xref: /openbmc/linux/arch/riscv/mm/init.c (revision e2942062)
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/tlbflush.h>
30 #include <asm/sections.h>
31 #include <asm/soc.h>
32 #include <asm/io.h>
33 #include <asm/ptdump.h>
34 #include <asm/numa.h>
35 
36 #include "../kernel/head.h"
37 
38 struct kernel_mapping kernel_map __ro_after_init;
39 EXPORT_SYMBOL(kernel_map);
40 #ifdef CONFIG_XIP_KERNEL
41 #define kernel_map	(*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
42 #endif
43 
44 #ifdef CONFIG_64BIT
45 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39;
46 #else
47 u64 satp_mode __ro_after_init = SATP_MODE_32;
48 #endif
49 EXPORT_SYMBOL(satp_mode);
50 
51 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
52 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
53 EXPORT_SYMBOL(pgtable_l4_enabled);
54 EXPORT_SYMBOL(pgtable_l5_enabled);
55 
56 phys_addr_t phys_ram_base __ro_after_init;
57 EXPORT_SYMBOL(phys_ram_base);
58 
59 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
60 							__page_aligned_bss;
61 EXPORT_SYMBOL(empty_zero_page);
62 
63 extern char _start[];
64 void *_dtb_early_va __initdata;
65 uintptr_t _dtb_early_pa __initdata;
66 
67 static phys_addr_t dma32_phys_limit __initdata;
68 
69 static void __init zone_sizes_init(void)
70 {
71 	unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
72 
73 #ifdef CONFIG_ZONE_DMA32
74 	max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
75 #endif
76 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
77 
78 	free_area_init(max_zone_pfns);
79 }
80 
81 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
82 
83 #define LOG2_SZ_1K  ilog2(SZ_1K)
84 #define LOG2_SZ_1M  ilog2(SZ_1M)
85 #define LOG2_SZ_1G  ilog2(SZ_1G)
86 #define LOG2_SZ_1T  ilog2(SZ_1T)
87 
88 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
89 {
90 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld kB)\n", name, b, t,
91 		  (((t) - (b)) >> LOG2_SZ_1K));
92 }
93 
94 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
95 {
96 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld MB)\n", name, b, t,
97 		  (((t) - (b)) >> LOG2_SZ_1M));
98 }
99 
100 static inline void print_mlg(char *name, unsigned long b, unsigned long t)
101 {
102 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld GB)\n", name, b, t,
103 		   (((t) - (b)) >> LOG2_SZ_1G));
104 }
105 
106 #ifdef CONFIG_64BIT
107 static inline void print_mlt(char *name, unsigned long b, unsigned long t)
108 {
109 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld TB)\n", name, b, t,
110 		   (((t) - (b)) >> LOG2_SZ_1T));
111 }
112 #else
113 #define print_mlt(n, b, t) do {} while (0)
114 #endif
115 
116 static inline void print_ml(char *name, unsigned long b, unsigned long t)
117 {
118 	unsigned long diff = t - b;
119 
120 	if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10)
121 		print_mlt(name, b, t);
122 	else if ((diff >> LOG2_SZ_1G) >= 10)
123 		print_mlg(name, b, t);
124 	else if ((diff >> LOG2_SZ_1M) >= 10)
125 		print_mlm(name, b, t);
126 	else
127 		print_mlk(name, b, t);
128 }
129 
130 static void __init print_vm_layout(void)
131 {
132 	pr_notice("Virtual kernel memory layout:\n");
133 	print_ml("fixmap", (unsigned long)FIXADDR_START,
134 		(unsigned long)FIXADDR_TOP);
135 	print_ml("pci io", (unsigned long)PCI_IO_START,
136 		(unsigned long)PCI_IO_END);
137 	print_ml("vmemmap", (unsigned long)VMEMMAP_START,
138 		(unsigned long)VMEMMAP_END);
139 	print_ml("vmalloc", (unsigned long)VMALLOC_START,
140 		(unsigned long)VMALLOC_END);
141 #ifdef CONFIG_64BIT
142 	print_ml("modules", (unsigned long)MODULES_VADDR,
143 		(unsigned long)MODULES_END);
144 #endif
145 	print_ml("lowmem", (unsigned long)PAGE_OFFSET,
146 		(unsigned long)high_memory);
147 	if (IS_ENABLED(CONFIG_64BIT)) {
148 #ifdef CONFIG_KASAN
149 		print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
150 #endif
151 
152 		print_ml("kernel", (unsigned long)kernel_map.virt_addr,
153 			 (unsigned long)ADDRESS_SPACE_END);
154 	}
155 }
156 #else
157 static void print_vm_layout(void) { }
158 #endif /* CONFIG_DEBUG_VM */
159 
160 void __init mem_init(void)
161 {
162 #ifdef CONFIG_FLATMEM
163 	BUG_ON(!mem_map);
164 #endif /* CONFIG_FLATMEM */
165 
166 	swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE);
167 	memblock_free_all();
168 
169 	print_vm_layout();
170 }
171 
172 /* Limit the memory size via mem. */
173 static phys_addr_t memory_limit;
174 
175 static int __init early_mem(char *p)
176 {
177 	u64 size;
178 
179 	if (!p)
180 		return 1;
181 
182 	size = memparse(p, &p) & PAGE_MASK;
183 	memory_limit = min_t(u64, size, memory_limit);
184 
185 	pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
186 
187 	return 0;
188 }
189 early_param("mem", early_mem);
190 
191 static void __init setup_bootmem(void)
192 {
193 	phys_addr_t vmlinux_end = __pa_symbol(&_end);
194 	phys_addr_t max_mapped_addr;
195 	phys_addr_t phys_ram_end, vmlinux_start;
196 
197 	if (IS_ENABLED(CONFIG_XIP_KERNEL))
198 		vmlinux_start = __pa_symbol(&_sdata);
199 	else
200 		vmlinux_start = __pa_symbol(&_start);
201 
202 	memblock_enforce_memory_limit(memory_limit);
203 
204 	/*
205 	 * Make sure we align the reservation on PMD_SIZE since we will
206 	 * map the kernel in the linear mapping as read-only: we do not want
207 	 * any allocation to happen between _end and the next pmd aligned page.
208 	 */
209 	if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
210 		vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
211 	/*
212 	 * Reserve from the start of the kernel to the end of the kernel
213 	 */
214 	memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
215 
216 	phys_ram_end = memblock_end_of_DRAM();
217 	if (!IS_ENABLED(CONFIG_XIP_KERNEL))
218 		phys_ram_base = memblock_start_of_DRAM();
219 
220 	/*
221 	 * In 64-bit, any use of __va/__pa before this point is wrong as we
222 	 * did not know the start of DRAM before.
223 	 */
224 	if (IS_ENABLED(CONFIG_64BIT))
225 		kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base;
226 
227 	/*
228 	 * memblock allocator is not aware of the fact that last 4K bytes of
229 	 * the addressable memory can not be mapped because of IS_ERR_VALUE
230 	 * macro. Make sure that last 4k bytes are not usable by memblock
231 	 * if end of dram is equal to maximum addressable memory.  For 64-bit
232 	 * kernel, this problem can't happen here as the end of the virtual
233 	 * address space is occupied by the kernel mapping then this check must
234 	 * be done as soon as the kernel mapping base address is determined.
235 	 */
236 	if (!IS_ENABLED(CONFIG_64BIT)) {
237 		max_mapped_addr = __pa(~(ulong)0);
238 		if (max_mapped_addr == (phys_ram_end - 1))
239 			memblock_set_current_limit(max_mapped_addr - 4096);
240 	}
241 
242 	min_low_pfn = PFN_UP(phys_ram_base);
243 	max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
244 	high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
245 
246 	dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
247 	set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
248 
249 	reserve_initrd_mem();
250 
251 	/*
252 	 * No allocation should be done before reserving the memory as defined
253 	 * in the device tree, otherwise the allocation could end up in a
254 	 * reserved region.
255 	 */
256 	early_init_fdt_scan_reserved_mem();
257 
258 	/*
259 	 * If DTB is built in, no need to reserve its memblock.
260 	 * Otherwise, do reserve it but avoid using
261 	 * early_init_fdt_reserve_self() since __pa() does
262 	 * not work for DTB pointers that are fixmap addresses
263 	 */
264 	if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
265 		memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
266 
267 	dma_contiguous_reserve(dma32_phys_limit);
268 	if (IS_ENABLED(CONFIG_64BIT))
269 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
270 }
271 
272 #ifdef CONFIG_MMU
273 struct pt_alloc_ops pt_ops __initdata;
274 
275 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
276 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
277 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
278 
279 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
280 
281 #ifdef CONFIG_XIP_KERNEL
282 #define pt_ops			(*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
283 #define trampoline_pg_dir      ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
284 #define fixmap_pte             ((pte_t *)XIP_FIXUP(fixmap_pte))
285 #define early_pg_dir           ((pgd_t *)XIP_FIXUP(early_pg_dir))
286 #endif /* CONFIG_XIP_KERNEL */
287 
288 static const pgprot_t protection_map[16] = {
289 	[VM_NONE]					= PAGE_NONE,
290 	[VM_READ]					= PAGE_READ,
291 	[VM_WRITE]					= PAGE_COPY,
292 	[VM_WRITE | VM_READ]				= PAGE_COPY,
293 	[VM_EXEC]					= PAGE_EXEC,
294 	[VM_EXEC | VM_READ]				= PAGE_READ_EXEC,
295 	[VM_EXEC | VM_WRITE]				= PAGE_COPY_EXEC,
296 	[VM_EXEC | VM_WRITE | VM_READ]			= PAGE_COPY_EXEC,
297 	[VM_SHARED]					= PAGE_NONE,
298 	[VM_SHARED | VM_READ]				= PAGE_READ,
299 	[VM_SHARED | VM_WRITE]				= PAGE_SHARED,
300 	[VM_SHARED | VM_WRITE | VM_READ]		= PAGE_SHARED,
301 	[VM_SHARED | VM_EXEC]				= PAGE_EXEC,
302 	[VM_SHARED | VM_EXEC | VM_READ]			= PAGE_READ_EXEC,
303 	[VM_SHARED | VM_EXEC | VM_WRITE]		= PAGE_SHARED_EXEC,
304 	[VM_SHARED | VM_EXEC | VM_WRITE | VM_READ]	= PAGE_SHARED_EXEC
305 };
306 DECLARE_VM_GET_PAGE_PROT
307 
308 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
309 {
310 	unsigned long addr = __fix_to_virt(idx);
311 	pte_t *ptep;
312 
313 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
314 
315 	ptep = &fixmap_pte[pte_index(addr)];
316 
317 	if (pgprot_val(prot))
318 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
319 	else
320 		pte_clear(&init_mm, addr, ptep);
321 	local_flush_tlb_page(addr);
322 }
323 
324 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
325 {
326 	return (pte_t *)((uintptr_t)pa);
327 }
328 
329 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
330 {
331 	clear_fixmap(FIX_PTE);
332 	return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
333 }
334 
335 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
336 {
337 	return (pte_t *) __va(pa);
338 }
339 
340 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
341 {
342 	/*
343 	 * We only create PMD or PGD early mappings so we
344 	 * should never reach here with MMU disabled.
345 	 */
346 	BUG();
347 }
348 
349 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
350 {
351 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
352 }
353 
354 static phys_addr_t __init alloc_pte_late(uintptr_t va)
355 {
356 	unsigned long vaddr;
357 
358 	vaddr = __get_free_page(GFP_KERNEL);
359 	BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page((void *)vaddr)));
360 
361 	return __pa(vaddr);
362 }
363 
364 static void __init create_pte_mapping(pte_t *ptep,
365 				      uintptr_t va, phys_addr_t pa,
366 				      phys_addr_t sz, pgprot_t prot)
367 {
368 	uintptr_t pte_idx = pte_index(va);
369 
370 	BUG_ON(sz != PAGE_SIZE);
371 
372 	if (pte_none(ptep[pte_idx]))
373 		ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
374 }
375 
376 #ifndef __PAGETABLE_PMD_FOLDED
377 
378 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
379 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
380 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
381 
382 #ifdef CONFIG_XIP_KERNEL
383 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
384 #define fixmap_pmd     ((pmd_t *)XIP_FIXUP(fixmap_pmd))
385 #define early_pmd      ((pmd_t *)XIP_FIXUP(early_pmd))
386 #endif /* CONFIG_XIP_KERNEL */
387 
388 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
389 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
390 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
391 
392 #ifdef CONFIG_XIP_KERNEL
393 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
394 #define fixmap_p4d     ((p4d_t *)XIP_FIXUP(fixmap_p4d))
395 #define early_p4d      ((p4d_t *)XIP_FIXUP(early_p4d))
396 #endif /* CONFIG_XIP_KERNEL */
397 
398 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
399 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
400 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
401 
402 #ifdef CONFIG_XIP_KERNEL
403 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
404 #define fixmap_pud     ((pud_t *)XIP_FIXUP(fixmap_pud))
405 #define early_pud      ((pud_t *)XIP_FIXUP(early_pud))
406 #endif /* CONFIG_XIP_KERNEL */
407 
408 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
409 {
410 	/* Before MMU is enabled */
411 	return (pmd_t *)((uintptr_t)pa);
412 }
413 
414 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
415 {
416 	clear_fixmap(FIX_PMD);
417 	return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
418 }
419 
420 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
421 {
422 	return (pmd_t *) __va(pa);
423 }
424 
425 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
426 {
427 	BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
428 
429 	return (uintptr_t)early_pmd;
430 }
431 
432 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
433 {
434 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
435 }
436 
437 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
438 {
439 	unsigned long vaddr;
440 
441 	vaddr = __get_free_page(GFP_KERNEL);
442 	BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page((void *)vaddr)));
443 
444 	return __pa(vaddr);
445 }
446 
447 static void __init create_pmd_mapping(pmd_t *pmdp,
448 				      uintptr_t va, phys_addr_t pa,
449 				      phys_addr_t sz, pgprot_t prot)
450 {
451 	pte_t *ptep;
452 	phys_addr_t pte_phys;
453 	uintptr_t pmd_idx = pmd_index(va);
454 
455 	if (sz == PMD_SIZE) {
456 		if (pmd_none(pmdp[pmd_idx]))
457 			pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
458 		return;
459 	}
460 
461 	if (pmd_none(pmdp[pmd_idx])) {
462 		pte_phys = pt_ops.alloc_pte(va);
463 		pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
464 		ptep = pt_ops.get_pte_virt(pte_phys);
465 		memset(ptep, 0, PAGE_SIZE);
466 	} else {
467 		pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
468 		ptep = pt_ops.get_pte_virt(pte_phys);
469 	}
470 
471 	create_pte_mapping(ptep, va, pa, sz, prot);
472 }
473 
474 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
475 {
476 	return (pud_t *)((uintptr_t)pa);
477 }
478 
479 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
480 {
481 	clear_fixmap(FIX_PUD);
482 	return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
483 }
484 
485 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
486 {
487 	return (pud_t *)__va(pa);
488 }
489 
490 static phys_addr_t __init alloc_pud_early(uintptr_t va)
491 {
492 	/* Only one PUD is available for early mapping */
493 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
494 
495 	return (uintptr_t)early_pud;
496 }
497 
498 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
499 {
500 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
501 }
502 
503 static phys_addr_t alloc_pud_late(uintptr_t va)
504 {
505 	unsigned long vaddr;
506 
507 	vaddr = __get_free_page(GFP_KERNEL);
508 	BUG_ON(!vaddr);
509 	return __pa(vaddr);
510 }
511 
512 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
513 {
514 	return (p4d_t *)((uintptr_t)pa);
515 }
516 
517 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
518 {
519 	clear_fixmap(FIX_P4D);
520 	return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
521 }
522 
523 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
524 {
525 	return (p4d_t *)__va(pa);
526 }
527 
528 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
529 {
530 	/* Only one P4D is available for early mapping */
531 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
532 
533 	return (uintptr_t)early_p4d;
534 }
535 
536 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
537 {
538 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
539 }
540 
541 static phys_addr_t alloc_p4d_late(uintptr_t va)
542 {
543 	unsigned long vaddr;
544 
545 	vaddr = __get_free_page(GFP_KERNEL);
546 	BUG_ON(!vaddr);
547 	return __pa(vaddr);
548 }
549 
550 static void __init create_pud_mapping(pud_t *pudp,
551 				      uintptr_t va, phys_addr_t pa,
552 				      phys_addr_t sz, pgprot_t prot)
553 {
554 	pmd_t *nextp;
555 	phys_addr_t next_phys;
556 	uintptr_t pud_index = pud_index(va);
557 
558 	if (sz == PUD_SIZE) {
559 		if (pud_val(pudp[pud_index]) == 0)
560 			pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
561 		return;
562 	}
563 
564 	if (pud_val(pudp[pud_index]) == 0) {
565 		next_phys = pt_ops.alloc_pmd(va);
566 		pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
567 		nextp = pt_ops.get_pmd_virt(next_phys);
568 		memset(nextp, 0, PAGE_SIZE);
569 	} else {
570 		next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
571 		nextp = pt_ops.get_pmd_virt(next_phys);
572 	}
573 
574 	create_pmd_mapping(nextp, va, pa, sz, prot);
575 }
576 
577 static void __init create_p4d_mapping(p4d_t *p4dp,
578 				      uintptr_t va, phys_addr_t pa,
579 				      phys_addr_t sz, pgprot_t prot)
580 {
581 	pud_t *nextp;
582 	phys_addr_t next_phys;
583 	uintptr_t p4d_index = p4d_index(va);
584 
585 	if (sz == P4D_SIZE) {
586 		if (p4d_val(p4dp[p4d_index]) == 0)
587 			p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
588 		return;
589 	}
590 
591 	if (p4d_val(p4dp[p4d_index]) == 0) {
592 		next_phys = pt_ops.alloc_pud(va);
593 		p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
594 		nextp = pt_ops.get_pud_virt(next_phys);
595 		memset(nextp, 0, PAGE_SIZE);
596 	} else {
597 		next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
598 		nextp = pt_ops.get_pud_virt(next_phys);
599 	}
600 
601 	create_pud_mapping(nextp, va, pa, sz, prot);
602 }
603 
604 #define pgd_next_t		p4d_t
605 #define alloc_pgd_next(__va)	(pgtable_l5_enabled ?			\
606 		pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ?		\
607 		pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
608 #define get_pgd_next_virt(__pa)	(pgtable_l5_enabled ?			\
609 		pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ?	\
610 		pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
611 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
612 				(pgtable_l5_enabled ?			\
613 		create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
614 				(pgtable_l4_enabled ?			\
615 		create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) :	\
616 		create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
617 #define fixmap_pgd_next		(pgtable_l5_enabled ?			\
618 		(uintptr_t)fixmap_p4d : (pgtable_l4_enabled ?		\
619 		(uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
620 #define trampoline_pgd_next	(pgtable_l5_enabled ?			\
621 		(uintptr_t)trampoline_p4d : (pgtable_l4_enabled ?	\
622 		(uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
623 #else
624 #define pgd_next_t		pte_t
625 #define alloc_pgd_next(__va)	pt_ops.alloc_pte(__va)
626 #define get_pgd_next_virt(__pa)	pt_ops.get_pte_virt(__pa)
627 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
628 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
629 #define fixmap_pgd_next		((uintptr_t)fixmap_pte)
630 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
631 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
632 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
633 #endif /* __PAGETABLE_PMD_FOLDED */
634 
635 void __init create_pgd_mapping(pgd_t *pgdp,
636 				      uintptr_t va, phys_addr_t pa,
637 				      phys_addr_t sz, pgprot_t prot)
638 {
639 	pgd_next_t *nextp;
640 	phys_addr_t next_phys;
641 	uintptr_t pgd_idx = pgd_index(va);
642 
643 	if (sz == PGDIR_SIZE) {
644 		if (pgd_val(pgdp[pgd_idx]) == 0)
645 			pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
646 		return;
647 	}
648 
649 	if (pgd_val(pgdp[pgd_idx]) == 0) {
650 		next_phys = alloc_pgd_next(va);
651 		pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
652 		nextp = get_pgd_next_virt(next_phys);
653 		memset(nextp, 0, PAGE_SIZE);
654 	} else {
655 		next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
656 		nextp = get_pgd_next_virt(next_phys);
657 	}
658 
659 	create_pgd_next_mapping(nextp, va, pa, sz, prot);
660 }
661 
662 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va,
663 				      phys_addr_t size)
664 {
665 	if (!(pa & (PGDIR_SIZE - 1)) && !(va & (PGDIR_SIZE - 1)) && size >= PGDIR_SIZE)
666 		return PGDIR_SIZE;
667 
668 	if (!(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
669 		return P4D_SIZE;
670 
671 	if (!(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
672 		return PUD_SIZE;
673 
674 	if (!(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
675 		return PMD_SIZE;
676 
677 	return PAGE_SIZE;
678 }
679 
680 #ifdef CONFIG_XIP_KERNEL
681 #define phys_ram_base  (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
682 extern char _xiprom[], _exiprom[], __data_loc;
683 
684 /* called from head.S with MMU off */
685 asmlinkage void __init __copy_data(void)
686 {
687 	void *from = (void *)(&__data_loc);
688 	void *to = (void *)CONFIG_PHYS_RAM_BASE;
689 	size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
690 
691 	memcpy(to, from, sz);
692 }
693 #endif
694 
695 #ifdef CONFIG_STRICT_KERNEL_RWX
696 static __init pgprot_t pgprot_from_va(uintptr_t va)
697 {
698 	if (is_va_kernel_text(va))
699 		return PAGE_KERNEL_READ_EXEC;
700 
701 	/*
702 	 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
703 	 * we must protect its linear mapping alias from being executed and
704 	 * written.
705 	 * And rodata section is marked readonly in mark_rodata_ro.
706 	 */
707 	if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
708 		return PAGE_KERNEL_READ;
709 
710 	return PAGE_KERNEL;
711 }
712 
713 void mark_rodata_ro(void)
714 {
715 	set_kernel_memory(__start_rodata, _data, set_memory_ro);
716 	if (IS_ENABLED(CONFIG_64BIT))
717 		set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
718 				  set_memory_ro);
719 
720 	debug_checkwx();
721 }
722 #else
723 static __init pgprot_t pgprot_from_va(uintptr_t va)
724 {
725 	if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
726 		return PAGE_KERNEL;
727 
728 	return PAGE_KERNEL_EXEC;
729 }
730 #endif /* CONFIG_STRICT_KERNEL_RWX */
731 
732 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
733 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
734 
735 static void __init disable_pgtable_l5(void)
736 {
737 	pgtable_l5_enabled = false;
738 	kernel_map.page_offset = PAGE_OFFSET_L4;
739 	satp_mode = SATP_MODE_48;
740 }
741 
742 static void __init disable_pgtable_l4(void)
743 {
744 	pgtable_l4_enabled = false;
745 	kernel_map.page_offset = PAGE_OFFSET_L3;
746 	satp_mode = SATP_MODE_39;
747 }
748 
749 static int __init print_no4lvl(char *p)
750 {
751 	pr_info("Disabled 4-level and 5-level paging");
752 	return 0;
753 }
754 early_param("no4lvl", print_no4lvl);
755 
756 static int __init print_no5lvl(char *p)
757 {
758 	pr_info("Disabled 5-level paging");
759 	return 0;
760 }
761 early_param("no5lvl", print_no5lvl);
762 
763 /*
764  * There is a simple way to determine if 4-level is supported by the
765  * underlying hardware: establish 1:1 mapping in 4-level page table mode
766  * then read SATP to see if the configuration was taken into account
767  * meaning sv48 is supported.
768  */
769 static __init void set_satp_mode(uintptr_t dtb_pa)
770 {
771 	u64 identity_satp, hw_satp;
772 	uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
773 	u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
774 
775 	if (satp_mode_cmdline == SATP_MODE_57) {
776 		disable_pgtable_l5();
777 	} else if (satp_mode_cmdline == SATP_MODE_48) {
778 		disable_pgtable_l5();
779 		disable_pgtable_l4();
780 		return;
781 	}
782 
783 	create_p4d_mapping(early_p4d,
784 			set_satp_mode_pmd, (uintptr_t)early_pud,
785 			P4D_SIZE, PAGE_TABLE);
786 	create_pud_mapping(early_pud,
787 			   set_satp_mode_pmd, (uintptr_t)early_pmd,
788 			   PUD_SIZE, PAGE_TABLE);
789 	/* Handle the case where set_satp_mode straddles 2 PMDs */
790 	create_pmd_mapping(early_pmd,
791 			   set_satp_mode_pmd, set_satp_mode_pmd,
792 			   PMD_SIZE, PAGE_KERNEL_EXEC);
793 	create_pmd_mapping(early_pmd,
794 			   set_satp_mode_pmd + PMD_SIZE,
795 			   set_satp_mode_pmd + PMD_SIZE,
796 			   PMD_SIZE, PAGE_KERNEL_EXEC);
797 retry:
798 	create_pgd_mapping(early_pg_dir,
799 			   set_satp_mode_pmd,
800 			   pgtable_l5_enabled ?
801 				(uintptr_t)early_p4d : (uintptr_t)early_pud,
802 			   PGDIR_SIZE, PAGE_TABLE);
803 
804 	identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
805 
806 	local_flush_tlb_all();
807 	csr_write(CSR_SATP, identity_satp);
808 	hw_satp = csr_swap(CSR_SATP, 0ULL);
809 	local_flush_tlb_all();
810 
811 	if (hw_satp != identity_satp) {
812 		if (pgtable_l5_enabled) {
813 			disable_pgtable_l5();
814 			memset(early_pg_dir, 0, PAGE_SIZE);
815 			goto retry;
816 		}
817 		disable_pgtable_l4();
818 	}
819 
820 	memset(early_pg_dir, 0, PAGE_SIZE);
821 	memset(early_p4d, 0, PAGE_SIZE);
822 	memset(early_pud, 0, PAGE_SIZE);
823 	memset(early_pmd, 0, PAGE_SIZE);
824 }
825 #endif
826 
827 /*
828  * setup_vm() is called from head.S with MMU-off.
829  *
830  * Following requirements should be honoured for setup_vm() to work
831  * correctly:
832  * 1) It should use PC-relative addressing for accessing kernel symbols.
833  *    To achieve this we always use GCC cmodel=medany.
834  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
835  *    so disable compiler instrumentation when FTRACE is enabled.
836  *
837  * Currently, the above requirements are honoured by using custom CFLAGS
838  * for init.o in mm/Makefile.
839  */
840 
841 #ifndef __riscv_cmodel_medany
842 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
843 #endif
844 
845 #ifdef CONFIG_RELOCATABLE
846 extern unsigned long __rela_dyn_start, __rela_dyn_end;
847 
848 static void __init relocate_kernel(void)
849 {
850 	Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
851 	/*
852 	 * This holds the offset between the linked virtual address and the
853 	 * relocated virtual address.
854 	 */
855 	uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
856 	/*
857 	 * This holds the offset between kernel linked virtual address and
858 	 * physical address.
859 	 */
860 	uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
861 
862 	for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
863 		Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
864 		Elf64_Addr relocated_addr = rela->r_addend;
865 
866 		if (rela->r_info != R_RISCV_RELATIVE)
867 			continue;
868 
869 		/*
870 		 * Make sure to not relocate vdso symbols like rt_sigreturn
871 		 * which are linked from the address 0 in vmlinux since
872 		 * vdso symbol addresses are actually used as an offset from
873 		 * mm->context.vdso in VDSO_OFFSET macro.
874 		 */
875 		if (relocated_addr >= KERNEL_LINK_ADDR)
876 			relocated_addr += reloc_offset;
877 
878 		*(Elf64_Addr *)addr = relocated_addr;
879 	}
880 }
881 #endif /* CONFIG_RELOCATABLE */
882 
883 #ifdef CONFIG_XIP_KERNEL
884 static void __init create_kernel_page_table(pgd_t *pgdir,
885 					    __always_unused bool early)
886 {
887 	uintptr_t va, end_va;
888 
889 	/* Map the flash resident part */
890 	end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
891 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
892 		create_pgd_mapping(pgdir, va,
893 				   kernel_map.xiprom + (va - kernel_map.virt_addr),
894 				   PMD_SIZE, PAGE_KERNEL_EXEC);
895 
896 	/* Map the data in RAM */
897 	end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
898 	for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
899 		create_pgd_mapping(pgdir, va,
900 				   kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
901 				   PMD_SIZE, PAGE_KERNEL);
902 }
903 #else
904 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
905 {
906 	uintptr_t va, end_va;
907 
908 	end_va = kernel_map.virt_addr + kernel_map.size;
909 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
910 		create_pgd_mapping(pgdir, va,
911 				   kernel_map.phys_addr + (va - kernel_map.virt_addr),
912 				   PMD_SIZE,
913 				   early ?
914 					PAGE_KERNEL_EXEC : pgprot_from_va(va));
915 }
916 #endif
917 
918 /*
919  * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
920  * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
921  * entry.
922  */
923 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
924 					       uintptr_t dtb_pa)
925 {
926 #ifndef CONFIG_BUILTIN_DTB
927 	uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
928 
929 	/* Make sure the fdt fixmap address is always aligned on PMD size */
930 	BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
931 
932 	/* In 32-bit only, the fdt lies in its own PGD */
933 	if (!IS_ENABLED(CONFIG_64BIT)) {
934 		create_pgd_mapping(early_pg_dir, fix_fdt_va,
935 				   pa, MAX_FDT_SIZE, PAGE_KERNEL);
936 	} else {
937 		create_pmd_mapping(fixmap_pmd, fix_fdt_va,
938 				   pa, PMD_SIZE, PAGE_KERNEL);
939 		create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
940 				   pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
941 	}
942 
943 	dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
944 #else
945 	/*
946 	 * For 64-bit kernel, __va can't be used since it would return a linear
947 	 * mapping address whereas dtb_early_va will be used before
948 	 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
949 	 * kernel is mapped in the linear mapping, that makes no difference.
950 	 */
951 	dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
952 #endif
953 
954 	dtb_early_pa = dtb_pa;
955 }
956 
957 /*
958  * MMU is not enabled, the page tables are allocated directly using
959  * early_pmd/pud/p4d and the address returned is the physical one.
960  */
961 static void __init pt_ops_set_early(void)
962 {
963 	pt_ops.alloc_pte = alloc_pte_early;
964 	pt_ops.get_pte_virt = get_pte_virt_early;
965 #ifndef __PAGETABLE_PMD_FOLDED
966 	pt_ops.alloc_pmd = alloc_pmd_early;
967 	pt_ops.get_pmd_virt = get_pmd_virt_early;
968 	pt_ops.alloc_pud = alloc_pud_early;
969 	pt_ops.get_pud_virt = get_pud_virt_early;
970 	pt_ops.alloc_p4d = alloc_p4d_early;
971 	pt_ops.get_p4d_virt = get_p4d_virt_early;
972 #endif
973 }
974 
975 /*
976  * MMU is enabled but page table setup is not complete yet.
977  * fixmap page table alloc functions must be used as a means to temporarily
978  * map the allocated physical pages since the linear mapping does not exist yet.
979  *
980  * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
981  * but it will be used as described above.
982  */
983 static void __init pt_ops_set_fixmap(void)
984 {
985 	pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
986 	pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
987 #ifndef __PAGETABLE_PMD_FOLDED
988 	pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
989 	pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
990 	pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
991 	pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
992 	pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
993 	pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
994 #endif
995 }
996 
997 /*
998  * MMU is enabled and page table setup is complete, so from now, we can use
999  * generic page allocation functions to setup page table.
1000  */
1001 static void __init pt_ops_set_late(void)
1002 {
1003 	pt_ops.alloc_pte = alloc_pte_late;
1004 	pt_ops.get_pte_virt = get_pte_virt_late;
1005 #ifndef __PAGETABLE_PMD_FOLDED
1006 	pt_ops.alloc_pmd = alloc_pmd_late;
1007 	pt_ops.get_pmd_virt = get_pmd_virt_late;
1008 	pt_ops.alloc_pud = alloc_pud_late;
1009 	pt_ops.get_pud_virt = get_pud_virt_late;
1010 	pt_ops.alloc_p4d = alloc_p4d_late;
1011 	pt_ops.get_p4d_virt = get_p4d_virt_late;
1012 #endif
1013 }
1014 
1015 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1016 {
1017 	pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1018 
1019 	kernel_map.virt_addr = KERNEL_LINK_ADDR;
1020 	kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1021 
1022 #ifdef CONFIG_XIP_KERNEL
1023 	kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1024 	kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1025 
1026 	phys_ram_base = CONFIG_PHYS_RAM_BASE;
1027 	kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1028 	kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
1029 
1030 	kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1031 #else
1032 	kernel_map.phys_addr = (uintptr_t)(&_start);
1033 	kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1034 #endif
1035 
1036 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1037 	set_satp_mode(dtb_pa);
1038 #endif
1039 
1040 	/*
1041 	 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1042 	 * where we have the system memory layout: this allows us to align
1043 	 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1044 	 * for the linear mapping. This is only possible because the kernel
1045 	 * mapping lies outside the linear mapping.
1046 	 * In 32-bit however, as the kernel resides in the linear mapping,
1047 	 * setup_vm_final can not change the mapping established here,
1048 	 * otherwise the same kernel addresses would get mapped to different
1049 	 * physical addresses (if the start of dram is different from the
1050 	 * kernel physical address start).
1051 	 */
1052 	kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1053 				0UL : PAGE_OFFSET - kernel_map.phys_addr;
1054 	kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1055 
1056 	/*
1057 	 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
1058 	 * kernel, whereas for 64-bit kernel, the end of the virtual address
1059 	 * space is occupied by the modules/BPF/kernel mappings which reduces
1060 	 * the available size of the linear mapping.
1061 	 */
1062 	memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
1063 
1064 	/* Sanity check alignment and size */
1065 	BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1066 	BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1067 
1068 #ifdef CONFIG_64BIT
1069 	/*
1070 	 * The last 4K bytes of the addressable memory can not be mapped because
1071 	 * of IS_ERR_VALUE macro.
1072 	 */
1073 	BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1074 #endif
1075 
1076 #ifdef CONFIG_RELOCATABLE
1077 	/*
1078 	 * Early page table uses only one PUD, which makes it possible
1079 	 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1080 	 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1081 	 * since a part of the kernel would not get mapped.
1082 	 */
1083 	BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1084 	relocate_kernel();
1085 #endif
1086 
1087 	apply_early_boot_alternatives();
1088 	pt_ops_set_early();
1089 
1090 	/* Setup early PGD for fixmap */
1091 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
1092 			   fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1093 
1094 #ifndef __PAGETABLE_PMD_FOLDED
1095 	/* Setup fixmap P4D and PUD */
1096 	if (pgtable_l5_enabled)
1097 		create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1098 				   (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1099 	/* Setup fixmap PUD and PMD */
1100 	if (pgtable_l4_enabled)
1101 		create_pud_mapping(fixmap_pud, FIXADDR_START,
1102 				   (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1103 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1104 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1105 	/* Setup trampoline PGD and PMD */
1106 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1107 			   trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1108 	if (pgtable_l5_enabled)
1109 		create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1110 				   (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1111 	if (pgtable_l4_enabled)
1112 		create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1113 				   (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1114 #ifdef CONFIG_XIP_KERNEL
1115 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1116 			   kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1117 #else
1118 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1119 			   kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1120 #endif
1121 #else
1122 	/* Setup trampoline PGD */
1123 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1124 			   kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1125 #endif
1126 
1127 	/*
1128 	 * Setup early PGD covering entire kernel which will allow
1129 	 * us to reach paging_init(). We map all memory banks later
1130 	 * in setup_vm_final() below.
1131 	 */
1132 	create_kernel_page_table(early_pg_dir, true);
1133 
1134 	/* Setup early mapping for FDT early scan */
1135 	create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1136 
1137 	/*
1138 	 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1139 	 * range can not span multiple pmds.
1140 	 */
1141 	BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1142 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1143 
1144 #ifndef __PAGETABLE_PMD_FOLDED
1145 	/*
1146 	 * Early ioremap fixmap is already created as it lies within first 2MB
1147 	 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1148 	 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1149 	 * the user if not.
1150 	 */
1151 	fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1152 	fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1153 	if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1154 		WARN_ON(1);
1155 		pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1156 			pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1157 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1158 			fix_to_virt(FIX_BTMAP_BEGIN));
1159 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
1160 			fix_to_virt(FIX_BTMAP_END));
1161 
1162 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
1163 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
1164 	}
1165 #endif
1166 
1167 	pt_ops_set_fixmap();
1168 }
1169 
1170 static void __init create_linear_mapping_range(phys_addr_t start,
1171 					       phys_addr_t end,
1172 					       uintptr_t fixed_map_size)
1173 {
1174 	phys_addr_t pa;
1175 	uintptr_t va, map_size;
1176 
1177 	for (pa = start; pa < end; pa += map_size) {
1178 		va = (uintptr_t)__va(pa);
1179 		map_size = fixed_map_size ? fixed_map_size :
1180 					    best_map_size(pa, va, end - pa);
1181 
1182 		create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1183 				   pgprot_from_va(va));
1184 	}
1185 }
1186 
1187 static void __init create_linear_mapping_page_table(void)
1188 {
1189 	phys_addr_t start, end;
1190 	phys_addr_t kfence_pool __maybe_unused;
1191 	u64 i;
1192 
1193 #ifdef CONFIG_STRICT_KERNEL_RWX
1194 	phys_addr_t ktext_start = __pa_symbol(_start);
1195 	phys_addr_t ktext_size = __init_data_begin - _start;
1196 	phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1197 	phys_addr_t krodata_size = _data - __start_rodata;
1198 
1199 	/* Isolate kernel text and rodata so they don't get mapped with a PUD */
1200 	memblock_mark_nomap(ktext_start,  ktext_size);
1201 	memblock_mark_nomap(krodata_start, krodata_size);
1202 #endif
1203 
1204 #ifdef CONFIG_KFENCE
1205 	/*
1206 	 *  kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1207 	 *  before we setup the linear mapping so that we avoid using hugepages
1208 	 *  for this region.
1209 	 */
1210 	kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1211 	BUG_ON(!kfence_pool);
1212 
1213 	memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1214 	__kfence_pool = __va(kfence_pool);
1215 #endif
1216 
1217 	/* Map all memory banks in the linear mapping */
1218 	for_each_mem_range(i, &start, &end) {
1219 		if (start >= end)
1220 			break;
1221 		if (start <= __pa(PAGE_OFFSET) &&
1222 		    __pa(PAGE_OFFSET) < end)
1223 			start = __pa(PAGE_OFFSET);
1224 		if (end >= __pa(PAGE_OFFSET) + memory_limit)
1225 			end = __pa(PAGE_OFFSET) + memory_limit;
1226 
1227 		create_linear_mapping_range(start, end, 0);
1228 	}
1229 
1230 #ifdef CONFIG_STRICT_KERNEL_RWX
1231 	create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0);
1232 	create_linear_mapping_range(krodata_start,
1233 				    krodata_start + krodata_size, 0);
1234 
1235 	memblock_clear_nomap(ktext_start,  ktext_size);
1236 	memblock_clear_nomap(krodata_start, krodata_size);
1237 #endif
1238 
1239 #ifdef CONFIG_KFENCE
1240 	create_linear_mapping_range(kfence_pool,
1241 				    kfence_pool + KFENCE_POOL_SIZE,
1242 				    PAGE_SIZE);
1243 
1244 	memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1245 #endif
1246 }
1247 
1248 static void __init setup_vm_final(void)
1249 {
1250 	/* Setup swapper PGD for fixmap */
1251 #if !defined(CONFIG_64BIT)
1252 	/*
1253 	 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1254 	 * directly in swapper_pg_dir in addition to the pgd entry that points
1255 	 * to fixmap_pte.
1256 	 */
1257 	unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1258 
1259 	set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1260 #endif
1261 	create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1262 			   __pa_symbol(fixmap_pgd_next),
1263 			   PGDIR_SIZE, PAGE_TABLE);
1264 
1265 	/* Map the linear mapping */
1266 	create_linear_mapping_page_table();
1267 
1268 	/* Map the kernel */
1269 	if (IS_ENABLED(CONFIG_64BIT))
1270 		create_kernel_page_table(swapper_pg_dir, false);
1271 
1272 #ifdef CONFIG_KASAN
1273 	kasan_swapper_init();
1274 #endif
1275 
1276 	/* Clear fixmap PTE and PMD mappings */
1277 	clear_fixmap(FIX_PTE);
1278 	clear_fixmap(FIX_PMD);
1279 	clear_fixmap(FIX_PUD);
1280 	clear_fixmap(FIX_P4D);
1281 
1282 	/* Move to swapper page table */
1283 	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1284 	local_flush_tlb_all();
1285 
1286 	pt_ops_set_late();
1287 }
1288 #else
1289 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1290 {
1291 	dtb_early_va = (void *)dtb_pa;
1292 	dtb_early_pa = dtb_pa;
1293 }
1294 
1295 static inline void setup_vm_final(void)
1296 {
1297 }
1298 #endif /* CONFIG_MMU */
1299 
1300 /*
1301  * reserve_crashkernel() - reserves memory for crash kernel
1302  *
1303  * This function reserves memory area given in "crashkernel=" kernel command
1304  * line parameter. The memory reserved is used by dump capture kernel when
1305  * primary kernel is crashing.
1306  */
1307 static void __init reserve_crashkernel(void)
1308 {
1309 	unsigned long long crash_base = 0;
1310 	unsigned long long crash_size = 0;
1311 	unsigned long search_start = memblock_start_of_DRAM();
1312 	unsigned long search_end = memblock_end_of_DRAM();
1313 
1314 	int ret = 0;
1315 
1316 	if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1317 		return;
1318 	/*
1319 	 * Don't reserve a region for a crash kernel on a crash kernel
1320 	 * since it doesn't make much sense and we have limited memory
1321 	 * resources.
1322 	 */
1323 	if (is_kdump_kernel()) {
1324 		pr_info("crashkernel: ignoring reservation request\n");
1325 		return;
1326 	}
1327 
1328 	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
1329 				&crash_size, &crash_base);
1330 	if (ret || !crash_size)
1331 		return;
1332 
1333 	crash_size = PAGE_ALIGN(crash_size);
1334 
1335 	if (crash_base) {
1336 		search_start = crash_base;
1337 		search_end = crash_base + crash_size;
1338 	}
1339 
1340 	/*
1341 	 * Current riscv boot protocol requires 2MB alignment for
1342 	 * RV64 and 4MB alignment for RV32 (hugepage size)
1343 	 *
1344 	 * Try to alloc from 32bit addressible physical memory so that
1345 	 * swiotlb can work on the crash kernel.
1346 	 */
1347 	crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1348 					       search_start,
1349 					       min(search_end, (unsigned long)(SZ_4G - 1)));
1350 	if (crash_base == 0) {
1351 		/* Try again without restricting region to 32bit addressible memory */
1352 		crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1353 						search_start, search_end);
1354 		if (crash_base == 0) {
1355 			pr_warn("crashkernel: couldn't allocate %lldKB\n",
1356 				crash_size >> 10);
1357 			return;
1358 		}
1359 	}
1360 
1361 	pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1362 		crash_base, crash_base + crash_size, crash_size >> 20);
1363 
1364 	crashk_res.start = crash_base;
1365 	crashk_res.end = crash_base + crash_size - 1;
1366 }
1367 
1368 void __init paging_init(void)
1369 {
1370 	setup_bootmem();
1371 	setup_vm_final();
1372 
1373 	/* Depend on that Linear Mapping is ready */
1374 	memblock_allow_resize();
1375 }
1376 
1377 void __init misc_mem_init(void)
1378 {
1379 	early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1380 	arch_numa_init();
1381 	sparse_init();
1382 	zone_sizes_init();
1383 	reserve_crashkernel();
1384 	memblock_dump_all();
1385 }
1386 
1387 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1388 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1389 			       struct vmem_altmap *altmap)
1390 {
1391 	return vmemmap_populate_basepages(start, end, node, NULL);
1392 }
1393 #endif
1394 
1395 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1396 /*
1397  * Pre-allocates page-table pages for a specific area in the kernel
1398  * page-table. Only the level which needs to be synchronized between
1399  * all page-tables is allocated because the synchronization can be
1400  * expensive.
1401  */
1402 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1403 					       const char *area)
1404 {
1405 	unsigned long addr;
1406 	const char *lvl;
1407 
1408 	for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1409 		pgd_t *pgd = pgd_offset_k(addr);
1410 		p4d_t *p4d;
1411 		pud_t *pud;
1412 		pmd_t *pmd;
1413 
1414 		lvl = "p4d";
1415 		p4d = p4d_alloc(&init_mm, pgd, addr);
1416 		if (!p4d)
1417 			goto failed;
1418 
1419 		if (pgtable_l5_enabled)
1420 			continue;
1421 
1422 		lvl = "pud";
1423 		pud = pud_alloc(&init_mm, p4d, addr);
1424 		if (!pud)
1425 			goto failed;
1426 
1427 		if (pgtable_l4_enabled)
1428 			continue;
1429 
1430 		lvl = "pmd";
1431 		pmd = pmd_alloc(&init_mm, pud, addr);
1432 		if (!pmd)
1433 			goto failed;
1434 	}
1435 	return;
1436 
1437 failed:
1438 	/*
1439 	 * The pages have to be there now or they will be missing in
1440 	 * process page-tables later.
1441 	 */
1442 	panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1443 }
1444 
1445 void __init pgtable_cache_init(void)
1446 {
1447 	preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1448 	if (IS_ENABLED(CONFIG_MODULES))
1449 		preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
1450 }
1451 #endif
1452