xref: /linux/arch/riscv/mm/init.c (revision 019b383d1132e4051de0d2e43254454b86538cf4)
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 
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 
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 
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 
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
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 
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 
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
158 static void print_vm_layout(void) { }
159 #endif /* CONFIG_DEBUG_VM */
160 
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 
176 static int __init early_mem(char *p)
177 {
178 	u64 size;
179 
180 	if (!p)
181 		return 1;
182 
183 	size = memparse(p, &p) & PAGE_MASK;
184 	memory_limit = min_t(u64, size, memory_limit);
185 
186 	pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
187 
188 	return 0;
189 }
190 early_param("mem", early_mem);
191 
192 static void __init setup_bootmem(void)
193 {
194 	phys_addr_t vmlinux_end = __pa_symbol(&_end);
195 	phys_addr_t max_mapped_addr;
196 	phys_addr_t phys_ram_end, vmlinux_start;
197 
198 	if (IS_ENABLED(CONFIG_XIP_KERNEL))
199 		vmlinux_start = __pa_symbol(&_sdata);
200 	else
201 		vmlinux_start = __pa_symbol(&_start);
202 
203 	memblock_enforce_memory_limit(memory_limit);
204 
205 	/*
206 	 * Make sure we align the reservation on PMD_SIZE since we will
207 	 * map the kernel in the linear mapping as read-only: we do not want
208 	 * any allocation to happen between _end and the next pmd aligned page.
209 	 */
210 	if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
211 		vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
212 	/*
213 	 * Reserve from the start of the kernel to the end of the kernel
214 	 */
215 	memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
216 
217 	phys_ram_end = memblock_end_of_DRAM();
218 
219 	/*
220 	 * Make sure we align the start of the memory on a PMD boundary so that
221 	 * at worst, we map the linear mapping with PMD mappings.
222 	 */
223 	if (!IS_ENABLED(CONFIG_XIP_KERNEL))
224 		phys_ram_base = memblock_start_of_DRAM() & PMD_MASK;
225 
226 	/*
227 	 * In 64-bit, any use of __va/__pa before this point is wrong as we
228 	 * did not know the start of DRAM before.
229 	 */
230 	if (IS_ENABLED(CONFIG_64BIT))
231 		kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base;
232 
233 	/*
234 	 * memblock allocator is not aware of the fact that last 4K bytes of
235 	 * the addressable memory can not be mapped because of IS_ERR_VALUE
236 	 * macro. Make sure that last 4k bytes are not usable by memblock
237 	 * if end of dram is equal to maximum addressable memory.  For 64-bit
238 	 * kernel, this problem can't happen here as the end of the virtual
239 	 * address space is occupied by the kernel mapping then this check must
240 	 * be done as soon as the kernel mapping base address is determined.
241 	 */
242 	if (!IS_ENABLED(CONFIG_64BIT)) {
243 		max_mapped_addr = __pa(~(ulong)0);
244 		if (max_mapped_addr == (phys_ram_end - 1))
245 			memblock_set_current_limit(max_mapped_addr - 4096);
246 	}
247 
248 	min_low_pfn = PFN_UP(phys_ram_base);
249 	max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
250 	high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
251 
252 	dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
253 	set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
254 
255 	reserve_initrd_mem();
256 
257 	/*
258 	 * No allocation should be done before reserving the memory as defined
259 	 * in the device tree, otherwise the allocation could end up in a
260 	 * reserved region.
261 	 */
262 	early_init_fdt_scan_reserved_mem();
263 
264 	/*
265 	 * If DTB is built in, no need to reserve its memblock.
266 	 * Otherwise, do reserve it but avoid using
267 	 * early_init_fdt_reserve_self() since __pa() does
268 	 * not work for DTB pointers that are fixmap addresses
269 	 */
270 	if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
271 		memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
272 
273 	dma_contiguous_reserve(dma32_phys_limit);
274 	if (IS_ENABLED(CONFIG_64BIT))
275 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
276 }
277 
278 #ifdef CONFIG_MMU
279 struct pt_alloc_ops pt_ops __initdata;
280 
281 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
282 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
283 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
284 
285 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
286 
287 #ifdef CONFIG_XIP_KERNEL
288 #define pt_ops			(*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
289 #define trampoline_pg_dir      ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
290 #define fixmap_pte             ((pte_t *)XIP_FIXUP(fixmap_pte))
291 #define early_pg_dir           ((pgd_t *)XIP_FIXUP(early_pg_dir))
292 #endif /* CONFIG_XIP_KERNEL */
293 
294 static const pgprot_t protection_map[16] = {
295 	[VM_NONE]					= PAGE_NONE,
296 	[VM_READ]					= PAGE_READ,
297 	[VM_WRITE]					= PAGE_COPY,
298 	[VM_WRITE | VM_READ]				= PAGE_COPY,
299 	[VM_EXEC]					= PAGE_EXEC,
300 	[VM_EXEC | VM_READ]				= PAGE_READ_EXEC,
301 	[VM_EXEC | VM_WRITE]				= PAGE_COPY_EXEC,
302 	[VM_EXEC | VM_WRITE | VM_READ]			= PAGE_COPY_EXEC,
303 	[VM_SHARED]					= PAGE_NONE,
304 	[VM_SHARED | VM_READ]				= PAGE_READ,
305 	[VM_SHARED | VM_WRITE]				= PAGE_SHARED,
306 	[VM_SHARED | VM_WRITE | VM_READ]		= PAGE_SHARED,
307 	[VM_SHARED | VM_EXEC]				= PAGE_EXEC,
308 	[VM_SHARED | VM_EXEC | VM_READ]			= PAGE_READ_EXEC,
309 	[VM_SHARED | VM_EXEC | VM_WRITE]		= PAGE_SHARED_EXEC,
310 	[VM_SHARED | VM_EXEC | VM_WRITE | VM_READ]	= PAGE_SHARED_EXEC
311 };
312 DECLARE_VM_GET_PAGE_PROT
313 
314 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
315 {
316 	unsigned long addr = __fix_to_virt(idx);
317 	pte_t *ptep;
318 
319 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
320 
321 	ptep = &fixmap_pte[pte_index(addr)];
322 
323 	if (pgprot_val(prot))
324 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
325 	else
326 		pte_clear(&init_mm, addr, ptep);
327 	local_flush_tlb_page(addr);
328 }
329 
330 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
331 {
332 	return (pte_t *)((uintptr_t)pa);
333 }
334 
335 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
336 {
337 	clear_fixmap(FIX_PTE);
338 	return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
339 }
340 
341 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
342 {
343 	return (pte_t *) __va(pa);
344 }
345 
346 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
347 {
348 	/*
349 	 * We only create PMD or PGD early mappings so we
350 	 * should never reach here with MMU disabled.
351 	 */
352 	BUG();
353 }
354 
355 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
356 {
357 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
358 }
359 
360 static phys_addr_t __init alloc_pte_late(uintptr_t va)
361 {
362 	struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
363 
364 	BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc));
365 	return __pa((pte_t *)ptdesc_address(ptdesc));
366 }
367 
368 static void __init create_pte_mapping(pte_t *ptep,
369 				      uintptr_t va, phys_addr_t pa,
370 				      phys_addr_t sz, pgprot_t prot)
371 {
372 	uintptr_t pte_idx = pte_index(va);
373 
374 	BUG_ON(sz != PAGE_SIZE);
375 
376 	if (pte_none(ptep[pte_idx]))
377 		ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
378 }
379 
380 #ifndef __PAGETABLE_PMD_FOLDED
381 
382 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
383 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
384 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
385 
386 #ifdef CONFIG_XIP_KERNEL
387 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
388 #define fixmap_pmd     ((pmd_t *)XIP_FIXUP(fixmap_pmd))
389 #define early_pmd      ((pmd_t *)XIP_FIXUP(early_pmd))
390 #endif /* CONFIG_XIP_KERNEL */
391 
392 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
393 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
394 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
395 
396 #ifdef CONFIG_XIP_KERNEL
397 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
398 #define fixmap_p4d     ((p4d_t *)XIP_FIXUP(fixmap_p4d))
399 #define early_p4d      ((p4d_t *)XIP_FIXUP(early_p4d))
400 #endif /* CONFIG_XIP_KERNEL */
401 
402 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
403 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
404 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
405 
406 #ifdef CONFIG_XIP_KERNEL
407 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
408 #define fixmap_pud     ((pud_t *)XIP_FIXUP(fixmap_pud))
409 #define early_pud      ((pud_t *)XIP_FIXUP(early_pud))
410 #endif /* CONFIG_XIP_KERNEL */
411 
412 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
413 {
414 	/* Before MMU is enabled */
415 	return (pmd_t *)((uintptr_t)pa);
416 }
417 
418 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
419 {
420 	clear_fixmap(FIX_PMD);
421 	return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
422 }
423 
424 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
425 {
426 	return (pmd_t *) __va(pa);
427 }
428 
429 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
430 {
431 	BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
432 
433 	return (uintptr_t)early_pmd;
434 }
435 
436 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
437 {
438 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
439 }
440 
441 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
442 {
443 	struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
444 
445 	BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc));
446 	return __pa((pmd_t *)ptdesc_address(ptdesc));
447 }
448 
449 static void __init create_pmd_mapping(pmd_t *pmdp,
450 				      uintptr_t va, phys_addr_t pa,
451 				      phys_addr_t sz, pgprot_t prot)
452 {
453 	pte_t *ptep;
454 	phys_addr_t pte_phys;
455 	uintptr_t pmd_idx = pmd_index(va);
456 
457 	if (sz == PMD_SIZE) {
458 		if (pmd_none(pmdp[pmd_idx]))
459 			pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
460 		return;
461 	}
462 
463 	if (pmd_none(pmdp[pmd_idx])) {
464 		pte_phys = pt_ops.alloc_pte(va);
465 		pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
466 		ptep = pt_ops.get_pte_virt(pte_phys);
467 		memset(ptep, 0, PAGE_SIZE);
468 	} else {
469 		pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
470 		ptep = pt_ops.get_pte_virt(pte_phys);
471 	}
472 
473 	create_pte_mapping(ptep, va, pa, sz, prot);
474 }
475 
476 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
477 {
478 	return (pud_t *)((uintptr_t)pa);
479 }
480 
481 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
482 {
483 	clear_fixmap(FIX_PUD);
484 	return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
485 }
486 
487 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
488 {
489 	return (pud_t *)__va(pa);
490 }
491 
492 static phys_addr_t __init alloc_pud_early(uintptr_t va)
493 {
494 	/* Only one PUD is available for early mapping */
495 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
496 
497 	return (uintptr_t)early_pud;
498 }
499 
500 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
501 {
502 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
503 }
504 
505 static phys_addr_t alloc_pud_late(uintptr_t va)
506 {
507 	unsigned long vaddr;
508 
509 	vaddr = __get_free_page(GFP_KERNEL);
510 	BUG_ON(!vaddr);
511 	return __pa(vaddr);
512 }
513 
514 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
515 {
516 	return (p4d_t *)((uintptr_t)pa);
517 }
518 
519 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
520 {
521 	clear_fixmap(FIX_P4D);
522 	return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
523 }
524 
525 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
526 {
527 	return (p4d_t *)__va(pa);
528 }
529 
530 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
531 {
532 	/* Only one P4D is available for early mapping */
533 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
534 
535 	return (uintptr_t)early_p4d;
536 }
537 
538 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
539 {
540 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
541 }
542 
543 static phys_addr_t alloc_p4d_late(uintptr_t va)
544 {
545 	unsigned long vaddr;
546 
547 	vaddr = __get_free_page(GFP_KERNEL);
548 	BUG_ON(!vaddr);
549 	return __pa(vaddr);
550 }
551 
552 static void __init create_pud_mapping(pud_t *pudp,
553 				      uintptr_t va, phys_addr_t pa,
554 				      phys_addr_t sz, pgprot_t prot)
555 {
556 	pmd_t *nextp;
557 	phys_addr_t next_phys;
558 	uintptr_t pud_index = pud_index(va);
559 
560 	if (sz == PUD_SIZE) {
561 		if (pud_val(pudp[pud_index]) == 0)
562 			pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
563 		return;
564 	}
565 
566 	if (pud_val(pudp[pud_index]) == 0) {
567 		next_phys = pt_ops.alloc_pmd(va);
568 		pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
569 		nextp = pt_ops.get_pmd_virt(next_phys);
570 		memset(nextp, 0, PAGE_SIZE);
571 	} else {
572 		next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
573 		nextp = pt_ops.get_pmd_virt(next_phys);
574 	}
575 
576 	create_pmd_mapping(nextp, va, pa, sz, prot);
577 }
578 
579 static void __init create_p4d_mapping(p4d_t *p4dp,
580 				      uintptr_t va, phys_addr_t pa,
581 				      phys_addr_t sz, pgprot_t prot)
582 {
583 	pud_t *nextp;
584 	phys_addr_t next_phys;
585 	uintptr_t p4d_index = p4d_index(va);
586 
587 	if (sz == P4D_SIZE) {
588 		if (p4d_val(p4dp[p4d_index]) == 0)
589 			p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
590 		return;
591 	}
592 
593 	if (p4d_val(p4dp[p4d_index]) == 0) {
594 		next_phys = pt_ops.alloc_pud(va);
595 		p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
596 		nextp = pt_ops.get_pud_virt(next_phys);
597 		memset(nextp, 0, PAGE_SIZE);
598 	} else {
599 		next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
600 		nextp = pt_ops.get_pud_virt(next_phys);
601 	}
602 
603 	create_pud_mapping(nextp, va, pa, sz, prot);
604 }
605 
606 #define pgd_next_t		p4d_t
607 #define alloc_pgd_next(__va)	(pgtable_l5_enabled ?			\
608 		pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ?		\
609 		pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
610 #define get_pgd_next_virt(__pa)	(pgtable_l5_enabled ?			\
611 		pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ?	\
612 		pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
613 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
614 				(pgtable_l5_enabled ?			\
615 		create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
616 				(pgtable_l4_enabled ?			\
617 		create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) :	\
618 		create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
619 #define fixmap_pgd_next		(pgtable_l5_enabled ?			\
620 		(uintptr_t)fixmap_p4d : (pgtable_l4_enabled ?		\
621 		(uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
622 #define trampoline_pgd_next	(pgtable_l5_enabled ?			\
623 		(uintptr_t)trampoline_p4d : (pgtable_l4_enabled ?	\
624 		(uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
625 #else
626 #define pgd_next_t		pte_t
627 #define alloc_pgd_next(__va)	pt_ops.alloc_pte(__va)
628 #define get_pgd_next_virt(__pa)	pt_ops.get_pte_virt(__pa)
629 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
630 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
631 #define fixmap_pgd_next		((uintptr_t)fixmap_pte)
632 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
633 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
634 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
635 #endif /* __PAGETABLE_PMD_FOLDED */
636 
637 void __init create_pgd_mapping(pgd_t *pgdp,
638 				      uintptr_t va, phys_addr_t pa,
639 				      phys_addr_t sz, pgprot_t prot)
640 {
641 	pgd_next_t *nextp;
642 	phys_addr_t next_phys;
643 	uintptr_t pgd_idx = pgd_index(va);
644 
645 	if (sz == PGDIR_SIZE) {
646 		if (pgd_val(pgdp[pgd_idx]) == 0)
647 			pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
648 		return;
649 	}
650 
651 	if (pgd_val(pgdp[pgd_idx]) == 0) {
652 		next_phys = alloc_pgd_next(va);
653 		pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
654 		nextp = get_pgd_next_virt(next_phys);
655 		memset(nextp, 0, PAGE_SIZE);
656 	} else {
657 		next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
658 		nextp = get_pgd_next_virt(next_phys);
659 	}
660 
661 	create_pgd_next_mapping(nextp, va, pa, sz, prot);
662 }
663 
664 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va,
665 				      phys_addr_t size)
666 {
667 	if (!(pa & (PGDIR_SIZE - 1)) && !(va & (PGDIR_SIZE - 1)) && size >= PGDIR_SIZE)
668 		return PGDIR_SIZE;
669 
670 	if (!(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
671 		return P4D_SIZE;
672 
673 	if (!(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
674 		return PUD_SIZE;
675 
676 	if (!(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
677 		return PMD_SIZE;
678 
679 	return PAGE_SIZE;
680 }
681 
682 #ifdef CONFIG_XIP_KERNEL
683 #define phys_ram_base  (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
684 extern char _xiprom[], _exiprom[], __data_loc;
685 
686 /* called from head.S with MMU off */
687 asmlinkage void __init __copy_data(void)
688 {
689 	void *from = (void *)(&__data_loc);
690 	void *to = (void *)CONFIG_PHYS_RAM_BASE;
691 	size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
692 
693 	memcpy(to, from, sz);
694 }
695 #endif
696 
697 #ifdef CONFIG_STRICT_KERNEL_RWX
698 static __init pgprot_t pgprot_from_va(uintptr_t va)
699 {
700 	if (is_va_kernel_text(va))
701 		return PAGE_KERNEL_READ_EXEC;
702 
703 	/*
704 	 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
705 	 * we must protect its linear mapping alias from being executed and
706 	 * written.
707 	 * And rodata section is marked readonly in mark_rodata_ro.
708 	 */
709 	if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
710 		return PAGE_KERNEL_READ;
711 
712 	return PAGE_KERNEL;
713 }
714 
715 void mark_rodata_ro(void)
716 {
717 	set_kernel_memory(__start_rodata, _data, set_memory_ro);
718 	if (IS_ENABLED(CONFIG_64BIT))
719 		set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
720 				  set_memory_ro);
721 
722 	debug_checkwx();
723 }
724 #else
725 static __init pgprot_t pgprot_from_va(uintptr_t va)
726 {
727 	if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
728 		return PAGE_KERNEL;
729 
730 	return PAGE_KERNEL_EXEC;
731 }
732 #endif /* CONFIG_STRICT_KERNEL_RWX */
733 
734 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
735 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
736 
737 static void __init disable_pgtable_l5(void)
738 {
739 	pgtable_l5_enabled = false;
740 	kernel_map.page_offset = PAGE_OFFSET_L4;
741 	satp_mode = SATP_MODE_48;
742 }
743 
744 static void __init disable_pgtable_l4(void)
745 {
746 	pgtable_l4_enabled = false;
747 	kernel_map.page_offset = PAGE_OFFSET_L3;
748 	satp_mode = SATP_MODE_39;
749 }
750 
751 static int __init print_no4lvl(char *p)
752 {
753 	pr_info("Disabled 4-level and 5-level paging");
754 	return 0;
755 }
756 early_param("no4lvl", print_no4lvl);
757 
758 static int __init print_no5lvl(char *p)
759 {
760 	pr_info("Disabled 5-level paging");
761 	return 0;
762 }
763 early_param("no5lvl", print_no5lvl);
764 
765 /*
766  * There is a simple way to determine if 4-level is supported by the
767  * underlying hardware: establish 1:1 mapping in 4-level page table mode
768  * then read SATP to see if the configuration was taken into account
769  * meaning sv48 is supported.
770  */
771 static __init void set_satp_mode(uintptr_t dtb_pa)
772 {
773 	u64 identity_satp, hw_satp;
774 	uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
775 	u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
776 
777 	if (satp_mode_cmdline == SATP_MODE_57) {
778 		disable_pgtable_l5();
779 	} else if (satp_mode_cmdline == SATP_MODE_48) {
780 		disable_pgtable_l5();
781 		disable_pgtable_l4();
782 		return;
783 	}
784 
785 	create_p4d_mapping(early_p4d,
786 			set_satp_mode_pmd, (uintptr_t)early_pud,
787 			P4D_SIZE, PAGE_TABLE);
788 	create_pud_mapping(early_pud,
789 			   set_satp_mode_pmd, (uintptr_t)early_pmd,
790 			   PUD_SIZE, PAGE_TABLE);
791 	/* Handle the case where set_satp_mode straddles 2 PMDs */
792 	create_pmd_mapping(early_pmd,
793 			   set_satp_mode_pmd, set_satp_mode_pmd,
794 			   PMD_SIZE, PAGE_KERNEL_EXEC);
795 	create_pmd_mapping(early_pmd,
796 			   set_satp_mode_pmd + PMD_SIZE,
797 			   set_satp_mode_pmd + PMD_SIZE,
798 			   PMD_SIZE, PAGE_KERNEL_EXEC);
799 retry:
800 	create_pgd_mapping(early_pg_dir,
801 			   set_satp_mode_pmd,
802 			   pgtable_l5_enabled ?
803 				(uintptr_t)early_p4d : (uintptr_t)early_pud,
804 			   PGDIR_SIZE, PAGE_TABLE);
805 
806 	identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
807 
808 	local_flush_tlb_all();
809 	csr_write(CSR_SATP, identity_satp);
810 	hw_satp = csr_swap(CSR_SATP, 0ULL);
811 	local_flush_tlb_all();
812 
813 	if (hw_satp != identity_satp) {
814 		if (pgtable_l5_enabled) {
815 			disable_pgtable_l5();
816 			memset(early_pg_dir, 0, PAGE_SIZE);
817 			goto retry;
818 		}
819 		disable_pgtable_l4();
820 	}
821 
822 	memset(early_pg_dir, 0, PAGE_SIZE);
823 	memset(early_p4d, 0, PAGE_SIZE);
824 	memset(early_pud, 0, PAGE_SIZE);
825 	memset(early_pmd, 0, PAGE_SIZE);
826 }
827 #endif
828 
829 /*
830  * setup_vm() is called from head.S with MMU-off.
831  *
832  * Following requirements should be honoured for setup_vm() to work
833  * correctly:
834  * 1) It should use PC-relative addressing for accessing kernel symbols.
835  *    To achieve this we always use GCC cmodel=medany.
836  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
837  *    so disable compiler instrumentation when FTRACE is enabled.
838  *
839  * Currently, the above requirements are honoured by using custom CFLAGS
840  * for init.o in mm/Makefile.
841  */
842 
843 #ifndef __riscv_cmodel_medany
844 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
845 #endif
846 
847 #ifdef CONFIG_RELOCATABLE
848 extern unsigned long __rela_dyn_start, __rela_dyn_end;
849 
850 static void __init relocate_kernel(void)
851 {
852 	Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
853 	/*
854 	 * This holds the offset between the linked virtual address and the
855 	 * relocated virtual address.
856 	 */
857 	uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
858 	/*
859 	 * This holds the offset between kernel linked virtual address and
860 	 * physical address.
861 	 */
862 	uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
863 
864 	for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
865 		Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
866 		Elf64_Addr relocated_addr = rela->r_addend;
867 
868 		if (rela->r_info != R_RISCV_RELATIVE)
869 			continue;
870 
871 		/*
872 		 * Make sure to not relocate vdso symbols like rt_sigreturn
873 		 * which are linked from the address 0 in vmlinux since
874 		 * vdso symbol addresses are actually used as an offset from
875 		 * mm->context.vdso in VDSO_OFFSET macro.
876 		 */
877 		if (relocated_addr >= KERNEL_LINK_ADDR)
878 			relocated_addr += reloc_offset;
879 
880 		*(Elf64_Addr *)addr = relocated_addr;
881 	}
882 }
883 #endif /* CONFIG_RELOCATABLE */
884 
885 #ifdef CONFIG_XIP_KERNEL
886 static void __init create_kernel_page_table(pgd_t *pgdir,
887 					    __always_unused bool early)
888 {
889 	uintptr_t va, end_va;
890 
891 	/* Map the flash resident part */
892 	end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
893 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
894 		create_pgd_mapping(pgdir, va,
895 				   kernel_map.xiprom + (va - kernel_map.virt_addr),
896 				   PMD_SIZE, PAGE_KERNEL_EXEC);
897 
898 	/* Map the data in RAM */
899 	end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
900 	for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
901 		create_pgd_mapping(pgdir, va,
902 				   kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
903 				   PMD_SIZE, PAGE_KERNEL);
904 }
905 #else
906 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
907 {
908 	uintptr_t va, end_va;
909 
910 	end_va = kernel_map.virt_addr + kernel_map.size;
911 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
912 		create_pgd_mapping(pgdir, va,
913 				   kernel_map.phys_addr + (va - kernel_map.virt_addr),
914 				   PMD_SIZE,
915 				   early ?
916 					PAGE_KERNEL_EXEC : pgprot_from_va(va));
917 }
918 #endif
919 
920 /*
921  * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
922  * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
923  * entry.
924  */
925 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
926 					       uintptr_t dtb_pa)
927 {
928 #ifndef CONFIG_BUILTIN_DTB
929 	uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
930 
931 	/* Make sure the fdt fixmap address is always aligned on PMD size */
932 	BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
933 
934 	/* In 32-bit only, the fdt lies in its own PGD */
935 	if (!IS_ENABLED(CONFIG_64BIT)) {
936 		create_pgd_mapping(early_pg_dir, fix_fdt_va,
937 				   pa, MAX_FDT_SIZE, PAGE_KERNEL);
938 	} else {
939 		create_pmd_mapping(fixmap_pmd, fix_fdt_va,
940 				   pa, PMD_SIZE, PAGE_KERNEL);
941 		create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
942 				   pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
943 	}
944 
945 	dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
946 #else
947 	/*
948 	 * For 64-bit kernel, __va can't be used since it would return a linear
949 	 * mapping address whereas dtb_early_va will be used before
950 	 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
951 	 * kernel is mapped in the linear mapping, that makes no difference.
952 	 */
953 	dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
954 #endif
955 
956 	dtb_early_pa = dtb_pa;
957 }
958 
959 /*
960  * MMU is not enabled, the page tables are allocated directly using
961  * early_pmd/pud/p4d and the address returned is the physical one.
962  */
963 static void __init pt_ops_set_early(void)
964 {
965 	pt_ops.alloc_pte = alloc_pte_early;
966 	pt_ops.get_pte_virt = get_pte_virt_early;
967 #ifndef __PAGETABLE_PMD_FOLDED
968 	pt_ops.alloc_pmd = alloc_pmd_early;
969 	pt_ops.get_pmd_virt = get_pmd_virt_early;
970 	pt_ops.alloc_pud = alloc_pud_early;
971 	pt_ops.get_pud_virt = get_pud_virt_early;
972 	pt_ops.alloc_p4d = alloc_p4d_early;
973 	pt_ops.get_p4d_virt = get_p4d_virt_early;
974 #endif
975 }
976 
977 /*
978  * MMU is enabled but page table setup is not complete yet.
979  * fixmap page table alloc functions must be used as a means to temporarily
980  * map the allocated physical pages since the linear mapping does not exist yet.
981  *
982  * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
983  * but it will be used as described above.
984  */
985 static void __init pt_ops_set_fixmap(void)
986 {
987 	pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
988 	pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
989 #ifndef __PAGETABLE_PMD_FOLDED
990 	pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
991 	pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
992 	pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
993 	pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
994 	pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
995 	pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
996 #endif
997 }
998 
999 /*
1000  * MMU is enabled and page table setup is complete, so from now, we can use
1001  * generic page allocation functions to setup page table.
1002  */
1003 static void __init pt_ops_set_late(void)
1004 {
1005 	pt_ops.alloc_pte = alloc_pte_late;
1006 	pt_ops.get_pte_virt = get_pte_virt_late;
1007 #ifndef __PAGETABLE_PMD_FOLDED
1008 	pt_ops.alloc_pmd = alloc_pmd_late;
1009 	pt_ops.get_pmd_virt = get_pmd_virt_late;
1010 	pt_ops.alloc_pud = alloc_pud_late;
1011 	pt_ops.get_pud_virt = get_pud_virt_late;
1012 	pt_ops.alloc_p4d = alloc_p4d_late;
1013 	pt_ops.get_p4d_virt = get_p4d_virt_late;
1014 #endif
1015 }
1016 
1017 #ifdef CONFIG_RANDOMIZE_BASE
1018 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa);
1019 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa);
1020 
1021 static int __init print_nokaslr(char *p)
1022 {
1023 	pr_info("Disabled KASLR");
1024 	return 0;
1025 }
1026 early_param("nokaslr", print_nokaslr);
1027 
1028 unsigned long kaslr_offset(void)
1029 {
1030 	return kernel_map.virt_offset;
1031 }
1032 #endif
1033 
1034 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1035 {
1036 	pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1037 
1038 #ifdef CONFIG_RANDOMIZE_BASE
1039 	if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) {
1040 		u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa);
1041 		u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
1042 		u32 nr_pos;
1043 
1044 		/*
1045 		 * Compute the number of positions available: we are limited
1046 		 * by the early page table that only has one PUD and we must
1047 		 * be aligned on PMD_SIZE.
1048 		 */
1049 		nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE;
1050 
1051 		kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE;
1052 	}
1053 #endif
1054 
1055 	kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset;
1056 	kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1057 
1058 #ifdef CONFIG_XIP_KERNEL
1059 	kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1060 	kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1061 
1062 	phys_ram_base = CONFIG_PHYS_RAM_BASE;
1063 	kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1064 	kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
1065 
1066 	kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1067 #else
1068 	kernel_map.phys_addr = (uintptr_t)(&_start);
1069 	kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1070 #endif
1071 
1072 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1073 	set_satp_mode(dtb_pa);
1074 #endif
1075 
1076 	/*
1077 	 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1078 	 * where we have the system memory layout: this allows us to align
1079 	 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1080 	 * for the linear mapping. This is only possible because the kernel
1081 	 * mapping lies outside the linear mapping.
1082 	 * In 32-bit however, as the kernel resides in the linear mapping,
1083 	 * setup_vm_final can not change the mapping established here,
1084 	 * otherwise the same kernel addresses would get mapped to different
1085 	 * physical addresses (if the start of dram is different from the
1086 	 * kernel physical address start).
1087 	 */
1088 	kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1089 				0UL : PAGE_OFFSET - kernel_map.phys_addr;
1090 	kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1091 
1092 	/*
1093 	 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
1094 	 * kernel, whereas for 64-bit kernel, the end of the virtual address
1095 	 * space is occupied by the modules/BPF/kernel mappings which reduces
1096 	 * the available size of the linear mapping.
1097 	 */
1098 	memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
1099 
1100 	/* Sanity check alignment and size */
1101 	BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1102 	BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1103 
1104 #ifdef CONFIG_64BIT
1105 	/*
1106 	 * The last 4K bytes of the addressable memory can not be mapped because
1107 	 * of IS_ERR_VALUE macro.
1108 	 */
1109 	BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1110 #endif
1111 
1112 #ifdef CONFIG_RELOCATABLE
1113 	/*
1114 	 * Early page table uses only one PUD, which makes it possible
1115 	 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1116 	 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1117 	 * since a part of the kernel would not get mapped.
1118 	 */
1119 	BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1120 	relocate_kernel();
1121 #endif
1122 
1123 	apply_early_boot_alternatives();
1124 	pt_ops_set_early();
1125 
1126 	/* Setup early PGD for fixmap */
1127 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
1128 			   fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1129 
1130 #ifndef __PAGETABLE_PMD_FOLDED
1131 	/* Setup fixmap P4D and PUD */
1132 	if (pgtable_l5_enabled)
1133 		create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1134 				   (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1135 	/* Setup fixmap PUD and PMD */
1136 	if (pgtable_l4_enabled)
1137 		create_pud_mapping(fixmap_pud, FIXADDR_START,
1138 				   (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1139 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1140 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1141 	/* Setup trampoline PGD and PMD */
1142 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1143 			   trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1144 	if (pgtable_l5_enabled)
1145 		create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1146 				   (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1147 	if (pgtable_l4_enabled)
1148 		create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1149 				   (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1150 #ifdef CONFIG_XIP_KERNEL
1151 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1152 			   kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1153 #else
1154 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1155 			   kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1156 #endif
1157 #else
1158 	/* Setup trampoline PGD */
1159 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1160 			   kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1161 #endif
1162 
1163 	/*
1164 	 * Setup early PGD covering entire kernel which will allow
1165 	 * us to reach paging_init(). We map all memory banks later
1166 	 * in setup_vm_final() below.
1167 	 */
1168 	create_kernel_page_table(early_pg_dir, true);
1169 
1170 	/* Setup early mapping for FDT early scan */
1171 	create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1172 
1173 	/*
1174 	 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1175 	 * range can not span multiple pmds.
1176 	 */
1177 	BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1178 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1179 
1180 #ifndef __PAGETABLE_PMD_FOLDED
1181 	/*
1182 	 * Early ioremap fixmap is already created as it lies within first 2MB
1183 	 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1184 	 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1185 	 * the user if not.
1186 	 */
1187 	fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1188 	fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1189 	if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1190 		WARN_ON(1);
1191 		pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1192 			pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1193 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1194 			fix_to_virt(FIX_BTMAP_BEGIN));
1195 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
1196 			fix_to_virt(FIX_BTMAP_END));
1197 
1198 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
1199 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
1200 	}
1201 #endif
1202 
1203 	pt_ops_set_fixmap();
1204 }
1205 
1206 static void __init create_linear_mapping_range(phys_addr_t start,
1207 					       phys_addr_t end,
1208 					       uintptr_t fixed_map_size)
1209 {
1210 	phys_addr_t pa;
1211 	uintptr_t va, map_size;
1212 
1213 	for (pa = start; pa < end; pa += map_size) {
1214 		va = (uintptr_t)__va(pa);
1215 		map_size = fixed_map_size ? fixed_map_size :
1216 					    best_map_size(pa, va, end - pa);
1217 
1218 		create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1219 				   pgprot_from_va(va));
1220 	}
1221 }
1222 
1223 static void __init create_linear_mapping_page_table(void)
1224 {
1225 	phys_addr_t start, end;
1226 	phys_addr_t kfence_pool __maybe_unused;
1227 	u64 i;
1228 
1229 #ifdef CONFIG_STRICT_KERNEL_RWX
1230 	phys_addr_t ktext_start = __pa_symbol(_start);
1231 	phys_addr_t ktext_size = __init_data_begin - _start;
1232 	phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1233 	phys_addr_t krodata_size = _data - __start_rodata;
1234 
1235 	/* Isolate kernel text and rodata so they don't get mapped with a PUD */
1236 	memblock_mark_nomap(ktext_start,  ktext_size);
1237 	memblock_mark_nomap(krodata_start, krodata_size);
1238 #endif
1239 
1240 #ifdef CONFIG_KFENCE
1241 	/*
1242 	 *  kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1243 	 *  before we setup the linear mapping so that we avoid using hugepages
1244 	 *  for this region.
1245 	 */
1246 	kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1247 	BUG_ON(!kfence_pool);
1248 
1249 	memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1250 	__kfence_pool = __va(kfence_pool);
1251 #endif
1252 
1253 	/* Map all memory banks in the linear mapping */
1254 	for_each_mem_range(i, &start, &end) {
1255 		if (start >= end)
1256 			break;
1257 		if (start <= __pa(PAGE_OFFSET) &&
1258 		    __pa(PAGE_OFFSET) < end)
1259 			start = __pa(PAGE_OFFSET);
1260 		if (end >= __pa(PAGE_OFFSET) + memory_limit)
1261 			end = __pa(PAGE_OFFSET) + memory_limit;
1262 
1263 		create_linear_mapping_range(start, end, 0);
1264 	}
1265 
1266 #ifdef CONFIG_STRICT_KERNEL_RWX
1267 	create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0);
1268 	create_linear_mapping_range(krodata_start,
1269 				    krodata_start + krodata_size, 0);
1270 
1271 	memblock_clear_nomap(ktext_start,  ktext_size);
1272 	memblock_clear_nomap(krodata_start, krodata_size);
1273 #endif
1274 
1275 #ifdef CONFIG_KFENCE
1276 	create_linear_mapping_range(kfence_pool,
1277 				    kfence_pool + KFENCE_POOL_SIZE,
1278 				    PAGE_SIZE);
1279 
1280 	memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1281 #endif
1282 }
1283 
1284 static void __init setup_vm_final(void)
1285 {
1286 	/* Setup swapper PGD for fixmap */
1287 #if !defined(CONFIG_64BIT)
1288 	/*
1289 	 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1290 	 * directly in swapper_pg_dir in addition to the pgd entry that points
1291 	 * to fixmap_pte.
1292 	 */
1293 	unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1294 
1295 	set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1296 #endif
1297 	create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1298 			   __pa_symbol(fixmap_pgd_next),
1299 			   PGDIR_SIZE, PAGE_TABLE);
1300 
1301 	/* Map the linear mapping */
1302 	create_linear_mapping_page_table();
1303 
1304 	/* Map the kernel */
1305 	if (IS_ENABLED(CONFIG_64BIT))
1306 		create_kernel_page_table(swapper_pg_dir, false);
1307 
1308 #ifdef CONFIG_KASAN
1309 	kasan_swapper_init();
1310 #endif
1311 
1312 	/* Clear fixmap PTE and PMD mappings */
1313 	clear_fixmap(FIX_PTE);
1314 	clear_fixmap(FIX_PMD);
1315 	clear_fixmap(FIX_PUD);
1316 	clear_fixmap(FIX_P4D);
1317 
1318 	/* Move to swapper page table */
1319 	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1320 	local_flush_tlb_all();
1321 
1322 	pt_ops_set_late();
1323 }
1324 #else
1325 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1326 {
1327 	dtb_early_va = (void *)dtb_pa;
1328 	dtb_early_pa = dtb_pa;
1329 }
1330 
1331 static inline void setup_vm_final(void)
1332 {
1333 }
1334 #endif /* CONFIG_MMU */
1335 
1336 /* Reserve 128M low memory by default for swiotlb buffer */
1337 #define DEFAULT_CRASH_KERNEL_LOW_SIZE	(128UL << 20)
1338 
1339 static int __init reserve_crashkernel_low(unsigned long long low_size)
1340 {
1341 	unsigned long long low_base;
1342 
1343 	low_base = memblock_phys_alloc_range(low_size, PMD_SIZE, 0, dma32_phys_limit);
1344 	if (!low_base) {
1345 		pr_err("cannot allocate crashkernel low memory (size:0x%llx).\n", low_size);
1346 		return -ENOMEM;
1347 	}
1348 
1349 	pr_info("crashkernel low memory reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
1350 		low_base, low_base + low_size, low_size >> 20);
1351 
1352 	crashk_low_res.start = low_base;
1353 	crashk_low_res.end = low_base + low_size - 1;
1354 
1355 	return 0;
1356 }
1357 
1358 /*
1359  * reserve_crashkernel() - reserves memory for crash kernel
1360  *
1361  * This function reserves memory area given in "crashkernel=" kernel command
1362  * line parameter. The memory reserved is used by dump capture kernel when
1363  * primary kernel is crashing.
1364  */
1365 static void __init reserve_crashkernel(void)
1366 {
1367 	unsigned long long crash_base = 0;
1368 	unsigned long long crash_size = 0;
1369 	unsigned long long crash_low_size = 0;
1370 	unsigned long search_start = memblock_start_of_DRAM();
1371 	unsigned long search_end = (unsigned long)dma32_phys_limit;
1372 	char *cmdline = boot_command_line;
1373 	bool fixed_base = false;
1374 	bool high = false;
1375 
1376 	int ret = 0;
1377 
1378 	if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1379 		return;
1380 	/*
1381 	 * Don't reserve a region for a crash kernel on a crash kernel
1382 	 * since it doesn't make much sense and we have limited memory
1383 	 * resources.
1384 	 */
1385 	if (is_kdump_kernel()) {
1386 		pr_info("crashkernel: ignoring reservation request\n");
1387 		return;
1388 	}
1389 
1390 	ret = parse_crashkernel(cmdline, memblock_phys_mem_size(),
1391 				&crash_size, &crash_base);
1392 	if (ret == -ENOENT) {
1393 		/* Fallback to crashkernel=X,[high,low] */
1394 		ret = parse_crashkernel_high(cmdline, 0, &crash_size, &crash_base);
1395 		if (ret || !crash_size)
1396 			return;
1397 
1398 		/*
1399 		 * crashkernel=Y,low is valid only when crashkernel=X,high
1400 		 * is passed.
1401 		 */
1402 		ret = parse_crashkernel_low(cmdline, 0, &crash_low_size, &crash_base);
1403 		if (ret == -ENOENT)
1404 			crash_low_size = DEFAULT_CRASH_KERNEL_LOW_SIZE;
1405 		else if (ret)
1406 			return;
1407 
1408 		search_start = (unsigned long)dma32_phys_limit;
1409 		search_end = memblock_end_of_DRAM();
1410 		high = true;
1411 	} else if (ret || !crash_size) {
1412 		/* Invalid argument value specified */
1413 		return;
1414 	}
1415 
1416 	crash_size = PAGE_ALIGN(crash_size);
1417 
1418 	if (crash_base) {
1419 		fixed_base = true;
1420 		search_start = crash_base;
1421 		search_end = crash_base + crash_size;
1422 	}
1423 
1424 	/*
1425 	 * Current riscv boot protocol requires 2MB alignment for
1426 	 * RV64 and 4MB alignment for RV32 (hugepage size)
1427 	 *
1428 	 * Try to alloc from 32bit addressible physical memory so that
1429 	 * swiotlb can work on the crash kernel.
1430 	 */
1431 	crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1432 					       search_start, search_end);
1433 	if (crash_base == 0) {
1434 		/*
1435 		 * For crashkernel=size[KMG]@offset[KMG], print out failure
1436 		 * message if can't reserve the specified region.
1437 		 */
1438 		if (fixed_base) {
1439 			pr_warn("crashkernel: allocating failed with given size@offset\n");
1440 			return;
1441 		}
1442 
1443 		if (high) {
1444 			/*
1445 			 * For crashkernel=size[KMG],high, if the first attempt was
1446 			 * for high memory, fall back to low memory.
1447 			 */
1448 			search_start = memblock_start_of_DRAM();
1449 			search_end = (unsigned long)dma32_phys_limit;
1450 		} else {
1451 			/*
1452 			 * For crashkernel=size[KMG], if the first attempt was for
1453 			 * low memory, fall back to high memory, the minimum required
1454 			 * low memory will be reserved later.
1455 			 */
1456 			search_start = (unsigned long)dma32_phys_limit;
1457 			search_end = memblock_end_of_DRAM();
1458 			crash_low_size = DEFAULT_CRASH_KERNEL_LOW_SIZE;
1459 		}
1460 
1461 		crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1462 						       search_start, search_end);
1463 		if (crash_base == 0) {
1464 			pr_warn("crashkernel: couldn't allocate %lldKB\n",
1465 				crash_size >> 10);
1466 			return;
1467 		}
1468 	}
1469 
1470 	if ((crash_base >= dma32_phys_limit) && crash_low_size &&
1471 	     reserve_crashkernel_low(crash_low_size)) {
1472 		memblock_phys_free(crash_base, crash_size);
1473 		return;
1474 	}
1475 
1476 	pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1477 		crash_base, crash_base + crash_size, crash_size >> 20);
1478 
1479 	crashk_res.start = crash_base;
1480 	crashk_res.end = crash_base + crash_size - 1;
1481 }
1482 
1483 void __init paging_init(void)
1484 {
1485 	setup_bootmem();
1486 	setup_vm_final();
1487 
1488 	/* Depend on that Linear Mapping is ready */
1489 	memblock_allow_resize();
1490 }
1491 
1492 void __init misc_mem_init(void)
1493 {
1494 	early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1495 	arch_numa_init();
1496 	sparse_init();
1497 	zone_sizes_init();
1498 	reserve_crashkernel();
1499 	memblock_dump_all();
1500 }
1501 
1502 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1503 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1504 			       struct vmem_altmap *altmap)
1505 {
1506 	return vmemmap_populate_basepages(start, end, node, NULL);
1507 }
1508 #endif
1509 
1510 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1511 /*
1512  * Pre-allocates page-table pages for a specific area in the kernel
1513  * page-table. Only the level which needs to be synchronized between
1514  * all page-tables is allocated because the synchronization can be
1515  * expensive.
1516  */
1517 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1518 					       const char *area)
1519 {
1520 	unsigned long addr;
1521 	const char *lvl;
1522 
1523 	for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1524 		pgd_t *pgd = pgd_offset_k(addr);
1525 		p4d_t *p4d;
1526 		pud_t *pud;
1527 		pmd_t *pmd;
1528 
1529 		lvl = "p4d";
1530 		p4d = p4d_alloc(&init_mm, pgd, addr);
1531 		if (!p4d)
1532 			goto failed;
1533 
1534 		if (pgtable_l5_enabled)
1535 			continue;
1536 
1537 		lvl = "pud";
1538 		pud = pud_alloc(&init_mm, p4d, addr);
1539 		if (!pud)
1540 			goto failed;
1541 
1542 		if (pgtable_l4_enabled)
1543 			continue;
1544 
1545 		lvl = "pmd";
1546 		pmd = pmd_alloc(&init_mm, pud, addr);
1547 		if (!pmd)
1548 			goto failed;
1549 	}
1550 	return;
1551 
1552 failed:
1553 	/*
1554 	 * The pages have to be there now or they will be missing in
1555 	 * process page-tables later.
1556 	 */
1557 	panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1558 }
1559 
1560 void __init pgtable_cache_init(void)
1561 {
1562 	preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1563 	if (IS_ENABLED(CONFIG_MODULES))
1564 		preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
1565 }
1566 #endif
1567