2 * Based on arch/arm/mm/mmu.c
4 * Copyright (C) 1995-2005 Russell King
5 * Copyright (C) 2012 ARM Ltd.
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program. If not, see <http://www.gnu.org/licenses/>.
20 #include <linux/export.h>
21 #include <linux/kernel.h>
22 #include <linux/errno.h>
23 #include <linux/init.h>
24 #include <linux/libfdt.h>
25 #include <linux/mman.h>
26 #include <linux/nodemask.h>
27 #include <linux/memblock.h>
30 #include <linux/slab.h>
31 #include <linux/stop_machine.h>
33 #include <asm/cputype.h>
34 #include <asm/fixmap.h>
35 #include <asm/kernel-pgtable.h>
36 #include <asm/sections.h>
37 #include <asm/setup.h>
38 #include <asm/sizes.h>
40 #include <asm/memblock.h>
41 #include <asm/mmu_context.h>
45 u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
48 * Empty_zero_page is a special page that is used for zero-initialized data
51 struct page *empty_zero_page;
52 EXPORT_SYMBOL(empty_zero_page);
54 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
55 unsigned long size, pgprot_t vma_prot)
58 return pgprot_noncached(vma_prot);
59 else if (file->f_flags & O_SYNC)
60 return pgprot_writecombine(vma_prot);
63 EXPORT_SYMBOL(phys_mem_access_prot);
65 static void __init *early_alloc(unsigned long sz)
70 phys = memblock_alloc(sz, sz);
78 * remap a PMD into pages
80 static void split_pmd(pmd_t *pmd, pte_t *pte)
82 unsigned long pfn = pmd_pfn(*pmd);
87 * Need to have the least restrictive permissions available
88 * permissions will be fixed up later
90 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
92 } while (pte++, i++, i < PTRS_PER_PTE);
95 static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
96 unsigned long end, unsigned long pfn,
98 void *(*alloc)(unsigned long size))
102 if (pmd_none(*pmd) || pmd_sect(*pmd)) {
103 pte = alloc(PTRS_PER_PTE * sizeof(pte_t));
106 __pmd_populate(pmd, __pa(pte), PMD_TYPE_TABLE);
109 BUG_ON(pmd_bad(*pmd));
111 pte = pte_offset_kernel(pmd, addr);
113 set_pte(pte, pfn_pte(pfn, prot));
115 } while (pte++, addr += PAGE_SIZE, addr != end);
118 static void split_pud(pud_t *old_pud, pmd_t *pmd)
120 unsigned long addr = pud_pfn(*old_pud) << PAGE_SHIFT;
121 pgprot_t prot = __pgprot(pud_val(*old_pud) ^ addr);
125 set_pmd(pmd, __pmd(addr | pgprot_val(prot)));
127 } while (pmd++, i++, i < PTRS_PER_PMD);
130 static void alloc_init_pmd(struct mm_struct *mm, pud_t *pud,
131 unsigned long addr, unsigned long end,
132 phys_addr_t phys, pgprot_t prot,
133 void *(*alloc)(unsigned long size))
139 * Check for initial section mappings in the pgd/pud and remove them.
141 if (pud_none(*pud) || pud_sect(*pud)) {
142 pmd = alloc(PTRS_PER_PMD * sizeof(pmd_t));
143 if (pud_sect(*pud)) {
145 * need to have the 1G of mappings continue to be
150 pud_populate(mm, pud, pmd);
153 BUG_ON(pud_bad(*pud));
155 pmd = pmd_offset(pud, addr);
157 next = pmd_addr_end(addr, end);
158 /* try section mapping first */
159 if (((addr | next | phys) & ~SECTION_MASK) == 0) {
161 set_pmd(pmd, __pmd(phys |
162 pgprot_val(mk_sect_prot(prot))));
164 * Check for previous table entries created during
165 * boot (__create_page_tables) and flush them.
167 if (!pmd_none(old_pmd)) {
169 if (pmd_table(old_pmd)) {
170 phys_addr_t table = __pa(pte_offset_map(&old_pmd, 0));
171 if (!WARN_ON_ONCE(slab_is_available()))
172 memblock_free(table, PAGE_SIZE);
176 alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
180 } while (pmd++, addr = next, addr != end);
183 static inline bool use_1G_block(unsigned long addr, unsigned long next,
186 if (PAGE_SHIFT != 12)
189 if (((addr | next | phys) & ~PUD_MASK) != 0)
195 static void alloc_init_pud(struct mm_struct *mm, pgd_t *pgd,
196 unsigned long addr, unsigned long end,
197 phys_addr_t phys, pgprot_t prot,
198 void *(*alloc)(unsigned long size))
203 if (pgd_none(*pgd)) {
204 pud = alloc(PTRS_PER_PUD * sizeof(pud_t));
205 pgd_populate(mm, pgd, pud);
207 BUG_ON(pgd_bad(*pgd));
209 pud = pud_offset(pgd, addr);
211 next = pud_addr_end(addr, end);
214 * For 4K granule only, attempt to put down a 1GB block
216 if (use_1G_block(addr, next, phys)) {
217 pud_t old_pud = *pud;
218 set_pud(pud, __pud(phys |
219 pgprot_val(mk_sect_prot(prot))));
222 * If we have an old value for a pud, it will
223 * be pointing to a pmd table that we no longer
224 * need (from swapper_pg_dir).
226 * Look up the old pmd table and free it.
228 if (!pud_none(old_pud)) {
230 if (pud_table(old_pud)) {
231 phys_addr_t table = __pa(pmd_offset(&old_pud, 0));
232 if (!WARN_ON_ONCE(slab_is_available()))
233 memblock_free(table, PAGE_SIZE);
237 alloc_init_pmd(mm, pud, addr, next, phys, prot, alloc);
240 } while (pud++, addr = next, addr != end);
244 * Create the page directory entries and any necessary page tables for the
245 * mapping specified by 'md'.
247 static void __create_mapping(struct mm_struct *mm, pgd_t *pgd,
248 phys_addr_t phys, unsigned long virt,
249 phys_addr_t size, pgprot_t prot,
250 void *(*alloc)(unsigned long size))
252 unsigned long addr, length, end, next;
255 * If the virtual and physical address don't have the same offset
256 * within a page, we cannot map the region as the caller expects.
258 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
262 addr = virt & PAGE_MASK;
263 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
267 next = pgd_addr_end(addr, end);
268 alloc_init_pud(mm, pgd, addr, next, phys, prot, alloc);
270 } while (pgd++, addr = next, addr != end);
273 static void *late_alloc(unsigned long size)
277 BUG_ON(size > PAGE_SIZE);
278 ptr = (void *)__get_free_page(PGALLOC_GFP);
283 static void __init create_mapping(phys_addr_t phys, unsigned long virt,
284 phys_addr_t size, pgprot_t prot)
286 if (virt < VMALLOC_START) {
287 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
291 __create_mapping(&init_mm, pgd_offset_k(virt), phys, virt,
292 size, prot, early_alloc);
295 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
296 unsigned long virt, phys_addr_t size,
299 __create_mapping(mm, pgd_offset(mm, virt), phys, virt, size, prot,
303 static void create_mapping_late(phys_addr_t phys, unsigned long virt,
304 phys_addr_t size, pgprot_t prot)
306 if (virt < VMALLOC_START) {
307 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
312 return __create_mapping(&init_mm, pgd_offset_k(virt),
313 phys, virt, size, prot, late_alloc);
316 #ifdef CONFIG_DEBUG_RODATA
317 static void __init __map_memblock(phys_addr_t start, phys_addr_t end)
320 * Set up the executable regions using the existing section mappings
321 * for now. This will get more fine grained later once all memory
324 unsigned long kernel_x_start = round_down(__pa(_stext), SWAPPER_BLOCK_SIZE);
325 unsigned long kernel_x_end = round_up(__pa(__init_end), SWAPPER_BLOCK_SIZE);
327 if (end < kernel_x_start) {
328 create_mapping(start, __phys_to_virt(start),
329 end - start, PAGE_KERNEL);
330 } else if (start >= kernel_x_end) {
331 create_mapping(start, __phys_to_virt(start),
332 end - start, PAGE_KERNEL);
334 if (start < kernel_x_start)
335 create_mapping(start, __phys_to_virt(start),
336 kernel_x_start - start,
338 create_mapping(kernel_x_start,
339 __phys_to_virt(kernel_x_start),
340 kernel_x_end - kernel_x_start,
342 if (kernel_x_end < end)
343 create_mapping(kernel_x_end,
344 __phys_to_virt(kernel_x_end),
351 static void __init __map_memblock(phys_addr_t start, phys_addr_t end)
353 create_mapping(start, __phys_to_virt(start), end - start,
358 static void __init map_mem(void)
360 struct memblock_region *reg;
364 * Temporarily limit the memblock range. We need to do this as
365 * create_mapping requires puds, pmds and ptes to be allocated from
366 * memory addressable from the initial direct kernel mapping.
368 * The initial direct kernel mapping, located at swapper_pg_dir, gives
369 * us PUD_SIZE (with SECTION maps) or PMD_SIZE (without SECTION maps,
370 * memory starting from PHYS_OFFSET (which must be aligned to 2MB as
371 * per Documentation/arm64/booting.txt).
373 limit = PHYS_OFFSET + SWAPPER_INIT_MAP_SIZE;
374 memblock_set_current_limit(limit);
376 /* map all the memory banks */
377 for_each_memblock(memory, reg) {
378 phys_addr_t start = reg->base;
379 phys_addr_t end = start + reg->size;
383 if (memblock_is_nomap(reg))
386 if (ARM64_SWAPPER_USES_SECTION_MAPS) {
388 * For the first memory bank align the start address and
389 * current memblock limit to prevent create_mapping() from
390 * allocating pte page tables from unmapped memory. With
391 * the section maps, if the first block doesn't end on section
392 * size boundary, create_mapping() will try to allocate a pte
393 * page, which may be returned from an unmapped area.
394 * When section maps are not used, the pte page table for the
395 * current limit is already present in swapper_pg_dir.
398 start = ALIGN(start, SECTION_SIZE);
400 limit = end & SECTION_MASK;
401 memblock_set_current_limit(limit);
404 __map_memblock(start, end);
407 /* Limit no longer required. */
408 memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
411 static void __init fixup_executable(void)
413 #ifdef CONFIG_DEBUG_RODATA
414 /* now that we are actually fully mapped, make the start/end more fine grained */
415 if (!IS_ALIGNED((unsigned long)_stext, SWAPPER_BLOCK_SIZE)) {
416 unsigned long aligned_start = round_down(__pa(_stext),
419 create_mapping(aligned_start, __phys_to_virt(aligned_start),
420 __pa(_stext) - aligned_start,
424 if (!IS_ALIGNED((unsigned long)__init_end, SWAPPER_BLOCK_SIZE)) {
425 unsigned long aligned_end = round_up(__pa(__init_end),
427 create_mapping(__pa(__init_end), (unsigned long)__init_end,
428 aligned_end - __pa(__init_end),
434 #ifdef CONFIG_DEBUG_RODATA
435 void mark_rodata_ro(void)
437 create_mapping_late(__pa(_stext), (unsigned long)_stext,
438 (unsigned long)_etext - (unsigned long)_stext,
444 void fixup_init(void)
446 create_mapping_late(__pa(__init_begin), (unsigned long)__init_begin,
447 (unsigned long)__init_end - (unsigned long)__init_begin,
452 * paging_init() sets up the page tables, initialises the zone memory
453 * maps and sets up the zero page.
455 void __init paging_init(void)
462 /* allocate the zero page. */
463 zero_page = early_alloc(PAGE_SIZE);
467 empty_zero_page = virt_to_page(zero_page);
469 /* Ensure the zero page is visible to the page table walker */
473 * TTBR0 is only used for the identity mapping at this stage. Make it
474 * point to zero page to avoid speculatively fetching new entries.
476 cpu_set_reserved_ttbr0();
477 local_flush_tlb_all();
478 cpu_set_default_tcr_t0sz();
482 * Check whether a kernel address is valid (derived from arch/x86/).
484 int kern_addr_valid(unsigned long addr)
491 if ((((long)addr) >> VA_BITS) != -1UL)
494 pgd = pgd_offset_k(addr);
498 pud = pud_offset(pgd, addr);
503 return pfn_valid(pud_pfn(*pud));
505 pmd = pmd_offset(pud, addr);
510 return pfn_valid(pmd_pfn(*pmd));
512 pte = pte_offset_kernel(pmd, addr);
516 return pfn_valid(pte_pfn(*pte));
518 #ifdef CONFIG_SPARSEMEM_VMEMMAP
519 #if !ARM64_SWAPPER_USES_SECTION_MAPS
520 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
522 return vmemmap_populate_basepages(start, end, node);
524 #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
525 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
527 unsigned long addr = start;
534 next = pmd_addr_end(addr, end);
536 pgd = vmemmap_pgd_populate(addr, node);
540 pud = vmemmap_pud_populate(pgd, addr, node);
544 pmd = pmd_offset(pud, addr);
545 if (pmd_none(*pmd)) {
548 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
552 set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
554 vmemmap_verify((pte_t *)pmd, node, addr, next);
555 } while (addr = next, addr != end);
559 #endif /* CONFIG_ARM64_64K_PAGES */
560 void vmemmap_free(unsigned long start, unsigned long end)
563 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
565 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
566 #if CONFIG_PGTABLE_LEVELS > 2
567 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss;
569 #if CONFIG_PGTABLE_LEVELS > 3
570 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss;
573 static inline pud_t * fixmap_pud(unsigned long addr)
575 pgd_t *pgd = pgd_offset_k(addr);
577 BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
579 return pud_offset(pgd, addr);
582 static inline pmd_t * fixmap_pmd(unsigned long addr)
584 pud_t *pud = fixmap_pud(addr);
586 BUG_ON(pud_none(*pud) || pud_bad(*pud));
588 return pmd_offset(pud, addr);
591 static inline pte_t * fixmap_pte(unsigned long addr)
593 pmd_t *pmd = fixmap_pmd(addr);
595 BUG_ON(pmd_none(*pmd) || pmd_bad(*pmd));
597 return pte_offset_kernel(pmd, addr);
600 void __init early_fixmap_init(void)
605 unsigned long addr = FIXADDR_START;
607 pgd = pgd_offset_k(addr);
608 pgd_populate(&init_mm, pgd, bm_pud);
609 pud = pud_offset(pgd, addr);
610 pud_populate(&init_mm, pud, bm_pmd);
611 pmd = pmd_offset(pud, addr);
612 pmd_populate_kernel(&init_mm, pmd, bm_pte);
615 * The boot-ioremap range spans multiple pmds, for which
616 * we are not preparted:
618 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
619 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
621 if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
622 || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
624 pr_warn("pmd %p != %p, %p\n",
625 pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
626 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
627 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
628 fix_to_virt(FIX_BTMAP_BEGIN));
629 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
630 fix_to_virt(FIX_BTMAP_END));
632 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
633 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
637 void __set_fixmap(enum fixed_addresses idx,
638 phys_addr_t phys, pgprot_t flags)
640 unsigned long addr = __fix_to_virt(idx);
643 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
645 pte = fixmap_pte(addr);
647 if (pgprot_val(flags)) {
648 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
650 pte_clear(&init_mm, addr, pte);
651 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
655 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
657 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
658 pgprot_t prot = PAGE_KERNEL_RO;
663 * Check whether the physical FDT address is set and meets the minimum
664 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
665 * at least 8 bytes so that we can always access the size field of the
666 * FDT header after mapping the first chunk, double check here if that
667 * is indeed the case.
669 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
670 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
674 * Make sure that the FDT region can be mapped without the need to
675 * allocate additional translation table pages, so that it is safe
676 * to call create_mapping() this early.
678 * On 64k pages, the FDT will be mapped using PTEs, so we need to
679 * be in the same PMD as the rest of the fixmap.
680 * On 4k pages, we'll use section mappings for the FDT so we only
681 * have to be in the same PUD.
683 BUILD_BUG_ON(dt_virt_base % SZ_2M);
685 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
686 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
688 offset = dt_phys % SWAPPER_BLOCK_SIZE;
689 dt_virt = (void *)dt_virt_base + offset;
691 /* map the first chunk so we can read the size from the header */
692 create_mapping(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
693 SWAPPER_BLOCK_SIZE, prot);
695 if (fdt_check_header(dt_virt) != 0)
698 size = fdt_totalsize(dt_virt);
699 if (size > MAX_FDT_SIZE)
702 if (offset + size > SWAPPER_BLOCK_SIZE)
703 create_mapping(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
704 round_up(offset + size, SWAPPER_BLOCK_SIZE), prot);
706 memblock_reserve(dt_phys, size);