blk: rq_data_dir() should not return a boolean
[cris-mirror.git] / arch / arm64 / mm / mmu.c
blob9211b8527f2580aeb561b8c7cc3cdc75f728571f
1 /*
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>
28 #include <linux/fs.h>
29 #include <linux/io.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/sections.h>
36 #include <asm/setup.h>
37 #include <asm/sizes.h>
38 #include <asm/tlb.h>
39 #include <asm/memblock.h>
40 #include <asm/mmu_context.h>
42 #include "mm.h"
44 u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
47 * Empty_zero_page is a special page that is used for zero-initialized data
48 * and COW.
50 struct page *empty_zero_page;
51 EXPORT_SYMBOL(empty_zero_page);
53 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
54 unsigned long size, pgprot_t vma_prot)
56 if (!pfn_valid(pfn))
57 return pgprot_noncached(vma_prot);
58 else if (file->f_flags & O_SYNC)
59 return pgprot_writecombine(vma_prot);
60 return vma_prot;
62 EXPORT_SYMBOL(phys_mem_access_prot);
64 static void __init *early_alloc(unsigned long sz)
66 void *ptr = __va(memblock_alloc(sz, sz));
67 BUG_ON(!ptr);
68 memset(ptr, 0, sz);
69 return ptr;
73 * remap a PMD into pages
75 static void split_pmd(pmd_t *pmd, pte_t *pte)
77 unsigned long pfn = pmd_pfn(*pmd);
78 int i = 0;
80 do {
82 * Need to have the least restrictive permissions available
83 * permissions will be fixed up later
85 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
86 pfn++;
87 } while (pte++, i++, i < PTRS_PER_PTE);
90 static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
91 unsigned long end, unsigned long pfn,
92 pgprot_t prot,
93 void *(*alloc)(unsigned long size))
95 pte_t *pte;
97 if (pmd_none(*pmd) || pmd_sect(*pmd)) {
98 pte = alloc(PTRS_PER_PTE * sizeof(pte_t));
99 if (pmd_sect(*pmd))
100 split_pmd(pmd, pte);
101 __pmd_populate(pmd, __pa(pte), PMD_TYPE_TABLE);
102 flush_tlb_all();
104 BUG_ON(pmd_bad(*pmd));
106 pte = pte_offset_kernel(pmd, addr);
107 do {
108 set_pte(pte, pfn_pte(pfn, prot));
109 pfn++;
110 } while (pte++, addr += PAGE_SIZE, addr != end);
113 void split_pud(pud_t *old_pud, pmd_t *pmd)
115 unsigned long addr = pud_pfn(*old_pud) << PAGE_SHIFT;
116 pgprot_t prot = __pgprot(pud_val(*old_pud) ^ addr);
117 int i = 0;
119 do {
120 set_pmd(pmd, __pmd(addr | pgprot_val(prot)));
121 addr += PMD_SIZE;
122 } while (pmd++, i++, i < PTRS_PER_PMD);
125 static void alloc_init_pmd(struct mm_struct *mm, pud_t *pud,
126 unsigned long addr, unsigned long end,
127 phys_addr_t phys, pgprot_t prot,
128 void *(*alloc)(unsigned long size))
130 pmd_t *pmd;
131 unsigned long next;
134 * Check for initial section mappings in the pgd/pud and remove them.
136 if (pud_none(*pud) || pud_sect(*pud)) {
137 pmd = alloc(PTRS_PER_PMD * sizeof(pmd_t));
138 if (pud_sect(*pud)) {
140 * need to have the 1G of mappings continue to be
141 * present
143 split_pud(pud, pmd);
145 pud_populate(mm, pud, pmd);
146 flush_tlb_all();
148 BUG_ON(pud_bad(*pud));
150 pmd = pmd_offset(pud, addr);
151 do {
152 next = pmd_addr_end(addr, end);
153 /* try section mapping first */
154 if (((addr | next | phys) & ~SECTION_MASK) == 0) {
155 pmd_t old_pmd =*pmd;
156 set_pmd(pmd, __pmd(phys |
157 pgprot_val(mk_sect_prot(prot))));
159 * Check for previous table entries created during
160 * boot (__create_page_tables) and flush them.
162 if (!pmd_none(old_pmd)) {
163 flush_tlb_all();
164 if (pmd_table(old_pmd)) {
165 phys_addr_t table = __pa(pte_offset_map(&old_pmd, 0));
166 if (!WARN_ON_ONCE(slab_is_available()))
167 memblock_free(table, PAGE_SIZE);
170 } else {
171 alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
172 prot, alloc);
174 phys += next - addr;
175 } while (pmd++, addr = next, addr != end);
178 static inline bool use_1G_block(unsigned long addr, unsigned long next,
179 unsigned long phys)
181 if (PAGE_SHIFT != 12)
182 return false;
184 if (((addr | next | phys) & ~PUD_MASK) != 0)
185 return false;
187 return true;
190 static void alloc_init_pud(struct mm_struct *mm, pgd_t *pgd,
191 unsigned long addr, unsigned long end,
192 phys_addr_t phys, pgprot_t prot,
193 void *(*alloc)(unsigned long size))
195 pud_t *pud;
196 unsigned long next;
198 if (pgd_none(*pgd)) {
199 pud = alloc(PTRS_PER_PUD * sizeof(pud_t));
200 pgd_populate(mm, pgd, pud);
202 BUG_ON(pgd_bad(*pgd));
204 pud = pud_offset(pgd, addr);
205 do {
206 next = pud_addr_end(addr, end);
209 * For 4K granule only, attempt to put down a 1GB block
211 if (use_1G_block(addr, next, phys)) {
212 pud_t old_pud = *pud;
213 set_pud(pud, __pud(phys |
214 pgprot_val(mk_sect_prot(prot))));
217 * If we have an old value for a pud, it will
218 * be pointing to a pmd table that we no longer
219 * need (from swapper_pg_dir).
221 * Look up the old pmd table and free it.
223 if (!pud_none(old_pud)) {
224 flush_tlb_all();
225 if (pud_table(old_pud)) {
226 phys_addr_t table = __pa(pmd_offset(&old_pud, 0));
227 if (!WARN_ON_ONCE(slab_is_available()))
228 memblock_free(table, PAGE_SIZE);
231 } else {
232 alloc_init_pmd(mm, pud, addr, next, phys, prot, alloc);
234 phys += next - addr;
235 } while (pud++, addr = next, addr != end);
239 * Create the page directory entries and any necessary page tables for the
240 * mapping specified by 'md'.
242 static void __create_mapping(struct mm_struct *mm, pgd_t *pgd,
243 phys_addr_t phys, unsigned long virt,
244 phys_addr_t size, pgprot_t prot,
245 void *(*alloc)(unsigned long size))
247 unsigned long addr, length, end, next;
249 addr = virt & PAGE_MASK;
250 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
252 end = addr + length;
253 do {
254 next = pgd_addr_end(addr, end);
255 alloc_init_pud(mm, pgd, addr, next, phys, prot, alloc);
256 phys += next - addr;
257 } while (pgd++, addr = next, addr != end);
260 static void *late_alloc(unsigned long size)
262 void *ptr;
264 BUG_ON(size > PAGE_SIZE);
265 ptr = (void *)__get_free_page(PGALLOC_GFP);
266 BUG_ON(!ptr);
267 return ptr;
270 static void __init create_mapping(phys_addr_t phys, unsigned long virt,
271 phys_addr_t size, pgprot_t prot)
273 if (virt < VMALLOC_START) {
274 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
275 &phys, virt);
276 return;
278 __create_mapping(&init_mm, pgd_offset_k(virt & PAGE_MASK), phys, virt,
279 size, prot, early_alloc);
282 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
283 unsigned long virt, phys_addr_t size,
284 pgprot_t prot)
286 __create_mapping(mm, pgd_offset(mm, virt), phys, virt, size, prot,
287 late_alloc);
290 static void create_mapping_late(phys_addr_t phys, unsigned long virt,
291 phys_addr_t size, pgprot_t prot)
293 if (virt < VMALLOC_START) {
294 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
295 &phys, virt);
296 return;
299 return __create_mapping(&init_mm, pgd_offset_k(virt & PAGE_MASK),
300 phys, virt, size, prot, late_alloc);
303 #ifdef CONFIG_DEBUG_RODATA
304 static void __init __map_memblock(phys_addr_t start, phys_addr_t end)
307 * Set up the executable regions using the existing section mappings
308 * for now. This will get more fine grained later once all memory
309 * is mapped
311 unsigned long kernel_x_start = round_down(__pa(_stext), SECTION_SIZE);
312 unsigned long kernel_x_end = round_up(__pa(__init_end), SECTION_SIZE);
314 if (end < kernel_x_start) {
315 create_mapping(start, __phys_to_virt(start),
316 end - start, PAGE_KERNEL);
317 } else if (start >= kernel_x_end) {
318 create_mapping(start, __phys_to_virt(start),
319 end - start, PAGE_KERNEL);
320 } else {
321 if (start < kernel_x_start)
322 create_mapping(start, __phys_to_virt(start),
323 kernel_x_start - start,
324 PAGE_KERNEL);
325 create_mapping(kernel_x_start,
326 __phys_to_virt(kernel_x_start),
327 kernel_x_end - kernel_x_start,
328 PAGE_KERNEL_EXEC);
329 if (kernel_x_end < end)
330 create_mapping(kernel_x_end,
331 __phys_to_virt(kernel_x_end),
332 end - kernel_x_end,
333 PAGE_KERNEL);
337 #else
338 static void __init __map_memblock(phys_addr_t start, phys_addr_t end)
340 create_mapping(start, __phys_to_virt(start), end - start,
341 PAGE_KERNEL_EXEC);
343 #endif
345 static void __init map_mem(void)
347 struct memblock_region *reg;
348 phys_addr_t limit;
351 * Temporarily limit the memblock range. We need to do this as
352 * create_mapping requires puds, pmds and ptes to be allocated from
353 * memory addressable from the initial direct kernel mapping.
355 * The initial direct kernel mapping, located at swapper_pg_dir, gives
356 * us PUD_SIZE (4K pages) or PMD_SIZE (64K pages) memory starting from
357 * PHYS_OFFSET (which must be aligned to 2MB as per
358 * Documentation/arm64/booting.txt).
360 if (IS_ENABLED(CONFIG_ARM64_64K_PAGES))
361 limit = PHYS_OFFSET + PMD_SIZE;
362 else
363 limit = PHYS_OFFSET + PUD_SIZE;
364 memblock_set_current_limit(limit);
366 /* map all the memory banks */
367 for_each_memblock(memory, reg) {
368 phys_addr_t start = reg->base;
369 phys_addr_t end = start + reg->size;
371 if (start >= end)
372 break;
374 #ifndef CONFIG_ARM64_64K_PAGES
376 * For the first memory bank align the start address and
377 * current memblock limit to prevent create_mapping() from
378 * allocating pte page tables from unmapped memory.
379 * When 64K pages are enabled, the pte page table for the
380 * first PGDIR_SIZE is already present in swapper_pg_dir.
382 if (start < limit)
383 start = ALIGN(start, PMD_SIZE);
384 if (end < limit) {
385 limit = end & PMD_MASK;
386 memblock_set_current_limit(limit);
388 #endif
389 __map_memblock(start, end);
392 /* Limit no longer required. */
393 memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
396 void __init fixup_executable(void)
398 #ifdef CONFIG_DEBUG_RODATA
399 /* now that we are actually fully mapped, make the start/end more fine grained */
400 if (!IS_ALIGNED((unsigned long)_stext, SECTION_SIZE)) {
401 unsigned long aligned_start = round_down(__pa(_stext),
402 SECTION_SIZE);
404 create_mapping(aligned_start, __phys_to_virt(aligned_start),
405 __pa(_stext) - aligned_start,
406 PAGE_KERNEL);
409 if (!IS_ALIGNED((unsigned long)__init_end, SECTION_SIZE)) {
410 unsigned long aligned_end = round_up(__pa(__init_end),
411 SECTION_SIZE);
412 create_mapping(__pa(__init_end), (unsigned long)__init_end,
413 aligned_end - __pa(__init_end),
414 PAGE_KERNEL);
416 #endif
419 #ifdef CONFIG_DEBUG_RODATA
420 void mark_rodata_ro(void)
422 create_mapping_late(__pa(_stext), (unsigned long)_stext,
423 (unsigned long)_etext - (unsigned long)_stext,
424 PAGE_KERNEL_EXEC | PTE_RDONLY);
427 #endif
429 void fixup_init(void)
431 create_mapping_late(__pa(__init_begin), (unsigned long)__init_begin,
432 (unsigned long)__init_end - (unsigned long)__init_begin,
433 PAGE_KERNEL);
437 * paging_init() sets up the page tables, initialises the zone memory
438 * maps and sets up the zero page.
440 void __init paging_init(void)
442 void *zero_page;
444 map_mem();
445 fixup_executable();
447 /* allocate the zero page. */
448 zero_page = early_alloc(PAGE_SIZE);
450 bootmem_init();
452 empty_zero_page = virt_to_page(zero_page);
455 * TTBR0 is only used for the identity mapping at this stage. Make it
456 * point to zero page to avoid speculatively fetching new entries.
458 cpu_set_reserved_ttbr0();
459 flush_tlb_all();
460 cpu_set_default_tcr_t0sz();
464 * Check whether a kernel address is valid (derived from arch/x86/).
466 int kern_addr_valid(unsigned long addr)
468 pgd_t *pgd;
469 pud_t *pud;
470 pmd_t *pmd;
471 pte_t *pte;
473 if ((((long)addr) >> VA_BITS) != -1UL)
474 return 0;
476 pgd = pgd_offset_k(addr);
477 if (pgd_none(*pgd))
478 return 0;
480 pud = pud_offset(pgd, addr);
481 if (pud_none(*pud))
482 return 0;
484 if (pud_sect(*pud))
485 return pfn_valid(pud_pfn(*pud));
487 pmd = pmd_offset(pud, addr);
488 if (pmd_none(*pmd))
489 return 0;
491 if (pmd_sect(*pmd))
492 return pfn_valid(pmd_pfn(*pmd));
494 pte = pte_offset_kernel(pmd, addr);
495 if (pte_none(*pte))
496 return 0;
498 return pfn_valid(pte_pfn(*pte));
500 #ifdef CONFIG_SPARSEMEM_VMEMMAP
501 #ifdef CONFIG_ARM64_64K_PAGES
502 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
504 return vmemmap_populate_basepages(start, end, node);
506 #else /* !CONFIG_ARM64_64K_PAGES */
507 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
509 unsigned long addr = start;
510 unsigned long next;
511 pgd_t *pgd;
512 pud_t *pud;
513 pmd_t *pmd;
515 do {
516 next = pmd_addr_end(addr, end);
518 pgd = vmemmap_pgd_populate(addr, node);
519 if (!pgd)
520 return -ENOMEM;
522 pud = vmemmap_pud_populate(pgd, addr, node);
523 if (!pud)
524 return -ENOMEM;
526 pmd = pmd_offset(pud, addr);
527 if (pmd_none(*pmd)) {
528 void *p = NULL;
530 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
531 if (!p)
532 return -ENOMEM;
534 set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
535 } else
536 vmemmap_verify((pte_t *)pmd, node, addr, next);
537 } while (addr = next, addr != end);
539 return 0;
541 #endif /* CONFIG_ARM64_64K_PAGES */
542 void vmemmap_free(unsigned long start, unsigned long end)
545 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
547 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
548 #if CONFIG_PGTABLE_LEVELS > 2
549 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss;
550 #endif
551 #if CONFIG_PGTABLE_LEVELS > 3
552 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss;
553 #endif
555 static inline pud_t * fixmap_pud(unsigned long addr)
557 pgd_t *pgd = pgd_offset_k(addr);
559 BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
561 return pud_offset(pgd, addr);
564 static inline pmd_t * fixmap_pmd(unsigned long addr)
566 pud_t *pud = fixmap_pud(addr);
568 BUG_ON(pud_none(*pud) || pud_bad(*pud));
570 return pmd_offset(pud, addr);
573 static inline pte_t * fixmap_pte(unsigned long addr)
575 pmd_t *pmd = fixmap_pmd(addr);
577 BUG_ON(pmd_none(*pmd) || pmd_bad(*pmd));
579 return pte_offset_kernel(pmd, addr);
582 void __init early_fixmap_init(void)
584 pgd_t *pgd;
585 pud_t *pud;
586 pmd_t *pmd;
587 unsigned long addr = FIXADDR_START;
589 pgd = pgd_offset_k(addr);
590 pgd_populate(&init_mm, pgd, bm_pud);
591 pud = pud_offset(pgd, addr);
592 pud_populate(&init_mm, pud, bm_pmd);
593 pmd = pmd_offset(pud, addr);
594 pmd_populate_kernel(&init_mm, pmd, bm_pte);
597 * The boot-ioremap range spans multiple pmds, for which
598 * we are not preparted:
600 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
601 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
603 if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
604 || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
605 WARN_ON(1);
606 pr_warn("pmd %p != %p, %p\n",
607 pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
608 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
609 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
610 fix_to_virt(FIX_BTMAP_BEGIN));
611 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
612 fix_to_virt(FIX_BTMAP_END));
614 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
615 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
619 void __set_fixmap(enum fixed_addresses idx,
620 phys_addr_t phys, pgprot_t flags)
622 unsigned long addr = __fix_to_virt(idx);
623 pte_t *pte;
625 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
627 pte = fixmap_pte(addr);
629 if (pgprot_val(flags)) {
630 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
631 } else {
632 pte_clear(&init_mm, addr, pte);
633 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
637 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
639 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
640 pgprot_t prot = PAGE_KERNEL | PTE_RDONLY;
641 int granularity, size, offset;
642 void *dt_virt;
645 * Check whether the physical FDT address is set and meets the minimum
646 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
647 * at least 8 bytes so that we can always access the size field of the
648 * FDT header after mapping the first chunk, double check here if that
649 * is indeed the case.
651 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
652 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
653 return NULL;
656 * Make sure that the FDT region can be mapped without the need to
657 * allocate additional translation table pages, so that it is safe
658 * to call create_mapping() this early.
660 * On 64k pages, the FDT will be mapped using PTEs, so we need to
661 * be in the same PMD as the rest of the fixmap.
662 * On 4k pages, we'll use section mappings for the FDT so we only
663 * have to be in the same PUD.
665 BUILD_BUG_ON(dt_virt_base % SZ_2M);
667 if (IS_ENABLED(CONFIG_ARM64_64K_PAGES)) {
668 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> PMD_SHIFT !=
669 __fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT);
671 granularity = PAGE_SIZE;
672 } else {
673 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> PUD_SHIFT !=
674 __fix_to_virt(FIX_BTMAP_BEGIN) >> PUD_SHIFT);
676 granularity = PMD_SIZE;
679 offset = dt_phys % granularity;
680 dt_virt = (void *)dt_virt_base + offset;
682 /* map the first chunk so we can read the size from the header */
683 create_mapping(round_down(dt_phys, granularity), dt_virt_base,
684 granularity, prot);
686 if (fdt_check_header(dt_virt) != 0)
687 return NULL;
689 size = fdt_totalsize(dt_virt);
690 if (size > MAX_FDT_SIZE)
691 return NULL;
693 if (offset + size > granularity)
694 create_mapping(round_down(dt_phys, granularity), dt_virt_base,
695 round_up(offset + size, granularity), prot);
697 memblock_reserve(dt_phys, size);
699 return dt_virt;