1/*
  2 * arch/arm64/mm/hugetlbpage.c
  3 *
  4 * Copyright (C) 2013 Linaro Ltd.
  5 *
  6 * Based on arch/x86/mm/hugetlbpage.c.
  7 *
  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.
 11 *
 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.
 16 */
 17
 18#include <linux/init.h>
 19#include <linux/fs.h>
 20#include <linux/mm.h>
 21#include <linux/hugetlb.h>
 22#include <linux/pagemap.h>
 23#include <linux/err.h>
 24#include <linux/sysctl.h>
 25#include <asm/mman.h>
 26#include <asm/tlb.h>
 27#include <asm/tlbflush.h>
 28#include <asm/pgalloc.h>
 29
 30int pmd_huge(pmd_t pmd)
 31{
 32	return pmd_val(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT);
 33}
 34
 35int pud_huge(pud_t pud)
 36{
 37#ifndef __PAGETABLE_PMD_FOLDED
 38	return pud_val(pud) && !(pud_val(pud) & PUD_TABLE_BIT);
 39#else
 40	return 0;
 41#endif
 42}
 43
 44static int find_num_contig(struct mm_struct *mm, unsigned long addr,
 45			   pte_t *ptep, pte_t pte, size_t *pgsize)
 46{
 47	pgd_t *pgd = pgd_offset(mm, addr);
 48	pud_t *pud;
 49	pmd_t *pmd;
 50
 51	*pgsize = PAGE_SIZE;
 52	if (!pte_cont(pte))
 53		return 1;
 54	pud = pud_offset(pgd, addr);
 55	pmd = pmd_offset(pud, addr);
 56	if ((pte_t *)pmd == ptep) {
 57		*pgsize = PMD_SIZE;
 58		return CONT_PMDS;
 59	}
 60	return CONT_PTES;
 61}
 62
 63void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
 64			    pte_t *ptep, pte_t pte)
 65{
 66	size_t pgsize;
 67	int i;
 68	int ncontig = find_num_contig(mm, addr, ptep, pte, &pgsize);
 69	unsigned long pfn;
 70	pgprot_t hugeprot;
 71
 72	if (ncontig == 1) {
 73		set_pte_at(mm, addr, ptep, pte);
 74		return;
 75	}
 76
 77	pfn = pte_pfn(pte);
 78	hugeprot = __pgprot(pte_val(pfn_pte(pfn, __pgprot(0))) ^ pte_val(pte));
 79	for (i = 0; i < ncontig; i++) {
 80		pr_debug("%s: set pte %p to 0x%llx\n", __func__, ptep,
 81			 pte_val(pfn_pte(pfn, hugeprot)));
 82		set_pte_at(mm, addr, ptep, pfn_pte(pfn, hugeprot));
 83		ptep++;
 84		pfn += pgsize >> PAGE_SHIFT;
 85		addr += pgsize;
 86	}
 87}
 88
 89pte_t *huge_pte_alloc(struct mm_struct *mm,
 90		      unsigned long addr, unsigned long sz)
 91{
 92	pgd_t *pgd;
 93	pud_t *pud;
 94	pte_t *pte = NULL;
 95
 96	pr_debug("%s: addr:0x%lx sz:0x%lx\n", __func__, addr, sz);
 97	pgd = pgd_offset(mm, addr);
 98	pud = pud_alloc(mm, pgd, addr);
 99	if (!pud)
100		return NULL;
101
102	if (sz == PUD_SIZE) {
103		pte = (pte_t *)pud;
104	} else if (sz == (PAGE_SIZE * CONT_PTES)) {
105		pmd_t *pmd = pmd_alloc(mm, pud, addr);
106
107		WARN_ON(addr & (sz - 1));
108		/*
109		 * Note that if this code were ever ported to the
110		 * 32-bit arm platform then it will cause trouble in
111		 * the case where CONFIG_HIGHPTE is set, since there
112		 * will be no pte_unmap() to correspond with this
113		 * pte_alloc_map().
114		 */
115		pte = pte_alloc_map(mm, pmd, addr);
116	} else if (sz == PMD_SIZE) {
117		if (IS_ENABLED(CONFIG_ARCH_WANT_HUGE_PMD_SHARE) &&
118		    pud_none(*pud))
119			pte = huge_pmd_share(mm, addr, pud);
120		else
121			pte = (pte_t *)pmd_alloc(mm, pud, addr);
122	} else if (sz == (PMD_SIZE * CONT_PMDS)) {
123		pmd_t *pmd;
124
125		pmd = pmd_alloc(mm, pud, addr);
126		WARN_ON(addr & (sz - 1));
127		return (pte_t *)pmd;
128	}
129
130	pr_debug("%s: addr:0x%lx sz:0x%lx ret pte=%p/0x%llx\n", __func__, addr,
131	       sz, pte, pte_val(*pte));
132	return pte;
133}
134
135pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
136{
137	pgd_t *pgd;
138	pud_t *pud;
139	pmd_t *pmd = NULL;
140	pte_t *pte = NULL;
141
142	pgd = pgd_offset(mm, addr);
143	pr_debug("%s: addr:0x%lx pgd:%p\n", __func__, addr, pgd);
144	if (!pgd_present(*pgd))
145		return NULL;
146	pud = pud_offset(pgd, addr);
147	if (!pud_present(*pud))
148		return NULL;
149
150	if (pud_huge(*pud))
151		return (pte_t *)pud;
152	pmd = pmd_offset(pud, addr);
153	if (!pmd_present(*pmd))
154		return NULL;
155
156	if (pte_cont(pmd_pte(*pmd))) {
157		pmd = pmd_offset(
158			pud, (addr & CONT_PMD_MASK));
159		return (pte_t *)pmd;
160	}
161	if (pmd_huge(*pmd))
162		return (pte_t *)pmd;
163	pte = pte_offset_kernel(pmd, addr);
164	if (pte_present(*pte) && pte_cont(*pte)) {
165		pte = pte_offset_kernel(
166			pmd, (addr & CONT_PTE_MASK));
167		return pte;
168	}
169	return NULL;
170}
171
172pte_t arch_make_huge_pte(pte_t entry, struct vm_area_struct *vma,
173			 struct page *page, int writable)
174{
175	size_t pagesize = huge_page_size(hstate_vma(vma));
176
177	if (pagesize == CONT_PTE_SIZE) {
178		entry = pte_mkcont(entry);
179	} else if (pagesize == CONT_PMD_SIZE) {
180		entry = pmd_pte(pmd_mkcont(pte_pmd(entry)));
181	} else if (pagesize != PUD_SIZE && pagesize != PMD_SIZE) {
182		pr_warn("%s: unrecognized huge page size 0x%lx\n",
183			__func__, pagesize);
184	}
185	return entry;
186}
187
188pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
189			      unsigned long addr, pte_t *ptep)
190{
191	pte_t pte;
192
193	if (pte_cont(*ptep)) {
194		int ncontig, i;
195		size_t pgsize;
196		pte_t *cpte;
197		bool is_dirty = false;
198
199		cpte = huge_pte_offset(mm, addr);
200		ncontig = find_num_contig(mm, addr, cpte, *cpte, &pgsize);
201		/* save the 1st pte to return */
202		pte = ptep_get_and_clear(mm, addr, cpte);
203		for (i = 1, addr += pgsize; i < ncontig; ++i, addr += pgsize) {
204			/*
205			 * If HW_AFDBM is enabled, then the HW could
206			 * turn on the dirty bit for any of the page
207			 * in the set, so check them all.
208			 */
209			++cpte;
210			if (pte_dirty(ptep_get_and_clear(mm, addr, cpte)))
211				is_dirty = true;
212		}
213		if (is_dirty)
214			return pte_mkdirty(pte);
215		else
216			return pte;
217	} else {
218		return ptep_get_and_clear(mm, addr, ptep);
219	}
220}
221
222int huge_ptep_set_access_flags(struct vm_area_struct *vma,
223			       unsigned long addr, pte_t *ptep,
224			       pte_t pte, int dirty)
225{
226	pte_t *cpte;
227
228	if (pte_cont(pte)) {
229		int ncontig, i, changed = 0;
230		size_t pgsize = 0;
231		unsigned long pfn = pte_pfn(pte);
232		/* Select all bits except the pfn */
233		pgprot_t hugeprot =
234			__pgprot(pte_val(pfn_pte(pfn, __pgprot(0))) ^
235				 pte_val(pte));
236
237		cpte = huge_pte_offset(vma->vm_mm, addr);
238		pfn = pte_pfn(*cpte);
239		ncontig = find_num_contig(vma->vm_mm, addr, cpte,
240					  *cpte, &pgsize);
241		for (i = 0; i < ncontig; ++i, ++cpte, addr += pgsize) {
242			changed |= ptep_set_access_flags(vma, addr, cpte,
243							pfn_pte(pfn,
244								hugeprot),
245							dirty);
246			pfn += pgsize >> PAGE_SHIFT;
247		}
248		return changed;
249	} else {
250		return ptep_set_access_flags(vma, addr, ptep, pte, dirty);
251	}
252}
253
254void huge_ptep_set_wrprotect(struct mm_struct *mm,
255			     unsigned long addr, pte_t *ptep)
256{
257	if (pte_cont(*ptep)) {
258		int ncontig, i;
259		pte_t *cpte;
260		size_t pgsize = 0;
261
262		cpte = huge_pte_offset(mm, addr);
263		ncontig = find_num_contig(mm, addr, cpte, *cpte, &pgsize);
264		for (i = 0; i < ncontig; ++i, ++cpte, addr += pgsize)
265			ptep_set_wrprotect(mm, addr, cpte);
266	} else {
267		ptep_set_wrprotect(mm, addr, ptep);
268	}
269}
270
271void huge_ptep_clear_flush(struct vm_area_struct *vma,
272			   unsigned long addr, pte_t *ptep)
273{
274	if (pte_cont(*ptep)) {
275		int ncontig, i;
276		pte_t *cpte;
277		size_t pgsize = 0;
278
279		cpte = huge_pte_offset(vma->vm_mm, addr);
280		ncontig = find_num_contig(vma->vm_mm, addr, cpte,
281					  *cpte, &pgsize);
282		for (i = 0; i < ncontig; ++i, ++cpte, addr += pgsize)
283			ptep_clear_flush(vma, addr, cpte);
284	} else {
285		ptep_clear_flush(vma, addr, ptep);
286	}
287}
288
289static __init int setup_hugepagesz(char *opt)
290{
291	unsigned long ps = memparse(opt, &opt);
292
293	if (ps == PMD_SIZE) {
294		hugetlb_add_hstate(PMD_SHIFT - PAGE_SHIFT);
295	} else if (ps == PUD_SIZE) {
296		hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT);
297	} else if (ps == (PAGE_SIZE * CONT_PTES)) {
298		hugetlb_add_hstate(CONT_PTE_SHIFT);
299	} else if (ps == (PMD_SIZE * CONT_PMDS)) {
300		hugetlb_add_hstate((PMD_SHIFT + CONT_PMD_SHIFT) - PAGE_SHIFT);
301	} else {
302		hugetlb_bad_size();
303		pr_err("hugepagesz: Unsupported page size %lu K\n", ps >> 10);
304		return 0;
305	}
306	return 1;
307}
308__setup("hugepagesz=", setup_hugepagesz);
309
310#ifdef CONFIG_ARM64_64K_PAGES
311static __init int add_default_hugepagesz(void)
312{
313	if (size_to_hstate(CONT_PTES * PAGE_SIZE) == NULL)
314		hugetlb_add_hstate(CONT_PTE_SHIFT);
315	return 0;
316}
317arch_initcall(add_default_hugepagesz);
318#endif