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 *pgd = pgd_offset(mm, addr);
62 pud = pud_offset(pgd, addr);
63 pmd = pmd_offset(pud, addr);
64 if ((pte_t *)pmd == 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 struct vm_area_struct vma = { .vm_mm = mm };
112 pte_t orig_pte = huge_ptep_get(ptep);
113 bool valid = pte_valid(orig_pte);
114 unsigned long i, saddr = addr;
116 for (i = 0; i < ncontig; i++, addr += pgsize, ptep++) {
117 pte_t pte = ptep_get_and_clear(mm, addr, ptep);
120 * If HW_AFDBM is enabled, then the HW could turn on
121 * the dirty or accessed bit for any page in the set,
125 orig_pte = pte_mkdirty(orig_pte);
128 orig_pte = pte_mkyoung(orig_pte);
132 flush_tlb_range(&vma, saddr, addr);
137 * Changing some bits of contiguous entries requires us to follow a
138 * Break-Before-Make approach, breaking the whole contiguous set
139 * before we can change any entries. See ARM DDI 0487A.k_iss10775,
140 * "Misprogramming of the Contiguous bit", page D4-1762.
142 * This helper performs the break step for use cases where the
143 * original pte is not needed.
145 static void clear_flush(struct mm_struct *mm,
148 unsigned long pgsize,
149 unsigned long ncontig)
151 struct vm_area_struct vma = { .vm_mm = mm };
152 unsigned long i, saddr = addr;
154 for (i = 0; i < ncontig; i++, addr += pgsize, ptep++)
155 pte_clear(mm, addr, ptep);
157 flush_tlb_range(&vma, saddr, addr);
160 void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
161 pte_t *ptep, pte_t pte)
166 unsigned long pfn, dpfn;
170 * Code needs to be expanded to handle huge swap and migration
171 * entries. Needed for HUGETLB and MEMORY_FAILURE.
173 WARN_ON(!pte_present(pte));
175 if (!pte_cont(pte)) {
176 set_pte_at(mm, addr, ptep, pte);
180 ncontig = find_num_contig(mm, addr, ptep, &pgsize);
182 dpfn = pgsize >> PAGE_SHIFT;
183 hugeprot = pte_pgprot(pte);
185 clear_flush(mm, addr, ptep, pgsize, ncontig);
187 for (i = 0; i < ncontig; i++, ptep++, addr += pgsize, pfn += dpfn) {
188 pr_debug("%s: set pte %p to 0x%llx\n", __func__, ptep,
189 pte_val(pfn_pte(pfn, hugeprot)));
190 set_pte_at(mm, addr, ptep, pfn_pte(pfn, hugeprot));
194 void set_huge_swap_pte_at(struct mm_struct *mm, unsigned long addr,
195 pte_t *ptep, pte_t pte, unsigned long sz)
200 ncontig = num_contig_ptes(sz, &pgsize);
202 for (i = 0; i < ncontig; i++, ptep++)
206 pte_t *huge_pte_alloc(struct mm_struct *mm,
207 unsigned long addr, unsigned long sz)
213 pr_debug("%s: addr:0x%lx sz:0x%lx\n", __func__, addr, sz);
214 pgd = pgd_offset(mm, addr);
215 pud = pud_alloc(mm, pgd, addr);
219 if (sz == PUD_SIZE) {
221 } else if (sz == (PAGE_SIZE * CONT_PTES)) {
222 pmd_t *pmd = pmd_alloc(mm, pud, addr);
224 WARN_ON(addr & (sz - 1));
226 * Note that if this code were ever ported to the
227 * 32-bit arm platform then it will cause trouble in
228 * the case where CONFIG_HIGHPTE is set, since there
229 * will be no pte_unmap() to correspond with this
232 pte = pte_alloc_map(mm, pmd, addr);
233 } else if (sz == PMD_SIZE) {
234 if (IS_ENABLED(CONFIG_ARCH_WANT_HUGE_PMD_SHARE) &&
236 pte = huge_pmd_share(mm, addr, pud);
238 pte = (pte_t *)pmd_alloc(mm, pud, addr);
239 } else if (sz == (PMD_SIZE * CONT_PMDS)) {
242 pmd = pmd_alloc(mm, pud, addr);
243 WARN_ON(addr & (sz - 1));
247 pr_debug("%s: addr:0x%lx sz:0x%lx ret pte=%p/0x%llx\n", __func__, addr,
248 sz, pte, pte_val(*pte));
252 pte_t *huge_pte_offset(struct mm_struct *mm,
253 unsigned long addr, unsigned long sz)
259 pgd = pgd_offset(mm, addr);
260 pr_debug("%s: addr:0x%lx pgd:%p\n", __func__, addr, pgd);
261 if (!pgd_present(*pgd))
264 pud = pud_offset(pgd, addr);
265 if (sz != PUD_SIZE && pud_none(*pud))
267 /* hugepage or swap? */
268 if (pud_huge(*pud) || !pud_present(*pud))
270 /* table; check the next level */
272 if (sz == CONT_PMD_SIZE)
273 addr &= CONT_PMD_MASK;
275 pmd = pmd_offset(pud, addr);
276 if (!(sz == PMD_SIZE || sz == CONT_PMD_SIZE) &&
279 if (pmd_huge(*pmd) || !pmd_present(*pmd))
282 if (sz == CONT_PTE_SIZE) {
283 pte_t *pte = pte_offset_kernel(pmd, (addr & CONT_PTE_MASK));
290 pte_t arch_make_huge_pte(pte_t entry, struct vm_area_struct *vma,
291 struct page *page, int writable)
293 size_t pagesize = huge_page_size(hstate_vma(vma));
295 if (pagesize == CONT_PTE_SIZE) {
296 entry = pte_mkcont(entry);
297 } else if (pagesize == CONT_PMD_SIZE) {
298 entry = pmd_pte(pmd_mkcont(pte_pmd(entry)));
299 } else if (pagesize != PUD_SIZE && pagesize != PMD_SIZE) {
300 pr_warn("%s: unrecognized huge page size 0x%lx\n",
306 void huge_pte_clear(struct mm_struct *mm, unsigned long addr,
307 pte_t *ptep, unsigned long sz)
312 ncontig = num_contig_ptes(sz, &pgsize);
314 for (i = 0; i < ncontig; i++, addr += pgsize, ptep++)
315 pte_clear(mm, addr, ptep);
318 pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
319 unsigned long addr, pte_t *ptep)
323 pte_t orig_pte = huge_ptep_get(ptep);
325 if (!pte_cont(orig_pte))
326 return ptep_get_and_clear(mm, addr, ptep);
328 ncontig = find_num_contig(mm, addr, ptep, &pgsize);
330 return get_clear_flush(mm, addr, ptep, pgsize, ncontig);
333 int huge_ptep_set_access_flags(struct vm_area_struct *vma,
334 unsigned long addr, pte_t *ptep,
335 pte_t pte, int dirty)
337 int ncontig, i, changed = 0;
339 unsigned long pfn = pte_pfn(pte), dpfn;
344 return ptep_set_access_flags(vma, addr, ptep, pte, dirty);
346 ncontig = find_num_contig(vma->vm_mm, addr, ptep, &pgsize);
347 dpfn = pgsize >> PAGE_SHIFT;
349 orig_pte = get_clear_flush(vma->vm_mm, addr, ptep, pgsize, ncontig);
350 if (!pte_same(orig_pte, pte))
353 /* Make sure we don't lose the dirty or young state */
354 if (pte_dirty(orig_pte))
355 pte = pte_mkdirty(pte);
357 if (pte_young(orig_pte))
358 pte = pte_mkyoung(pte);
360 hugeprot = pte_pgprot(pte);
361 for (i = 0; i < ncontig; i++, ptep++, addr += pgsize, pfn += dpfn)
362 set_pte_at(vma->vm_mm, addr, ptep, pfn_pte(pfn, hugeprot));
367 void huge_ptep_set_wrprotect(struct mm_struct *mm,
368 unsigned long addr, pte_t *ptep)
370 unsigned long pfn, dpfn;
376 if (!pte_cont(*ptep)) {
377 ptep_set_wrprotect(mm, addr, ptep);
381 ncontig = find_num_contig(mm, addr, ptep, &pgsize);
382 dpfn = pgsize >> PAGE_SHIFT;
384 pte = get_clear_flush(mm, addr, ptep, pgsize, ncontig);
385 pte = pte_wrprotect(pte);
387 hugeprot = pte_pgprot(pte);
390 for (i = 0; i < ncontig; i++, ptep++, addr += pgsize, pfn += dpfn)
391 set_pte_at(mm, addr, ptep, pfn_pte(pfn, hugeprot));
394 void huge_ptep_clear_flush(struct vm_area_struct *vma,
395 unsigned long addr, pte_t *ptep)
400 if (!pte_cont(*ptep)) {
401 ptep_clear_flush(vma, addr, ptep);
405 ncontig = find_num_contig(vma->vm_mm, addr, ptep, &pgsize);
406 clear_flush(vma->vm_mm, addr, ptep, pgsize, ncontig);
409 static __init int setup_hugepagesz(char *opt)
411 unsigned long ps = memparse(opt, &opt);
414 #ifdef CONFIG_ARM64_4K_PAGES
417 case PMD_SIZE * CONT_PMDS:
419 case PAGE_SIZE * CONT_PTES:
420 hugetlb_add_hstate(ilog2(ps) - PAGE_SHIFT);
425 pr_err("hugepagesz: Unsupported page size %lu K\n", ps >> 10);
428 __setup("hugepagesz=", setup_hugepagesz);
430 #ifdef CONFIG_ARM64_64K_PAGES
431 static __init int add_default_hugepagesz(void)
433 if (size_to_hstate(CONT_PTES * PAGE_SIZE) == NULL)
434 hugetlb_add_hstate(CONT_PTE_SHIFT);
437 arch_initcall(add_default_hugepagesz);