xref: /linux/arch/riscv/mm/init.c (revision 8e07e0e3964ca4e23ce7b68e2096fe660a888942)
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 #ifdef CONFIG_64BIT
53 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
54 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
55 EXPORT_SYMBOL(pgtable_l4_enabled);
56 EXPORT_SYMBOL(pgtable_l5_enabled);
57 #endif
58 
59 phys_addr_t phys_ram_base __ro_after_init;
60 EXPORT_SYMBOL(phys_ram_base);
61 
62 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
63 							__page_aligned_bss;
64 EXPORT_SYMBOL(empty_zero_page);
65 
66 extern char _start[];
67 void *_dtb_early_va __initdata;
68 uintptr_t _dtb_early_pa __initdata;
69 
70 phys_addr_t dma32_phys_limit __initdata;
71 
72 static void __init zone_sizes_init(void)
73 {
74 	unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
75 
76 #ifdef CONFIG_ZONE_DMA32
77 	max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
78 #endif
79 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
80 
81 	free_area_init(max_zone_pfns);
82 }
83 
84 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
85 
86 #define LOG2_SZ_1K  ilog2(SZ_1K)
87 #define LOG2_SZ_1M  ilog2(SZ_1M)
88 #define LOG2_SZ_1G  ilog2(SZ_1G)
89 #define LOG2_SZ_1T  ilog2(SZ_1T)
90 
91 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
92 {
93 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld kB)\n", name, b, t,
94 		  (((t) - (b)) >> LOG2_SZ_1K));
95 }
96 
97 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
98 {
99 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld MB)\n", name, b, t,
100 		  (((t) - (b)) >> LOG2_SZ_1M));
101 }
102 
103 static inline void print_mlg(char *name, unsigned long b, unsigned long t)
104 {
105 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld GB)\n", name, b, t,
106 		   (((t) - (b)) >> LOG2_SZ_1G));
107 }
108 
109 #ifdef CONFIG_64BIT
110 static inline void print_mlt(char *name, unsigned long b, unsigned long t)
111 {
112 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld TB)\n", name, b, t,
113 		   (((t) - (b)) >> LOG2_SZ_1T));
114 }
115 #else
116 #define print_mlt(n, b, t) do {} while (0)
117 #endif
118 
119 static inline void print_ml(char *name, unsigned long b, unsigned long t)
120 {
121 	unsigned long diff = t - b;
122 
123 	if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10)
124 		print_mlt(name, b, t);
125 	else if ((diff >> LOG2_SZ_1G) >= 10)
126 		print_mlg(name, b, t);
127 	else if ((diff >> LOG2_SZ_1M) >= 10)
128 		print_mlm(name, b, t);
129 	else
130 		print_mlk(name, b, t);
131 }
132 
133 static void __init print_vm_layout(void)
134 {
135 	pr_notice("Virtual kernel memory layout:\n");
136 	print_ml("fixmap", (unsigned long)FIXADDR_START,
137 		(unsigned long)FIXADDR_TOP);
138 	print_ml("pci io", (unsigned long)PCI_IO_START,
139 		(unsigned long)PCI_IO_END);
140 	print_ml("vmemmap", (unsigned long)VMEMMAP_START,
141 		(unsigned long)VMEMMAP_END);
142 	print_ml("vmalloc", (unsigned long)VMALLOC_START,
143 		(unsigned long)VMALLOC_END);
144 #ifdef CONFIG_64BIT
145 	print_ml("modules", (unsigned long)MODULES_VADDR,
146 		(unsigned long)MODULES_END);
147 #endif
148 	print_ml("lowmem", (unsigned long)PAGE_OFFSET,
149 		(unsigned long)high_memory);
150 	if (IS_ENABLED(CONFIG_64BIT)) {
151 #ifdef CONFIG_KASAN
152 		print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
153 #endif
154 
155 		print_ml("kernel", (unsigned long)kernel_map.virt_addr,
156 			 (unsigned long)ADDRESS_SPACE_END);
157 	}
158 }
159 #else
160 static void print_vm_layout(void) { }
161 #endif /* CONFIG_DEBUG_VM */
162 
163 void __init mem_init(void)
164 {
165 #ifdef CONFIG_FLATMEM
166 	BUG_ON(!mem_map);
167 #endif /* CONFIG_FLATMEM */
168 
169 	swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE);
170 	memblock_free_all();
171 
172 	print_vm_layout();
173 }
174 
175 /* Limit the memory size via mem. */
176 static phys_addr_t memory_limit;
177 
178 static int __init early_mem(char *p)
179 {
180 	u64 size;
181 
182 	if (!p)
183 		return 1;
184 
185 	size = memparse(p, &p) & PAGE_MASK;
186 	memory_limit = min_t(u64, size, memory_limit);
187 
188 	pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
189 
190 	return 0;
191 }
192 early_param("mem", early_mem);
193 
194 static void __init setup_bootmem(void)
195 {
196 	phys_addr_t vmlinux_end = __pa_symbol(&_end);
197 	phys_addr_t max_mapped_addr;
198 	phys_addr_t phys_ram_end, vmlinux_start;
199 
200 	if (IS_ENABLED(CONFIG_XIP_KERNEL))
201 		vmlinux_start = __pa_symbol(&_sdata);
202 	else
203 		vmlinux_start = __pa_symbol(&_start);
204 
205 	memblock_enforce_memory_limit(memory_limit);
206 
207 	/*
208 	 * Make sure we align the reservation on PMD_SIZE since we will
209 	 * map the kernel in the linear mapping as read-only: we do not want
210 	 * any allocation to happen between _end and the next pmd aligned page.
211 	 */
212 	if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
213 		vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
214 	/*
215 	 * Reserve from the start of the kernel to the end of the kernel
216 	 */
217 	memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
218 
219 	phys_ram_end = memblock_end_of_DRAM();
220 
221 	/*
222 	 * Make sure we align the start of the memory on a PMD boundary so that
223 	 * at worst, we map the linear mapping with PMD mappings.
224 	 */
225 	if (!IS_ENABLED(CONFIG_XIP_KERNEL))
226 		phys_ram_base = memblock_start_of_DRAM() & PMD_MASK;
227 
228 	/*
229 	 * In 64-bit, any use of __va/__pa before this point is wrong as we
230 	 * did not know the start of DRAM before.
231 	 */
232 	if (IS_ENABLED(CONFIG_64BIT))
233 		kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base;
234 
235 	/*
236 	 * memblock allocator is not aware of the fact that last 4K bytes of
237 	 * the addressable memory can not be mapped because of IS_ERR_VALUE
238 	 * macro. Make sure that last 4k bytes are not usable by memblock
239 	 * if end of dram is equal to maximum addressable memory.  For 64-bit
240 	 * kernel, this problem can't happen here as the end of the virtual
241 	 * address space is occupied by the kernel mapping then this check must
242 	 * be done as soon as the kernel mapping base address is determined.
243 	 */
244 	if (!IS_ENABLED(CONFIG_64BIT)) {
245 		max_mapped_addr = __pa(~(ulong)0);
246 		if (max_mapped_addr == (phys_ram_end - 1))
247 			memblock_set_current_limit(max_mapped_addr - 4096);
248 	}
249 
250 	min_low_pfn = PFN_UP(phys_ram_base);
251 	max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
252 	high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
253 
254 	dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
255 	set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
256 
257 	reserve_initrd_mem();
258 
259 	/*
260 	 * No allocation should be done before reserving the memory as defined
261 	 * in the device tree, otherwise the allocation could end up in a
262 	 * reserved region.
263 	 */
264 	early_init_fdt_scan_reserved_mem();
265 
266 	/*
267 	 * If DTB is built in, no need to reserve its memblock.
268 	 * Otherwise, do reserve it but avoid using
269 	 * early_init_fdt_reserve_self() since __pa() does
270 	 * not work for DTB pointers that are fixmap addresses
271 	 */
272 	if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
273 		memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
274 
275 	dma_contiguous_reserve(dma32_phys_limit);
276 	if (IS_ENABLED(CONFIG_64BIT))
277 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
278 }
279 
280 #ifdef CONFIG_MMU
281 struct pt_alloc_ops pt_ops __initdata;
282 
283 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
284 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
285 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
286 
287 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
288 
289 #ifdef CONFIG_XIP_KERNEL
290 #define pt_ops			(*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
291 #define trampoline_pg_dir      ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
292 #define fixmap_pte             ((pte_t *)XIP_FIXUP(fixmap_pte))
293 #define early_pg_dir           ((pgd_t *)XIP_FIXUP(early_pg_dir))
294 #endif /* CONFIG_XIP_KERNEL */
295 
296 static const pgprot_t protection_map[16] = {
297 	[VM_NONE]					= PAGE_NONE,
298 	[VM_READ]					= PAGE_READ,
299 	[VM_WRITE]					= PAGE_COPY,
300 	[VM_WRITE | VM_READ]				= PAGE_COPY,
301 	[VM_EXEC]					= PAGE_EXEC,
302 	[VM_EXEC | VM_READ]				= PAGE_READ_EXEC,
303 	[VM_EXEC | VM_WRITE]				= PAGE_COPY_EXEC,
304 	[VM_EXEC | VM_WRITE | VM_READ]			= PAGE_COPY_EXEC,
305 	[VM_SHARED]					= PAGE_NONE,
306 	[VM_SHARED | VM_READ]				= PAGE_READ,
307 	[VM_SHARED | VM_WRITE]				= PAGE_SHARED,
308 	[VM_SHARED | VM_WRITE | VM_READ]		= PAGE_SHARED,
309 	[VM_SHARED | VM_EXEC]				= PAGE_EXEC,
310 	[VM_SHARED | VM_EXEC | VM_READ]			= PAGE_READ_EXEC,
311 	[VM_SHARED | VM_EXEC | VM_WRITE]		= PAGE_SHARED_EXEC,
312 	[VM_SHARED | VM_EXEC | VM_WRITE | VM_READ]	= PAGE_SHARED_EXEC
313 };
314 DECLARE_VM_GET_PAGE_PROT
315 
316 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
317 {
318 	unsigned long addr = __fix_to_virt(idx);
319 	pte_t *ptep;
320 
321 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
322 
323 	ptep = &fixmap_pte[pte_index(addr)];
324 
325 	if (pgprot_val(prot))
326 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
327 	else
328 		pte_clear(&init_mm, addr, ptep);
329 	local_flush_tlb_page(addr);
330 }
331 
332 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
333 {
334 	return (pte_t *)((uintptr_t)pa);
335 }
336 
337 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
338 {
339 	clear_fixmap(FIX_PTE);
340 	return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
341 }
342 
343 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
344 {
345 	return (pte_t *) __va(pa);
346 }
347 
348 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
349 {
350 	/*
351 	 * We only create PMD or PGD early mappings so we
352 	 * should never reach here with MMU disabled.
353 	 */
354 	BUG();
355 }
356 
357 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
358 {
359 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
360 }
361 
362 static phys_addr_t __init alloc_pte_late(uintptr_t va)
363 {
364 	struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
365 
366 	BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc));
367 	return __pa((pte_t *)ptdesc_address(ptdesc));
368 }
369 
370 static void __init create_pte_mapping(pte_t *ptep,
371 				      uintptr_t va, phys_addr_t pa,
372 				      phys_addr_t sz, pgprot_t prot)
373 {
374 	uintptr_t pte_idx = pte_index(va);
375 
376 	BUG_ON(sz != PAGE_SIZE);
377 
378 	if (pte_none(ptep[pte_idx]))
379 		ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
380 }
381 
382 #ifndef __PAGETABLE_PMD_FOLDED
383 
384 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
385 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
386 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
387 
388 #ifdef CONFIG_XIP_KERNEL
389 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
390 #define fixmap_pmd     ((pmd_t *)XIP_FIXUP(fixmap_pmd))
391 #define early_pmd      ((pmd_t *)XIP_FIXUP(early_pmd))
392 #endif /* CONFIG_XIP_KERNEL */
393 
394 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
395 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
396 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
397 
398 #ifdef CONFIG_XIP_KERNEL
399 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
400 #define fixmap_p4d     ((p4d_t *)XIP_FIXUP(fixmap_p4d))
401 #define early_p4d      ((p4d_t *)XIP_FIXUP(early_p4d))
402 #endif /* CONFIG_XIP_KERNEL */
403 
404 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
405 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
406 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
407 
408 #ifdef CONFIG_XIP_KERNEL
409 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
410 #define fixmap_pud     ((pud_t *)XIP_FIXUP(fixmap_pud))
411 #define early_pud      ((pud_t *)XIP_FIXUP(early_pud))
412 #endif /* CONFIG_XIP_KERNEL */
413 
414 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
415 {
416 	/* Before MMU is enabled */
417 	return (pmd_t *)((uintptr_t)pa);
418 }
419 
420 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
421 {
422 	clear_fixmap(FIX_PMD);
423 	return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
424 }
425 
426 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
427 {
428 	return (pmd_t *) __va(pa);
429 }
430 
431 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
432 {
433 	BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
434 
435 	return (uintptr_t)early_pmd;
436 }
437 
438 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
439 {
440 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
441 }
442 
443 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
444 {
445 	struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
446 
447 	BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc));
448 	return __pa((pmd_t *)ptdesc_address(ptdesc));
449 }
450 
451 static void __init create_pmd_mapping(pmd_t *pmdp,
452 				      uintptr_t va, phys_addr_t pa,
453 				      phys_addr_t sz, pgprot_t prot)
454 {
455 	pte_t *ptep;
456 	phys_addr_t pte_phys;
457 	uintptr_t pmd_idx = pmd_index(va);
458 
459 	if (sz == PMD_SIZE) {
460 		if (pmd_none(pmdp[pmd_idx]))
461 			pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
462 		return;
463 	}
464 
465 	if (pmd_none(pmdp[pmd_idx])) {
466 		pte_phys = pt_ops.alloc_pte(va);
467 		pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
468 		ptep = pt_ops.get_pte_virt(pte_phys);
469 		memset(ptep, 0, PAGE_SIZE);
470 	} else {
471 		pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
472 		ptep = pt_ops.get_pte_virt(pte_phys);
473 	}
474 
475 	create_pte_mapping(ptep, va, pa, sz, prot);
476 }
477 
478 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
479 {
480 	return (pud_t *)((uintptr_t)pa);
481 }
482 
483 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
484 {
485 	clear_fixmap(FIX_PUD);
486 	return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
487 }
488 
489 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
490 {
491 	return (pud_t *)__va(pa);
492 }
493 
494 static phys_addr_t __init alloc_pud_early(uintptr_t va)
495 {
496 	/* Only one PUD is available for early mapping */
497 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
498 
499 	return (uintptr_t)early_pud;
500 }
501 
502 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
503 {
504 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
505 }
506 
507 static phys_addr_t alloc_pud_late(uintptr_t va)
508 {
509 	unsigned long vaddr;
510 
511 	vaddr = __get_free_page(GFP_KERNEL);
512 	BUG_ON(!vaddr);
513 	return __pa(vaddr);
514 }
515 
516 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
517 {
518 	return (p4d_t *)((uintptr_t)pa);
519 }
520 
521 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
522 {
523 	clear_fixmap(FIX_P4D);
524 	return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
525 }
526 
527 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
528 {
529 	return (p4d_t *)__va(pa);
530 }
531 
532 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
533 {
534 	/* Only one P4D is available for early mapping */
535 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
536 
537 	return (uintptr_t)early_p4d;
538 }
539 
540 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
541 {
542 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
543 }
544 
545 static phys_addr_t alloc_p4d_late(uintptr_t va)
546 {
547 	unsigned long vaddr;
548 
549 	vaddr = __get_free_page(GFP_KERNEL);
550 	BUG_ON(!vaddr);
551 	return __pa(vaddr);
552 }
553 
554 static void __init create_pud_mapping(pud_t *pudp,
555 				      uintptr_t va, phys_addr_t pa,
556 				      phys_addr_t sz, pgprot_t prot)
557 {
558 	pmd_t *nextp;
559 	phys_addr_t next_phys;
560 	uintptr_t pud_index = pud_index(va);
561 
562 	if (sz == PUD_SIZE) {
563 		if (pud_val(pudp[pud_index]) == 0)
564 			pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
565 		return;
566 	}
567 
568 	if (pud_val(pudp[pud_index]) == 0) {
569 		next_phys = pt_ops.alloc_pmd(va);
570 		pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
571 		nextp = pt_ops.get_pmd_virt(next_phys);
572 		memset(nextp, 0, PAGE_SIZE);
573 	} else {
574 		next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
575 		nextp = pt_ops.get_pmd_virt(next_phys);
576 	}
577 
578 	create_pmd_mapping(nextp, va, pa, sz, prot);
579 }
580 
581 static void __init create_p4d_mapping(p4d_t *p4dp,
582 				      uintptr_t va, phys_addr_t pa,
583 				      phys_addr_t sz, pgprot_t prot)
584 {
585 	pud_t *nextp;
586 	phys_addr_t next_phys;
587 	uintptr_t p4d_index = p4d_index(va);
588 
589 	if (sz == P4D_SIZE) {
590 		if (p4d_val(p4dp[p4d_index]) == 0)
591 			p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
592 		return;
593 	}
594 
595 	if (p4d_val(p4dp[p4d_index]) == 0) {
596 		next_phys = pt_ops.alloc_pud(va);
597 		p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
598 		nextp = pt_ops.get_pud_virt(next_phys);
599 		memset(nextp, 0, PAGE_SIZE);
600 	} else {
601 		next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
602 		nextp = pt_ops.get_pud_virt(next_phys);
603 	}
604 
605 	create_pud_mapping(nextp, va, pa, sz, prot);
606 }
607 
608 #define pgd_next_t		p4d_t
609 #define alloc_pgd_next(__va)	(pgtable_l5_enabled ?			\
610 		pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ?		\
611 		pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
612 #define get_pgd_next_virt(__pa)	(pgtable_l5_enabled ?			\
613 		pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ?	\
614 		pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
615 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
616 				(pgtable_l5_enabled ?			\
617 		create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
618 				(pgtable_l4_enabled ?			\
619 		create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) :	\
620 		create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
621 #define fixmap_pgd_next		(pgtable_l5_enabled ?			\
622 		(uintptr_t)fixmap_p4d : (pgtable_l4_enabled ?		\
623 		(uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
624 #define trampoline_pgd_next	(pgtable_l5_enabled ?			\
625 		(uintptr_t)trampoline_p4d : (pgtable_l4_enabled ?	\
626 		(uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
627 #else
628 #define pgd_next_t		pte_t
629 #define alloc_pgd_next(__va)	pt_ops.alloc_pte(__va)
630 #define get_pgd_next_virt(__pa)	pt_ops.get_pte_virt(__pa)
631 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
632 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
633 #define fixmap_pgd_next		((uintptr_t)fixmap_pte)
634 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
635 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
636 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
637 #endif /* __PAGETABLE_PMD_FOLDED */
638 
639 void __init create_pgd_mapping(pgd_t *pgdp,
640 				      uintptr_t va, phys_addr_t pa,
641 				      phys_addr_t sz, pgprot_t prot)
642 {
643 	pgd_next_t *nextp;
644 	phys_addr_t next_phys;
645 	uintptr_t pgd_idx = pgd_index(va);
646 
647 	if (sz == PGDIR_SIZE) {
648 		if (pgd_val(pgdp[pgd_idx]) == 0)
649 			pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
650 		return;
651 	}
652 
653 	if (pgd_val(pgdp[pgd_idx]) == 0) {
654 		next_phys = alloc_pgd_next(va);
655 		pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
656 		nextp = get_pgd_next_virt(next_phys);
657 		memset(nextp, 0, PAGE_SIZE);
658 	} else {
659 		next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
660 		nextp = get_pgd_next_virt(next_phys);
661 	}
662 
663 	create_pgd_next_mapping(nextp, va, pa, sz, prot);
664 }
665 
666 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va,
667 				      phys_addr_t size)
668 {
669 	if (pgtable_l5_enabled &&
670 	    !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
671 		return P4D_SIZE;
672 
673 	if (pgtable_l4_enabled &&
674 	    !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
675 		return PUD_SIZE;
676 
677 	if (IS_ENABLED(CONFIG_64BIT) &&
678 	    !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
679 		return PMD_SIZE;
680 
681 	return PAGE_SIZE;
682 }
683 
684 #ifdef CONFIG_XIP_KERNEL
685 #define phys_ram_base  (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
686 extern char _xiprom[], _exiprom[], __data_loc;
687 
688 /* called from head.S with MMU off */
689 asmlinkage void __init __copy_data(void)
690 {
691 	void *from = (void *)(&__data_loc);
692 	void *to = (void *)CONFIG_PHYS_RAM_BASE;
693 	size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
694 
695 	memcpy(to, from, sz);
696 }
697 #endif
698 
699 #ifdef CONFIG_STRICT_KERNEL_RWX
700 static __init pgprot_t pgprot_from_va(uintptr_t va)
701 {
702 	if (is_va_kernel_text(va))
703 		return PAGE_KERNEL_READ_EXEC;
704 
705 	/*
706 	 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
707 	 * we must protect its linear mapping alias from being executed and
708 	 * written.
709 	 * And rodata section is marked readonly in mark_rodata_ro.
710 	 */
711 	if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
712 		return PAGE_KERNEL_READ;
713 
714 	return PAGE_KERNEL;
715 }
716 
717 void mark_rodata_ro(void)
718 {
719 	set_kernel_memory(__start_rodata, _data, set_memory_ro);
720 	if (IS_ENABLED(CONFIG_64BIT))
721 		set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
722 				  set_memory_ro);
723 
724 	debug_checkwx();
725 }
726 #else
727 static __init pgprot_t pgprot_from_va(uintptr_t va)
728 {
729 	if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
730 		return PAGE_KERNEL;
731 
732 	return PAGE_KERNEL_EXEC;
733 }
734 #endif /* CONFIG_STRICT_KERNEL_RWX */
735 
736 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
737 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
738 
739 static void __init disable_pgtable_l5(void)
740 {
741 	pgtable_l5_enabled = false;
742 	kernel_map.page_offset = PAGE_OFFSET_L4;
743 	satp_mode = SATP_MODE_48;
744 }
745 
746 static void __init disable_pgtable_l4(void)
747 {
748 	pgtable_l4_enabled = false;
749 	kernel_map.page_offset = PAGE_OFFSET_L3;
750 	satp_mode = SATP_MODE_39;
751 }
752 
753 static int __init print_no4lvl(char *p)
754 {
755 	pr_info("Disabled 4-level and 5-level paging");
756 	return 0;
757 }
758 early_param("no4lvl", print_no4lvl);
759 
760 static int __init print_no5lvl(char *p)
761 {
762 	pr_info("Disabled 5-level paging");
763 	return 0;
764 }
765 early_param("no5lvl", print_no5lvl);
766 
767 /*
768  * There is a simple way to determine if 4-level is supported by the
769  * underlying hardware: establish 1:1 mapping in 4-level page table mode
770  * then read SATP to see if the configuration was taken into account
771  * meaning sv48 is supported.
772  */
773 static __init void set_satp_mode(uintptr_t dtb_pa)
774 {
775 	u64 identity_satp, hw_satp;
776 	uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
777 	u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
778 
779 	if (satp_mode_cmdline == SATP_MODE_57) {
780 		disable_pgtable_l5();
781 	} else if (satp_mode_cmdline == SATP_MODE_48) {
782 		disable_pgtable_l5();
783 		disable_pgtable_l4();
784 		return;
785 	}
786 
787 	create_p4d_mapping(early_p4d,
788 			set_satp_mode_pmd, (uintptr_t)early_pud,
789 			P4D_SIZE, PAGE_TABLE);
790 	create_pud_mapping(early_pud,
791 			   set_satp_mode_pmd, (uintptr_t)early_pmd,
792 			   PUD_SIZE, PAGE_TABLE);
793 	/* Handle the case where set_satp_mode straddles 2 PMDs */
794 	create_pmd_mapping(early_pmd,
795 			   set_satp_mode_pmd, set_satp_mode_pmd,
796 			   PMD_SIZE, PAGE_KERNEL_EXEC);
797 	create_pmd_mapping(early_pmd,
798 			   set_satp_mode_pmd + PMD_SIZE,
799 			   set_satp_mode_pmd + PMD_SIZE,
800 			   PMD_SIZE, PAGE_KERNEL_EXEC);
801 retry:
802 	create_pgd_mapping(early_pg_dir,
803 			   set_satp_mode_pmd,
804 			   pgtable_l5_enabled ?
805 				(uintptr_t)early_p4d : (uintptr_t)early_pud,
806 			   PGDIR_SIZE, PAGE_TABLE);
807 
808 	identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
809 
810 	local_flush_tlb_all();
811 	csr_write(CSR_SATP, identity_satp);
812 	hw_satp = csr_swap(CSR_SATP, 0ULL);
813 	local_flush_tlb_all();
814 
815 	if (hw_satp != identity_satp) {
816 		if (pgtable_l5_enabled) {
817 			disable_pgtable_l5();
818 			memset(early_pg_dir, 0, PAGE_SIZE);
819 			goto retry;
820 		}
821 		disable_pgtable_l4();
822 	}
823 
824 	memset(early_pg_dir, 0, PAGE_SIZE);
825 	memset(early_p4d, 0, PAGE_SIZE);
826 	memset(early_pud, 0, PAGE_SIZE);
827 	memset(early_pmd, 0, PAGE_SIZE);
828 }
829 #endif
830 
831 /*
832  * setup_vm() is called from head.S with MMU-off.
833  *
834  * Following requirements should be honoured for setup_vm() to work
835  * correctly:
836  * 1) It should use PC-relative addressing for accessing kernel symbols.
837  *    To achieve this we always use GCC cmodel=medany.
838  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
839  *    so disable compiler instrumentation when FTRACE is enabled.
840  *
841  * Currently, the above requirements are honoured by using custom CFLAGS
842  * for init.o in mm/Makefile.
843  */
844 
845 #ifndef __riscv_cmodel_medany
846 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
847 #endif
848 
849 #ifdef CONFIG_RELOCATABLE
850 extern unsigned long __rela_dyn_start, __rela_dyn_end;
851 
852 static void __init relocate_kernel(void)
853 {
854 	Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
855 	/*
856 	 * This holds the offset between the linked virtual address and the
857 	 * relocated virtual address.
858 	 */
859 	uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
860 	/*
861 	 * This holds the offset between kernel linked virtual address and
862 	 * physical address.
863 	 */
864 	uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
865 
866 	for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
867 		Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
868 		Elf64_Addr relocated_addr = rela->r_addend;
869 
870 		if (rela->r_info != R_RISCV_RELATIVE)
871 			continue;
872 
873 		/*
874 		 * Make sure to not relocate vdso symbols like rt_sigreturn
875 		 * which are linked from the address 0 in vmlinux since
876 		 * vdso symbol addresses are actually used as an offset from
877 		 * mm->context.vdso in VDSO_OFFSET macro.
878 		 */
879 		if (relocated_addr >= KERNEL_LINK_ADDR)
880 			relocated_addr += reloc_offset;
881 
882 		*(Elf64_Addr *)addr = relocated_addr;
883 	}
884 }
885 #endif /* CONFIG_RELOCATABLE */
886 
887 #ifdef CONFIG_XIP_KERNEL
888 static void __init create_kernel_page_table(pgd_t *pgdir,
889 					    __always_unused bool early)
890 {
891 	uintptr_t va, end_va;
892 
893 	/* Map the flash resident part */
894 	end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
895 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
896 		create_pgd_mapping(pgdir, va,
897 				   kernel_map.xiprom + (va - kernel_map.virt_addr),
898 				   PMD_SIZE, PAGE_KERNEL_EXEC);
899 
900 	/* Map the data in RAM */
901 	end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
902 	for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
903 		create_pgd_mapping(pgdir, va,
904 				   kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
905 				   PMD_SIZE, PAGE_KERNEL);
906 }
907 #else
908 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
909 {
910 	uintptr_t va, end_va;
911 
912 	end_va = kernel_map.virt_addr + kernel_map.size;
913 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
914 		create_pgd_mapping(pgdir, va,
915 				   kernel_map.phys_addr + (va - kernel_map.virt_addr),
916 				   PMD_SIZE,
917 				   early ?
918 					PAGE_KERNEL_EXEC : pgprot_from_va(va));
919 }
920 #endif
921 
922 /*
923  * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
924  * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
925  * entry.
926  */
927 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
928 					       uintptr_t dtb_pa)
929 {
930 #ifndef CONFIG_BUILTIN_DTB
931 	uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
932 
933 	/* Make sure the fdt fixmap address is always aligned on PMD size */
934 	BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
935 
936 	/* In 32-bit only, the fdt lies in its own PGD */
937 	if (!IS_ENABLED(CONFIG_64BIT)) {
938 		create_pgd_mapping(early_pg_dir, fix_fdt_va,
939 				   pa, MAX_FDT_SIZE, PAGE_KERNEL);
940 	} else {
941 		create_pmd_mapping(fixmap_pmd, fix_fdt_va,
942 				   pa, PMD_SIZE, PAGE_KERNEL);
943 		create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
944 				   pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
945 	}
946 
947 	dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
948 #else
949 	/*
950 	 * For 64-bit kernel, __va can't be used since it would return a linear
951 	 * mapping address whereas dtb_early_va will be used before
952 	 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
953 	 * kernel is mapped in the linear mapping, that makes no difference.
954 	 */
955 	dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
956 #endif
957 
958 	dtb_early_pa = dtb_pa;
959 }
960 
961 /*
962  * MMU is not enabled, the page tables are allocated directly using
963  * early_pmd/pud/p4d and the address returned is the physical one.
964  */
965 static void __init pt_ops_set_early(void)
966 {
967 	pt_ops.alloc_pte = alloc_pte_early;
968 	pt_ops.get_pte_virt = get_pte_virt_early;
969 #ifndef __PAGETABLE_PMD_FOLDED
970 	pt_ops.alloc_pmd = alloc_pmd_early;
971 	pt_ops.get_pmd_virt = get_pmd_virt_early;
972 	pt_ops.alloc_pud = alloc_pud_early;
973 	pt_ops.get_pud_virt = get_pud_virt_early;
974 	pt_ops.alloc_p4d = alloc_p4d_early;
975 	pt_ops.get_p4d_virt = get_p4d_virt_early;
976 #endif
977 }
978 
979 /*
980  * MMU is enabled but page table setup is not complete yet.
981  * fixmap page table alloc functions must be used as a means to temporarily
982  * map the allocated physical pages since the linear mapping does not exist yet.
983  *
984  * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
985  * but it will be used as described above.
986  */
987 static void __init pt_ops_set_fixmap(void)
988 {
989 	pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
990 	pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
991 #ifndef __PAGETABLE_PMD_FOLDED
992 	pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
993 	pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
994 	pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
995 	pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
996 	pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
997 	pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
998 #endif
999 }
1000 
1001 /*
1002  * MMU is enabled and page table setup is complete, so from now, we can use
1003  * generic page allocation functions to setup page table.
1004  */
1005 static void __init pt_ops_set_late(void)
1006 {
1007 	pt_ops.alloc_pte = alloc_pte_late;
1008 	pt_ops.get_pte_virt = get_pte_virt_late;
1009 #ifndef __PAGETABLE_PMD_FOLDED
1010 	pt_ops.alloc_pmd = alloc_pmd_late;
1011 	pt_ops.get_pmd_virt = get_pmd_virt_late;
1012 	pt_ops.alloc_pud = alloc_pud_late;
1013 	pt_ops.get_pud_virt = get_pud_virt_late;
1014 	pt_ops.alloc_p4d = alloc_p4d_late;
1015 	pt_ops.get_p4d_virt = get_p4d_virt_late;
1016 #endif
1017 }
1018 
1019 #ifdef CONFIG_RANDOMIZE_BASE
1020 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa);
1021 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa);
1022 
1023 static int __init print_nokaslr(char *p)
1024 {
1025 	pr_info("Disabled KASLR");
1026 	return 0;
1027 }
1028 early_param("nokaslr", print_nokaslr);
1029 
1030 unsigned long kaslr_offset(void)
1031 {
1032 	return kernel_map.virt_offset;
1033 }
1034 #endif
1035 
1036 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1037 {
1038 	pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1039 
1040 #ifdef CONFIG_RANDOMIZE_BASE
1041 	if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) {
1042 		u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa);
1043 		u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
1044 		u32 nr_pos;
1045 
1046 		/*
1047 		 * Compute the number of positions available: we are limited
1048 		 * by the early page table that only has one PUD and we must
1049 		 * be aligned on PMD_SIZE.
1050 		 */
1051 		nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE;
1052 
1053 		kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE;
1054 	}
1055 #endif
1056 
1057 	kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset;
1058 	kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1059 
1060 #ifdef CONFIG_XIP_KERNEL
1061 	kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1062 	kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1063 
1064 	phys_ram_base = CONFIG_PHYS_RAM_BASE;
1065 	kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1066 	kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
1067 
1068 	kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1069 #else
1070 	kernel_map.phys_addr = (uintptr_t)(&_start);
1071 	kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1072 #endif
1073 
1074 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1075 	set_satp_mode(dtb_pa);
1076 #endif
1077 
1078 	/*
1079 	 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1080 	 * where we have the system memory layout: this allows us to align
1081 	 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1082 	 * for the linear mapping. This is only possible because the kernel
1083 	 * mapping lies outside the linear mapping.
1084 	 * In 32-bit however, as the kernel resides in the linear mapping,
1085 	 * setup_vm_final can not change the mapping established here,
1086 	 * otherwise the same kernel addresses would get mapped to different
1087 	 * physical addresses (if the start of dram is different from the
1088 	 * kernel physical address start).
1089 	 */
1090 	kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1091 				0UL : PAGE_OFFSET - kernel_map.phys_addr;
1092 	kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1093 
1094 	/*
1095 	 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
1096 	 * kernel, whereas for 64-bit kernel, the end of the virtual address
1097 	 * space is occupied by the modules/BPF/kernel mappings which reduces
1098 	 * the available size of the linear mapping.
1099 	 */
1100 	memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
1101 
1102 	/* Sanity check alignment and size */
1103 	BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1104 	BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1105 
1106 #ifdef CONFIG_64BIT
1107 	/*
1108 	 * The last 4K bytes of the addressable memory can not be mapped because
1109 	 * of IS_ERR_VALUE macro.
1110 	 */
1111 	BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1112 #endif
1113 
1114 #ifdef CONFIG_RELOCATABLE
1115 	/*
1116 	 * Early page table uses only one PUD, which makes it possible
1117 	 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1118 	 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1119 	 * since a part of the kernel would not get mapped.
1120 	 */
1121 	BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1122 	relocate_kernel();
1123 #endif
1124 
1125 	apply_early_boot_alternatives();
1126 	pt_ops_set_early();
1127 
1128 	/* Setup early PGD for fixmap */
1129 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
1130 			   fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1131 
1132 #ifndef __PAGETABLE_PMD_FOLDED
1133 	/* Setup fixmap P4D and PUD */
1134 	if (pgtable_l5_enabled)
1135 		create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1136 				   (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1137 	/* Setup fixmap PUD and PMD */
1138 	if (pgtable_l4_enabled)
1139 		create_pud_mapping(fixmap_pud, FIXADDR_START,
1140 				   (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1141 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1142 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1143 	/* Setup trampoline PGD and PMD */
1144 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1145 			   trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1146 	if (pgtable_l5_enabled)
1147 		create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1148 				   (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1149 	if (pgtable_l4_enabled)
1150 		create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1151 				   (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1152 #ifdef CONFIG_XIP_KERNEL
1153 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1154 			   kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1155 #else
1156 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1157 			   kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1158 #endif
1159 #else
1160 	/* Setup trampoline PGD */
1161 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1162 			   kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1163 #endif
1164 
1165 	/*
1166 	 * Setup early PGD covering entire kernel which will allow
1167 	 * us to reach paging_init(). We map all memory banks later
1168 	 * in setup_vm_final() below.
1169 	 */
1170 	create_kernel_page_table(early_pg_dir, true);
1171 
1172 	/* Setup early mapping for FDT early scan */
1173 	create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1174 
1175 	/*
1176 	 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1177 	 * range can not span multiple pmds.
1178 	 */
1179 	BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1180 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1181 
1182 #ifndef __PAGETABLE_PMD_FOLDED
1183 	/*
1184 	 * Early ioremap fixmap is already created as it lies within first 2MB
1185 	 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1186 	 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1187 	 * the user if not.
1188 	 */
1189 	fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1190 	fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1191 	if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1192 		WARN_ON(1);
1193 		pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1194 			pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1195 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1196 			fix_to_virt(FIX_BTMAP_BEGIN));
1197 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
1198 			fix_to_virt(FIX_BTMAP_END));
1199 
1200 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
1201 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
1202 	}
1203 #endif
1204 
1205 	pt_ops_set_fixmap();
1206 }
1207 
1208 static void __init create_linear_mapping_range(phys_addr_t start,
1209 					       phys_addr_t end,
1210 					       uintptr_t fixed_map_size)
1211 {
1212 	phys_addr_t pa;
1213 	uintptr_t va, map_size;
1214 
1215 	for (pa = start; pa < end; pa += map_size) {
1216 		va = (uintptr_t)__va(pa);
1217 		map_size = fixed_map_size ? fixed_map_size :
1218 					    best_map_size(pa, va, end - pa);
1219 
1220 		create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1221 				   pgprot_from_va(va));
1222 	}
1223 }
1224 
1225 static void __init create_linear_mapping_page_table(void)
1226 {
1227 	phys_addr_t start, end;
1228 	phys_addr_t kfence_pool __maybe_unused;
1229 	u64 i;
1230 
1231 #ifdef CONFIG_STRICT_KERNEL_RWX
1232 	phys_addr_t ktext_start = __pa_symbol(_start);
1233 	phys_addr_t ktext_size = __init_data_begin - _start;
1234 	phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1235 	phys_addr_t krodata_size = _data - __start_rodata;
1236 
1237 	/* Isolate kernel text and rodata so they don't get mapped with a PUD */
1238 	memblock_mark_nomap(ktext_start,  ktext_size);
1239 	memblock_mark_nomap(krodata_start, krodata_size);
1240 #endif
1241 
1242 #ifdef CONFIG_KFENCE
1243 	/*
1244 	 *  kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1245 	 *  before we setup the linear mapping so that we avoid using hugepages
1246 	 *  for this region.
1247 	 */
1248 	kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1249 	BUG_ON(!kfence_pool);
1250 
1251 	memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1252 	__kfence_pool = __va(kfence_pool);
1253 #endif
1254 
1255 	/* Map all memory banks in the linear mapping */
1256 	for_each_mem_range(i, &start, &end) {
1257 		if (start >= end)
1258 			break;
1259 		if (start <= __pa(PAGE_OFFSET) &&
1260 		    __pa(PAGE_OFFSET) < end)
1261 			start = __pa(PAGE_OFFSET);
1262 		if (end >= __pa(PAGE_OFFSET) + memory_limit)
1263 			end = __pa(PAGE_OFFSET) + memory_limit;
1264 
1265 		create_linear_mapping_range(start, end, 0);
1266 	}
1267 
1268 #ifdef CONFIG_STRICT_KERNEL_RWX
1269 	create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0);
1270 	create_linear_mapping_range(krodata_start,
1271 				    krodata_start + krodata_size, 0);
1272 
1273 	memblock_clear_nomap(ktext_start,  ktext_size);
1274 	memblock_clear_nomap(krodata_start, krodata_size);
1275 #endif
1276 
1277 #ifdef CONFIG_KFENCE
1278 	create_linear_mapping_range(kfence_pool,
1279 				    kfence_pool + KFENCE_POOL_SIZE,
1280 				    PAGE_SIZE);
1281 
1282 	memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1283 #endif
1284 }
1285 
1286 static void __init setup_vm_final(void)
1287 {
1288 	/* Setup swapper PGD for fixmap */
1289 #if !defined(CONFIG_64BIT)
1290 	/*
1291 	 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1292 	 * directly in swapper_pg_dir in addition to the pgd entry that points
1293 	 * to fixmap_pte.
1294 	 */
1295 	unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1296 
1297 	set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1298 #endif
1299 	create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1300 			   __pa_symbol(fixmap_pgd_next),
1301 			   PGDIR_SIZE, PAGE_TABLE);
1302 
1303 	/* Map the linear mapping */
1304 	create_linear_mapping_page_table();
1305 
1306 	/* Map the kernel */
1307 	if (IS_ENABLED(CONFIG_64BIT))
1308 		create_kernel_page_table(swapper_pg_dir, false);
1309 
1310 #ifdef CONFIG_KASAN
1311 	kasan_swapper_init();
1312 #endif
1313 
1314 	/* Clear fixmap PTE and PMD mappings */
1315 	clear_fixmap(FIX_PTE);
1316 	clear_fixmap(FIX_PMD);
1317 	clear_fixmap(FIX_PUD);
1318 	clear_fixmap(FIX_P4D);
1319 
1320 	/* Move to swapper page table */
1321 	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1322 	local_flush_tlb_all();
1323 
1324 	pt_ops_set_late();
1325 }
1326 #else
1327 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1328 {
1329 	dtb_early_va = (void *)dtb_pa;
1330 	dtb_early_pa = dtb_pa;
1331 }
1332 
1333 static inline void setup_vm_final(void)
1334 {
1335 }
1336 #endif /* CONFIG_MMU */
1337 
1338 /*
1339  * reserve_crashkernel() - reserves memory for crash kernel
1340  *
1341  * This function reserves memory area given in "crashkernel=" kernel command
1342  * line parameter. The memory reserved is used by dump capture kernel when
1343  * primary kernel is crashing.
1344  */
1345 static void __init arch_reserve_crashkernel(void)
1346 {
1347 	unsigned long long low_size = 0;
1348 	unsigned long long crash_base, crash_size;
1349 	char *cmdline = boot_command_line;
1350 	bool high = false;
1351 	int ret;
1352 
1353 	if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1354 		return;
1355 
1356 	ret = parse_crashkernel(cmdline, memblock_phys_mem_size(),
1357 				&crash_size, &crash_base,
1358 				&low_size, &high);
1359 	if (ret)
1360 		return;
1361 
1362 	reserve_crashkernel_generic(cmdline, crash_size, crash_base,
1363 				    low_size, high);
1364 }
1365 
1366 void __init paging_init(void)
1367 {
1368 	setup_bootmem();
1369 	setup_vm_final();
1370 
1371 	/* Depend on that Linear Mapping is ready */
1372 	memblock_allow_resize();
1373 }
1374 
1375 void __init misc_mem_init(void)
1376 {
1377 	early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1378 	arch_numa_init();
1379 	sparse_init();
1380 	zone_sizes_init();
1381 	arch_reserve_crashkernel();
1382 	memblock_dump_all();
1383 }
1384 
1385 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1386 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1387 			       struct vmem_altmap *altmap)
1388 {
1389 	return vmemmap_populate_basepages(start, end, node, NULL);
1390 }
1391 #endif
1392 
1393 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1394 /*
1395  * Pre-allocates page-table pages for a specific area in the kernel
1396  * page-table. Only the level which needs to be synchronized between
1397  * all page-tables is allocated because the synchronization can be
1398  * expensive.
1399  */
1400 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1401 					       const char *area)
1402 {
1403 	unsigned long addr;
1404 	const char *lvl;
1405 
1406 	for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1407 		pgd_t *pgd = pgd_offset_k(addr);
1408 		p4d_t *p4d;
1409 		pud_t *pud;
1410 		pmd_t *pmd;
1411 
1412 		lvl = "p4d";
1413 		p4d = p4d_alloc(&init_mm, pgd, addr);
1414 		if (!p4d)
1415 			goto failed;
1416 
1417 		if (pgtable_l5_enabled)
1418 			continue;
1419 
1420 		lvl = "pud";
1421 		pud = pud_alloc(&init_mm, p4d, addr);
1422 		if (!pud)
1423 			goto failed;
1424 
1425 		if (pgtable_l4_enabled)
1426 			continue;
1427 
1428 		lvl = "pmd";
1429 		pmd = pmd_alloc(&init_mm, pud, addr);
1430 		if (!pmd)
1431 			goto failed;
1432 	}
1433 	return;
1434 
1435 failed:
1436 	/*
1437 	 * The pages have to be there now or they will be missing in
1438 	 * process page-tables later.
1439 	 */
1440 	panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1441 }
1442 
1443 void __init pgtable_cache_init(void)
1444 {
1445 	preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1446 	if (IS_ENABLED(CONFIG_MODULES))
1447 		preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
1448 }
1449 #endif
1450