2 * arch/arm64/mm/hugetlbpage.c
4 * Copyright (C) 2013 Linaro Ltd.
6 * Based on arch/x86/mm/hugetlbpage.c.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 as
10 * published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
18 #include <linux/init.h>
21 #include <linux/hugetlb.h>
22 #include <linux/pagemap.h>
23 #include <linux/err.h>
24 #include <linux/sysctl.h>
27 #include <asm/tlbflush.h>
28 #include <asm/pgalloc.h>
30 int pmd_huge(pmd_t pmd)
32 return pmd_val(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT);
35 int pud_huge(pud_t pud)
37 #ifndef __PAGETABLE_PMD_FOLDED
38 return pud_val(pud) && !(pud_val(pud) & PUD_TABLE_BIT);
45 * Select all bits except the pfn
47 static inline pgprot_t pte_pgprot(pte_t pte)
49 unsigned long pfn = pte_pfn(pte);
51 return __pgprot(pte_val(pfn_pte(pfn, __pgprot(0))) ^ pte_val(pte));
54 static int find_num_contig(struct mm_struct *mm, unsigned long addr,
55 pte_t *ptep, size_t *pgsize)
57 pgd_t *pgdp = pgd_offset(mm, addr);
62 pudp = pud_offset(pgdp, addr);
63 pmdp = pmd_offset(pudp, addr);
64 if ((pte_t *)pmdp == ptep) {
71 static inline int num_contig_ptes(unsigned long size, size_t *pgsize)
78 #ifdef CONFIG_ARM64_4K_PAGES
86 contig_ptes = CONT_PMDS;
90 contig_ptes = CONT_PTES;
98 * Changing some bits of contiguous entries requires us to follow a
99 * Break-Before-Make approach, breaking the whole contiguous set
100 * before we can change any entries. See ARM DDI 0487A.k_iss10775,
101 * "Misprogramming of the Contiguous bit", page D4-1762.
103 * This helper performs the break step.
105 static pte_t get_clear_flush(struct mm_struct *mm,
108 unsigned long pgsize,
109 unsigned long ncontig)
111 pte_t orig_pte = huge_ptep_get(ptep);
112 bool valid = pte_valid(orig_pte);
113 unsigned long i, saddr = addr;
115 for (i = 0; i < ncontig; i++, addr += pgsize, ptep++) {
116 pte_t pte = ptep_get_and_clear(mm, addr, ptep);
119 * If HW_AFDBM is enabled, then the HW could turn on
120 * the dirty bit for any page in the set, so check
121 * them all. All hugetlb entries are already young.
124 orig_pte = pte_mkdirty(orig_pte);
128 struct vm_area_struct vma = TLB_FLUSH_VMA(mm, 0);
129 flush_tlb_range(&vma, saddr, addr);
135 * Changing some bits of contiguous entries requires us to follow a
136 * Break-Before-Make approach, breaking the whole contiguous set
137 * before we can change any entries. See ARM DDI 0487A.k_iss10775,
138 * "Misprogramming of the Contiguous bit", page D4-1762.
140 * This helper performs the break step for use cases where the
141 * original pte is not needed.
143 static void clear_flush(struct mm_struct *mm,
146 unsigned long pgsize,
147 unsigned long ncontig)
149 struct vm_area_struct vma = TLB_FLUSH_VMA(mm, 0);
150 unsigned long i, saddr = addr;
152 for (i = 0; i < ncontig; i++, addr += pgsize, ptep++)
153 pte_clear(mm, addr, ptep);
155 flush_tlb_range(&vma, saddr, addr);
158 void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
159 pte_t *ptep, pte_t pte)
164 unsigned long pfn, dpfn;
168 * Code needs to be expanded to handle huge swap and migration
169 * entries. Needed for HUGETLB and MEMORY_FAILURE.
171 WARN_ON(!pte_present(pte));
173 if (!pte_cont(pte)) {
174 set_pte_at(mm, addr, ptep, pte);
178 ncontig = find_num_contig(mm, addr, ptep, &pgsize);
180 dpfn = pgsize >> PAGE_SHIFT;
181 hugeprot = pte_pgprot(pte);
183 clear_flush(mm, addr, ptep, pgsize, ncontig);
185 for (i = 0; i < ncontig; i++, ptep++, addr += pgsize, pfn += dpfn)
186 set_pte_at(mm, addr, ptep, pfn_pte(pfn, hugeprot));
189 void set_huge_swap_pte_at(struct mm_struct *mm, unsigned long addr,
190 pte_t *ptep, pte_t pte, unsigned long sz)
195 ncontig = num_contig_ptes(sz, &pgsize);
197 for (i = 0; i < ncontig; i++, ptep++)
201 pte_t *huge_pte_alloc(struct mm_struct *mm,
202 unsigned long addr, unsigned long sz)
209 pgdp = pgd_offset(mm, addr);
210 pudp = pud_alloc(mm, pgdp, addr);
214 if (sz == PUD_SIZE) {
215 ptep = (pte_t *)pudp;
216 } else if (sz == (PAGE_SIZE * CONT_PTES)) {
217 pmdp = pmd_alloc(mm, pudp, addr);
219 WARN_ON(addr & (sz - 1));
221 * Note that if this code were ever ported to the
222 * 32-bit arm platform then it will cause trouble in
223 * the case where CONFIG_HIGHPTE is set, since there
224 * will be no pte_unmap() to correspond with this
227 ptep = pte_alloc_map(mm, pmdp, addr);
228 } else if (sz == PMD_SIZE) {
229 if (IS_ENABLED(CONFIG_ARCH_WANT_HUGE_PMD_SHARE) &&
230 pud_none(READ_ONCE(*pudp)))
231 ptep = huge_pmd_share(mm, addr, pudp);
233 ptep = (pte_t *)pmd_alloc(mm, pudp, addr);
234 } else if (sz == (PMD_SIZE * CONT_PMDS)) {
235 pmdp = pmd_alloc(mm, pudp, addr);
236 WARN_ON(addr & (sz - 1));
237 return (pte_t *)pmdp;
243 pte_t *huge_pte_offset(struct mm_struct *mm,
244 unsigned long addr, unsigned long sz)
250 pgdp = pgd_offset(mm, addr);
251 if (!pgd_present(READ_ONCE(*pgdp)))
254 pudp = pud_offset(pgdp, addr);
255 pud = READ_ONCE(*pudp);
256 if (sz != PUD_SIZE && pud_none(pud))
258 /* hugepage or swap? */
259 if (pud_huge(pud) || !pud_present(pud))
260 return (pte_t *)pudp;
261 /* table; check the next level */
263 if (sz == CONT_PMD_SIZE)
264 addr &= CONT_PMD_MASK;
266 pmdp = pmd_offset(pudp, addr);
267 pmd = READ_ONCE(*pmdp);
268 if (!(sz == PMD_SIZE || sz == CONT_PMD_SIZE) &&
271 if (pmd_huge(pmd) || !pmd_present(pmd))
272 return (pte_t *)pmdp;
274 if (sz == CONT_PTE_SIZE)
275 return pte_offset_kernel(pmdp, (addr & CONT_PTE_MASK));
280 pte_t arch_make_huge_pte(pte_t entry, struct vm_area_struct *vma,
281 struct page *page, int writable)
283 size_t pagesize = huge_page_size(hstate_vma(vma));
285 if (pagesize == CONT_PTE_SIZE) {
286 entry = pte_mkcont(entry);
287 } else if (pagesize == CONT_PMD_SIZE) {
288 entry = pmd_pte(pmd_mkcont(pte_pmd(entry)));
289 } else if (pagesize != PUD_SIZE && pagesize != PMD_SIZE) {
290 pr_warn("%s: unrecognized huge page size 0x%lx\n",
296 void huge_pte_clear(struct mm_struct *mm, unsigned long addr,
297 pte_t *ptep, unsigned long sz)
302 ncontig = num_contig_ptes(sz, &pgsize);
304 for (i = 0; i < ncontig; i++, addr += pgsize, ptep++)
305 pte_clear(mm, addr, ptep);
308 pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
309 unsigned long addr, pte_t *ptep)
313 pte_t orig_pte = huge_ptep_get(ptep);
315 if (!pte_cont(orig_pte))
316 return ptep_get_and_clear(mm, addr, ptep);
318 ncontig = find_num_contig(mm, addr, ptep, &pgsize);
320 return get_clear_flush(mm, addr, ptep, pgsize, ncontig);
323 int huge_ptep_set_access_flags(struct vm_area_struct *vma,
324 unsigned long addr, pte_t *ptep,
325 pte_t pte, int dirty)
327 int ncontig, i, changed = 0;
329 unsigned long pfn = pte_pfn(pte), dpfn;
334 return ptep_set_access_flags(vma, addr, ptep, pte, dirty);
336 ncontig = find_num_contig(vma->vm_mm, addr, ptep, &pgsize);
337 dpfn = pgsize >> PAGE_SHIFT;
339 orig_pte = get_clear_flush(vma->vm_mm, addr, ptep, pgsize, ncontig);
340 if (!pte_same(orig_pte, pte))
343 /* Make sure we don't lose the dirty state */
344 if (pte_dirty(orig_pte))
345 pte = pte_mkdirty(pte);
347 hugeprot = pte_pgprot(pte);
348 for (i = 0; i < ncontig; i++, ptep++, addr += pgsize, pfn += dpfn)
349 set_pte_at(vma->vm_mm, addr, ptep, pfn_pte(pfn, hugeprot));
354 void huge_ptep_set_wrprotect(struct mm_struct *mm,
355 unsigned long addr, pte_t *ptep)
357 unsigned long pfn, dpfn;
363 if (!pte_cont(READ_ONCE(*ptep))) {
364 ptep_set_wrprotect(mm, addr, ptep);
368 ncontig = find_num_contig(mm, addr, ptep, &pgsize);
369 dpfn = pgsize >> PAGE_SHIFT;
371 pte = get_clear_flush(mm, addr, ptep, pgsize, ncontig);
372 pte = pte_wrprotect(pte);
374 hugeprot = pte_pgprot(pte);
377 for (i = 0; i < ncontig; i++, ptep++, addr += pgsize, pfn += dpfn)
378 set_pte_at(mm, addr, ptep, pfn_pte(pfn, hugeprot));
381 void huge_ptep_clear_flush(struct vm_area_struct *vma,
382 unsigned long addr, pte_t *ptep)
387 if (!pte_cont(READ_ONCE(*ptep))) {
388 ptep_clear_flush(vma, addr, ptep);
392 ncontig = find_num_contig(vma->vm_mm, addr, ptep, &pgsize);
393 clear_flush(vma->vm_mm, addr, ptep, pgsize, ncontig);
396 static __init int setup_hugepagesz(char *opt)
398 unsigned long ps = memparse(opt, &opt);
401 #ifdef CONFIG_ARM64_4K_PAGES
404 case PMD_SIZE * CONT_PMDS:
406 case PAGE_SIZE * CONT_PTES:
407 hugetlb_add_hstate(ilog2(ps) - PAGE_SHIFT);
412 pr_err("hugepagesz: Unsupported page size %lu K\n", ps >> 10);
415 __setup("hugepagesz=", setup_hugepagesz);
417 #ifdef CONFIG_ARM64_64K_PAGES
418 static __init int add_default_hugepagesz(void)
420 if (size_to_hstate(CONT_PTES * PAGE_SIZE) == NULL)
421 hugetlb_add_hstate(CONT_PTE_SHIFT);
424 arch_initcall(add_default_hugepagesz);