1 // SPDX-License-Identifier: GPL-2.0
3 * SPARC64 Huge TLB page support.
5 * Copyright (C) 2002, 2003, 2006 David S. Miller (davem@davemloft.net)
10 #include <linux/sched/mm.h>
11 #include <linux/hugetlb.h>
12 #include <linux/pagemap.h>
13 #include <linux/sysctl.h>
16 #include <asm/pgalloc.h>
18 #include <asm/tlbflush.h>
19 #include <asm/cacheflush.h>
20 #include <asm/mmu_context.h>
22 /* Slightly simplified from the non-hugepage variant because by
23 * definition we don't have to worry about any page coloring stuff
26 static unsigned long hugetlb_get_unmapped_area_bottomup(struct file *filp,
32 struct hstate *h = hstate_file(filp);
33 unsigned long task_size = TASK_SIZE;
34 struct vm_unmapped_area_info info;
36 if (test_thread_flag(TIF_32BIT))
37 task_size = STACK_TOP32;
41 info.low_limit = TASK_UNMAPPED_BASE;
42 info.high_limit = min(task_size, VA_EXCLUDE_START);
43 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
44 info.align_offset = 0;
45 addr = vm_unmapped_area(&info);
47 if ((addr & ~PAGE_MASK) && task_size > VA_EXCLUDE_END) {
48 VM_BUG_ON(addr != -ENOMEM);
49 info.low_limit = VA_EXCLUDE_END;
50 info.high_limit = task_size;
51 addr = vm_unmapped_area(&info);
58 hugetlb_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
59 const unsigned long len,
60 const unsigned long pgoff,
61 const unsigned long flags)
63 struct hstate *h = hstate_file(filp);
64 struct mm_struct *mm = current->mm;
65 unsigned long addr = addr0;
66 struct vm_unmapped_area_info info;
68 /* This should only ever run for 32-bit processes. */
69 BUG_ON(!test_thread_flag(TIF_32BIT));
71 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
73 info.low_limit = PAGE_SIZE;
74 info.high_limit = mm->mmap_base;
75 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
76 info.align_offset = 0;
77 addr = vm_unmapped_area(&info);
80 * A failed mmap() very likely causes application failure,
81 * so fall back to the bottom-up function here. This scenario
82 * can happen with large stack limits and large mmap()
85 if (addr & ~PAGE_MASK) {
86 VM_BUG_ON(addr != -ENOMEM);
88 info.low_limit = TASK_UNMAPPED_BASE;
89 info.high_limit = STACK_TOP32;
90 addr = vm_unmapped_area(&info);
97 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
98 unsigned long len, unsigned long pgoff, unsigned long flags)
100 struct hstate *h = hstate_file(file);
101 struct mm_struct *mm = current->mm;
102 struct vm_area_struct *vma;
103 unsigned long task_size = TASK_SIZE;
105 if (test_thread_flag(TIF_32BIT))
106 task_size = STACK_TOP32;
108 if (len & ~huge_page_mask(h))
113 if (flags & MAP_FIXED) {
114 if (prepare_hugepage_range(file, addr, len))
120 addr = ALIGN(addr, huge_page_size(h));
121 vma = find_vma(mm, addr);
122 if (task_size - len >= addr &&
123 (!vma || addr + len <= vm_start_gap(vma)))
126 if (mm->get_unmapped_area == arch_get_unmapped_area)
127 return hugetlb_get_unmapped_area_bottomup(file, addr, len,
130 return hugetlb_get_unmapped_area_topdown(file, addr, len,
134 static pte_t sun4u_hugepage_shift_to_tte(pte_t entry, unsigned int shift)
139 static pte_t sun4v_hugepage_shift_to_tte(pte_t entry, unsigned int shift)
141 unsigned long hugepage_size = _PAGE_SZ4MB_4V;
143 pte_val(entry) = pte_val(entry) & ~_PAGE_SZALL_4V;
146 case HPAGE_16GB_SHIFT:
147 hugepage_size = _PAGE_SZ16GB_4V;
148 pte_val(entry) |= _PAGE_PUD_HUGE;
150 case HPAGE_2GB_SHIFT:
151 hugepage_size = _PAGE_SZ2GB_4V;
152 pte_val(entry) |= _PAGE_PMD_HUGE;
154 case HPAGE_256MB_SHIFT:
155 hugepage_size = _PAGE_SZ256MB_4V;
156 pte_val(entry) |= _PAGE_PMD_HUGE;
159 pte_val(entry) |= _PAGE_PMD_HUGE;
161 case HPAGE_64K_SHIFT:
162 hugepage_size = _PAGE_SZ64K_4V;
165 WARN_ONCE(1, "unsupported hugepage shift=%u\n", shift);
168 pte_val(entry) = pte_val(entry) | hugepage_size;
172 static pte_t hugepage_shift_to_tte(pte_t entry, unsigned int shift)
174 if (tlb_type == hypervisor)
175 return sun4v_hugepage_shift_to_tte(entry, shift);
177 return sun4u_hugepage_shift_to_tte(entry, shift);
180 pte_t arch_make_huge_pte(pte_t entry, unsigned int shift, vm_flags_t flags)
184 entry = pte_mkhuge(entry);
185 pte = hugepage_shift_to_tte(entry, shift);
187 #ifdef CONFIG_SPARC64
188 /* If this vma has ADI enabled on it, turn on TTE.mcd
190 if (flags & VM_SPARC_ADI)
191 return pte_mkmcd(pte);
193 return pte_mknotmcd(pte);
199 static unsigned int sun4v_huge_tte_to_shift(pte_t entry)
201 unsigned long tte_szbits = pte_val(entry) & _PAGE_SZALL_4V;
204 switch (tte_szbits) {
205 case _PAGE_SZ16GB_4V:
206 shift = HPAGE_16GB_SHIFT;
209 shift = HPAGE_2GB_SHIFT;
211 case _PAGE_SZ256MB_4V:
212 shift = HPAGE_256MB_SHIFT;
215 shift = REAL_HPAGE_SHIFT;
218 shift = HPAGE_64K_SHIFT;
227 static unsigned int sun4u_huge_tte_to_shift(pte_t entry)
229 unsigned long tte_szbits = pte_val(entry) & _PAGE_SZALL_4U;
232 switch (tte_szbits) {
233 case _PAGE_SZ256MB_4U:
234 shift = HPAGE_256MB_SHIFT;
237 shift = REAL_HPAGE_SHIFT;
240 shift = HPAGE_64K_SHIFT;
249 static unsigned long tte_to_shift(pte_t entry)
251 if (tlb_type == hypervisor)
252 return sun4v_huge_tte_to_shift(entry);
254 return sun4u_huge_tte_to_shift(entry);
257 static unsigned int huge_tte_to_shift(pte_t entry)
259 unsigned long shift = tte_to_shift(entry);
261 if (shift == PAGE_SHIFT)
262 WARN_ONCE(1, "tto_to_shift: invalid hugepage tte=0x%lx\n",
268 static unsigned long huge_tte_to_size(pte_t pte)
270 unsigned long size = 1UL << huge_tte_to_shift(pte);
272 if (size == REAL_HPAGE_SIZE)
277 unsigned long pud_leaf_size(pud_t pud) { return 1UL << tte_to_shift(*(pte_t *)&pud); }
278 unsigned long pmd_leaf_size(pmd_t pmd) { return 1UL << tte_to_shift(*(pte_t *)&pmd); }
279 unsigned long pte_leaf_size(pte_t pte) { return 1UL << tte_to_shift(pte); }
281 pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
282 unsigned long addr, unsigned long sz)
289 pgd = pgd_offset(mm, addr);
290 p4d = p4d_offset(pgd, addr);
291 pud = pud_alloc(mm, p4d, addr);
296 pmd = pmd_alloc(mm, pud, addr);
301 return pte_alloc_huge(mm, pmd, addr);
304 pte_t *huge_pte_offset(struct mm_struct *mm,
305 unsigned long addr, unsigned long sz)
312 pgd = pgd_offset(mm, addr);
315 p4d = p4d_offset(pgd, addr);
318 pud = pud_offset(p4d, addr);
321 if (is_hugetlb_pud(*pud))
323 pmd = pmd_offset(pud, addr);
326 if (is_hugetlb_pmd(*pmd))
328 return pte_offset_huge(pmd, addr);
331 void __set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
332 pte_t *ptep, pte_t entry)
334 unsigned int nptes, orig_shift, shift;
335 unsigned long i, size;
338 size = huge_tte_to_size(entry);
341 if (size >= PUD_SIZE)
343 else if (size >= PMD_SIZE)
348 nptes = size >> shift;
350 if (!pte_present(*ptep) && pte_present(entry))
351 mm->context.hugetlb_pte_count += nptes;
355 orig_shift = pte_none(orig) ? PAGE_SHIFT : huge_tte_to_shift(orig);
357 for (i = 0; i < nptes; i++)
358 ptep[i] = __pte(pte_val(entry) + (i << shift));
360 maybe_tlb_batch_add(mm, addr, ptep, orig, 0, orig_shift);
361 /* An HPAGE_SIZE'ed page is composed of two REAL_HPAGE_SIZE'ed pages */
362 if (size == HPAGE_SIZE)
363 maybe_tlb_batch_add(mm, addr + REAL_HPAGE_SIZE, ptep, orig, 0,
367 void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
368 pte_t *ptep, pte_t entry, unsigned long sz)
370 __set_huge_pte_at(mm, addr, ptep, entry);
373 pte_t huge_ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
376 unsigned int i, nptes, orig_shift, shift;
381 size = huge_tte_to_size(entry);
384 if (size >= PUD_SIZE)
386 else if (size >= PMD_SIZE)
391 nptes = size >> shift;
392 orig_shift = pte_none(entry) ? PAGE_SHIFT : huge_tte_to_shift(entry);
394 if (pte_present(entry))
395 mm->context.hugetlb_pte_count -= nptes;
398 for (i = 0; i < nptes; i++)
399 ptep[i] = __pte(0UL);
401 maybe_tlb_batch_add(mm, addr, ptep, entry, 0, orig_shift);
402 /* An HPAGE_SIZE'ed page is composed of two REAL_HPAGE_SIZE'ed pages */
403 if (size == HPAGE_SIZE)
404 maybe_tlb_batch_add(mm, addr + REAL_HPAGE_SIZE, ptep, entry, 0,
410 int pmd_huge(pmd_t pmd)
412 return !pmd_none(pmd) &&
413 (pmd_val(pmd) & (_PAGE_VALID|_PAGE_PMD_HUGE)) != _PAGE_VALID;
416 int pud_huge(pud_t pud)
418 return !pud_none(pud) &&
419 (pud_val(pud) & (_PAGE_VALID|_PAGE_PUD_HUGE)) != _PAGE_VALID;
422 static void hugetlb_free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
425 pgtable_t token = pmd_pgtable(*pmd);
428 pte_free_tlb(tlb, token, addr);
429 mm_dec_nr_ptes(tlb->mm);
432 static void hugetlb_free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
433 unsigned long addr, unsigned long end,
434 unsigned long floor, unsigned long ceiling)
441 pmd = pmd_offset(pud, addr);
443 next = pmd_addr_end(addr, end);
446 if (is_hugetlb_pmd(*pmd))
449 hugetlb_free_pte_range(tlb, pmd, addr);
450 } while (pmd++, addr = next, addr != end);
460 if (end - 1 > ceiling - 1)
463 pmd = pmd_offset(pud, start);
465 pmd_free_tlb(tlb, pmd, start);
466 mm_dec_nr_pmds(tlb->mm);
469 static void hugetlb_free_pud_range(struct mmu_gather *tlb, p4d_t *p4d,
470 unsigned long addr, unsigned long end,
471 unsigned long floor, unsigned long ceiling)
478 pud = pud_offset(p4d, addr);
480 next = pud_addr_end(addr, end);
481 if (pud_none_or_clear_bad(pud))
483 if (is_hugetlb_pud(*pud))
486 hugetlb_free_pmd_range(tlb, pud, addr, next, floor,
488 } while (pud++, addr = next, addr != end);
494 ceiling &= PGDIR_MASK;
498 if (end - 1 > ceiling - 1)
501 pud = pud_offset(p4d, start);
503 pud_free_tlb(tlb, pud, start);
504 mm_dec_nr_puds(tlb->mm);
507 void hugetlb_free_pgd_range(struct mmu_gather *tlb,
508 unsigned long addr, unsigned long end,
509 unsigned long floor, unsigned long ceiling)
526 if (end - 1 > ceiling - 1)
531 pgd = pgd_offset(tlb->mm, addr);
532 p4d = p4d_offset(pgd, addr);
534 next = p4d_addr_end(addr, end);
535 if (p4d_none_or_clear_bad(p4d))
537 hugetlb_free_pud_range(tlb, p4d, addr, next, floor, ceiling);
538 } while (p4d++, addr = next, addr != end);