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, struct vm_area_struct *vma,
181 struct page *page, int writeable)
183 unsigned int shift = huge_page_shift(hstate_vma(vma));
186 pte = hugepage_shift_to_tte(entry, shift);
188 #ifdef CONFIG_SPARC64
189 /* If this vma has ADI enabled on it, turn on TTE.mcd
191 if (vma->vm_flags & VM_SPARC_ADI)
192 return pte_mkmcd(pte);
194 return pte_mknotmcd(pte);
200 static unsigned int sun4v_huge_tte_to_shift(pte_t entry)
202 unsigned long tte_szbits = pte_val(entry) & _PAGE_SZALL_4V;
205 switch (tte_szbits) {
206 case _PAGE_SZ16GB_4V:
207 shift = HPAGE_16GB_SHIFT;
210 shift = HPAGE_2GB_SHIFT;
212 case _PAGE_SZ256MB_4V:
213 shift = HPAGE_256MB_SHIFT;
216 shift = REAL_HPAGE_SHIFT;
219 shift = HPAGE_64K_SHIFT;
228 static unsigned int sun4u_huge_tte_to_shift(pte_t entry)
230 unsigned long tte_szbits = pte_val(entry) & _PAGE_SZALL_4U;
233 switch (tte_szbits) {
234 case _PAGE_SZ256MB_4U:
235 shift = HPAGE_256MB_SHIFT;
238 shift = REAL_HPAGE_SHIFT;
241 shift = HPAGE_64K_SHIFT;
250 static unsigned int huge_tte_to_shift(pte_t entry)
254 if (tlb_type == hypervisor)
255 shift = sun4v_huge_tte_to_shift(entry);
257 shift = sun4u_huge_tte_to_shift(entry);
259 if (shift == PAGE_SHIFT)
260 WARN_ONCE(1, "tto_to_shift: invalid hugepage tte=0x%lx\n",
266 static unsigned long huge_tte_to_size(pte_t pte)
268 unsigned long size = 1UL << huge_tte_to_shift(pte);
270 if (size == REAL_HPAGE_SIZE)
275 pte_t *huge_pte_alloc(struct mm_struct *mm,
276 unsigned long addr, unsigned long sz)
283 pgd = pgd_offset(mm, addr);
284 p4d = p4d_offset(pgd, addr);
285 pud = pud_alloc(mm, p4d, addr);
290 pmd = pmd_alloc(mm, pud, addr);
295 return pte_alloc_map(mm, pmd, addr);
298 pte_t *huge_pte_offset(struct mm_struct *mm,
299 unsigned long addr, unsigned long sz)
306 pgd = pgd_offset(mm, addr);
309 p4d = p4d_offset(pgd, addr);
312 pud = pud_offset(p4d, addr);
315 if (is_hugetlb_pud(*pud))
317 pmd = pmd_offset(pud, addr);
320 if (is_hugetlb_pmd(*pmd))
322 return pte_offset_map(pmd, addr);
325 void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
326 pte_t *ptep, pte_t entry)
328 unsigned int nptes, orig_shift, shift;
329 unsigned long i, size;
332 size = huge_tte_to_size(entry);
335 if (size >= PUD_SIZE)
337 else if (size >= PMD_SIZE)
342 nptes = size >> shift;
344 if (!pte_present(*ptep) && pte_present(entry))
345 mm->context.hugetlb_pte_count += nptes;
349 orig_shift = pte_none(orig) ? PAGE_SHIFT : huge_tte_to_shift(orig);
351 for (i = 0; i < nptes; i++)
352 ptep[i] = __pte(pte_val(entry) + (i << shift));
354 maybe_tlb_batch_add(mm, addr, ptep, orig, 0, orig_shift);
355 /* An HPAGE_SIZE'ed page is composed of two REAL_HPAGE_SIZE'ed pages */
356 if (size == HPAGE_SIZE)
357 maybe_tlb_batch_add(mm, addr + REAL_HPAGE_SIZE, ptep, orig, 0,
361 pte_t huge_ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
364 unsigned int i, nptes, orig_shift, shift;
369 size = huge_tte_to_size(entry);
372 if (size >= PUD_SIZE)
374 else if (size >= PMD_SIZE)
379 nptes = size >> shift;
380 orig_shift = pte_none(entry) ? PAGE_SHIFT : huge_tte_to_shift(entry);
382 if (pte_present(entry))
383 mm->context.hugetlb_pte_count -= nptes;
386 for (i = 0; i < nptes; i++)
387 ptep[i] = __pte(0UL);
389 maybe_tlb_batch_add(mm, addr, ptep, entry, 0, orig_shift);
390 /* An HPAGE_SIZE'ed page is composed of two REAL_HPAGE_SIZE'ed pages */
391 if (size == HPAGE_SIZE)
392 maybe_tlb_batch_add(mm, addr + REAL_HPAGE_SIZE, ptep, entry, 0,
398 int pmd_huge(pmd_t pmd)
400 return !pmd_none(pmd) &&
401 (pmd_val(pmd) & (_PAGE_VALID|_PAGE_PMD_HUGE)) != _PAGE_VALID;
404 int pud_huge(pud_t pud)
406 return !pud_none(pud) &&
407 (pud_val(pud) & (_PAGE_VALID|_PAGE_PUD_HUGE)) != _PAGE_VALID;
410 static void hugetlb_free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
413 pgtable_t token = pmd_pgtable(*pmd);
416 pte_free_tlb(tlb, token, addr);
417 mm_dec_nr_ptes(tlb->mm);
420 static void hugetlb_free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
421 unsigned long addr, unsigned long end,
422 unsigned long floor, unsigned long ceiling)
429 pmd = pmd_offset(pud, addr);
431 next = pmd_addr_end(addr, end);
434 if (is_hugetlb_pmd(*pmd))
437 hugetlb_free_pte_range(tlb, pmd, addr);
438 } while (pmd++, addr = next, addr != end);
448 if (end - 1 > ceiling - 1)
451 pmd = pmd_offset(pud, start);
453 pmd_free_tlb(tlb, pmd, start);
454 mm_dec_nr_pmds(tlb->mm);
457 static void hugetlb_free_pud_range(struct mmu_gather *tlb, p4d_t *p4d,
458 unsigned long addr, unsigned long end,
459 unsigned long floor, unsigned long ceiling)
466 pud = pud_offset(p4d, addr);
468 next = pud_addr_end(addr, end);
469 if (pud_none_or_clear_bad(pud))
471 if (is_hugetlb_pud(*pud))
474 hugetlb_free_pmd_range(tlb, pud, addr, next, floor,
476 } while (pud++, addr = next, addr != end);
482 ceiling &= PGDIR_MASK;
486 if (end - 1 > ceiling - 1)
489 pud = pud_offset(p4d, start);
491 pud_free_tlb(tlb, pud, start);
492 mm_dec_nr_puds(tlb->mm);
495 void hugetlb_free_pgd_range(struct mmu_gather *tlb,
496 unsigned long addr, unsigned long end,
497 unsigned long floor, unsigned long ceiling)
514 if (end - 1 > ceiling - 1)
519 pgd = pgd_offset(tlb->mm, addr);
520 p4d = p4d_offset(pgd, addr);
522 next = p4d_addr_end(addr, end);
523 if (p4d_none_or_clear_bad(p4d))
525 hugetlb_free_pud_range(tlb, p4d, addr, next, floor, ceiling);
526 } while (p4d++, addr = next, addr != end);