1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/hugetlb.h>
5 #ifdef CONFIG_RISCV_ISA_SVNAPOT
6 pte_t huge_ptep_get(pte_t *ptep)
10 pte_t orig_pte = ptep_get(ptep);
12 if (!pte_present(orig_pte) || !pte_napot(orig_pte))
15 pte_num = napot_pte_num(napot_cont_order(orig_pte));
17 for (i = 0; i < pte_num; i++, ptep++) {
18 pte_t pte = ptep_get(ptep);
21 orig_pte = pte_mkdirty(orig_pte);
24 orig_pte = pte_mkyoung(orig_pte);
30 pte_t *huge_pte_alloc(struct mm_struct *mm,
31 struct vm_area_struct *vma,
42 pgd = pgd_offset(mm, addr);
43 p4d = p4d_alloc(mm, pgd, addr);
47 pud = pud_alloc(mm, p4d, addr);
57 if (want_pmd_share(vma, addr) && pud_none(*pud))
58 pte = huge_pmd_share(mm, vma, addr, pud);
60 pte = (pte_t *)pmd_alloc(mm, pud, addr);
64 pmd = pmd_alloc(mm, pud, addr);
68 for_each_napot_order(order) {
69 if (napot_cont_size(order) == sz) {
70 pte = pte_alloc_huge(mm, pmd, addr & napot_cont_mask(order));
77 pte_t pteval = ptep_get_lockless(pte);
79 WARN_ON_ONCE(pte_present(pteval) && !pte_huge(pteval));
84 pte_t *huge_pte_offset(struct mm_struct *mm,
95 pgd = pgd_offset(mm, addr);
96 if (!pgd_present(*pgd))
99 p4d = p4d_offset(pgd, addr);
100 if (!p4d_present(*p4d))
103 pud = pud_offset(p4d, addr);
105 /* must be pud huge, non-present or none */
108 if (!pud_present(*pud))
111 pmd = pmd_offset(pud, addr);
113 /* must be pmd huge, non-present or none */
116 if (!pmd_present(*pmd))
119 for_each_napot_order(order) {
120 if (napot_cont_size(order) == sz) {
121 pte = pte_offset_huge(pmd, addr & napot_cont_mask(order));
128 static pte_t get_clear_contig(struct mm_struct *mm,
131 unsigned long pte_num)
133 pte_t orig_pte = ptep_get(ptep);
136 for (i = 0; i < pte_num; i++, addr += PAGE_SIZE, ptep++) {
137 pte_t pte = ptep_get_and_clear(mm, addr, ptep);
140 orig_pte = pte_mkdirty(orig_pte);
143 orig_pte = pte_mkyoung(orig_pte);
149 static pte_t get_clear_contig_flush(struct mm_struct *mm,
152 unsigned long pte_num)
154 pte_t orig_pte = get_clear_contig(mm, addr, ptep, pte_num);
155 struct vm_area_struct vma = TLB_FLUSH_VMA(mm, 0);
156 bool valid = !pte_none(orig_pte);
159 flush_tlb_range(&vma, addr, addr + (PAGE_SIZE * pte_num));
164 pte_t arch_make_huge_pte(pte_t entry, unsigned int shift, vm_flags_t flags)
168 for_each_napot_order(order) {
169 if (shift == napot_cont_shift(order)) {
170 entry = pte_mknapot(entry, order);
174 if (order == NAPOT_ORDER_MAX)
175 entry = pte_mkhuge(entry);
180 static void clear_flush(struct mm_struct *mm,
183 unsigned long pgsize,
184 unsigned long ncontig)
186 struct vm_area_struct vma = TLB_FLUSH_VMA(mm, 0);
187 unsigned long i, saddr = addr;
189 for (i = 0; i < ncontig; i++, addr += pgsize, ptep++)
190 ptep_get_and_clear(mm, addr, ptep);
192 flush_tlb_range(&vma, saddr, addr);
196 * When dealing with NAPOT mappings, the privileged specification indicates that
197 * "if an update needs to be made, the OS generally should first mark all of the
198 * PTEs invalid, then issue SFENCE.VMA instruction(s) covering all 4 KiB regions
199 * within the range, [...] then update the PTE(s), as described in Section
200 * 4.2.1.". That's the equivalent of the Break-Before-Make approach used by
203 void set_huge_pte_at(struct mm_struct *mm,
209 unsigned long hugepage_shift, pgsize;
212 if (sz >= PGDIR_SIZE)
213 hugepage_shift = PGDIR_SHIFT;
214 else if (sz >= P4D_SIZE)
215 hugepage_shift = P4D_SHIFT;
216 else if (sz >= PUD_SIZE)
217 hugepage_shift = PUD_SHIFT;
218 else if (sz >= PMD_SIZE)
219 hugepage_shift = PMD_SHIFT;
221 hugepage_shift = PAGE_SHIFT;
223 pte_num = sz >> hugepage_shift;
224 pgsize = 1 << hugepage_shift;
226 if (!pte_present(pte)) {
227 for (i = 0; i < pte_num; i++, ptep++, addr += pgsize)
228 set_ptes(mm, addr, ptep, pte, 1);
232 if (!pte_napot(pte)) {
233 set_ptes(mm, addr, ptep, pte, 1);
237 clear_flush(mm, addr, ptep, pgsize, pte_num);
239 for (i = 0; i < pte_num; i++, ptep++, addr += pgsize)
240 set_pte_at(mm, addr, ptep, pte);
243 int huge_ptep_set_access_flags(struct vm_area_struct *vma,
249 struct mm_struct *mm = vma->vm_mm;
255 return ptep_set_access_flags(vma, addr, ptep, pte, dirty);
257 order = napot_cont_order(pte);
258 pte_num = napot_pte_num(order);
259 ptep = huge_pte_offset(mm, addr, napot_cont_size(order));
260 orig_pte = get_clear_contig_flush(mm, addr, ptep, pte_num);
262 if (pte_dirty(orig_pte))
263 pte = pte_mkdirty(pte);
265 if (pte_young(orig_pte))
266 pte = pte_mkyoung(pte);
268 for (i = 0; i < pte_num; i++, addr += PAGE_SIZE, ptep++)
269 set_pte_at(mm, addr, ptep, pte);
274 pte_t huge_ptep_get_and_clear(struct mm_struct *mm,
278 pte_t orig_pte = ptep_get(ptep);
281 if (!pte_napot(orig_pte))
282 return ptep_get_and_clear(mm, addr, ptep);
284 pte_num = napot_pte_num(napot_cont_order(orig_pte));
286 return get_clear_contig(mm, addr, ptep, pte_num);
289 void huge_ptep_set_wrprotect(struct mm_struct *mm,
293 pte_t pte = ptep_get(ptep);
298 if (!pte_napot(pte)) {
299 ptep_set_wrprotect(mm, addr, ptep);
303 order = napot_cont_order(pte);
304 pte_num = napot_pte_num(order);
305 ptep = huge_pte_offset(mm, addr, napot_cont_size(order));
306 orig_pte = get_clear_contig_flush(mm, addr, ptep, pte_num);
308 orig_pte = pte_wrprotect(orig_pte);
310 for (i = 0; i < pte_num; i++, addr += PAGE_SIZE, ptep++)
311 set_pte_at(mm, addr, ptep, orig_pte);
314 pte_t huge_ptep_clear_flush(struct vm_area_struct *vma,
318 pte_t pte = ptep_get(ptep);
322 return ptep_clear_flush(vma, addr, ptep);
324 pte_num = napot_pte_num(napot_cont_order(pte));
326 return get_clear_contig_flush(vma->vm_mm, addr, ptep, pte_num);
329 void huge_pte_clear(struct mm_struct *mm,
334 pte_t pte = READ_ONCE(*ptep);
337 if (!pte_napot(pte)) {
338 pte_clear(mm, addr, ptep);
342 pte_num = napot_pte_num(napot_cont_order(pte));
343 for (i = 0; i < pte_num; i++, addr += PAGE_SIZE, ptep++)
344 pte_clear(mm, addr, ptep);
347 static __init bool is_napot_size(unsigned long size)
354 for_each_napot_order(order) {
355 if (size == napot_cont_size(order))
361 static __init int napot_hugetlbpages_init(void)
366 for_each_napot_order(order)
367 hugetlb_add_hstate(order);
371 arch_initcall(napot_hugetlbpages_init);
375 static __init bool is_napot_size(unsigned long size)
380 #endif /*CONFIG_RISCV_ISA_SVNAPOT*/
382 int pud_huge(pud_t pud)
384 return pud_leaf(pud);
387 int pmd_huge(pmd_t pmd)
389 return pmd_leaf(pmd);
392 bool __init arch_hugetlb_valid_size(unsigned long size)
394 if (size == HPAGE_SIZE)
396 else if (IS_ENABLED(CONFIG_64BIT) && size == PUD_SIZE)
398 else if (is_napot_size(size))
404 #ifdef CONFIG_CONTIG_ALLOC
405 static __init int gigantic_pages_init(void)
407 /* With CONTIG_ALLOC, we can allocate gigantic pages at runtime */
408 if (IS_ENABLED(CONFIG_64BIT))
409 hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT);
412 arch_initcall(gigantic_pages_init);