xref: /openbmc/linux/arch/riscv/mm/init.c (revision e8069f5a)
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 	memblock_allow_resize();
271 }
272 
273 #ifdef CONFIG_MMU
274 struct pt_alloc_ops pt_ops __initdata;
275 
276 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
277 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
278 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
279 
280 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
281 
282 #ifdef CONFIG_XIP_KERNEL
283 #define pt_ops			(*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
284 #define trampoline_pg_dir      ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
285 #define fixmap_pte             ((pte_t *)XIP_FIXUP(fixmap_pte))
286 #define early_pg_dir           ((pgd_t *)XIP_FIXUP(early_pg_dir))
287 #endif /* CONFIG_XIP_KERNEL */
288 
289 static const pgprot_t protection_map[16] = {
290 	[VM_NONE]					= PAGE_NONE,
291 	[VM_READ]					= PAGE_READ,
292 	[VM_WRITE]					= PAGE_COPY,
293 	[VM_WRITE | VM_READ]				= PAGE_COPY,
294 	[VM_EXEC]					= PAGE_EXEC,
295 	[VM_EXEC | VM_READ]				= PAGE_READ_EXEC,
296 	[VM_EXEC | VM_WRITE]				= PAGE_COPY_EXEC,
297 	[VM_EXEC | VM_WRITE | VM_READ]			= PAGE_COPY_EXEC,
298 	[VM_SHARED]					= PAGE_NONE,
299 	[VM_SHARED | VM_READ]				= PAGE_READ,
300 	[VM_SHARED | VM_WRITE]				= PAGE_SHARED,
301 	[VM_SHARED | VM_WRITE | VM_READ]		= PAGE_SHARED,
302 	[VM_SHARED | VM_EXEC]				= PAGE_EXEC,
303 	[VM_SHARED | VM_EXEC | VM_READ]			= PAGE_READ_EXEC,
304 	[VM_SHARED | VM_EXEC | VM_WRITE]		= PAGE_SHARED_EXEC,
305 	[VM_SHARED | VM_EXEC | VM_WRITE | VM_READ]	= PAGE_SHARED_EXEC
306 };
307 DECLARE_VM_GET_PAGE_PROT
308 
309 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
310 {
311 	unsigned long addr = __fix_to_virt(idx);
312 	pte_t *ptep;
313 
314 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
315 
316 	ptep = &fixmap_pte[pte_index(addr)];
317 
318 	if (pgprot_val(prot))
319 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
320 	else
321 		pte_clear(&init_mm, addr, ptep);
322 	local_flush_tlb_page(addr);
323 }
324 
325 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
326 {
327 	return (pte_t *)((uintptr_t)pa);
328 }
329 
330 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
331 {
332 	clear_fixmap(FIX_PTE);
333 	return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
334 }
335 
336 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
337 {
338 	return (pte_t *) __va(pa);
339 }
340 
341 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
342 {
343 	/*
344 	 * We only create PMD or PGD early mappings so we
345 	 * should never reach here with MMU disabled.
346 	 */
347 	BUG();
348 }
349 
350 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
351 {
352 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
353 }
354 
355 static phys_addr_t __init alloc_pte_late(uintptr_t va)
356 {
357 	unsigned long vaddr;
358 
359 	vaddr = __get_free_page(GFP_KERNEL);
360 	BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)));
361 
362 	return __pa(vaddr);
363 }
364 
365 static void __init create_pte_mapping(pte_t *ptep,
366 				      uintptr_t va, phys_addr_t pa,
367 				      phys_addr_t sz, pgprot_t prot)
368 {
369 	uintptr_t pte_idx = pte_index(va);
370 
371 	BUG_ON(sz != PAGE_SIZE);
372 
373 	if (pte_none(ptep[pte_idx]))
374 		ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
375 }
376 
377 #ifndef __PAGETABLE_PMD_FOLDED
378 
379 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
380 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
381 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
382 
383 #ifdef CONFIG_XIP_KERNEL
384 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
385 #define fixmap_pmd     ((pmd_t *)XIP_FIXUP(fixmap_pmd))
386 #define early_pmd      ((pmd_t *)XIP_FIXUP(early_pmd))
387 #endif /* CONFIG_XIP_KERNEL */
388 
389 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
390 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
391 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
392 
393 #ifdef CONFIG_XIP_KERNEL
394 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
395 #define fixmap_p4d     ((p4d_t *)XIP_FIXUP(fixmap_p4d))
396 #define early_p4d      ((p4d_t *)XIP_FIXUP(early_p4d))
397 #endif /* CONFIG_XIP_KERNEL */
398 
399 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
400 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
401 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
402 
403 #ifdef CONFIG_XIP_KERNEL
404 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
405 #define fixmap_pud     ((pud_t *)XIP_FIXUP(fixmap_pud))
406 #define early_pud      ((pud_t *)XIP_FIXUP(early_pud))
407 #endif /* CONFIG_XIP_KERNEL */
408 
409 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
410 {
411 	/* Before MMU is enabled */
412 	return (pmd_t *)((uintptr_t)pa);
413 }
414 
415 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
416 {
417 	clear_fixmap(FIX_PMD);
418 	return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
419 }
420 
421 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
422 {
423 	return (pmd_t *) __va(pa);
424 }
425 
426 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
427 {
428 	BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
429 
430 	return (uintptr_t)early_pmd;
431 }
432 
433 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
434 {
435 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
436 }
437 
438 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
439 {
440 	unsigned long vaddr;
441 
442 	vaddr = __get_free_page(GFP_KERNEL);
443 	BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page(vaddr)));
444 
445 	return __pa(vaddr);
446 }
447 
448 static void __init create_pmd_mapping(pmd_t *pmdp,
449 				      uintptr_t va, phys_addr_t pa,
450 				      phys_addr_t sz, pgprot_t prot)
451 {
452 	pte_t *ptep;
453 	phys_addr_t pte_phys;
454 	uintptr_t pmd_idx = pmd_index(va);
455 
456 	if (sz == PMD_SIZE) {
457 		if (pmd_none(pmdp[pmd_idx]))
458 			pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
459 		return;
460 	}
461 
462 	if (pmd_none(pmdp[pmd_idx])) {
463 		pte_phys = pt_ops.alloc_pte(va);
464 		pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
465 		ptep = pt_ops.get_pte_virt(pte_phys);
466 		memset(ptep, 0, PAGE_SIZE);
467 	} else {
468 		pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
469 		ptep = pt_ops.get_pte_virt(pte_phys);
470 	}
471 
472 	create_pte_mapping(ptep, va, pa, sz, prot);
473 }
474 
475 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
476 {
477 	return (pud_t *)((uintptr_t)pa);
478 }
479 
480 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
481 {
482 	clear_fixmap(FIX_PUD);
483 	return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
484 }
485 
486 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
487 {
488 	return (pud_t *)__va(pa);
489 }
490 
491 static phys_addr_t __init alloc_pud_early(uintptr_t va)
492 {
493 	/* Only one PUD is available for early mapping */
494 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
495 
496 	return (uintptr_t)early_pud;
497 }
498 
499 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
500 {
501 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
502 }
503 
504 static phys_addr_t alloc_pud_late(uintptr_t va)
505 {
506 	unsigned long vaddr;
507 
508 	vaddr = __get_free_page(GFP_KERNEL);
509 	BUG_ON(!vaddr);
510 	return __pa(vaddr);
511 }
512 
513 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
514 {
515 	return (p4d_t *)((uintptr_t)pa);
516 }
517 
518 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
519 {
520 	clear_fixmap(FIX_P4D);
521 	return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
522 }
523 
524 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
525 {
526 	return (p4d_t *)__va(pa);
527 }
528 
529 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
530 {
531 	/* Only one P4D is available for early mapping */
532 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
533 
534 	return (uintptr_t)early_p4d;
535 }
536 
537 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
538 {
539 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
540 }
541 
542 static phys_addr_t alloc_p4d_late(uintptr_t va)
543 {
544 	unsigned long vaddr;
545 
546 	vaddr = __get_free_page(GFP_KERNEL);
547 	BUG_ON(!vaddr);
548 	return __pa(vaddr);
549 }
550 
551 static void __init create_pud_mapping(pud_t *pudp,
552 				      uintptr_t va, phys_addr_t pa,
553 				      phys_addr_t sz, pgprot_t prot)
554 {
555 	pmd_t *nextp;
556 	phys_addr_t next_phys;
557 	uintptr_t pud_index = pud_index(va);
558 
559 	if (sz == PUD_SIZE) {
560 		if (pud_val(pudp[pud_index]) == 0)
561 			pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
562 		return;
563 	}
564 
565 	if (pud_val(pudp[pud_index]) == 0) {
566 		next_phys = pt_ops.alloc_pmd(va);
567 		pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
568 		nextp = pt_ops.get_pmd_virt(next_phys);
569 		memset(nextp, 0, PAGE_SIZE);
570 	} else {
571 		next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
572 		nextp = pt_ops.get_pmd_virt(next_phys);
573 	}
574 
575 	create_pmd_mapping(nextp, va, pa, sz, prot);
576 }
577 
578 static void __init create_p4d_mapping(p4d_t *p4dp,
579 				      uintptr_t va, phys_addr_t pa,
580 				      phys_addr_t sz, pgprot_t prot)
581 {
582 	pud_t *nextp;
583 	phys_addr_t next_phys;
584 	uintptr_t p4d_index = p4d_index(va);
585 
586 	if (sz == P4D_SIZE) {
587 		if (p4d_val(p4dp[p4d_index]) == 0)
588 			p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
589 		return;
590 	}
591 
592 	if (p4d_val(p4dp[p4d_index]) == 0) {
593 		next_phys = pt_ops.alloc_pud(va);
594 		p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
595 		nextp = pt_ops.get_pud_virt(next_phys);
596 		memset(nextp, 0, PAGE_SIZE);
597 	} else {
598 		next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
599 		nextp = pt_ops.get_pud_virt(next_phys);
600 	}
601 
602 	create_pud_mapping(nextp, va, pa, sz, prot);
603 }
604 
605 #define pgd_next_t		p4d_t
606 #define alloc_pgd_next(__va)	(pgtable_l5_enabled ?			\
607 		pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ?		\
608 		pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
609 #define get_pgd_next_virt(__pa)	(pgtable_l5_enabled ?			\
610 		pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ?	\
611 		pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
612 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
613 				(pgtable_l5_enabled ?			\
614 		create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
615 				(pgtable_l4_enabled ?			\
616 		create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) :	\
617 		create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
618 #define fixmap_pgd_next		(pgtable_l5_enabled ?			\
619 		(uintptr_t)fixmap_p4d : (pgtable_l4_enabled ?		\
620 		(uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
621 #define trampoline_pgd_next	(pgtable_l5_enabled ?			\
622 		(uintptr_t)trampoline_p4d : (pgtable_l4_enabled ?	\
623 		(uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
624 #else
625 #define pgd_next_t		pte_t
626 #define alloc_pgd_next(__va)	pt_ops.alloc_pte(__va)
627 #define get_pgd_next_virt(__pa)	pt_ops.get_pte_virt(__pa)
628 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
629 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
630 #define fixmap_pgd_next		((uintptr_t)fixmap_pte)
631 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
632 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
633 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
634 #endif /* __PAGETABLE_PMD_FOLDED */
635 
636 void __init create_pgd_mapping(pgd_t *pgdp,
637 				      uintptr_t va, phys_addr_t pa,
638 				      phys_addr_t sz, pgprot_t prot)
639 {
640 	pgd_next_t *nextp;
641 	phys_addr_t next_phys;
642 	uintptr_t pgd_idx = pgd_index(va);
643 
644 	if (sz == PGDIR_SIZE) {
645 		if (pgd_val(pgdp[pgd_idx]) == 0)
646 			pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
647 		return;
648 	}
649 
650 	if (pgd_val(pgdp[pgd_idx]) == 0) {
651 		next_phys = alloc_pgd_next(va);
652 		pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
653 		nextp = get_pgd_next_virt(next_phys);
654 		memset(nextp, 0, PAGE_SIZE);
655 	} else {
656 		next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
657 		nextp = get_pgd_next_virt(next_phys);
658 	}
659 
660 	create_pgd_next_mapping(nextp, va, pa, sz, prot);
661 }
662 
663 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va,
664 				      phys_addr_t size)
665 {
666 	if (!(pa & (PGDIR_SIZE - 1)) && !(va & (PGDIR_SIZE - 1)) && size >= PGDIR_SIZE)
667 		return PGDIR_SIZE;
668 
669 	if (!(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
670 		return P4D_SIZE;
671 
672 	if (!(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
673 		return PUD_SIZE;
674 
675 	if (!(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
676 		return PMD_SIZE;
677 
678 	return PAGE_SIZE;
679 }
680 
681 #ifdef CONFIG_XIP_KERNEL
682 #define phys_ram_base  (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
683 extern char _xiprom[], _exiprom[], __data_loc;
684 
685 /* called from head.S with MMU off */
686 asmlinkage void __init __copy_data(void)
687 {
688 	void *from = (void *)(&__data_loc);
689 	void *to = (void *)CONFIG_PHYS_RAM_BASE;
690 	size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
691 
692 	memcpy(to, from, sz);
693 }
694 #endif
695 
696 #ifdef CONFIG_STRICT_KERNEL_RWX
697 static __init pgprot_t pgprot_from_va(uintptr_t va)
698 {
699 	if (is_va_kernel_text(va))
700 		return PAGE_KERNEL_READ_EXEC;
701 
702 	/*
703 	 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
704 	 * we must protect its linear mapping alias from being executed and
705 	 * written.
706 	 * And rodata section is marked readonly in mark_rodata_ro.
707 	 */
708 	if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
709 		return PAGE_KERNEL_READ;
710 
711 	return PAGE_KERNEL;
712 }
713 
714 void mark_rodata_ro(void)
715 {
716 	set_kernel_memory(__start_rodata, _data, set_memory_ro);
717 	if (IS_ENABLED(CONFIG_64BIT))
718 		set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
719 				  set_memory_ro);
720 
721 	debug_checkwx();
722 }
723 #else
724 static __init pgprot_t pgprot_from_va(uintptr_t va)
725 {
726 	if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
727 		return PAGE_KERNEL;
728 
729 	return PAGE_KERNEL_EXEC;
730 }
731 #endif /* CONFIG_STRICT_KERNEL_RWX */
732 
733 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
734 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
735 
736 static void __init disable_pgtable_l5(void)
737 {
738 	pgtable_l5_enabled = false;
739 	kernel_map.page_offset = PAGE_OFFSET_L4;
740 	satp_mode = SATP_MODE_48;
741 }
742 
743 static void __init disable_pgtable_l4(void)
744 {
745 	pgtable_l4_enabled = false;
746 	kernel_map.page_offset = PAGE_OFFSET_L3;
747 	satp_mode = SATP_MODE_39;
748 }
749 
750 static int __init print_no4lvl(char *p)
751 {
752 	pr_info("Disabled 4-level and 5-level paging");
753 	return 0;
754 }
755 early_param("no4lvl", print_no4lvl);
756 
757 static int __init print_no5lvl(char *p)
758 {
759 	pr_info("Disabled 5-level paging");
760 	return 0;
761 }
762 early_param("no5lvl", print_no5lvl);
763 
764 /*
765  * There is a simple way to determine if 4-level is supported by the
766  * underlying hardware: establish 1:1 mapping in 4-level page table mode
767  * then read SATP to see if the configuration was taken into account
768  * meaning sv48 is supported.
769  */
770 static __init void set_satp_mode(uintptr_t dtb_pa)
771 {
772 	u64 identity_satp, hw_satp;
773 	uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
774 	u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
775 
776 	if (satp_mode_cmdline == SATP_MODE_57) {
777 		disable_pgtable_l5();
778 	} else if (satp_mode_cmdline == SATP_MODE_48) {
779 		disable_pgtable_l5();
780 		disable_pgtable_l4();
781 		return;
782 	}
783 
784 	create_p4d_mapping(early_p4d,
785 			set_satp_mode_pmd, (uintptr_t)early_pud,
786 			P4D_SIZE, PAGE_TABLE);
787 	create_pud_mapping(early_pud,
788 			   set_satp_mode_pmd, (uintptr_t)early_pmd,
789 			   PUD_SIZE, PAGE_TABLE);
790 	/* Handle the case where set_satp_mode straddles 2 PMDs */
791 	create_pmd_mapping(early_pmd,
792 			   set_satp_mode_pmd, set_satp_mode_pmd,
793 			   PMD_SIZE, PAGE_KERNEL_EXEC);
794 	create_pmd_mapping(early_pmd,
795 			   set_satp_mode_pmd + PMD_SIZE,
796 			   set_satp_mode_pmd + PMD_SIZE,
797 			   PMD_SIZE, PAGE_KERNEL_EXEC);
798 retry:
799 	create_pgd_mapping(early_pg_dir,
800 			   set_satp_mode_pmd,
801 			   pgtable_l5_enabled ?
802 				(uintptr_t)early_p4d : (uintptr_t)early_pud,
803 			   PGDIR_SIZE, PAGE_TABLE);
804 
805 	identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
806 
807 	local_flush_tlb_all();
808 	csr_write(CSR_SATP, identity_satp);
809 	hw_satp = csr_swap(CSR_SATP, 0ULL);
810 	local_flush_tlb_all();
811 
812 	if (hw_satp != identity_satp) {
813 		if (pgtable_l5_enabled) {
814 			disable_pgtable_l5();
815 			memset(early_pg_dir, 0, PAGE_SIZE);
816 			goto retry;
817 		}
818 		disable_pgtable_l4();
819 	}
820 
821 	memset(early_pg_dir, 0, PAGE_SIZE);
822 	memset(early_p4d, 0, PAGE_SIZE);
823 	memset(early_pud, 0, PAGE_SIZE);
824 	memset(early_pmd, 0, PAGE_SIZE);
825 }
826 #endif
827 
828 /*
829  * setup_vm() is called from head.S with MMU-off.
830  *
831  * Following requirements should be honoured for setup_vm() to work
832  * correctly:
833  * 1) It should use PC-relative addressing for accessing kernel symbols.
834  *    To achieve this we always use GCC cmodel=medany.
835  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
836  *    so disable compiler instrumentation when FTRACE is enabled.
837  *
838  * Currently, the above requirements are honoured by using custom CFLAGS
839  * for init.o in mm/Makefile.
840  */
841 
842 #ifndef __riscv_cmodel_medany
843 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
844 #endif
845 
846 #ifdef CONFIG_RELOCATABLE
847 extern unsigned long __rela_dyn_start, __rela_dyn_end;
848 
849 static void __init relocate_kernel(void)
850 {
851 	Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
852 	/*
853 	 * This holds the offset between the linked virtual address and the
854 	 * relocated virtual address.
855 	 */
856 	uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
857 	/*
858 	 * This holds the offset between kernel linked virtual address and
859 	 * physical address.
860 	 */
861 	uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
862 
863 	for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
864 		Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
865 		Elf64_Addr relocated_addr = rela->r_addend;
866 
867 		if (rela->r_info != R_RISCV_RELATIVE)
868 			continue;
869 
870 		/*
871 		 * Make sure to not relocate vdso symbols like rt_sigreturn
872 		 * which are linked from the address 0 in vmlinux since
873 		 * vdso symbol addresses are actually used as an offset from
874 		 * mm->context.vdso in VDSO_OFFSET macro.
875 		 */
876 		if (relocated_addr >= KERNEL_LINK_ADDR)
877 			relocated_addr += reloc_offset;
878 
879 		*(Elf64_Addr *)addr = relocated_addr;
880 	}
881 }
882 #endif /* CONFIG_RELOCATABLE */
883 
884 #ifdef CONFIG_XIP_KERNEL
885 static void __init create_kernel_page_table(pgd_t *pgdir,
886 					    __always_unused bool early)
887 {
888 	uintptr_t va, end_va;
889 
890 	/* Map the flash resident part */
891 	end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
892 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
893 		create_pgd_mapping(pgdir, va,
894 				   kernel_map.xiprom + (va - kernel_map.virt_addr),
895 				   PMD_SIZE, PAGE_KERNEL_EXEC);
896 
897 	/* Map the data in RAM */
898 	end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
899 	for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
900 		create_pgd_mapping(pgdir, va,
901 				   kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
902 				   PMD_SIZE, PAGE_KERNEL);
903 }
904 #else
905 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
906 {
907 	uintptr_t va, end_va;
908 
909 	end_va = kernel_map.virt_addr + kernel_map.size;
910 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
911 		create_pgd_mapping(pgdir, va,
912 				   kernel_map.phys_addr + (va - kernel_map.virt_addr),
913 				   PMD_SIZE,
914 				   early ?
915 					PAGE_KERNEL_EXEC : pgprot_from_va(va));
916 }
917 #endif
918 
919 /*
920  * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
921  * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
922  * entry.
923  */
924 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
925 					       uintptr_t dtb_pa)
926 {
927 #ifndef CONFIG_BUILTIN_DTB
928 	uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
929 
930 	/* Make sure the fdt fixmap address is always aligned on PMD size */
931 	BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
932 
933 	/* In 32-bit only, the fdt lies in its own PGD */
934 	if (!IS_ENABLED(CONFIG_64BIT)) {
935 		create_pgd_mapping(early_pg_dir, fix_fdt_va,
936 				   pa, MAX_FDT_SIZE, PAGE_KERNEL);
937 	} else {
938 		create_pmd_mapping(fixmap_pmd, fix_fdt_va,
939 				   pa, PMD_SIZE, PAGE_KERNEL);
940 		create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
941 				   pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
942 	}
943 
944 	dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
945 #else
946 	/*
947 	 * For 64-bit kernel, __va can't be used since it would return a linear
948 	 * mapping address whereas dtb_early_va will be used before
949 	 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
950 	 * kernel is mapped in the linear mapping, that makes no difference.
951 	 */
952 	dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
953 #endif
954 
955 	dtb_early_pa = dtb_pa;
956 }
957 
958 /*
959  * MMU is not enabled, the page tables are allocated directly using
960  * early_pmd/pud/p4d and the address returned is the physical one.
961  */
962 static void __init pt_ops_set_early(void)
963 {
964 	pt_ops.alloc_pte = alloc_pte_early;
965 	pt_ops.get_pte_virt = get_pte_virt_early;
966 #ifndef __PAGETABLE_PMD_FOLDED
967 	pt_ops.alloc_pmd = alloc_pmd_early;
968 	pt_ops.get_pmd_virt = get_pmd_virt_early;
969 	pt_ops.alloc_pud = alloc_pud_early;
970 	pt_ops.get_pud_virt = get_pud_virt_early;
971 	pt_ops.alloc_p4d = alloc_p4d_early;
972 	pt_ops.get_p4d_virt = get_p4d_virt_early;
973 #endif
974 }
975 
976 /*
977  * MMU is enabled but page table setup is not complete yet.
978  * fixmap page table alloc functions must be used as a means to temporarily
979  * map the allocated physical pages since the linear mapping does not exist yet.
980  *
981  * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
982  * but it will be used as described above.
983  */
984 static void __init pt_ops_set_fixmap(void)
985 {
986 	pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
987 	pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
988 #ifndef __PAGETABLE_PMD_FOLDED
989 	pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
990 	pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
991 	pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
992 	pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
993 	pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
994 	pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
995 #endif
996 }
997 
998 /*
999  * MMU is enabled and page table setup is complete, so from now, we can use
1000  * generic page allocation functions to setup page table.
1001  */
1002 static void __init pt_ops_set_late(void)
1003 {
1004 	pt_ops.alloc_pte = alloc_pte_late;
1005 	pt_ops.get_pte_virt = get_pte_virt_late;
1006 #ifndef __PAGETABLE_PMD_FOLDED
1007 	pt_ops.alloc_pmd = alloc_pmd_late;
1008 	pt_ops.get_pmd_virt = get_pmd_virt_late;
1009 	pt_ops.alloc_pud = alloc_pud_late;
1010 	pt_ops.get_pud_virt = get_pud_virt_late;
1011 	pt_ops.alloc_p4d = alloc_p4d_late;
1012 	pt_ops.get_p4d_virt = get_p4d_virt_late;
1013 #endif
1014 }
1015 
1016 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1017 {
1018 	pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1019 
1020 	kernel_map.virt_addr = KERNEL_LINK_ADDR;
1021 	kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1022 
1023 #ifdef CONFIG_XIP_KERNEL
1024 	kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1025 	kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1026 
1027 	phys_ram_base = CONFIG_PHYS_RAM_BASE;
1028 	kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1029 	kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
1030 
1031 	kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1032 #else
1033 	kernel_map.phys_addr = (uintptr_t)(&_start);
1034 	kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1035 #endif
1036 
1037 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1038 	set_satp_mode(dtb_pa);
1039 #endif
1040 
1041 	/*
1042 	 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1043 	 * where we have the system memory layout: this allows us to align
1044 	 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1045 	 * for the linear mapping. This is only possible because the kernel
1046 	 * mapping lies outside the linear mapping.
1047 	 * In 32-bit however, as the kernel resides in the linear mapping,
1048 	 * setup_vm_final can not change the mapping established here,
1049 	 * otherwise the same kernel addresses would get mapped to different
1050 	 * physical addresses (if the start of dram is different from the
1051 	 * kernel physical address start).
1052 	 */
1053 	kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1054 				0UL : PAGE_OFFSET - kernel_map.phys_addr;
1055 	kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1056 
1057 	/*
1058 	 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
1059 	 * kernel, whereas for 64-bit kernel, the end of the virtual address
1060 	 * space is occupied by the modules/BPF/kernel mappings which reduces
1061 	 * the available size of the linear mapping.
1062 	 */
1063 	memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
1064 
1065 	/* Sanity check alignment and size */
1066 	BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1067 	BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1068 
1069 #ifdef CONFIG_64BIT
1070 	/*
1071 	 * The last 4K bytes of the addressable memory can not be mapped because
1072 	 * of IS_ERR_VALUE macro.
1073 	 */
1074 	BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1075 #endif
1076 
1077 #ifdef CONFIG_RELOCATABLE
1078 	/*
1079 	 * Early page table uses only one PUD, which makes it possible
1080 	 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1081 	 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1082 	 * since a part of the kernel would not get mapped.
1083 	 */
1084 	BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1085 	relocate_kernel();
1086 #endif
1087 
1088 	apply_early_boot_alternatives();
1089 	pt_ops_set_early();
1090 
1091 	/* Setup early PGD for fixmap */
1092 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
1093 			   fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1094 
1095 #ifndef __PAGETABLE_PMD_FOLDED
1096 	/* Setup fixmap P4D and PUD */
1097 	if (pgtable_l5_enabled)
1098 		create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1099 				   (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1100 	/* Setup fixmap PUD and PMD */
1101 	if (pgtable_l4_enabled)
1102 		create_pud_mapping(fixmap_pud, FIXADDR_START,
1103 				   (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1104 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1105 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1106 	/* Setup trampoline PGD and PMD */
1107 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1108 			   trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1109 	if (pgtable_l5_enabled)
1110 		create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1111 				   (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1112 	if (pgtable_l4_enabled)
1113 		create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1114 				   (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1115 #ifdef CONFIG_XIP_KERNEL
1116 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1117 			   kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1118 #else
1119 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1120 			   kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1121 #endif
1122 #else
1123 	/* Setup trampoline PGD */
1124 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1125 			   kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1126 #endif
1127 
1128 	/*
1129 	 * Setup early PGD covering entire kernel which will allow
1130 	 * us to reach paging_init(). We map all memory banks later
1131 	 * in setup_vm_final() below.
1132 	 */
1133 	create_kernel_page_table(early_pg_dir, true);
1134 
1135 	/* Setup early mapping for FDT early scan */
1136 	create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1137 
1138 	/*
1139 	 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1140 	 * range can not span multiple pmds.
1141 	 */
1142 	BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1143 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1144 
1145 #ifndef __PAGETABLE_PMD_FOLDED
1146 	/*
1147 	 * Early ioremap fixmap is already created as it lies within first 2MB
1148 	 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1149 	 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1150 	 * the user if not.
1151 	 */
1152 	fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1153 	fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1154 	if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1155 		WARN_ON(1);
1156 		pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1157 			pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1158 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1159 			fix_to_virt(FIX_BTMAP_BEGIN));
1160 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
1161 			fix_to_virt(FIX_BTMAP_END));
1162 
1163 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
1164 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
1165 	}
1166 #endif
1167 
1168 	pt_ops_set_fixmap();
1169 }
1170 
1171 static void __init create_linear_mapping_range(phys_addr_t start,
1172 					       phys_addr_t end,
1173 					       uintptr_t fixed_map_size)
1174 {
1175 	phys_addr_t pa;
1176 	uintptr_t va, map_size;
1177 
1178 	for (pa = start; pa < end; pa += map_size) {
1179 		va = (uintptr_t)__va(pa);
1180 		map_size = fixed_map_size ? fixed_map_size :
1181 					    best_map_size(pa, va, end - pa);
1182 
1183 		create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1184 				   pgprot_from_va(va));
1185 	}
1186 }
1187 
1188 static void __init create_linear_mapping_page_table(void)
1189 {
1190 	phys_addr_t start, end;
1191 	phys_addr_t kfence_pool __maybe_unused;
1192 	u64 i;
1193 
1194 #ifdef CONFIG_STRICT_KERNEL_RWX
1195 	phys_addr_t ktext_start = __pa_symbol(_start);
1196 	phys_addr_t ktext_size = __init_data_begin - _start;
1197 	phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1198 	phys_addr_t krodata_size = _data - __start_rodata;
1199 
1200 	/* Isolate kernel text and rodata so they don't get mapped with a PUD */
1201 	memblock_mark_nomap(ktext_start,  ktext_size);
1202 	memblock_mark_nomap(krodata_start, krodata_size);
1203 #endif
1204 
1205 #ifdef CONFIG_KFENCE
1206 	/*
1207 	 *  kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1208 	 *  before we setup the linear mapping so that we avoid using hugepages
1209 	 *  for this region.
1210 	 */
1211 	kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1212 	BUG_ON(!kfence_pool);
1213 
1214 	memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1215 	__kfence_pool = __va(kfence_pool);
1216 #endif
1217 
1218 	/* Map all memory banks in the linear mapping */
1219 	for_each_mem_range(i, &start, &end) {
1220 		if (start >= end)
1221 			break;
1222 		if (start <= __pa(PAGE_OFFSET) &&
1223 		    __pa(PAGE_OFFSET) < end)
1224 			start = __pa(PAGE_OFFSET);
1225 		if (end >= __pa(PAGE_OFFSET) + memory_limit)
1226 			end = __pa(PAGE_OFFSET) + memory_limit;
1227 
1228 		create_linear_mapping_range(start, end, 0);
1229 	}
1230 
1231 #ifdef CONFIG_STRICT_KERNEL_RWX
1232 	create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0);
1233 	create_linear_mapping_range(krodata_start,
1234 				    krodata_start + krodata_size, 0);
1235 
1236 	memblock_clear_nomap(ktext_start,  ktext_size);
1237 	memblock_clear_nomap(krodata_start, krodata_size);
1238 #endif
1239 
1240 #ifdef CONFIG_KFENCE
1241 	create_linear_mapping_range(kfence_pool,
1242 				    kfence_pool + KFENCE_POOL_SIZE,
1243 				    PAGE_SIZE);
1244 
1245 	memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1246 #endif
1247 }
1248 
1249 static void __init setup_vm_final(void)
1250 {
1251 	/* Setup swapper PGD for fixmap */
1252 #if !defined(CONFIG_64BIT)
1253 	/*
1254 	 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1255 	 * directly in swapper_pg_dir in addition to the pgd entry that points
1256 	 * to fixmap_pte.
1257 	 */
1258 	unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1259 
1260 	set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1261 #endif
1262 	create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1263 			   __pa_symbol(fixmap_pgd_next),
1264 			   PGDIR_SIZE, PAGE_TABLE);
1265 
1266 	/* Map the linear mapping */
1267 	create_linear_mapping_page_table();
1268 
1269 	/* Map the kernel */
1270 	if (IS_ENABLED(CONFIG_64BIT))
1271 		create_kernel_page_table(swapper_pg_dir, false);
1272 
1273 #ifdef CONFIG_KASAN
1274 	kasan_swapper_init();
1275 #endif
1276 
1277 	/* Clear fixmap PTE and PMD mappings */
1278 	clear_fixmap(FIX_PTE);
1279 	clear_fixmap(FIX_PMD);
1280 	clear_fixmap(FIX_PUD);
1281 	clear_fixmap(FIX_P4D);
1282 
1283 	/* Move to swapper page table */
1284 	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1285 	local_flush_tlb_all();
1286 
1287 	pt_ops_set_late();
1288 }
1289 #else
1290 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1291 {
1292 	dtb_early_va = (void *)dtb_pa;
1293 	dtb_early_pa = dtb_pa;
1294 }
1295 
1296 static inline void setup_vm_final(void)
1297 {
1298 }
1299 #endif /* CONFIG_MMU */
1300 
1301 /*
1302  * reserve_crashkernel() - reserves memory for crash kernel
1303  *
1304  * This function reserves memory area given in "crashkernel=" kernel command
1305  * line parameter. The memory reserved is used by dump capture kernel when
1306  * primary kernel is crashing.
1307  */
1308 static void __init reserve_crashkernel(void)
1309 {
1310 	unsigned long long crash_base = 0;
1311 	unsigned long long crash_size = 0;
1312 	unsigned long search_start = memblock_start_of_DRAM();
1313 	unsigned long search_end = memblock_end_of_DRAM();
1314 
1315 	int ret = 0;
1316 
1317 	if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1318 		return;
1319 	/*
1320 	 * Don't reserve a region for a crash kernel on a crash kernel
1321 	 * since it doesn't make much sense and we have limited memory
1322 	 * resources.
1323 	 */
1324 	if (is_kdump_kernel()) {
1325 		pr_info("crashkernel: ignoring reservation request\n");
1326 		return;
1327 	}
1328 
1329 	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
1330 				&crash_size, &crash_base);
1331 	if (ret || !crash_size)
1332 		return;
1333 
1334 	crash_size = PAGE_ALIGN(crash_size);
1335 
1336 	if (crash_base) {
1337 		search_start = crash_base;
1338 		search_end = crash_base + crash_size;
1339 	}
1340 
1341 	/*
1342 	 * Current riscv boot protocol requires 2MB alignment for
1343 	 * RV64 and 4MB alignment for RV32 (hugepage size)
1344 	 *
1345 	 * Try to alloc from 32bit addressible physical memory so that
1346 	 * swiotlb can work on the crash kernel.
1347 	 */
1348 	crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1349 					       search_start,
1350 					       min(search_end, (unsigned long) SZ_4G));
1351 	if (crash_base == 0) {
1352 		/* Try again without restricting region to 32bit addressible memory */
1353 		crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1354 						search_start, search_end);
1355 		if (crash_base == 0) {
1356 			pr_warn("crashkernel: couldn't allocate %lldKB\n",
1357 				crash_size >> 10);
1358 			return;
1359 		}
1360 	}
1361 
1362 	pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1363 		crash_base, crash_base + crash_size, crash_size >> 20);
1364 
1365 	crashk_res.start = crash_base;
1366 	crashk_res.end = crash_base + crash_size - 1;
1367 }
1368 
1369 void __init paging_init(void)
1370 {
1371 	setup_bootmem();
1372 	setup_vm_final();
1373 }
1374 
1375 void __init misc_mem_init(void)
1376 {
1377 	early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1378 	arch_numa_init();
1379 	sparse_init();
1380 	zone_sizes_init();
1381 	reserve_crashkernel();
1382 	memblock_dump_all();
1383 }
1384 
1385 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1386 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1387 			       struct vmem_altmap *altmap)
1388 {
1389 	return vmemmap_populate_basepages(start, end, node, NULL);
1390 }
1391 #endif
1392 
1393 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1394 /*
1395  * Pre-allocates page-table pages for a specific area in the kernel
1396  * page-table. Only the level which needs to be synchronized between
1397  * all page-tables is allocated because the synchronization can be
1398  * expensive.
1399  */
1400 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1401 					       const char *area)
1402 {
1403 	unsigned long addr;
1404 	const char *lvl;
1405 
1406 	for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1407 		pgd_t *pgd = pgd_offset_k(addr);
1408 		p4d_t *p4d;
1409 		pud_t *pud;
1410 		pmd_t *pmd;
1411 
1412 		lvl = "p4d";
1413 		p4d = p4d_alloc(&init_mm, pgd, addr);
1414 		if (!p4d)
1415 			goto failed;
1416 
1417 		if (pgtable_l5_enabled)
1418 			continue;
1419 
1420 		lvl = "pud";
1421 		pud = pud_alloc(&init_mm, p4d, addr);
1422 		if (!pud)
1423 			goto failed;
1424 
1425 		if (pgtable_l4_enabled)
1426 			continue;
1427 
1428 		lvl = "pmd";
1429 		pmd = pmd_alloc(&init_mm, pud, addr);
1430 		if (!pmd)
1431 			goto failed;
1432 	}
1433 	return;
1434 
1435 failed:
1436 	/*
1437 	 * The pages have to be there now or they will be missing in
1438 	 * process page-tables later.
1439 	 */
1440 	panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1441 }
1442 
1443 void __init pgtable_cache_init(void)
1444 {
1445 	preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1446 	if (IS_ENABLED(CONFIG_MODULES))
1447 		preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
1448 }
1449 #endif
1450