Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
[deliverable/linux.git] / arch / arm64 / mm / mmu.c
CommitLineData
c1cc1552
CM
1/*
2 * Based on arch/arm/mm/mmu.c
3 *
4 * Copyright (C) 1995-2005 Russell King
5 * Copyright (C) 2012 ARM Ltd.
6 *
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.
10 *
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.
15 *
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/>.
18 */
19
20#include <linux/export.h>
21#include <linux/kernel.h>
22#include <linux/errno.h>
23#include <linux/init.h>
61bd93ce 24#include <linux/libfdt.h>
c1cc1552
CM
25#include <linux/mman.h>
26#include <linux/nodemask.h>
27#include <linux/memblock.h>
28#include <linux/fs.h>
2475ff9d 29#include <linux/io.h>
41089357 30#include <linux/slab.h>
da141706 31#include <linux/stop_machine.h>
c1cc1552 32
21ab99c2 33#include <asm/barrier.h>
c1cc1552 34#include <asm/cputype.h>
af86e597 35#include <asm/fixmap.h>
068a17a5 36#include <asm/kasan.h>
b433dce0 37#include <asm/kernel-pgtable.h>
c1cc1552
CM
38#include <asm/sections.h>
39#include <asm/setup.h>
40#include <asm/sizes.h>
41#include <asm/tlb.h>
c79b954b 42#include <asm/memblock.h>
c1cc1552
CM
43#include <asm/mmu_context.h>
44
45#include "mm.h"
46
dd006da2
AB
47u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
48
a7f8de16
AB
49u64 kimage_voffset __read_mostly;
50EXPORT_SYMBOL(kimage_voffset);
51
c1cc1552
CM
52/*
53 * Empty_zero_page is a special page that is used for zero-initialized data
54 * and COW.
55 */
5227cfa7 56unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
c1cc1552
CM
57EXPORT_SYMBOL(empty_zero_page);
58
f9040773
AB
59static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
60static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
61static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
62
c1cc1552
CM
63pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
64 unsigned long size, pgprot_t vma_prot)
65{
66 if (!pfn_valid(pfn))
67 return pgprot_noncached(vma_prot);
68 else if (file->f_flags & O_SYNC)
69 return pgprot_writecombine(vma_prot);
70 return vma_prot;
71}
72EXPORT_SYMBOL(phys_mem_access_prot);
73
f4710445 74static phys_addr_t __init early_pgtable_alloc(void)
c1cc1552 75{
7142392d
SP
76 phys_addr_t phys;
77 void *ptr;
78
21ab99c2 79 phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
f4710445
MR
80
81 /*
82 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
83 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
84 * any level of table.
85 */
86 ptr = pte_set_fixmap(phys);
87
21ab99c2
MR
88 memset(ptr, 0, PAGE_SIZE);
89
f4710445
MR
90 /*
91 * Implicit barriers also ensure the zeroed page is visible to the page
92 * table walker
93 */
94 pte_clear_fixmap();
95
96 return phys;
c1cc1552
CM
97}
98
da141706 99static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
667c2759 100 unsigned long end, unsigned long pfn,
da141706 101 pgprot_t prot,
f4710445 102 phys_addr_t (*pgtable_alloc)(void))
c1cc1552
CM
103{
104 pte_t *pte;
105
4133af6c
CM
106 BUG_ON(pmd_sect(*pmd));
107 if (pmd_none(*pmd)) {
132233a7
LA
108 phys_addr_t pte_phys;
109 BUG_ON(!pgtable_alloc);
110 pte_phys = pgtable_alloc();
f4710445 111 pte = pte_set_fixmap(pte_phys);
f4710445 112 __pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
f4710445 113 pte_clear_fixmap();
c1cc1552 114 }
a1c76574 115 BUG_ON(pmd_bad(*pmd));
c1cc1552 116
f4710445 117 pte = pte_set_fixmap_offset(pmd, addr);
c1cc1552 118 do {
667c2759
CM
119 set_pte(pte, pfn_pte(pfn, prot));
120 pfn++;
121 } while (pte++, addr += PAGE_SIZE, addr != end);
f4710445
MR
122
123 pte_clear_fixmap();
c1cc1552
CM
124}
125
11509a30 126static void alloc_init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
da141706 127 phys_addr_t phys, pgprot_t prot,
53e1b329
AB
128 phys_addr_t (*pgtable_alloc)(void),
129 bool allow_block_mappings)
c1cc1552
CM
130{
131 pmd_t *pmd;
132 unsigned long next;
133
134 /*
135 * Check for initial section mappings in the pgd/pud and remove them.
136 */
4133af6c
CM
137 BUG_ON(pud_sect(*pud));
138 if (pud_none(*pud)) {
132233a7
LA
139 phys_addr_t pmd_phys;
140 BUG_ON(!pgtable_alloc);
141 pmd_phys = pgtable_alloc();
f4710445 142 pmd = pmd_set_fixmap(pmd_phys);
f4710445 143 __pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
f4710445 144 pmd_clear_fixmap();
c1cc1552 145 }
a1c76574 146 BUG_ON(pud_bad(*pud));
c1cc1552 147
f4710445 148 pmd = pmd_set_fixmap_offset(pud, addr);
c1cc1552
CM
149 do {
150 next = pmd_addr_end(addr, end);
151 /* try section mapping first */
83863f25 152 if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
53e1b329 153 allow_block_mappings) {
a55f9929 154 pmd_t old_pmd =*pmd;
c661cb1c 155 pmd_set_huge(pmd, phys, prot);
a55f9929
CM
156 /*
157 * Check for previous table entries created during
158 * boot (__create_page_tables) and flush them.
159 */
523d6e9f 160 if (!pmd_none(old_pmd)) {
a55f9929 161 flush_tlb_all();
523d6e9f 162 if (pmd_table(old_pmd)) {
316b39db 163 phys_addr_t table = pmd_page_paddr(old_pmd);
41089357
CM
164 if (!WARN_ON_ONCE(slab_is_available()))
165 memblock_free(table, PAGE_SIZE);
523d6e9f 166 }
167 }
a55f9929 168 } else {
667c2759 169 alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
21ab99c2 170 prot, pgtable_alloc);
a55f9929 171 }
c1cc1552
CM
172 phys += next - addr;
173 } while (pmd++, addr = next, addr != end);
f4710445
MR
174
175 pmd_clear_fixmap();
c1cc1552
CM
176}
177
da141706
LA
178static inline bool use_1G_block(unsigned long addr, unsigned long next,
179 unsigned long phys)
180{
181 if (PAGE_SHIFT != 12)
182 return false;
183
184 if (((addr | next | phys) & ~PUD_MASK) != 0)
185 return false;
186
187 return true;
188}
189
11509a30 190static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
da141706 191 phys_addr_t phys, pgprot_t prot,
53e1b329
AB
192 phys_addr_t (*pgtable_alloc)(void),
193 bool allow_block_mappings)
c1cc1552 194{
c79b954b 195 pud_t *pud;
c1cc1552
CM
196 unsigned long next;
197
c79b954b 198 if (pgd_none(*pgd)) {
132233a7
LA
199 phys_addr_t pud_phys;
200 BUG_ON(!pgtable_alloc);
201 pud_phys = pgtable_alloc();
f4710445 202 __pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
c79b954b
JL
203 }
204 BUG_ON(pgd_bad(*pgd));
205
f4710445 206 pud = pud_set_fixmap_offset(pgd, addr);
c1cc1552
CM
207 do {
208 next = pud_addr_end(addr, end);
206a2a73
SC
209
210 /*
211 * For 4K granule only, attempt to put down a 1GB block
212 */
53e1b329 213 if (use_1G_block(addr, next, phys) && allow_block_mappings) {
206a2a73 214 pud_t old_pud = *pud;
c661cb1c 215 pud_set_huge(pud, phys, prot);
206a2a73
SC
216
217 /*
218 * If we have an old value for a pud, it will
219 * be pointing to a pmd table that we no longer
220 * need (from swapper_pg_dir).
221 *
222 * Look up the old pmd table and free it.
223 */
224 if (!pud_none(old_pud)) {
206a2a73 225 flush_tlb_all();
523d6e9f 226 if (pud_table(old_pud)) {
316b39db 227 phys_addr_t table = pud_page_paddr(old_pud);
41089357
CM
228 if (!WARN_ON_ONCE(slab_is_available()))
229 memblock_free(table, PAGE_SIZE);
523d6e9f 230 }
206a2a73
SC
231 }
232 } else {
11509a30 233 alloc_init_pmd(pud, addr, next, phys, prot,
53e1b329 234 pgtable_alloc, allow_block_mappings);
206a2a73 235 }
c1cc1552
CM
236 phys += next - addr;
237 } while (pud++, addr = next, addr != end);
f4710445
MR
238
239 pud_clear_fixmap();
c1cc1552
CM
240}
241
40f87d31
AB
242static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
243 unsigned long virt, phys_addr_t size,
244 pgprot_t prot,
245 phys_addr_t (*pgtable_alloc)(void),
246 bool allow_block_mappings)
c1cc1552
CM
247{
248 unsigned long addr, length, end, next;
40f87d31 249 pgd_t *pgd = pgd_offset_raw(pgdir, virt);
c1cc1552 250
cc5d2b3b
MR
251 /*
252 * If the virtual and physical address don't have the same offset
253 * within a page, we cannot map the region as the caller expects.
254 */
255 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
256 return;
257
9c4e08a3 258 phys &= PAGE_MASK;
c1cc1552
CM
259 addr = virt & PAGE_MASK;
260 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
261
c1cc1552
CM
262 end = addr + length;
263 do {
264 next = pgd_addr_end(addr, end);
53e1b329
AB
265 alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc,
266 allow_block_mappings);
c1cc1552
CM
267 phys += next - addr;
268 } while (pgd++, addr = next, addr != end);
269}
270
1378dc3d 271static phys_addr_t pgd_pgtable_alloc(void)
da141706 272{
21ab99c2 273 void *ptr = (void *)__get_free_page(PGALLOC_GFP);
1378dc3d
AB
274 if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
275 BUG();
21ab99c2
MR
276
277 /* Ensure the zeroed page is visible to the page table walker */
278 dsb(ishst);
f4710445 279 return __pa(ptr);
da141706
LA
280}
281
132233a7
LA
282/*
283 * This function can only be used to modify existing table entries,
284 * without allocating new levels of table. Note that this permits the
285 * creation of new section or page entries.
286 */
287static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
da141706 288 phys_addr_t size, pgprot_t prot)
d7ecbddf
MS
289{
290 if (virt < VMALLOC_START) {
291 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
292 &phys, virt);
293 return;
294 }
53e1b329 295 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, true);
d7ecbddf
MS
296}
297
8ce837ce
AB
298void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
299 unsigned long virt, phys_addr_t size,
53e1b329 300 pgprot_t prot, bool allow_block_mappings)
8ce837ce 301{
1378dc3d
AB
302 BUG_ON(mm == &init_mm);
303
11509a30 304 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
1378dc3d 305 pgd_pgtable_alloc, allow_block_mappings);
d7ecbddf
MS
306}
307
da141706
LA
308static void create_mapping_late(phys_addr_t phys, unsigned long virt,
309 phys_addr_t size, pgprot_t prot)
310{
311 if (virt < VMALLOC_START) {
312 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
313 &phys, virt);
314 return;
315 }
316
11509a30 317 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
258a1605 318 NULL, !debug_pagealloc_enabled());
da141706
LA
319}
320
068a17a5 321static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, phys_addr_t end)
da141706 322{
7eb90f2f 323 unsigned long kernel_start = __pa(_text);
9fdc14c5 324 unsigned long kernel_end = __pa(__init_begin);
068a17a5 325
da141706 326 /*
f9040773
AB
327 * Take care not to create a writable alias for the
328 * read-only text and rodata sections of the kernel image.
da141706 329 */
068a17a5 330
9fdc14c5 331 /* No overlap with the kernel text/rodata */
068a17a5
MR
332 if (end < kernel_start || start >= kernel_end) {
333 __create_pgd_mapping(pgd, start, __phys_to_virt(start),
334 end - start, PAGE_KERNEL,
53e1b329
AB
335 early_pgtable_alloc,
336 !debug_pagealloc_enabled());
068a17a5 337 return;
da141706
LA
338 }
339
068a17a5 340 /*
9fdc14c5 341 * This block overlaps the kernel text/rodata mappings.
f9040773 342 * Map the portion(s) which don't overlap.
068a17a5
MR
343 */
344 if (start < kernel_start)
345 __create_pgd_mapping(pgd, start,
346 __phys_to_virt(start),
347 kernel_start - start, PAGE_KERNEL,
53e1b329
AB
348 early_pgtable_alloc,
349 !debug_pagealloc_enabled());
068a17a5
MR
350 if (kernel_end < end)
351 __create_pgd_mapping(pgd, kernel_end,
352 __phys_to_virt(kernel_end),
353 end - kernel_end, PAGE_KERNEL,
53e1b329
AB
354 early_pgtable_alloc,
355 !debug_pagealloc_enabled());
f9040773
AB
356
357 /*
9fdc14c5 358 * Map the linear alias of the [_text, __init_begin) interval as
f9040773
AB
359 * read-only/non-executable. This makes the contents of the
360 * region accessible to subsystems such as hibernate, but
361 * protects it from inadvertent modification or execution.
362 */
363 __create_pgd_mapping(pgd, kernel_start, __phys_to_virt(kernel_start),
364 kernel_end - kernel_start, PAGE_KERNEL_RO,
53e1b329 365 early_pgtable_alloc, !debug_pagealloc_enabled());
da141706 366}
da141706 367
068a17a5 368static void __init map_mem(pgd_t *pgd)
c1cc1552
CM
369{
370 struct memblock_region *reg;
f6bc87c3 371
c1cc1552
CM
372 /* map all the memory banks */
373 for_each_memblock(memory, reg) {
374 phys_addr_t start = reg->base;
375 phys_addr_t end = start + reg->size;
376
377 if (start >= end)
378 break;
68709f45
AB
379 if (memblock_is_nomap(reg))
380 continue;
c1cc1552 381
068a17a5 382 __map_memblock(pgd, start, end);
c1cc1552
CM
383 }
384}
385
da141706
LA
386void mark_rodata_ro(void)
387{
2f39b5f9 388 unsigned long section_size;
f9040773 389
9fdc14c5 390 section_size = (unsigned long)_etext - (unsigned long)_text;
7eb90f2f 391 create_mapping_late(__pa(_text), (unsigned long)_text,
2f39b5f9
JL
392 section_size, PAGE_KERNEL_ROX);
393 /*
9fdc14c5
AB
394 * mark .rodata as read only. Use __init_begin rather than __end_rodata
395 * to cover NOTES and EXCEPTION_TABLE.
2f39b5f9 396 */
9fdc14c5 397 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
2f39b5f9
JL
398 create_mapping_late(__pa(__start_rodata), (unsigned long)__start_rodata,
399 section_size, PAGE_KERNEL_RO);
da141706 400}
da141706
LA
401
402void fixup_init(void)
403{
f9040773
AB
404 /*
405 * Unmap the __init region but leave the VM area in place. This
406 * prevents the region from being reused for kernel modules, which
407 * is not supported by kallsyms.
408 */
409 unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin));
da141706
LA
410}
411
2c09ec06
AB
412static void __init map_kernel_segment(pgd_t *pgd, void *va_start, void *va_end,
413 pgprot_t prot, struct vm_struct *vma)
068a17a5
MR
414{
415 phys_addr_t pa_start = __pa(va_start);
416 unsigned long size = va_end - va_start;
417
418 BUG_ON(!PAGE_ALIGNED(pa_start));
419 BUG_ON(!PAGE_ALIGNED(size));
420
421 __create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
53e1b329 422 early_pgtable_alloc, !debug_pagealloc_enabled());
f9040773
AB
423
424 vma->addr = va_start;
425 vma->phys_addr = pa_start;
426 vma->size = size;
427 vma->flags = VM_MAP;
428 vma->caller = __builtin_return_address(0);
429
430 vm_area_add_early(vma);
068a17a5
MR
431}
432
433/*
434 * Create fine-grained mappings for the kernel.
435 */
436static void __init map_kernel(pgd_t *pgd)
437{
2f39b5f9 438 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_init, vmlinux_data;
068a17a5 439
9fdc14c5
AB
440 map_kernel_segment(pgd, _text, _etext, PAGE_KERNEL_EXEC, &vmlinux_text);
441 map_kernel_segment(pgd, __start_rodata, __init_begin, PAGE_KERNEL, &vmlinux_rodata);
2c09ec06
AB
442 map_kernel_segment(pgd, __init_begin, __init_end, PAGE_KERNEL_EXEC,
443 &vmlinux_init);
444 map_kernel_segment(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data);
068a17a5 445
f9040773
AB
446 if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
447 /*
448 * The fixmap falls in a separate pgd to the kernel, and doesn't
449 * live in the carveout for the swapper_pg_dir. We can simply
450 * re-use the existing dir for the fixmap.
451 */
452 set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
453 *pgd_offset_k(FIXADDR_START));
454 } else if (CONFIG_PGTABLE_LEVELS > 3) {
455 /*
456 * The fixmap shares its top level pgd entry with the kernel
457 * mapping. This can really only occur when we are running
458 * with 16k/4 levels, so we can simply reuse the pud level
459 * entry instead.
460 */
461 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
462 set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
463 __pud(__pa(bm_pmd) | PUD_TYPE_TABLE));
464 pud_clear_fixmap();
465 } else {
466 BUG();
467 }
068a17a5
MR
468
469 kasan_copy_shadow(pgd);
470}
471
c1cc1552
CM
472/*
473 * paging_init() sets up the page tables, initialises the zone memory
474 * maps and sets up the zero page.
475 */
476void __init paging_init(void)
477{
068a17a5
MR
478 phys_addr_t pgd_phys = early_pgtable_alloc();
479 pgd_t *pgd = pgd_set_fixmap(pgd_phys);
480
481 map_kernel(pgd);
482 map_mem(pgd);
483
484 /*
485 * We want to reuse the original swapper_pg_dir so we don't have to
486 * communicate the new address to non-coherent secondaries in
487 * secondary_entry, and so cpu_switch_mm can generate the address with
488 * adrp+add rather than a load from some global variable.
489 *
490 * To do this we need to go via a temporary pgd.
491 */
492 cpu_replace_ttbr1(__va(pgd_phys));
493 memcpy(swapper_pg_dir, pgd, PAGE_SIZE);
494 cpu_replace_ttbr1(swapper_pg_dir);
495
496 pgd_clear_fixmap();
497 memblock_free(pgd_phys, PAGE_SIZE);
498
499 /*
500 * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
501 * allocated with it.
502 */
503 memblock_free(__pa(swapper_pg_dir) + PAGE_SIZE,
504 SWAPPER_DIR_SIZE - PAGE_SIZE);
c1cc1552
CM
505}
506
c1cc1552
CM
507/*
508 * Check whether a kernel address is valid (derived from arch/x86/).
509 */
510int kern_addr_valid(unsigned long addr)
511{
512 pgd_t *pgd;
513 pud_t *pud;
514 pmd_t *pmd;
515 pte_t *pte;
516
517 if ((((long)addr) >> VA_BITS) != -1UL)
518 return 0;
519
520 pgd = pgd_offset_k(addr);
521 if (pgd_none(*pgd))
522 return 0;
523
524 pud = pud_offset(pgd, addr);
525 if (pud_none(*pud))
526 return 0;
527
206a2a73
SC
528 if (pud_sect(*pud))
529 return pfn_valid(pud_pfn(*pud));
530
c1cc1552
CM
531 pmd = pmd_offset(pud, addr);
532 if (pmd_none(*pmd))
533 return 0;
534
da6e4cb6
DA
535 if (pmd_sect(*pmd))
536 return pfn_valid(pmd_pfn(*pmd));
537
c1cc1552
CM
538 pte = pte_offset_kernel(pmd, addr);
539 if (pte_none(*pte))
540 return 0;
541
542 return pfn_valid(pte_pfn(*pte));
543}
544#ifdef CONFIG_SPARSEMEM_VMEMMAP
b433dce0 545#if !ARM64_SWAPPER_USES_SECTION_MAPS
0aad818b 546int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
c1cc1552 547{
0aad818b 548 return vmemmap_populate_basepages(start, end, node);
c1cc1552 549}
b433dce0 550#else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
0aad818b 551int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
c1cc1552 552{
0aad818b 553 unsigned long addr = start;
c1cc1552
CM
554 unsigned long next;
555 pgd_t *pgd;
556 pud_t *pud;
557 pmd_t *pmd;
558
559 do {
560 next = pmd_addr_end(addr, end);
561
562 pgd = vmemmap_pgd_populate(addr, node);
563 if (!pgd)
564 return -ENOMEM;
565
566 pud = vmemmap_pud_populate(pgd, addr, node);
567 if (!pud)
568 return -ENOMEM;
569
570 pmd = pmd_offset(pud, addr);
571 if (pmd_none(*pmd)) {
572 void *p = NULL;
573
574 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
575 if (!p)
576 return -ENOMEM;
577
a501e324 578 set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
c1cc1552
CM
579 } else
580 vmemmap_verify((pte_t *)pmd, node, addr, next);
581 } while (addr = next, addr != end);
582
583 return 0;
584}
585#endif /* CONFIG_ARM64_64K_PAGES */
0aad818b 586void vmemmap_free(unsigned long start, unsigned long end)
0197518c
TC
587{
588}
c1cc1552 589#endif /* CONFIG_SPARSEMEM_VMEMMAP */
af86e597 590
af86e597
LA
591static inline pud_t * fixmap_pud(unsigned long addr)
592{
593 pgd_t *pgd = pgd_offset_k(addr);
594
595 BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
596
157962f5 597 return pud_offset_kimg(pgd, addr);
af86e597
LA
598}
599
600static inline pmd_t * fixmap_pmd(unsigned long addr)
601{
602 pud_t *pud = fixmap_pud(addr);
603
604 BUG_ON(pud_none(*pud) || pud_bad(*pud));
605
157962f5 606 return pmd_offset_kimg(pud, addr);
af86e597
LA
607}
608
609static inline pte_t * fixmap_pte(unsigned long addr)
610{
157962f5 611 return &bm_pte[pte_index(addr)];
af86e597
LA
612}
613
614void __init early_fixmap_init(void)
615{
616 pgd_t *pgd;
617 pud_t *pud;
618 pmd_t *pmd;
619 unsigned long addr = FIXADDR_START;
620
621 pgd = pgd_offset_k(addr);
f80fb3a3
AB
622 if (CONFIG_PGTABLE_LEVELS > 3 &&
623 !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa(bm_pud))) {
f9040773
AB
624 /*
625 * We only end up here if the kernel mapping and the fixmap
626 * share the top level pgd entry, which should only happen on
627 * 16k/4 levels configurations.
628 */
629 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
630 pud = pud_offset_kimg(pgd, addr);
631 } else {
632 pgd_populate(&init_mm, pgd, bm_pud);
633 pud = fixmap_pud(addr);
634 }
af86e597 635 pud_populate(&init_mm, pud, bm_pmd);
157962f5 636 pmd = fixmap_pmd(addr);
af86e597
LA
637 pmd_populate_kernel(&init_mm, pmd, bm_pte);
638
639 /*
640 * The boot-ioremap range spans multiple pmds, for which
157962f5 641 * we are not prepared:
af86e597
LA
642 */
643 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
644 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
645
646 if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
647 || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
648 WARN_ON(1);
649 pr_warn("pmd %p != %p, %p\n",
650 pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
651 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
652 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
653 fix_to_virt(FIX_BTMAP_BEGIN));
654 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
655 fix_to_virt(FIX_BTMAP_END));
656
657 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
658 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
659 }
660}
661
662void __set_fixmap(enum fixed_addresses idx,
663 phys_addr_t phys, pgprot_t flags)
664{
665 unsigned long addr = __fix_to_virt(idx);
666 pte_t *pte;
667
b63dbef9 668 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
af86e597
LA
669
670 pte = fixmap_pte(addr);
671
672 if (pgprot_val(flags)) {
673 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
674 } else {
675 pte_clear(&init_mm, addr, pte);
676 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
677 }
678}
61bd93ce 679
f80fb3a3 680void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
61bd93ce
AB
681{
682 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
f80fb3a3 683 int offset;
61bd93ce
AB
684 void *dt_virt;
685
686 /*
687 * Check whether the physical FDT address is set and meets the minimum
688 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
04a84810
AB
689 * at least 8 bytes so that we can always access the magic and size
690 * fields of the FDT header after mapping the first chunk, double check
691 * here if that is indeed the case.
61bd93ce
AB
692 */
693 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
694 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
695 return NULL;
696
697 /*
698 * Make sure that the FDT region can be mapped without the need to
699 * allocate additional translation table pages, so that it is safe
132233a7 700 * to call create_mapping_noalloc() this early.
61bd93ce
AB
701 *
702 * On 64k pages, the FDT will be mapped using PTEs, so we need to
703 * be in the same PMD as the rest of the fixmap.
704 * On 4k pages, we'll use section mappings for the FDT so we only
705 * have to be in the same PUD.
706 */
707 BUILD_BUG_ON(dt_virt_base % SZ_2M);
708
b433dce0
SP
709 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
710 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
61bd93ce 711
b433dce0 712 offset = dt_phys % SWAPPER_BLOCK_SIZE;
61bd93ce
AB
713 dt_virt = (void *)dt_virt_base + offset;
714
715 /* map the first chunk so we can read the size from the header */
132233a7
LA
716 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
717 dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
61bd93ce 718
04a84810 719 if (fdt_magic(dt_virt) != FDT_MAGIC)
61bd93ce
AB
720 return NULL;
721
f80fb3a3
AB
722 *size = fdt_totalsize(dt_virt);
723 if (*size > MAX_FDT_SIZE)
61bd93ce
AB
724 return NULL;
725
f80fb3a3 726 if (offset + *size > SWAPPER_BLOCK_SIZE)
132233a7 727 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
f80fb3a3 728 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
61bd93ce 729
f80fb3a3
AB
730 return dt_virt;
731}
61bd93ce 732
f80fb3a3
AB
733void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
734{
735 void *dt_virt;
736 int size;
737
738 dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
739 if (!dt_virt)
740 return NULL;
741
742 memblock_reserve(dt_phys, size);
61bd93ce
AB
743 return dt_virt;
744}
324420bf
AB
745
746int __init arch_ioremap_pud_supported(void)
747{
748 /* only 4k granule supports level 1 block mappings */
749 return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
750}
751
752int __init arch_ioremap_pmd_supported(void)
753{
754 return 1;
755}
756
757int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
758{
759 BUG_ON(phys & ~PUD_MASK);
760 set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
761 return 1;
762}
763
764int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
765{
766 BUG_ON(phys & ~PMD_MASK);
767 set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
768 return 1;
769}
770
771int pud_clear_huge(pud_t *pud)
772{
773 if (!pud_sect(*pud))
774 return 0;
775 pud_clear(pud);
776 return 1;
777}
778
779int pmd_clear_huge(pmd_t *pmd)
780{
781 if (!pmd_sect(*pmd))
782 return 0;
783 pmd_clear(pmd);
784 return 1;
785}
This page took 0.217001 seconds and 5 git commands to generate.