]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - arch/arm64/mm/mmu.c
Merge branch 'WIP.x86/boot' into x86/boot, to pick up ready branch
[mirror_ubuntu-bionic-kernel.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
5a9e3e15 20#include <linux/cache.h>
c1cc1552
CM
21#include <linux/export.h>
22#include <linux/kernel.h>
23#include <linux/errno.h>
24#include <linux/init.h>
61bd93ce 25#include <linux/libfdt.h>
c1cc1552
CM
26#include <linux/mman.h>
27#include <linux/nodemask.h>
28#include <linux/memblock.h>
29#include <linux/fs.h>
2475ff9d 30#include <linux/io.h>
2077be67 31#include <linux/mm.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 43#include <asm/mmu_context.h>
1404d6f1 44#include <asm/ptdump.h>
c1cc1552 45
dd006da2
AB
46u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
47
5a9e3e15 48u64 kimage_voffset __ro_after_init;
a7f8de16
AB
49EXPORT_SYMBOL(kimage_voffset);
50
c1cc1552
CM
51/*
52 * Empty_zero_page is a special page that is used for zero-initialized data
53 * and COW.
54 */
5227cfa7 55unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
c1cc1552
CM
56EXPORT_SYMBOL(empty_zero_page);
57
f9040773
AB
58static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
59static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
60static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
61
c1cc1552
CM
62pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
63 unsigned long size, pgprot_t vma_prot)
64{
65 if (!pfn_valid(pfn))
66 return pgprot_noncached(vma_prot);
67 else if (file->f_flags & O_SYNC)
68 return pgprot_writecombine(vma_prot);
69 return vma_prot;
70}
71EXPORT_SYMBOL(phys_mem_access_prot);
72
f4710445 73static phys_addr_t __init early_pgtable_alloc(void)
c1cc1552 74{
7142392d
SP
75 phys_addr_t phys;
76 void *ptr;
77
21ab99c2 78 phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
f4710445
MR
79
80 /*
81 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
82 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
83 * any level of table.
84 */
85 ptr = pte_set_fixmap(phys);
86
21ab99c2
MR
87 memset(ptr, 0, PAGE_SIZE);
88
f4710445
MR
89 /*
90 * Implicit barriers also ensure the zeroed page is visible to the page
91 * table walker
92 */
93 pte_clear_fixmap();
94
95 return phys;
c1cc1552
CM
96}
97
e98216b5
AB
98static bool pgattr_change_is_safe(u64 old, u64 new)
99{
100 /*
101 * The following mapping attributes may be updated in live
102 * kernel mappings without the need for break-before-make.
103 */
104 static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE;
105
106 return old == 0 || new == 0 || ((old ^ new) & ~mask) == 0;
107}
108
da141706 109static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
667c2759 110 unsigned long end, unsigned long pfn,
da141706 111 pgprot_t prot,
d81bbe6d 112 phys_addr_t (*pgtable_alloc)(void))
c1cc1552
CM
113{
114 pte_t *pte;
115
4133af6c
CM
116 BUG_ON(pmd_sect(*pmd));
117 if (pmd_none(*pmd)) {
132233a7
LA
118 phys_addr_t pte_phys;
119 BUG_ON(!pgtable_alloc);
120 pte_phys = pgtable_alloc();
f4710445 121 pte = pte_set_fixmap(pte_phys);
f4710445 122 __pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
f4710445 123 pte_clear_fixmap();
c1cc1552 124 }
a1c76574 125 BUG_ON(pmd_bad(*pmd));
c1cc1552 126
f4710445 127 pte = pte_set_fixmap_offset(pmd, addr);
c1cc1552 128 do {
e98216b5
AB
129 pte_t old_pte = *pte;
130
d81bbe6d 131 set_pte(pte, pfn_pte(pfn, prot));
667c2759 132 pfn++;
e98216b5
AB
133
134 /*
135 * After the PTE entry has been populated once, we
136 * only allow updates to the permission attributes.
137 */
138 BUG_ON(!pgattr_change_is_safe(pte_val(old_pte), pte_val(*pte)));
139
667c2759 140 } while (pte++, addr += PAGE_SIZE, addr != end);
f4710445
MR
141
142 pte_clear_fixmap();
c1cc1552
CM
143}
144
11509a30 145static void alloc_init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
da141706 146 phys_addr_t phys, pgprot_t prot,
53e1b329 147 phys_addr_t (*pgtable_alloc)(void),
f14c66ce 148 bool page_mappings_only)
c1cc1552
CM
149{
150 pmd_t *pmd;
151 unsigned long next;
152
153 /*
154 * Check for initial section mappings in the pgd/pud and remove them.
155 */
4133af6c
CM
156 BUG_ON(pud_sect(*pud));
157 if (pud_none(*pud)) {
132233a7
LA
158 phys_addr_t pmd_phys;
159 BUG_ON(!pgtable_alloc);
160 pmd_phys = pgtable_alloc();
f4710445 161 pmd = pmd_set_fixmap(pmd_phys);
f4710445 162 __pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
f4710445 163 pmd_clear_fixmap();
c1cc1552 164 }
a1c76574 165 BUG_ON(pud_bad(*pud));
c1cc1552 166
f4710445 167 pmd = pmd_set_fixmap_offset(pud, addr);
c1cc1552 168 do {
e98216b5
AB
169 pmd_t old_pmd = *pmd;
170
c1cc1552 171 next = pmd_addr_end(addr, end);
e98216b5 172
c1cc1552 173 /* try section mapping first */
83863f25 174 if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
f14c66ce 175 !page_mappings_only) {
d81bbe6d 176 pmd_set_huge(pmd, phys, prot);
e98216b5 177
a55f9929 178 /*
e98216b5
AB
179 * After the PMD entry has been populated once, we
180 * only allow updates to the permission attributes.
a55f9929 181 */
e98216b5
AB
182 BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
183 pmd_val(*pmd)));
a55f9929 184 } else {
667c2759 185 alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
d81bbe6d 186 prot, pgtable_alloc);
e98216b5
AB
187
188 BUG_ON(pmd_val(old_pmd) != 0 &&
189 pmd_val(old_pmd) != pmd_val(*pmd));
a55f9929 190 }
c1cc1552
CM
191 phys += next - addr;
192 } while (pmd++, addr = next, addr != end);
f4710445
MR
193
194 pmd_clear_fixmap();
c1cc1552
CM
195}
196
da141706
LA
197static inline bool use_1G_block(unsigned long addr, unsigned long next,
198 unsigned long phys)
199{
200 if (PAGE_SHIFT != 12)
201 return false;
202
203 if (((addr | next | phys) & ~PUD_MASK) != 0)
204 return false;
205
206 return true;
207}
208
11509a30 209static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
da141706 210 phys_addr_t phys, pgprot_t prot,
53e1b329 211 phys_addr_t (*pgtable_alloc)(void),
f14c66ce 212 bool page_mappings_only)
c1cc1552 213{
c79b954b 214 pud_t *pud;
c1cc1552
CM
215 unsigned long next;
216
c79b954b 217 if (pgd_none(*pgd)) {
132233a7
LA
218 phys_addr_t pud_phys;
219 BUG_ON(!pgtable_alloc);
220 pud_phys = pgtable_alloc();
f4710445 221 __pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
c79b954b
JL
222 }
223 BUG_ON(pgd_bad(*pgd));
224
f4710445 225 pud = pud_set_fixmap_offset(pgd, addr);
c1cc1552 226 do {
e98216b5
AB
227 pud_t old_pud = *pud;
228
c1cc1552 229 next = pud_addr_end(addr, end);
206a2a73
SC
230
231 /*
232 * For 4K granule only, attempt to put down a 1GB block
233 */
f14c66ce 234 if (use_1G_block(addr, next, phys) && !page_mappings_only) {
c661cb1c 235 pud_set_huge(pud, phys, prot);
206a2a73
SC
236
237 /*
e98216b5
AB
238 * After the PUD entry has been populated once, we
239 * only allow updates to the permission attributes.
206a2a73 240 */
e98216b5
AB
241 BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
242 pud_val(*pud)));
206a2a73 243 } else {
11509a30 244 alloc_init_pmd(pud, addr, next, phys, prot,
f14c66ce 245 pgtable_alloc, page_mappings_only);
e98216b5
AB
246
247 BUG_ON(pud_val(old_pud) != 0 &&
248 pud_val(old_pud) != pud_val(*pud));
206a2a73 249 }
c1cc1552
CM
250 phys += next - addr;
251 } while (pud++, addr = next, addr != end);
f4710445
MR
252
253 pud_clear_fixmap();
c1cc1552
CM
254}
255
40f87d31
AB
256static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
257 unsigned long virt, phys_addr_t size,
258 pgprot_t prot,
259 phys_addr_t (*pgtable_alloc)(void),
f14c66ce 260 bool page_mappings_only)
c1cc1552
CM
261{
262 unsigned long addr, length, end, next;
40f87d31 263 pgd_t *pgd = pgd_offset_raw(pgdir, virt);
c1cc1552 264
cc5d2b3b
MR
265 /*
266 * If the virtual and physical address don't have the same offset
267 * within a page, we cannot map the region as the caller expects.
268 */
269 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
270 return;
271
9c4e08a3 272 phys &= PAGE_MASK;
c1cc1552
CM
273 addr = virt & PAGE_MASK;
274 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
275
c1cc1552
CM
276 end = addr + length;
277 do {
278 next = pgd_addr_end(addr, end);
53e1b329 279 alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc,
f14c66ce 280 page_mappings_only);
c1cc1552
CM
281 phys += next - addr;
282 } while (pgd++, addr = next, addr != end);
283}
284
1378dc3d 285static phys_addr_t pgd_pgtable_alloc(void)
da141706 286{
21ab99c2 287 void *ptr = (void *)__get_free_page(PGALLOC_GFP);
1378dc3d
AB
288 if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
289 BUG();
21ab99c2
MR
290
291 /* Ensure the zeroed page is visible to the page table walker */
292 dsb(ishst);
f4710445 293 return __pa(ptr);
da141706
LA
294}
295
132233a7
LA
296/*
297 * This function can only be used to modify existing table entries,
298 * without allocating new levels of table. Note that this permits the
299 * creation of new section or page entries.
300 */
301static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
da141706 302 phys_addr_t size, pgprot_t prot)
d7ecbddf
MS
303{
304 if (virt < VMALLOC_START) {
305 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
306 &phys, virt);
307 return;
308 }
f14c66ce 309 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, false);
d7ecbddf
MS
310}
311
8ce837ce
AB
312void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
313 unsigned long virt, phys_addr_t size,
f14c66ce 314 pgprot_t prot, bool page_mappings_only)
8ce837ce 315{
1378dc3d
AB
316 BUG_ON(mm == &init_mm);
317
11509a30 318 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
f14c66ce 319 pgd_pgtable_alloc, page_mappings_only);
d7ecbddf
MS
320}
321
da141706
LA
322static void create_mapping_late(phys_addr_t phys, unsigned long virt,
323 phys_addr_t size, pgprot_t prot)
324{
325 if (virt < VMALLOC_START) {
326 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
327 &phys, virt);
328 return;
329 }
330
11509a30 331 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
f14c66ce 332 NULL, debug_pagealloc_enabled());
da141706
LA
333}
334
068a17a5 335static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, phys_addr_t end)
da141706 336{
eac8017f
MC
337 phys_addr_t kernel_start = __pa_symbol(_text);
338 phys_addr_t kernel_end = __pa_symbol(__init_begin);
068a17a5 339
da141706 340 /*
f9040773
AB
341 * Take care not to create a writable alias for the
342 * read-only text and rodata sections of the kernel image.
da141706 343 */
068a17a5 344
9fdc14c5 345 /* No overlap with the kernel text/rodata */
068a17a5
MR
346 if (end < kernel_start || start >= kernel_end) {
347 __create_pgd_mapping(pgd, start, __phys_to_virt(start),
348 end - start, PAGE_KERNEL,
53e1b329 349 early_pgtable_alloc,
f14c66ce 350 debug_pagealloc_enabled());
068a17a5 351 return;
da141706
LA
352 }
353
068a17a5 354 /*
9fdc14c5 355 * This block overlaps the kernel text/rodata mappings.
f9040773 356 * Map the portion(s) which don't overlap.
068a17a5
MR
357 */
358 if (start < kernel_start)
359 __create_pgd_mapping(pgd, start,
360 __phys_to_virt(start),
361 kernel_start - start, PAGE_KERNEL,
53e1b329 362 early_pgtable_alloc,
f14c66ce 363 debug_pagealloc_enabled());
068a17a5
MR
364 if (kernel_end < end)
365 __create_pgd_mapping(pgd, kernel_end,
366 __phys_to_virt(kernel_end),
367 end - kernel_end, PAGE_KERNEL,
53e1b329 368 early_pgtable_alloc,
f14c66ce 369 debug_pagealloc_enabled());
f9040773
AB
370
371 /*
9fdc14c5 372 * Map the linear alias of the [_text, __init_begin) interval as
f9040773
AB
373 * read-only/non-executable. This makes the contents of the
374 * region accessible to subsystems such as hibernate, but
375 * protects it from inadvertent modification or execution.
376 */
377 __create_pgd_mapping(pgd, kernel_start, __phys_to_virt(kernel_start),
378 kernel_end - kernel_start, PAGE_KERNEL_RO,
f14c66ce 379 early_pgtable_alloc, debug_pagealloc_enabled());
da141706 380}
da141706 381
068a17a5 382static void __init map_mem(pgd_t *pgd)
c1cc1552
CM
383{
384 struct memblock_region *reg;
f6bc87c3 385
c1cc1552
CM
386 /* map all the memory banks */
387 for_each_memblock(memory, reg) {
388 phys_addr_t start = reg->base;
389 phys_addr_t end = start + reg->size;
390
391 if (start >= end)
392 break;
68709f45
AB
393 if (memblock_is_nomap(reg))
394 continue;
c1cc1552 395
068a17a5 396 __map_memblock(pgd, start, end);
c1cc1552
CM
397 }
398}
399
da141706
LA
400void mark_rodata_ro(void)
401{
2f39b5f9 402 unsigned long section_size;
f9040773 403
9fdc14c5 404 section_size = (unsigned long)_etext - (unsigned long)_text;
2077be67 405 create_mapping_late(__pa_symbol(_text), (unsigned long)_text,
2f39b5f9
JL
406 section_size, PAGE_KERNEL_ROX);
407 /*
9fdc14c5
AB
408 * mark .rodata as read only. Use __init_begin rather than __end_rodata
409 * to cover NOTES and EXCEPTION_TABLE.
2f39b5f9 410 */
9fdc14c5 411 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
2077be67 412 create_mapping_late(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
2f39b5f9 413 section_size, PAGE_KERNEL_RO);
e98216b5
AB
414
415 /* flush the TLBs after updating live kernel mappings */
416 flush_tlb_all();
1404d6f1
LA
417
418 debug_checkwx();
da141706 419}
da141706 420
2c09ec06
AB
421static void __init map_kernel_segment(pgd_t *pgd, void *va_start, void *va_end,
422 pgprot_t prot, struct vm_struct *vma)
068a17a5 423{
2077be67 424 phys_addr_t pa_start = __pa_symbol(va_start);
068a17a5
MR
425 unsigned long size = va_end - va_start;
426
427 BUG_ON(!PAGE_ALIGNED(pa_start));
428 BUG_ON(!PAGE_ALIGNED(size));
429
430 __create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
f14c66ce 431 early_pgtable_alloc, debug_pagealloc_enabled());
f9040773
AB
432
433 vma->addr = va_start;
434 vma->phys_addr = pa_start;
435 vma->size = size;
436 vma->flags = VM_MAP;
437 vma->caller = __builtin_return_address(0);
438
439 vm_area_add_early(vma);
068a17a5
MR
440}
441
442/*
443 * Create fine-grained mappings for the kernel.
444 */
445static void __init map_kernel(pgd_t *pgd)
446{
2f39b5f9 447 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_init, vmlinux_data;
068a17a5 448
9fdc14c5
AB
449 map_kernel_segment(pgd, _text, _etext, PAGE_KERNEL_EXEC, &vmlinux_text);
450 map_kernel_segment(pgd, __start_rodata, __init_begin, PAGE_KERNEL, &vmlinux_rodata);
2c09ec06
AB
451 map_kernel_segment(pgd, __init_begin, __init_end, PAGE_KERNEL_EXEC,
452 &vmlinux_init);
453 map_kernel_segment(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data);
068a17a5 454
f9040773
AB
455 if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
456 /*
457 * The fixmap falls in a separate pgd to the kernel, and doesn't
458 * live in the carveout for the swapper_pg_dir. We can simply
459 * re-use the existing dir for the fixmap.
460 */
461 set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
462 *pgd_offset_k(FIXADDR_START));
463 } else if (CONFIG_PGTABLE_LEVELS > 3) {
464 /*
465 * The fixmap shares its top level pgd entry with the kernel
466 * mapping. This can really only occur when we are running
467 * with 16k/4 levels, so we can simply reuse the pud level
468 * entry instead.
469 */
470 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
471 set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
2077be67 472 __pud(__pa_symbol(bm_pmd) | PUD_TYPE_TABLE));
f9040773
AB
473 pud_clear_fixmap();
474 } else {
475 BUG();
476 }
068a17a5
MR
477
478 kasan_copy_shadow(pgd);
479}
480
c1cc1552
CM
481/*
482 * paging_init() sets up the page tables, initialises the zone memory
483 * maps and sets up the zero page.
484 */
485void __init paging_init(void)
486{
068a17a5
MR
487 phys_addr_t pgd_phys = early_pgtable_alloc();
488 pgd_t *pgd = pgd_set_fixmap(pgd_phys);
489
490 map_kernel(pgd);
491 map_mem(pgd);
492
493 /*
494 * We want to reuse the original swapper_pg_dir so we don't have to
495 * communicate the new address to non-coherent secondaries in
496 * secondary_entry, and so cpu_switch_mm can generate the address with
497 * adrp+add rather than a load from some global variable.
498 *
499 * To do this we need to go via a temporary pgd.
500 */
501 cpu_replace_ttbr1(__va(pgd_phys));
12f043ff 502 memcpy(swapper_pg_dir, pgd, PGD_SIZE);
2077be67 503 cpu_replace_ttbr1(lm_alias(swapper_pg_dir));
068a17a5
MR
504
505 pgd_clear_fixmap();
506 memblock_free(pgd_phys, PAGE_SIZE);
507
508 /*
509 * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
510 * allocated with it.
511 */
2077be67 512 memblock_free(__pa_symbol(swapper_pg_dir) + PAGE_SIZE,
068a17a5 513 SWAPPER_DIR_SIZE - PAGE_SIZE);
c1cc1552
CM
514}
515
c1cc1552
CM
516/*
517 * Check whether a kernel address is valid (derived from arch/x86/).
518 */
519int kern_addr_valid(unsigned long addr)
520{
521 pgd_t *pgd;
522 pud_t *pud;
523 pmd_t *pmd;
524 pte_t *pte;
525
526 if ((((long)addr) >> VA_BITS) != -1UL)
527 return 0;
528
529 pgd = pgd_offset_k(addr);
530 if (pgd_none(*pgd))
531 return 0;
532
533 pud = pud_offset(pgd, addr);
534 if (pud_none(*pud))
535 return 0;
536
206a2a73
SC
537 if (pud_sect(*pud))
538 return pfn_valid(pud_pfn(*pud));
539
c1cc1552
CM
540 pmd = pmd_offset(pud, addr);
541 if (pmd_none(*pmd))
542 return 0;
543
da6e4cb6
DA
544 if (pmd_sect(*pmd))
545 return pfn_valid(pmd_pfn(*pmd));
546
c1cc1552
CM
547 pte = pte_offset_kernel(pmd, addr);
548 if (pte_none(*pte))
549 return 0;
550
551 return pfn_valid(pte_pfn(*pte));
552}
553#ifdef CONFIG_SPARSEMEM_VMEMMAP
b433dce0 554#if !ARM64_SWAPPER_USES_SECTION_MAPS
0aad818b 555int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
c1cc1552 556{
0aad818b 557 return vmemmap_populate_basepages(start, end, node);
c1cc1552 558}
b433dce0 559#else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
0aad818b 560int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
c1cc1552 561{
0aad818b 562 unsigned long addr = start;
c1cc1552
CM
563 unsigned long next;
564 pgd_t *pgd;
565 pud_t *pud;
566 pmd_t *pmd;
567
568 do {
569 next = pmd_addr_end(addr, end);
570
571 pgd = vmemmap_pgd_populate(addr, node);
572 if (!pgd)
573 return -ENOMEM;
574
575 pud = vmemmap_pud_populate(pgd, addr, node);
576 if (!pud)
577 return -ENOMEM;
578
579 pmd = pmd_offset(pud, addr);
580 if (pmd_none(*pmd)) {
581 void *p = NULL;
582
583 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
584 if (!p)
585 return -ENOMEM;
586
a501e324 587 set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
c1cc1552
CM
588 } else
589 vmemmap_verify((pte_t *)pmd, node, addr, next);
590 } while (addr = next, addr != end);
591
592 return 0;
593}
594#endif /* CONFIG_ARM64_64K_PAGES */
0aad818b 595void vmemmap_free(unsigned long start, unsigned long end)
0197518c
TC
596{
597}
c1cc1552 598#endif /* CONFIG_SPARSEMEM_VMEMMAP */
af86e597 599
af86e597
LA
600static inline pud_t * fixmap_pud(unsigned long addr)
601{
602 pgd_t *pgd = pgd_offset_k(addr);
603
604 BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
605
157962f5 606 return pud_offset_kimg(pgd, addr);
af86e597
LA
607}
608
609static inline pmd_t * fixmap_pmd(unsigned long addr)
610{
611 pud_t *pud = fixmap_pud(addr);
612
613 BUG_ON(pud_none(*pud) || pud_bad(*pud));
614
157962f5 615 return pmd_offset_kimg(pud, addr);
af86e597
LA
616}
617
618static inline pte_t * fixmap_pte(unsigned long addr)
619{
157962f5 620 return &bm_pte[pte_index(addr)];
af86e597
LA
621}
622
2077be67
LA
623/*
624 * The p*d_populate functions call virt_to_phys implicitly so they can't be used
625 * directly on kernel symbols (bm_p*d). This function is called too early to use
626 * lm_alias so __p*d_populate functions must be used to populate with the
627 * physical address from __pa_symbol.
628 */
af86e597
LA
629void __init early_fixmap_init(void)
630{
631 pgd_t *pgd;
632 pud_t *pud;
633 pmd_t *pmd;
634 unsigned long addr = FIXADDR_START;
635
636 pgd = pgd_offset_k(addr);
f80fb3a3 637 if (CONFIG_PGTABLE_LEVELS > 3 &&
2077be67 638 !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa_symbol(bm_pud))) {
f9040773
AB
639 /*
640 * We only end up here if the kernel mapping and the fixmap
641 * share the top level pgd entry, which should only happen on
642 * 16k/4 levels configurations.
643 */
644 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
645 pud = pud_offset_kimg(pgd, addr);
646 } else {
2077be67
LA
647 if (pgd_none(*pgd))
648 __pgd_populate(pgd, __pa_symbol(bm_pud), PUD_TYPE_TABLE);
f9040773
AB
649 pud = fixmap_pud(addr);
650 }
2077be67
LA
651 if (pud_none(*pud))
652 __pud_populate(pud, __pa_symbol(bm_pmd), PMD_TYPE_TABLE);
157962f5 653 pmd = fixmap_pmd(addr);
2077be67 654 __pmd_populate(pmd, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
af86e597
LA
655
656 /*
657 * The boot-ioremap range spans multiple pmds, for which
157962f5 658 * we are not prepared:
af86e597
LA
659 */
660 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
661 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
662
663 if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
664 || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
665 WARN_ON(1);
666 pr_warn("pmd %p != %p, %p\n",
667 pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
668 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
669 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
670 fix_to_virt(FIX_BTMAP_BEGIN));
671 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
672 fix_to_virt(FIX_BTMAP_END));
673
674 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
675 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
676 }
677}
678
679void __set_fixmap(enum fixed_addresses idx,
680 phys_addr_t phys, pgprot_t flags)
681{
682 unsigned long addr = __fix_to_virt(idx);
683 pte_t *pte;
684
b63dbef9 685 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
af86e597
LA
686
687 pte = fixmap_pte(addr);
688
689 if (pgprot_val(flags)) {
690 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
691 } else {
692 pte_clear(&init_mm, addr, pte);
693 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
694 }
695}
61bd93ce 696
f80fb3a3 697void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
61bd93ce
AB
698{
699 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
f80fb3a3 700 int offset;
61bd93ce
AB
701 void *dt_virt;
702
703 /*
704 * Check whether the physical FDT address is set and meets the minimum
705 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
04a84810
AB
706 * at least 8 bytes so that we can always access the magic and size
707 * fields of the FDT header after mapping the first chunk, double check
708 * here if that is indeed the case.
61bd93ce
AB
709 */
710 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
711 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
712 return NULL;
713
714 /*
715 * Make sure that the FDT region can be mapped without the need to
716 * allocate additional translation table pages, so that it is safe
132233a7 717 * to call create_mapping_noalloc() this early.
61bd93ce
AB
718 *
719 * On 64k pages, the FDT will be mapped using PTEs, so we need to
720 * be in the same PMD as the rest of the fixmap.
721 * On 4k pages, we'll use section mappings for the FDT so we only
722 * have to be in the same PUD.
723 */
724 BUILD_BUG_ON(dt_virt_base % SZ_2M);
725
b433dce0
SP
726 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
727 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
61bd93ce 728
b433dce0 729 offset = dt_phys % SWAPPER_BLOCK_SIZE;
61bd93ce
AB
730 dt_virt = (void *)dt_virt_base + offset;
731
732 /* map the first chunk so we can read the size from the header */
132233a7
LA
733 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
734 dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
61bd93ce 735
04a84810 736 if (fdt_magic(dt_virt) != FDT_MAGIC)
61bd93ce
AB
737 return NULL;
738
f80fb3a3
AB
739 *size = fdt_totalsize(dt_virt);
740 if (*size > MAX_FDT_SIZE)
61bd93ce
AB
741 return NULL;
742
f80fb3a3 743 if (offset + *size > SWAPPER_BLOCK_SIZE)
132233a7 744 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
f80fb3a3 745 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
61bd93ce 746
f80fb3a3
AB
747 return dt_virt;
748}
61bd93ce 749
f80fb3a3
AB
750void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
751{
752 void *dt_virt;
753 int size;
754
755 dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
756 if (!dt_virt)
757 return NULL;
758
759 memblock_reserve(dt_phys, size);
61bd93ce
AB
760 return dt_virt;
761}
324420bf
AB
762
763int __init arch_ioremap_pud_supported(void)
764{
765 /* only 4k granule supports level 1 block mappings */
766 return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
767}
768
769int __init arch_ioremap_pmd_supported(void)
770{
771 return 1;
772}
773
774int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
775{
776 BUG_ON(phys & ~PUD_MASK);
777 set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
778 return 1;
779}
780
781int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
782{
783 BUG_ON(phys & ~PMD_MASK);
784 set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
785 return 1;
786}
787
788int pud_clear_huge(pud_t *pud)
789{
790 if (!pud_sect(*pud))
791 return 0;
792 pud_clear(pud);
793 return 1;
794}
795
796int pmd_clear_huge(pmd_t *pmd)
797{
798 if (!pmd_sect(*pmd))
799 return 0;
800 pmd_clear(pmd);
801 return 1;
802}