i = 0; i < batch->index; i++)
264 kmem_cache_free(PGT_CACHE(PTE_T_ORDER), batch->ptes[i]);
265
266 free_page((unsigned long)batch);
267 }
268
hugepd_free(struct mmu_gather * tlb,void * hugepte)269 static void hugepd_free(struct mmu_gather *tlb, void *hugepte)
270 {
271 struct hugepd_freelist **batchp;
272
273 batchp = &get_cpu_var(hugepd_freelist_cur);
274
275 if (atomic_read(&tlb->mm->mm_users) < 2 ||
276 mm_is_thread_local(tlb->mm)) {
277 kmem_cache_free(PGT_CACHE(PTE_T_ORDER), hugepte);
278 put_cpu_var(hugepd_freelist_cur);
279 return;
280 }
281
282 if (*batchp == NULL) {
283 *batchp = (struct hugepd_freelist *)__get_free_page(GFP_ATOMIC);
284 (*batchp)->index = 0;
285 }
286
287 (*batchp)->ptes[(*batchp)->index++] = hugepte;
288 if ((*batchp)->index == HUGEPD_FREELIST_SIZE) {
289 call_rcu(&(*batchp)->rcu, hugepd_free_rcu_callback);
290 *batchp = NULL;
291 }
292 put_cpu_var(hugepd_freelist_cur);
293 }
294 #else
hugepd_free(struct mmu_gather * tlb,void * hugepte)295 static inline void hugepd_free(struct mmu_gather *tlb, void *hugepte) {}
296 #endif
297
free_hugepd_range(struct mmu_gather * tlb,hugepd_t * hpdp,int pdshift,unsigned long start,unsigned long end,unsigned long floor,unsigned long ceiling)298 static void free_hugepd_range(struct mmu_gather *tlb, hugepd_t *hpdp, int pdshift,
299 unsigned long start, unsigned long end,
300 unsigned long floor, unsigned long ceiling)
301 {
302 pte_t *hugepte = hugepd_page(*hpdp);
303 int i;
304
305 unsigned long pdmask = ~((1UL << pdshift) - 1);
306 unsigned int num_hugepd = 1;
307 unsigned int shift = hugepd_shift(*hpdp);
308
309 /* Note: On fsl the hpdp may be the first of several */
310 if (shift > pdshift)
311 num_hugepd = 1 << (shift - pdshift);
312
313 start &= pdmask;
314 if (start < floor)
315 return;
316 if (ceiling) {
317 ceiling &= pdmask;
318 if (! ceiling)
319 return;
320 }
321 if (end - 1 > ceiling - 1)
322 return;
323
324 for (i = 0; i < num_hugepd; i++, hpdp++)
325 *hpdp = __hugepd(0);
326
327 if (shift >= pdshift)
328 hugepd_free(tlb, hugepte);
329 else
330 pgtable_free_tlb(tlb, hugepte,
331 get_hugepd_cache_index(pdshift - shift));
332 }
333
hugetlb_free_pte_range(struct mmu_gather * tlb,pmd_t * pmd,unsigned long addr,unsigned long end,unsigned long floor,unsigned long ceiling)334 static void hugetlb_free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
335 unsigned long addr, unsigned long end,
336 unsigned long floor, unsigned long ceiling)
337 {
338 unsigned long start = addr;
339 pgtable_t token = pmd_pgtable(*pmd);
340
341 start &= PMD_MASK;
342 if (start < floor)
343 return;
344 if (ceiling) {
345 ceiling &= PMD_MASK;
346 if (!ceiling)
347 return;
348 }
349 if (end - 1 > ceiling - 1)
350 return;
351
352 pmd_clear(pmd);
353 pte_free_tlb(tlb, token, addr);
354 mm_dec_nr_ptes(tlb->mm);
355 }
356
hugetlb_free_pmd_range(struct mmu_gather * tlb,pud_t * pud,unsigned long addr,unsigned long end,unsigned long floor,unsigned long ceiling)357 static void hugetlb_free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
358 unsigned long addr, unsigned long end,
359 unsigned long floor, unsigned long ceiling)
360 {
361 pmd_t *pmd;
362 unsigned long next;
363 unsigned long start;
364
365 start = addr;
366 do {
367 unsigned long more;
368
369 pmd = pmd_offset(pud, addr);
370 next = pmd_addr_end(addr, end);
371 if (!is_hugepd(__hugepd(pmd_val(*pmd)))) {
372 if (pmd_none_or_clear_bad(pmd))
373 continue;
374
375 /*
376 * if it is not hugepd pointer, we should already find
377 * it cleared.
378 */
379 WARN_ON(!IS_ENABLED(CONFIG_PPC_8xx));
380
381 hugetlb_free_pte_range(tlb, pmd, addr, end, floor, ceiling);
382
383 continue;
384 }
385 /*
386 * Increment next by the size of the huge mapping since
387 * there may be more than one entry at this level for a
388 * single hugepage, but all of them point to
389 * the same kmem cache that holds the hugepte.
390 */
391 more = addr + (1 << hugepd_shift(*(hugepd_t *)pmd));
392 if (more > next)
393 next = more;
394
395 free_hugepd_range(tlb, (hugepd_t *)pmd, PMD_SHIFT,
396 addr, next, floor, ceiling);
397 } while (addr = next, addr != end);
398
399 start &= PUD_MASK;
400 if (start < floor)
401 return;
402 if (ceiling) {
403 ceiling &= PUD_MASK;
404 if (!ceiling)
405 return;
406 }
407 if (end - 1 > ceiling - 1)
408 return;
409
410 pmd = pmd_offset(pud, start);
411 pud_clear(pud);
412 pmd_free_tlb(tlb, pmd, start);
413 mm_dec_nr_pmds(tlb->mm);
414 }
415
hugetlb_free_pud_range(struct mmu_gather * tlb,p4d_t * p4d,unsigned long addr,unsigned long end,unsigned long floor,unsigned long ceiling)416 static void hugetlb_free_pud_range(struct mmu_gather *tlb, p4d_t *p4d,
417 unsigned long addr, unsigned long end,
418 unsigned long floor, unsigned long ceiling)
419 {
420 pud_t *pud;
421 unsigned long next;
422 unsigned long start;
423
424 start = addr;
425 do {
426 pud = pud_offset(p4d, addr);
427 next = pud_addr_end(addr, end);
428 if (!is_hugepd(__hugepd(pud_val(*pud)))) {
429 if (pud_none_or_clear_bad(pud))
430 continue;
431 hugetlb_free_pmd_range(tlb, pud, addr, next, floor,
432 ceiling);
433 } else {
434 unsigned long more;
435 /*
436 * Increment next by the size of the huge mapping since
437 * there may be more than one entry at this level for a
438 * single hugepage, but all of them point to
439 * the same kmem cache that holds the hugepte.
440 */
441 more = addr + (1 << hugepd_shift(*(hugepd_t *)pud));
442 if (more > next)
443 next = more;
444
445 free_hugepd_range(tlb, (hugepd_t *)pud, PUD_SHIFT,
446 addr, next, floor, ceiling);
447 }
448 } while (addr = next, addr != end);
449
450 start &= PGDIR_MASK;
451 if (start < floor)
452 return;
453 if (ceiling) {
454 ceiling &= PGDIR_MASK;
455 if (!ceiling)
456 return;
457 }
458 if (end - 1 > ceiling - 1)
459 return;
460
461 pud = pud_offset(p4d, start);
462 p4d_clear(p4d);
463 pud_free_tlb(tlb, pud, start);
464 mm_dec_nr_puds(tlb->mm);
465 }
466
467 /*
468 * This function frees user-level page tables of a process.
469 */
hugetlb_free_pgd_range(struct mmu_gather * tlb,unsigned long addr,unsigned long end,unsigned long floor,unsigned long ceiling)470 void hugetlb_free_pgd_range(struct mmu_gather *tlb,
471 unsigned long addr, unsigned long end,
472 unsigned long floor, unsigned long ceiling)
473 {
474 pgd_t *pgd;
475 p4d_t *p4d;
476 unsigned long next;
477
478 /*
479 * Because there are a number of different possible pagetable
480 * layouts for hugepage ranges, we limit knowledge of how
481 * things should be laid out to the allocation path
482 * (huge_pte_alloc(), above). Everything else works out the
483 * structure as it goes from information in the hugepd
484 * pointers. That means that we can't here use the
485 * optimization used in the normal page free_pgd_range(), of
486 * checking whether we're actually covering a large enough
487 * range to have to do anything at the top level of the walk
488 * instead of at the bottom.
489 *
490 * To make sense of this, you should probably go read the big
491 * block comment at the top of the normal free_pgd_range(),
492 * too.
493 */
494
495 do {
496 next = pgd_addr_end(addr, end);
497 pgd = pgd_offset(tlb->mm, addr);
498 p4d = p4d_offset(pgd, addr);
499 if (!is_hugepd(__hugepd(pgd_val(*pgd)))) {
500 if (p4d_none_or_clear_bad(p4d))
501 continue;
502 hugetlb_free_pud_range(tlb, p4d, addr, next, floor, ceiling);
503 } else {
504 unsigned long more;
505 /*
506 * Increment next by the size of the huge mapping since
507 * there may be more than one entry at the pgd level
508 * for a single hugepage, but all of them point to the
509 * same kmem cache that holds the hugepte.
510 */
511 more = addr + (1 << hugepd_shift(*(hugepd_t *)pgd));
512 if (more > next)
513 next = more;
514
515 free_hugepd_range(tlb, (hugepd_t *)p4d, PGDIR_SHIFT,
516 addr, next, floor, ceiling);
517 }
518 } while (addr = next, addr != end);
519 }
520
follow_huge_pd(struct vm_area_struct * vma,unsigned long address,hugepd_t hpd,int flags,int pdshift)521 struct page *follow_huge_pd(struct vm_area_struct *vma,
522 unsigned long address, hugepd_t hpd,
523 int flags, int pdshift)
524 {
525 pte_t *ptep;
526 spinlock_t *ptl;
527 struct page *page = NULL;
528 unsigned long mask;
529 int shift = hugepd_shift(hpd);
530 struct mm_struct *mm = vma->vm_mm;
531
532 retry:
533 /*
534 * hugepage directory entries are protected by mm->page_table_lock
535 * Use this instead of huge_pte_lockptr
536 */
537 ptl = &mm->page_table_lock;
538 spin_lock(ptl);
539
540 ptep = hugepte_offset(hpd, address, pdshift);
541 if (pte_present(*ptep)) {
542 mask = (1UL << shift) - 1;
543 page = pte_page(*ptep);
544 page += ((address & mask) >> PAGE_SHIFT);
545 if (flags & FOLL_GET)
546 get_page(page);
547 } else {
548 if (is_hugetlb_entry_migration(*ptep)) {
549 spin_unlock(ptl);
550 __migration_entry_wait(mm, ptep, ptl);
551 goto retry;
552 }
553 }
554 spin_unlock(ptl);
555 return page;
556 }
557
558 #ifdef CONFIG_PPC_MM_SLICES
hugetlb_get_unmapped_area(struct file * file,unsigned long addr,unsigned long len,unsigned long pgoff,unsigned long flags)559 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
560 unsigned long len, unsigned long pgoff,
561 unsigned long flags)
562 {
563 struct hstate *hstate = hstate_file(file);
564 int mmu_psize = shift_to_mmu_psize(huge_page_shift(hstate));
565
566 #ifdef CONFIG_PPC_RADIX_MMU
567 if (radix_enabled())
568 return radix__hugetlb_get_unmapped_area(file, addr, len,
569 pgoff, flags);
570 #endif
571 return slice_get_unmapped_area(addr, len, flags, mmu_psize, 1);
572 }
573 #endif
574
vma_mmu_pagesize(struct vm_area_struct * vma)575 unsigned long vma_mmu_pagesize(struct vm_area_struct *vma)
576 {
577 /* With radix we don't use slice, so derive it from vma*/
578 if (IS_ENABLED(CONFIG_PPC_MM_SLICES) && !radix_enabled()) {
579 unsigned int psize = get_slice_psize(vma->vm_mm, vma->vm_start);
580
581 return 1UL << mmu_psize_to_shift(psize);
582 }
583 return vma_kernel_pagesize(vma);
584 }
585
arch_hugetlb_valid_size(unsigned long size)586 bool __init arch_hugetlb_valid_size(unsigned long size)
587 {
588 int shift = __ffs(size);
589 int mmu_psize;
590
591 /* Check that it is a page size supported by the hardware and
592 * that it fits within pagetable and slice limits. */
593 if (size <= PAGE_SIZE || !is_power_of_2(size))
594 return false;
595
596 mmu_psize = check_and_get_huge_psize(shift);
597 if (mmu_psize < 0)
598 return false;
599
600 BUG_ON(mmu_psize_defs[mmu_psize].shift != shift);
601
602 return true;
603 }
604
add_huge_page_size(unsigned long long size)605 static int __init add_huge_page_size(unsigned long long size)
606 {
607 int shift = __ffs(size);
608
609 if (!arch_hugetlb_valid_size((unsigned long)size))
610 return -EINVAL;
611
612 hugetlb_add_hstate(shift - PAGE_SHIFT);
613 return 0;
614 }
615
hugetlbpage_init(void)616 static int __init hugetlbpage_init(void)
617 {
618 bool configured = false;
619 int psize;
620
621 if (hugetlb_disabled) {
622 pr_info("HugeTLB support is disabled!\n");
623 return 0;
624 }
625
626 if (IS_ENABLED(CONFIG_PPC_BOOK3S_64) && !radix_enabled() &&
627 !mmu_has_feature(MMU_FTR_16M_PAGE))
628 return -ENODEV;
629
630 for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) {
631 unsigned shift;
632 unsigned pdshift;
633
634 if (!mmu_psize_defs[psize].shift)
635 continue;
636
637 shift = mmu_psize_to_shift(psize);
638
639 #ifdef CONFIG_PPC_BOOK3S_64
640 if (shift > PGDIR_SHIFT)
641 continue;
642 else if (shift > PUD_SHIFT)
643 pdshift = PGDIR_SHIFT;
644 else if (shift > PMD_SHIFT)
645 pdshift = PUD_SHIFT;
646 else
647 pdshift = PMD_SHIFT;
648 #else
649 if (shift < PUD_SHIFT)
650 pdshift = PMD_SHIFT;
651 else if (shift < PGDIR_SHIFT)
652 pdshift = PUD_SHIFT;
653 else
654 pdshift = PGDIR_SHIFT;
655 #endif
656
657 if (add_huge_page_size(1ULL << shift) < 0)
658 continue;
659 /*
660 * if we have pdshift and shift value same, we don't
661 * use pgt cache for hugepd.
662 */
663 if (pdshift > shift) {
664 if (!IS_ENABLED(CONFIG_PPC_8xx))
665 pgtable_cache_add(pdshift - shift);
666 } else if (IS_ENABLED(CONFIG_PPC_FSL_BOOK3E) ||
667 IS_ENABLED(CONFIG_PPC_8xx)) {
668 pgtable_cache_add(PTE_T_ORDER);
669 }
670
671 configured = true;
672 }
673
674 if (configured) {
675 if (IS_ENABLED(CONFIG_HUGETLB_PAGE_SIZE_VARIABLE))
676 hugetlbpage_init_default();
677 } else
678 pr_info("Failed to initialize. Disabling HugeTLB");
679
680 return 0;
681 }
682
683 arch_initcall(hugetlbpage_init);
684
flush_dcache_icache_hugepage(struct page * page)685 void flush_dcache_icache_hugepage(struct page *page)
686 {
687 int i;
688 void *start;
689
690 BUG_ON(!PageCompound(page));
691
692 for (i = 0; i < compound_nr(page); i++) {
693 if (!PageHighMem(page)) {
694 __flush_dcache_icache(page_address(page+i));
695 } else {
696 start = kmap_atomic(page+i);
697 __flush_dcache_icache(start);
698 kunmap_atomic(start);
699 }
700 }
701 }
702
gigantic_hugetlb_cma_reserve(void)703 void __init gigantic_hugetlb_cma_reserve(void)
704 {
705 unsigned long order = 0;
706
707 if (radix_enabled())
708 order = PUD_SHIFT - PAGE_SHIFT;
709 else if (!firmware_has_feature(FW_FEATURE_LPAR) && mmu_psize_defs[MMU_PAGE_16G].shift)
710 /*
711 * For pseries we do use ibm,expected#pages for reserving 16G pages.
712 */
713 order = mmu_psize_to_shift(MMU_PAGE_16G) - PAGE_SHIFT;
714
715 if (order) {
716 VM_WARN_ON(order < MAX_ORDER);
717 hugetlb_cma_reserve(order);
718 }
719 }
720