1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 2012 Regents of the University of California
4 * Copyright (C) 2019 Western Digital Corporation or its affiliates.
5 * Copyright (C) 2020 FORTH-ICS/CARV
6 * Nick Kossifidis <mick@ics.forth.gr>
7 */
8
9 #include <linux/init.h>
10 #include <linux/mm.h>
11 #include <linux/memblock.h>
12 #include <linux/initrd.h>
13 #include <linux/swap.h>
14 #include <linux/swiotlb.h>
15 #include <linux/sizes.h>
16 #include <linux/of_fdt.h>
17 #include <linux/of_reserved_mem.h>
18 #include <linux/libfdt.h>
19 #include <linux/set_memory.h>
20 #include <linux/dma-map-ops.h>
21 #include <linux/crash_dump.h>
22 #include <linux/hugetlb.h>
23 #ifdef CONFIG_RELOCATABLE
24 #include <linux/elf.h>
25 #endif
26 #include <linux/kfence.h>
27
28 #include <asm/fixmap.h>
29 #include <asm/io.h>
30 #include <asm/numa.h>
31 #include <asm/pgtable.h>
32 #include <asm/ptdump.h>
33 #include <asm/sections.h>
34 #include <asm/soc.h>
35 #include <asm/tlbflush.h>
36
37 #include "../kernel/head.h"
38
39 struct kernel_mapping kernel_map __ro_after_init;
40 EXPORT_SYMBOL(kernel_map);
41 #ifdef CONFIG_XIP_KERNEL
42 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
43 #endif
44
45 #ifdef CONFIG_64BIT
46 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39;
47 #else
48 u64 satp_mode __ro_after_init = SATP_MODE_32;
49 #endif
50 EXPORT_SYMBOL(satp_mode);
51
52 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
53 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
54 EXPORT_SYMBOL(pgtable_l4_enabled);
55 EXPORT_SYMBOL(pgtable_l5_enabled);
56
57 phys_addr_t phys_ram_base __ro_after_init;
58 EXPORT_SYMBOL(phys_ram_base);
59
60 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
61 __page_aligned_bss;
62 EXPORT_SYMBOL(empty_zero_page);
63
64 extern char _start[];
65 void *_dtb_early_va __initdata;
66 uintptr_t _dtb_early_pa __initdata;
67
68 static phys_addr_t dma32_phys_limit __initdata;
69
zone_sizes_init(void)70 static void __init zone_sizes_init(void)
71 {
72 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
73
74 #ifdef CONFIG_ZONE_DMA32
75 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
76 #endif
77 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
78
79 free_area_init(max_zone_pfns);
80 }
81
82 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
83
84 #define LOG2_SZ_1K ilog2(SZ_1K)
85 #define LOG2_SZ_1M ilog2(SZ_1M)
86 #define LOG2_SZ_1G ilog2(SZ_1G)
87 #define LOG2_SZ_1T ilog2(SZ_1T)
88
print_mlk(char * name,unsigned long b,unsigned long t)89 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
90 {
91 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
92 (((t) - (b)) >> LOG2_SZ_1K));
93 }
94
print_mlm(char * name,unsigned long b,unsigned long t)95 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
96 {
97 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
98 (((t) - (b)) >> LOG2_SZ_1M));
99 }
100
print_mlg(char * name,unsigned long b,unsigned long t)101 static inline void print_mlg(char *name, unsigned long b, unsigned long t)
102 {
103 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t,
104 (((t) - (b)) >> LOG2_SZ_1G));
105 }
106
107 #ifdef CONFIG_64BIT
print_mlt(char * name,unsigned long b,unsigned long t)108 static inline void print_mlt(char *name, unsigned long b, unsigned long t)
109 {
110 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t,
111 (((t) - (b)) >> LOG2_SZ_1T));
112 }
113 #else
114 #define print_mlt(n, b, t) do {} while (0)
115 #endif
116
print_ml(char * name,unsigned long b,unsigned long t)117 static inline void print_ml(char *name, unsigned long b, unsigned long t)
118 {
119 unsigned long diff = t - b;
120
121 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10)
122 print_mlt(name, b, t);
123 else if ((diff >> LOG2_SZ_1G) >= 10)
124 print_mlg(name, b, t);
125 else if ((diff >> LOG2_SZ_1M) >= 10)
126 print_mlm(name, b, t);
127 else
128 print_mlk(name, b, t);
129 }
130
print_vm_layout(void)131 static void __init print_vm_layout(void)
132 {
133 pr_notice("Virtual kernel memory layout:\n");
134 print_ml("fixmap", (unsigned long)FIXADDR_START,
135 (unsigned long)FIXADDR_TOP);
136 print_ml("pci io", (unsigned long)PCI_IO_START,
137 (unsigned long)PCI_IO_END);
138 print_ml("vmemmap", (unsigned long)VMEMMAP_START,
139 (unsigned long)VMEMMAP_END);
140 print_ml("vmalloc", (unsigned long)VMALLOC_START,
141 (unsigned long)VMALLOC_END);
142 #ifdef CONFIG_64BIT
143 print_ml("modules", (unsigned long)MODULES_VADDR,
144 (unsigned long)MODULES_END);
145 #endif
146 print_ml("lowmem", (unsigned long)PAGE_OFFSET,
147 (unsigned long)high_memory);
148 if (IS_ENABLED(CONFIG_64BIT)) {
149 #ifdef CONFIG_KASAN
150 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
151 #endif
152
153 print_ml("kernel", (unsigned long)kernel_map.virt_addr,
154 (unsigned long)ADDRESS_SPACE_END);
155 }
156 }
157 #else
print_vm_layout(void)158 static void print_vm_layout(void) { }
159 #endif /* CONFIG_DEBUG_VM */
160
mem_init(void)161 void __init mem_init(void)
162 {
163 #ifdef CONFIG_FLATMEM
164 BUG_ON(!mem_map);
165 #endif /* CONFIG_FLATMEM */
166
167 swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE);
168 memblock_free_all();
169
170 print_vm_layout();
171 }
172
173 /* Limit the memory size via mem. */
174 static phys_addr_t memory_limit;
175 #ifdef CONFIG_XIP_KERNEL
176 #define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit))
177 #endif /* CONFIG_XIP_KERNEL */
178
early_mem(char * p)179 static int __init early_mem(char *p)
180 {
181 u64 size;
182
183 if (!p)
184 return 1;
185
186 size = memparse(p, &p) & PAGE_MASK;
187 memory_limit = min_t(u64, size, memory_limit);
188
189 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
190
191 return 0;
192 }
193 early_param("mem", early_mem);
194
setup_bootmem(void)195 static void __init setup_bootmem(void)
196 {
197 phys_addr_t vmlinux_end = __pa_symbol(&_end);
198 phys_addr_t max_mapped_addr;
199 phys_addr_t phys_ram_end, vmlinux_start;
200
201 if (IS_ENABLED(CONFIG_XIP_KERNEL))
202 vmlinux_start = __pa_symbol(&_sdata);
203 else
204 vmlinux_start = __pa_symbol(&_start);
205
206 memblock_enforce_memory_limit(memory_limit);
207
208 /*
209 * Make sure we align the reservation on PMD_SIZE since we will
210 * map the kernel in the linear mapping as read-only: we do not want
211 * any allocation to happen between _end and the next pmd aligned page.
212 */
213 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
214 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
215 /*
216 * Reserve from the start of the kernel to the end of the kernel
217 */
218 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
219
220 /*
221 * Make sure we align the start of the memory on a PMD boundary so that
222 * at worst, we map the linear mapping with PMD mappings.
223 */
224 if (!IS_ENABLED(CONFIG_XIP_KERNEL))
225 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK;
226
227 /*
228 * In 64-bit, any use of __va/__pa before this point is wrong as we
229 * did not know the start of DRAM before.
230 */
231 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU))
232 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base;
233
234 /*
235 * The size of the linear page mapping may restrict the amount of
236 * usable RAM.
237 */
238 if (IS_ENABLED(CONFIG_64BIT)) {
239 max_mapped_addr = __pa(PAGE_OFFSET) + KERN_VIRT_SIZE;
240 memblock_cap_memory_range(phys_ram_base,
241 max_mapped_addr - phys_ram_base);
242 }
243
244 /*
245 * Reserve physical address space that would be mapped to virtual
246 * addresses greater than (void *)(-PAGE_SIZE) because:
247 * - This memory would overlap with ERR_PTR
248 * - This memory belongs to high memory, which is not supported
249 *
250 * This is not applicable to 64-bit kernel, because virtual addresses
251 * after (void *)(-PAGE_SIZE) are not linearly mapped: they are
252 * occupied by kernel mapping. Also it is unrealistic for high memory
253 * to exist on 64-bit platforms.
254 */
255 if (!IS_ENABLED(CONFIG_64BIT)) {
256 max_mapped_addr = __va_to_pa_nodebug(-PAGE_SIZE);
257 memblock_reserve(max_mapped_addr, (phys_addr_t)-max_mapped_addr);
258 }
259
260 phys_ram_end = memblock_end_of_DRAM();
261 min_low_pfn = PFN_UP(phys_ram_base);
262 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
263 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
264
265 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
266 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
267
268 reserve_initrd_mem();
269
270 /*
271 * No allocation should be done before reserving the memory as defined
272 * in the device tree, otherwise the allocation could end up in a
273 * reserved region.
274 */
275 early_init_fdt_scan_reserved_mem();
276
277 /*
278 * If DTB is built in, no need to reserve its memblock.
279 * Otherwise, do reserve it but avoid using
280 * early_init_fdt_reserve_self() since __pa() does
281 * not work for DTB pointers that are fixmap addresses
282 */
283 if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
284 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
285
286 dma_contiguous_reserve(dma32_phys_limit);
287 if (IS_ENABLED(CONFIG_64BIT))
288 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
289 }
290
291 #ifdef CONFIG_MMU
292 struct pt_alloc_ops pt_ops __initdata;
293
294 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
295 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
296 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
297
298 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
299
300 #ifdef CONFIG_XIP_KERNEL
301 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
302 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
303 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
304 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
305 #endif /* CONFIG_XIP_KERNEL */
306
307 static const pgprot_t protection_map[16] = {
308 [VM_NONE] = PAGE_NONE,
309 [VM_READ] = PAGE_READ,
310 [VM_WRITE] = PAGE_COPY,
311 [VM_WRITE | VM_READ] = PAGE_COPY,
312 [VM_EXEC] = PAGE_EXEC,
313 [VM_EXEC | VM_READ] = PAGE_READ_EXEC,
314 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC,
315 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC,
316 [VM_SHARED] = PAGE_NONE,
317 [VM_SHARED | VM_READ] = PAGE_READ,
318 [VM_SHARED | VM_WRITE] = PAGE_SHARED,
319 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED,
320 [VM_SHARED | VM_EXEC] = PAGE_EXEC,
321 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC,
322 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC,
323 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC
324 };
325 DECLARE_VM_GET_PAGE_PROT
326
__set_fixmap(enum fixed_addresses idx,phys_addr_t phys,pgprot_t prot)327 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
328 {
329 unsigned long addr = __fix_to_virt(idx);
330 pte_t *ptep;
331
332 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
333
334 ptep = &fixmap_pte[pte_index(addr)];
335
336 if (pgprot_val(prot))
337 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
338 else
339 pte_clear(&init_mm, addr, ptep);
340 local_flush_tlb_page(addr);
341 }
342
get_pte_virt_early(phys_addr_t pa)343 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
344 {
345 return (pte_t *)((uintptr_t)pa);
346 }
347
get_pte_virt_fixmap(phys_addr_t pa)348 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
349 {
350 clear_fixmap(FIX_PTE);
351 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
352 }
353
get_pte_virt_late(phys_addr_t pa)354 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
355 {
356 return (pte_t *) __va(pa);
357 }
358
alloc_pte_early(uintptr_t va)359 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
360 {
361 /*
362 * We only create PMD or PGD early mappings so we
363 * should never reach here with MMU disabled.
364 */
365 BUG();
366 }
367
alloc_pte_fixmap(uintptr_t va)368 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
369 {
370 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
371 }
372
alloc_pte_late(uintptr_t va)373 static phys_addr_t __init alloc_pte_late(uintptr_t va)
374 {
375 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
376
377 BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc));
378 return __pa((pte_t *)ptdesc_address(ptdesc));
379 }
380
create_pte_mapping(pte_t * ptep,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)381 static void __init create_pte_mapping(pte_t *ptep,
382 uintptr_t va, phys_addr_t pa,
383 phys_addr_t sz, pgprot_t prot)
384 {
385 uintptr_t pte_idx = pte_index(va);
386
387 BUG_ON(sz != PAGE_SIZE);
388
389 if (pte_none(ptep[pte_idx]))
390 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
391 }
392
393 #ifndef __PAGETABLE_PMD_FOLDED
394
395 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
396 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
397 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
398
399 #ifdef CONFIG_XIP_KERNEL
400 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
401 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
402 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
403 #endif /* CONFIG_XIP_KERNEL */
404
405 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
406 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
407 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
408
409 #ifdef CONFIG_XIP_KERNEL
410 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
411 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d))
412 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d))
413 #endif /* CONFIG_XIP_KERNEL */
414
415 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
416 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
417 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
418
419 #ifdef CONFIG_XIP_KERNEL
420 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
421 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
422 #define early_pud ((pud_t *)XIP_FIXUP(early_pud))
423 #endif /* CONFIG_XIP_KERNEL */
424
get_pmd_virt_early(phys_addr_t pa)425 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
426 {
427 /* Before MMU is enabled */
428 return (pmd_t *)((uintptr_t)pa);
429 }
430
get_pmd_virt_fixmap(phys_addr_t pa)431 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
432 {
433 clear_fixmap(FIX_PMD);
434 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
435 }
436
get_pmd_virt_late(phys_addr_t pa)437 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
438 {
439 return (pmd_t *) __va(pa);
440 }
441
alloc_pmd_early(uintptr_t va)442 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
443 {
444 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
445
446 return (uintptr_t)early_pmd;
447 }
448
alloc_pmd_fixmap(uintptr_t va)449 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
450 {
451 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
452 }
453
alloc_pmd_late(uintptr_t va)454 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
455 {
456 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
457
458 BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc));
459 return __pa((pmd_t *)ptdesc_address(ptdesc));
460 }
461
create_pmd_mapping(pmd_t * pmdp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)462 static void __init create_pmd_mapping(pmd_t *pmdp,
463 uintptr_t va, phys_addr_t pa,
464 phys_addr_t sz, pgprot_t prot)
465 {
466 pte_t *ptep;
467 phys_addr_t pte_phys;
468 uintptr_t pmd_idx = pmd_index(va);
469
470 if (sz == PMD_SIZE) {
471 if (pmd_none(pmdp[pmd_idx]))
472 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
473 return;
474 }
475
476 if (pmd_none(pmdp[pmd_idx])) {
477 pte_phys = pt_ops.alloc_pte(va);
478 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
479 ptep = pt_ops.get_pte_virt(pte_phys);
480 memset(ptep, 0, PAGE_SIZE);
481 } else {
482 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
483 ptep = pt_ops.get_pte_virt(pte_phys);
484 }
485
486 create_pte_mapping(ptep, va, pa, sz, prot);
487 }
488
get_pud_virt_early(phys_addr_t pa)489 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
490 {
491 return (pud_t *)((uintptr_t)pa);
492 }
493
get_pud_virt_fixmap(phys_addr_t pa)494 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
495 {
496 clear_fixmap(FIX_PUD);
497 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
498 }
499
get_pud_virt_late(phys_addr_t pa)500 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
501 {
502 return (pud_t *)__va(pa);
503 }
504
alloc_pud_early(uintptr_t va)505 static phys_addr_t __init alloc_pud_early(uintptr_t va)
506 {
507 /* Only one PUD is available for early mapping */
508 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
509
510 return (uintptr_t)early_pud;
511 }
512
alloc_pud_fixmap(uintptr_t va)513 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
514 {
515 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
516 }
517
alloc_pud_late(uintptr_t va)518 static phys_addr_t alloc_pud_late(uintptr_t va)
519 {
520 unsigned long vaddr;
521
522 vaddr = __get_free_page(GFP_KERNEL);
523 BUG_ON(!vaddr);
524 return __pa(vaddr);
525 }
526
get_p4d_virt_early(phys_addr_t pa)527 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
528 {
529 return (p4d_t *)((uintptr_t)pa);
530 }
531
get_p4d_virt_fixmap(phys_addr_t pa)532 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
533 {
534 clear_fixmap(FIX_P4D);
535 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
536 }
537
get_p4d_virt_late(phys_addr_t pa)538 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
539 {
540 return (p4d_t *)__va(pa);
541 }
542
alloc_p4d_early(uintptr_t va)543 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
544 {
545 /* Only one P4D is available for early mapping */
546 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
547
548 return (uintptr_t)early_p4d;
549 }
550
alloc_p4d_fixmap(uintptr_t va)551 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
552 {
553 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
554 }
555
alloc_p4d_late(uintptr_t va)556 static phys_addr_t alloc_p4d_late(uintptr_t va)
557 {
558 unsigned long vaddr;
559
560 vaddr = __get_free_page(GFP_KERNEL);
561 BUG_ON(!vaddr);
562 return __pa(vaddr);
563 }
564
create_pud_mapping(pud_t * pudp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)565 static void __init create_pud_mapping(pud_t *pudp,
566 uintptr_t va, phys_addr_t pa,
567 phys_addr_t sz, pgprot_t prot)
568 {
569 pmd_t *nextp;
570 phys_addr_t next_phys;
571 uintptr_t pud_index = pud_index(va);
572
573 if (sz == PUD_SIZE) {
574 if (pud_val(pudp[pud_index]) == 0)
575 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
576 return;
577 }
578
579 if (pud_val(pudp[pud_index]) == 0) {
580 next_phys = pt_ops.alloc_pmd(va);
581 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
582 nextp = pt_ops.get_pmd_virt(next_phys);
583 memset(nextp, 0, PAGE_SIZE);
584 } else {
585 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
586 nextp = pt_ops.get_pmd_virt(next_phys);
587 }
588
589 create_pmd_mapping(nextp, va, pa, sz, prot);
590 }
591
create_p4d_mapping(p4d_t * p4dp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)592 static void __init create_p4d_mapping(p4d_t *p4dp,
593 uintptr_t va, phys_addr_t pa,
594 phys_addr_t sz, pgprot_t prot)
595 {
596 pud_t *nextp;
597 phys_addr_t next_phys;
598 uintptr_t p4d_index = p4d_index(va);
599
600 if (sz == P4D_SIZE) {
601 if (p4d_val(p4dp[p4d_index]) == 0)
602 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
603 return;
604 }
605
606 if (p4d_val(p4dp[p4d_index]) == 0) {
607 next_phys = pt_ops.alloc_pud(va);
608 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
609 nextp = pt_ops.get_pud_virt(next_phys);
610 memset(nextp, 0, PAGE_SIZE);
611 } else {
612 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
613 nextp = pt_ops.get_pud_virt(next_phys);
614 }
615
616 create_pud_mapping(nextp, va, pa, sz, prot);
617 }
618
619 #define pgd_next_t p4d_t
620 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \
621 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \
622 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
623 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \
624 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \
625 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
626 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
627 (pgtable_l5_enabled ? \
628 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
629 (pgtable_l4_enabled ? \
630 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \
631 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
632 #define fixmap_pgd_next (pgtable_l5_enabled ? \
633 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \
634 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
635 #define trampoline_pgd_next (pgtable_l5_enabled ? \
636 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \
637 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
638 #else
639 #define pgd_next_t pte_t
640 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
641 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
642 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
643 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
644 #define fixmap_pgd_next ((uintptr_t)fixmap_pte)
645 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
646 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
647 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
648 #endif /* __PAGETABLE_PMD_FOLDED */
649
create_pgd_mapping(pgd_t * pgdp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)650 void __init create_pgd_mapping(pgd_t *pgdp,
651 uintptr_t va, phys_addr_t pa,
652 phys_addr_t sz, pgprot_t prot)
653 {
654 pgd_next_t *nextp;
655 phys_addr_t next_phys;
656 uintptr_t pgd_idx = pgd_index(va);
657
658 if (sz == PGDIR_SIZE) {
659 if (pgd_val(pgdp[pgd_idx]) == 0)
660 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
661 return;
662 }
663
664 if (pgd_val(pgdp[pgd_idx]) == 0) {
665 next_phys = alloc_pgd_next(va);
666 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
667 nextp = get_pgd_next_virt(next_phys);
668 memset(nextp, 0, PAGE_SIZE);
669 } else {
670 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
671 nextp = get_pgd_next_virt(next_phys);
672 }
673
674 create_pgd_next_mapping(nextp, va, pa, sz, prot);
675 }
676
best_map_size(phys_addr_t pa,uintptr_t va,phys_addr_t size)677 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va,
678 phys_addr_t size)
679 {
680 if (debug_pagealloc_enabled())
681 return PAGE_SIZE;
682
683 if (pgtable_l5_enabled &&
684 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
685 return P4D_SIZE;
686
687 if (pgtable_l4_enabled &&
688 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
689 return PUD_SIZE;
690
691 if (IS_ENABLED(CONFIG_64BIT) &&
692 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
693 return PMD_SIZE;
694
695 return PAGE_SIZE;
696 }
697
698 #ifdef CONFIG_XIP_KERNEL
699 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
700 extern char _xiprom[], _exiprom[], __data_loc;
701
702 /* called from head.S with MMU off */
__copy_data(void)703 asmlinkage void __init __copy_data(void)
704 {
705 void *from = (void *)(&__data_loc);
706 void *to = (void *)CONFIG_PHYS_RAM_BASE;
707 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
708
709 memcpy(to, from, sz);
710 }
711 #endif
712
713 #ifdef CONFIG_STRICT_KERNEL_RWX
pgprot_from_va(uintptr_t va)714 static __init pgprot_t pgprot_from_va(uintptr_t va)
715 {
716 if (is_va_kernel_text(va))
717 return PAGE_KERNEL_READ_EXEC;
718
719 /*
720 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
721 * we must protect its linear mapping alias from being executed and
722 * written.
723 * And rodata section is marked readonly in mark_rodata_ro.
724 */
725 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
726 return PAGE_KERNEL_READ;
727
728 return PAGE_KERNEL;
729 }
730
mark_rodata_ro(void)731 void mark_rodata_ro(void)
732 {
733 set_kernel_memory(__start_rodata, _data, set_memory_ro);
734 if (IS_ENABLED(CONFIG_64BIT))
735 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
736 set_memory_ro);
737
738 debug_checkwx();
739 }
740 #else
pgprot_from_va(uintptr_t va)741 static __init pgprot_t pgprot_from_va(uintptr_t va)
742 {
743 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
744 return PAGE_KERNEL;
745
746 return PAGE_KERNEL_EXEC;
747 }
748 #endif /* CONFIG_STRICT_KERNEL_RWX */
749
750 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
751 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
752
disable_pgtable_l5(void)753 static void __init disable_pgtable_l5(void)
754 {
755 pgtable_l5_enabled = false;
756 kernel_map.page_offset = PAGE_OFFSET_L4;
757 satp_mode = SATP_MODE_48;
758 }
759
disable_pgtable_l4(void)760 static void __init disable_pgtable_l4(void)
761 {
762 pgtable_l4_enabled = false;
763 kernel_map.page_offset = PAGE_OFFSET_L3;
764 satp_mode = SATP_MODE_39;
765 }
766
print_no4lvl(char * p)767 static int __init print_no4lvl(char *p)
768 {
769 pr_info("Disabled 4-level and 5-level paging");
770 return 0;
771 }
772 early_param("no4lvl", print_no4lvl);
773
print_no5lvl(char * p)774 static int __init print_no5lvl(char *p)
775 {
776 pr_info("Disabled 5-level paging");
777 return 0;
778 }
779 early_param("no5lvl", print_no5lvl);
780
781 /*
782 * There is a simple way to determine if 4-level is supported by the
783 * underlying hardware: establish 1:1 mapping in 4-level page table mode
784 * then read SATP to see if the configuration was taken into account
785 * meaning sv48 is supported.
786 */
set_satp_mode(uintptr_t dtb_pa)787 static __init void set_satp_mode(uintptr_t dtb_pa)
788 {
789 u64 identity_satp, hw_satp;
790 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
791 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
792
793 if (satp_mode_cmdline == SATP_MODE_57) {
794 disable_pgtable_l5();
795 } else if (satp_mode_cmdline == SATP_MODE_48) {
796 disable_pgtable_l5();
797 disable_pgtable_l4();
798 return;
799 }
800
801 create_p4d_mapping(early_p4d,
802 set_satp_mode_pmd, (uintptr_t)early_pud,
803 P4D_SIZE, PAGE_TABLE);
804 create_pud_mapping(early_pud,
805 set_satp_mode_pmd, (uintptr_t)early_pmd,
806 PUD_SIZE, PAGE_TABLE);
807 /* Handle the case where set_satp_mode straddles 2 PMDs */
808 create_pmd_mapping(early_pmd,
809 set_satp_mode_pmd, set_satp_mode_pmd,
810 PMD_SIZE, PAGE_KERNEL_EXEC);
811 create_pmd_mapping(early_pmd,
812 set_satp_mode_pmd + PMD_SIZE,
813 set_satp_mode_pmd + PMD_SIZE,
814 PMD_SIZE, PAGE_KERNEL_EXEC);
815 retry:
816 create_pgd_mapping(early_pg_dir,
817 set_satp_mode_pmd,
818 pgtable_l5_enabled ?
819 (uintptr_t)early_p4d : (uintptr_t)early_pud,
820 PGDIR_SIZE, PAGE_TABLE);
821
822 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
823
824 local_flush_tlb_all();
825 csr_write(CSR_SATP, identity_satp);
826 hw_satp = csr_swap(CSR_SATP, 0ULL);
827 local_flush_tlb_all();
828
829 if (hw_satp != identity_satp) {
830 if (pgtable_l5_enabled) {
831 disable_pgtable_l5();
832 memset(early_pg_dir, 0, PAGE_SIZE);
833 goto retry;
834 }
835 disable_pgtable_l4();
836 }
837
838 memset(early_pg_dir, 0, PAGE_SIZE);
839 memset(early_p4d, 0, PAGE_SIZE);
840 memset(early_pud, 0, PAGE_SIZE);
841 memset(early_pmd, 0, PAGE_SIZE);
842 }
843 #endif
844
845 /*
846 * setup_vm() is called from head.S with MMU-off.
847 *
848 * Following requirements should be honoured for setup_vm() to work
849 * correctly:
850 * 1) It should use PC-relative addressing for accessing kernel symbols.
851 * To achieve this we always use GCC cmodel=medany.
852 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
853 * so disable compiler instrumentation when FTRACE is enabled.
854 *
855 * Currently, the above requirements are honoured by using custom CFLAGS
856 * for init.o in mm/Makefile.
857 */
858
859 #ifndef __riscv_cmodel_medany
860 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
861 #endif
862
863 #ifdef CONFIG_RELOCATABLE
864 extern unsigned long __rela_dyn_start, __rela_dyn_end;
865
relocate_kernel(void)866 static void __init relocate_kernel(void)
867 {
868 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
869 /*
870 * This holds the offset between the linked virtual address and the
871 * relocated virtual address.
872 */
873 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
874 /*
875 * This holds the offset between kernel linked virtual address and
876 * physical address.
877 */
878 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
879
880 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
881 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
882 Elf64_Addr relocated_addr = rela->r_addend;
883
884 if (rela->r_info != R_RISCV_RELATIVE)
885 continue;
886
887 /*
888 * Make sure to not relocate vdso symbols like rt_sigreturn
889 * which are linked from the address 0 in vmlinux since
890 * vdso symbol addresses are actually used as an offset from
891 * mm->context.vdso in VDSO_OFFSET macro.
892 */
893 if (relocated_addr >= KERNEL_LINK_ADDR)
894 relocated_addr += reloc_offset;
895
896 *(Elf64_Addr *)addr = relocated_addr;
897 }
898 }
899 #endif /* CONFIG_RELOCATABLE */
900
901 #ifdef CONFIG_XIP_KERNEL
create_kernel_page_table(pgd_t * pgdir,__always_unused bool early)902 static void __init create_kernel_page_table(pgd_t *pgdir,
903 __always_unused bool early)
904 {
905 uintptr_t va, end_va;
906
907 /* Map the flash resident part */
908 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
909 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
910 create_pgd_mapping(pgdir, va,
911 kernel_map.xiprom + (va - kernel_map.virt_addr),
912 PMD_SIZE, PAGE_KERNEL_EXEC);
913
914 /* Map the data in RAM */
915 end_va = kernel_map.virt_addr + kernel_map.size;
916 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
917 create_pgd_mapping(pgdir, va,
918 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
919 PMD_SIZE, PAGE_KERNEL);
920 }
921 #else
create_kernel_page_table(pgd_t * pgdir,bool early)922 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
923 {
924 uintptr_t va, end_va;
925
926 end_va = kernel_map.virt_addr + kernel_map.size;
927 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
928 create_pgd_mapping(pgdir, va,
929 kernel_map.phys_addr + (va - kernel_map.virt_addr),
930 PMD_SIZE,
931 early ?
932 PAGE_KERNEL_EXEC : pgprot_from_va(va));
933 }
934 #endif
935
936 /*
937 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
938 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
939 * entry.
940 */
create_fdt_early_page_table(uintptr_t fix_fdt_va,uintptr_t dtb_pa)941 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
942 uintptr_t dtb_pa)
943 {
944 #ifndef CONFIG_BUILTIN_DTB
945 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
946
947 /* Make sure the fdt fixmap address is always aligned on PMD size */
948 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
949
950 /* In 32-bit only, the fdt lies in its own PGD */
951 if (!IS_ENABLED(CONFIG_64BIT)) {
952 create_pgd_mapping(early_pg_dir, fix_fdt_va,
953 pa, MAX_FDT_SIZE, PAGE_KERNEL);
954 } else {
955 create_pmd_mapping(fixmap_pmd, fix_fdt_va,
956 pa, PMD_SIZE, PAGE_KERNEL);
957 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
958 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
959 }
960
961 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
962 #else
963 /*
964 * For 64-bit kernel, __va can't be used since it would return a linear
965 * mapping address whereas dtb_early_va will be used before
966 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
967 * kernel is mapped in the linear mapping, that makes no difference.
968 */
969 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa);
970 #endif
971
972 dtb_early_pa = dtb_pa;
973 }
974
975 /*
976 * MMU is not enabled, the page tables are allocated directly using
977 * early_pmd/pud/p4d and the address returned is the physical one.
978 */
pt_ops_set_early(void)979 static void __init pt_ops_set_early(void)
980 {
981 pt_ops.alloc_pte = alloc_pte_early;
982 pt_ops.get_pte_virt = get_pte_virt_early;
983 #ifndef __PAGETABLE_PMD_FOLDED
984 pt_ops.alloc_pmd = alloc_pmd_early;
985 pt_ops.get_pmd_virt = get_pmd_virt_early;
986 pt_ops.alloc_pud = alloc_pud_early;
987 pt_ops.get_pud_virt = get_pud_virt_early;
988 pt_ops.alloc_p4d = alloc_p4d_early;
989 pt_ops.get_p4d_virt = get_p4d_virt_early;
990 #endif
991 }
992
993 /*
994 * MMU is enabled but page table setup is not complete yet.
995 * fixmap page table alloc functions must be used as a means to temporarily
996 * map the allocated physical pages since the linear mapping does not exist yet.
997 *
998 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
999 * but it will be used as described above.
1000 */
pt_ops_set_fixmap(void)1001 static void __init pt_ops_set_fixmap(void)
1002 {
1003 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
1004 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
1005 #ifndef __PAGETABLE_PMD_FOLDED
1006 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
1007 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
1008 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
1009 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
1010 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
1011 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
1012 #endif
1013 }
1014
1015 /*
1016 * MMU is enabled and page table setup is complete, so from now, we can use
1017 * generic page allocation functions to setup page table.
1018 */
pt_ops_set_late(void)1019 static void __init pt_ops_set_late(void)
1020 {
1021 pt_ops.alloc_pte = alloc_pte_late;
1022 pt_ops.get_pte_virt = get_pte_virt_late;
1023 #ifndef __PAGETABLE_PMD_FOLDED
1024 pt_ops.alloc_pmd = alloc_pmd_late;
1025 pt_ops.get_pmd_virt = get_pmd_virt_late;
1026 pt_ops.alloc_pud = alloc_pud_late;
1027 pt_ops.get_pud_virt = get_pud_virt_late;
1028 pt_ops.alloc_p4d = alloc_p4d_late;
1029 pt_ops.get_p4d_virt = get_p4d_virt_late;
1030 #endif
1031 }
1032
1033 #ifdef CONFIG_RANDOMIZE_BASE
1034 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa);
1035 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa);
1036
print_nokaslr(char * p)1037 static int __init print_nokaslr(char *p)
1038 {
1039 pr_info("Disabled KASLR");
1040 return 0;
1041 }
1042 early_param("nokaslr", print_nokaslr);
1043
kaslr_offset(void)1044 unsigned long kaslr_offset(void)
1045 {
1046 return kernel_map.virt_offset;
1047 }
1048 #endif
1049
setup_vm(uintptr_t dtb_pa)1050 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1051 {
1052 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1053
1054 #ifdef CONFIG_RANDOMIZE_BASE
1055 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) {
1056 u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa);
1057 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
1058 u32 nr_pos;
1059
1060 /*
1061 * Compute the number of positions available: we are limited
1062 * by the early page table that only has one PUD and we must
1063 * be aligned on PMD_SIZE.
1064 */
1065 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE;
1066
1067 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE;
1068 }
1069 #endif
1070
1071 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset;
1072
1073 #ifdef CONFIG_XIP_KERNEL
1074 #ifdef CONFIG_64BIT
1075 kernel_map.page_offset = PAGE_OFFSET_L3;
1076 #else
1077 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1078 #endif
1079 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1080 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1081
1082 phys_ram_base = CONFIG_PHYS_RAM_BASE;
1083 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1084 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
1085
1086 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1087 #else
1088 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1089 kernel_map.phys_addr = (uintptr_t)(&_start);
1090 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1091 #endif
1092
1093 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1094 set_satp_mode(dtb_pa);
1095 #endif
1096
1097 /*
1098 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1099 * where we have the system memory layout: this allows us to align
1100 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1101 * for the linear mapping. This is only possible because the kernel
1102 * mapping lies outside the linear mapping.
1103 * In 32-bit however, as the kernel resides in the linear mapping,
1104 * setup_vm_final can not change the mapping established here,
1105 * otherwise the same kernel addresses would get mapped to different
1106 * physical addresses (if the start of dram is different from the
1107 * kernel physical address start).
1108 */
1109 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1110 0UL : PAGE_OFFSET - kernel_map.phys_addr;
1111 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1112
1113 /*
1114 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
1115 * kernel, whereas for 64-bit kernel, the end of the virtual address
1116 * space is occupied by the modules/BPF/kernel mappings which reduces
1117 * the available size of the linear mapping.
1118 */
1119 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
1120
1121 /* Sanity check alignment and size */
1122 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1123 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1124
1125 #ifdef CONFIG_64BIT
1126 /*
1127 * The last 4K bytes of the addressable memory can not be mapped because
1128 * of IS_ERR_VALUE macro.
1129 */
1130 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1131 #endif
1132
1133 #ifdef CONFIG_RELOCATABLE
1134 /*
1135 * Early page table uses only one PUD, which makes it possible
1136 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1137 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1138 * since a part of the kernel would not get mapped.
1139 */
1140 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1141 relocate_kernel();
1142 #endif
1143
1144 apply_early_boot_alternatives();
1145 pt_ops_set_early();
1146
1147 /* Setup early PGD for fixmap */
1148 create_pgd_mapping(early_pg_dir, FIXADDR_START,
1149 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1150
1151 #ifndef __PAGETABLE_PMD_FOLDED
1152 /* Setup fixmap P4D and PUD */
1153 if (pgtable_l5_enabled)
1154 create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1155 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1156 /* Setup fixmap PUD and PMD */
1157 if (pgtable_l4_enabled)
1158 create_pud_mapping(fixmap_pud, FIXADDR_START,
1159 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1160 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1161 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1162 /* Setup trampoline PGD and PMD */
1163 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1164 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1165 if (pgtable_l5_enabled)
1166 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1167 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1168 if (pgtable_l4_enabled)
1169 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1170 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1171 #ifdef CONFIG_XIP_KERNEL
1172 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1173 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1174 #else
1175 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1176 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1177 #endif
1178 #else
1179 /* Setup trampoline PGD */
1180 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1181 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1182 #endif
1183
1184 /*
1185 * Setup early PGD covering entire kernel which will allow
1186 * us to reach paging_init(). We map all memory banks later
1187 * in setup_vm_final() below.
1188 */
1189 create_kernel_page_table(early_pg_dir, true);
1190
1191 /* Setup early mapping for FDT early scan */
1192 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1193
1194 /*
1195 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1196 * range can not span multiple pmds.
1197 */
1198 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1199 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1200
1201 #ifndef __PAGETABLE_PMD_FOLDED
1202 /*
1203 * Early ioremap fixmap is already created as it lies within first 2MB
1204 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1205 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1206 * the user if not.
1207 */
1208 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1209 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1210 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1211 WARN_ON(1);
1212 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1213 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1214 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1215 fix_to_virt(FIX_BTMAP_BEGIN));
1216 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1217 fix_to_virt(FIX_BTMAP_END));
1218
1219 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1220 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1221 }
1222 #endif
1223
1224 pt_ops_set_fixmap();
1225 }
1226
create_linear_mapping_range(phys_addr_t start,phys_addr_t end,uintptr_t fixed_map_size)1227 static void __init create_linear_mapping_range(phys_addr_t start,
1228 phys_addr_t end,
1229 uintptr_t fixed_map_size)
1230 {
1231 phys_addr_t pa;
1232 uintptr_t va, map_size;
1233
1234 for (pa = start; pa < end; pa += map_size) {
1235 va = (uintptr_t)__va(pa);
1236 map_size = fixed_map_size ? fixed_map_size :
1237 best_map_size(pa, va, end - pa);
1238
1239 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1240 pgprot_from_va(va));
1241 }
1242 }
1243
create_linear_mapping_page_table(void)1244 static void __init create_linear_mapping_page_table(void)
1245 {
1246 phys_addr_t start, end;
1247 phys_addr_t kfence_pool __maybe_unused;
1248 u64 i;
1249
1250 #ifdef CONFIG_STRICT_KERNEL_RWX
1251 phys_addr_t ktext_start = __pa_symbol(_start);
1252 phys_addr_t ktext_size = __init_data_begin - _start;
1253 phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1254 phys_addr_t krodata_size = _data - __start_rodata;
1255
1256 /* Isolate kernel text and rodata so they don't get mapped with a PUD */
1257 memblock_mark_nomap(ktext_start, ktext_size);
1258 memblock_mark_nomap(krodata_start, krodata_size);
1259 #endif
1260
1261 #ifdef CONFIG_KFENCE
1262 /*
1263 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1264 * before we setup the linear mapping so that we avoid using hugepages
1265 * for this region.
1266 */
1267 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1268 BUG_ON(!kfence_pool);
1269
1270 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1271 __kfence_pool = __va(kfence_pool);
1272 #endif
1273
1274 /* Map all memory banks in the linear mapping */
1275 for_each_mem_range(i, &start, &end) {
1276 if (start >= end)
1277 break;
1278 if (start <= __pa(PAGE_OFFSET) &&
1279 __pa(PAGE_OFFSET) < end)
1280 start = __pa(PAGE_OFFSET);
1281
1282 create_linear_mapping_range(start, end, 0);
1283 }
1284
1285 #ifdef CONFIG_STRICT_KERNEL_RWX
1286 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0);
1287 create_linear_mapping_range(krodata_start,
1288 krodata_start + krodata_size, 0);
1289
1290 memblock_clear_nomap(ktext_start, ktext_size);
1291 memblock_clear_nomap(krodata_start, krodata_size);
1292 #endif
1293
1294 #ifdef CONFIG_KFENCE
1295 create_linear_mapping_range(kfence_pool,
1296 kfence_pool + KFENCE_POOL_SIZE,
1297 PAGE_SIZE);
1298
1299 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1300 #endif
1301 }
1302
setup_vm_final(void)1303 static void __init setup_vm_final(void)
1304 {
1305 /* Setup swapper PGD for fixmap */
1306 #if !defined(CONFIG_64BIT)
1307 /*
1308 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1309 * directly in swapper_pg_dir in addition to the pgd entry that points
1310 * to fixmap_pte.
1311 */
1312 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1313
1314 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1315 #endif
1316 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1317 __pa_symbol(fixmap_pgd_next),
1318 PGDIR_SIZE, PAGE_TABLE);
1319
1320 /* Map the linear mapping */
1321 create_linear_mapping_page_table();
1322
1323 /* Map the kernel */
1324 if (IS_ENABLED(CONFIG_64BIT))
1325 create_kernel_page_table(swapper_pg_dir, false);
1326
1327 #ifdef CONFIG_KASAN
1328 kasan_swapper_init();
1329 #endif
1330
1331 /* Clear fixmap PTE and PMD mappings */
1332 clear_fixmap(FIX_PTE);
1333 clear_fixmap(FIX_PMD);
1334 clear_fixmap(FIX_PUD);
1335 clear_fixmap(FIX_P4D);
1336
1337 /* Move to swapper page table */
1338 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1339 local_flush_tlb_all();
1340
1341 pt_ops_set_late();
1342 }
1343 #else
setup_vm(uintptr_t dtb_pa)1344 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1345 {
1346 dtb_early_va = (void *)dtb_pa;
1347 dtb_early_pa = dtb_pa;
1348 }
1349
setup_vm_final(void)1350 static inline void setup_vm_final(void)
1351 {
1352 }
1353 #endif /* CONFIG_MMU */
1354
1355 /* Reserve 128M low memory by default for swiotlb buffer */
1356 #define DEFAULT_CRASH_KERNEL_LOW_SIZE (128UL << 20)
1357
reserve_crashkernel_low(unsigned long long low_size)1358 static int __init reserve_crashkernel_low(unsigned long long low_size)
1359 {
1360 unsigned long long low_base;
1361
1362 low_base = memblock_phys_alloc_range(low_size, PMD_SIZE, 0, dma32_phys_limit);
1363 if (!low_base) {
1364 pr_err("cannot allocate crashkernel low memory (size:0x%llx).\n", low_size);
1365 return -ENOMEM;
1366 }
1367
1368 pr_info("crashkernel low memory reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
1369 low_base, low_base + low_size, low_size >> 20);
1370
1371 crashk_low_res.start = low_base;
1372 crashk_low_res.end = low_base + low_size - 1;
1373
1374 return 0;
1375 }
1376
1377 /*
1378 * reserve_crashkernel() - reserves memory for crash kernel
1379 *
1380 * This function reserves memory area given in "crashkernel=" kernel command
1381 * line parameter. The memory reserved is used by dump capture kernel when
1382 * primary kernel is crashing.
1383 */
reserve_crashkernel(void)1384 static void __init reserve_crashkernel(void)
1385 {
1386 unsigned long long crash_base = 0;
1387 unsigned long long crash_size = 0;
1388 unsigned long long crash_low_size = 0;
1389 unsigned long search_start = memblock_start_of_DRAM();
1390 unsigned long search_end = (unsigned long)dma32_phys_limit;
1391 char *cmdline = boot_command_line;
1392 bool fixed_base = false;
1393 bool high = false;
1394
1395 int ret = 0;
1396
1397 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1398 return;
1399 /*
1400 * Don't reserve a region for a crash kernel on a crash kernel
1401 * since it doesn't make much sense and we have limited memory
1402 * resources.
1403 */
1404 if (is_kdump_kernel()) {
1405 pr_info("crashkernel: ignoring reservation request\n");
1406 return;
1407 }
1408
1409 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(),
1410 &crash_size, &crash_base);
1411 if (ret == -ENOENT) {
1412 /* Fallback to crashkernel=X,[high,low] */
1413 ret = parse_crashkernel_high(cmdline, 0, &crash_size, &crash_base);
1414 if (ret || !crash_size)
1415 return;
1416
1417 /*
1418 * crashkernel=Y,low is valid only when crashkernel=X,high
1419 * is passed.
1420 */
1421 ret = parse_crashkernel_low(cmdline, 0, &crash_low_size, &crash_base);
1422 if (ret == -ENOENT)
1423 crash_low_size = DEFAULT_CRASH_KERNEL_LOW_SIZE;
1424 else if (ret)
1425 return;
1426
1427 search_start = (unsigned long)dma32_phys_limit;
1428 search_end = memblock_end_of_DRAM();
1429 high = true;
1430 } else if (ret || !crash_size) {
1431 /* Invalid argument value specified */
1432 return;
1433 }
1434
1435 crash_size = PAGE_ALIGN(crash_size);
1436
1437 if (crash_base) {
1438 fixed_base = true;
1439 search_start = crash_base;
1440 search_end = crash_base + crash_size;
1441 }
1442
1443 /*
1444 * Current riscv boot protocol requires 2MB alignment for
1445 * RV64 and 4MB alignment for RV32 (hugepage size)
1446 *
1447 * Try to alloc from 32bit addressible physical memory so that
1448 * swiotlb can work on the crash kernel.
1449 */
1450 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1451 search_start, search_end);
1452 if (crash_base == 0) {
1453 /*
1454 * For crashkernel=size[KMG]@offset[KMG], print out failure
1455 * message if can't reserve the specified region.
1456 */
1457 if (fixed_base) {
1458 pr_warn("crashkernel: allocating failed with given size@offset\n");
1459 return;
1460 }
1461
1462 if (high) {
1463 /*
1464 * For crashkernel=size[KMG],high, if the first attempt was
1465 * for high memory, fall back to low memory.
1466 */
1467 search_start = memblock_start_of_DRAM();
1468 search_end = (unsigned long)dma32_phys_limit;
1469 } else {
1470 /*
1471 * For crashkernel=size[KMG], if the first attempt was for
1472 * low memory, fall back to high memory, the minimum required
1473 * low memory will be reserved later.
1474 */
1475 search_start = (unsigned long)dma32_phys_limit;
1476 search_end = memblock_end_of_DRAM();
1477 crash_low_size = DEFAULT_CRASH_KERNEL_LOW_SIZE;
1478 }
1479
1480 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1481 search_start, search_end);
1482 if (crash_base == 0) {
1483 pr_warn("crashkernel: couldn't allocate %lldKB\n",
1484 crash_size >> 10);
1485 return;
1486 }
1487 }
1488
1489 if ((crash_base >= dma32_phys_limit) && crash_low_size &&
1490 reserve_crashkernel_low(crash_low_size)) {
1491 memblock_phys_free(crash_base, crash_size);
1492 return;
1493 }
1494
1495 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1496 crash_base, crash_base + crash_size, crash_size >> 20);
1497
1498 crashk_res.start = crash_base;
1499 crashk_res.end = crash_base + crash_size - 1;
1500 }
1501
paging_init(void)1502 void __init paging_init(void)
1503 {
1504 setup_bootmem();
1505 setup_vm_final();
1506
1507 /* Depend on that Linear Mapping is ready */
1508 memblock_allow_resize();
1509 }
1510
misc_mem_init(void)1511 void __init misc_mem_init(void)
1512 {
1513 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1514 arch_numa_init();
1515 sparse_init();
1516 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1517 /* The entire VMEMMAP region has been populated. Flush TLB for this region */
1518 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END);
1519 #endif
1520 zone_sizes_init();
1521 reserve_crashkernel();
1522 memblock_dump_all();
1523 }
1524
1525 #ifdef CONFIG_SPARSEMEM_VMEMMAP
vmemmap_populate(unsigned long start,unsigned long end,int node,struct vmem_altmap * altmap)1526 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1527 struct vmem_altmap *altmap)
1528 {
1529 return vmemmap_populate_basepages(start, end, node, NULL);
1530 }
1531 #endif
1532
1533 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1534 /*
1535 * Pre-allocates page-table pages for a specific area in the kernel
1536 * page-table. Only the level which needs to be synchronized between
1537 * all page-tables is allocated because the synchronization can be
1538 * expensive.
1539 */
preallocate_pgd_pages_range(unsigned long start,unsigned long end,const char * area)1540 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1541 const char *area)
1542 {
1543 unsigned long addr;
1544 const char *lvl;
1545
1546 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1547 pgd_t *pgd = pgd_offset_k(addr);
1548 p4d_t *p4d;
1549 pud_t *pud;
1550 pmd_t *pmd;
1551
1552 lvl = "p4d";
1553 p4d = p4d_alloc(&init_mm, pgd, addr);
1554 if (!p4d)
1555 goto failed;
1556
1557 if (pgtable_l5_enabled)
1558 continue;
1559
1560 lvl = "pud";
1561 pud = pud_alloc(&init_mm, p4d, addr);
1562 if (!pud)
1563 goto failed;
1564
1565 if (pgtable_l4_enabled)
1566 continue;
1567
1568 lvl = "pmd";
1569 pmd = pmd_alloc(&init_mm, pud, addr);
1570 if (!pmd)
1571 goto failed;
1572 }
1573 return;
1574
1575 failed:
1576 /*
1577 * The pages have to be there now or they will be missing in
1578 * process page-tables later.
1579 */
1580 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1581 }
1582
pgtable_cache_init(void)1583 void __init pgtable_cache_init(void)
1584 {
1585 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1586 if (IS_ENABLED(CONFIG_MODULES))
1587 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
1588 }
1589 #endif
1590