xref: /openbmc/linux/arch/riscv/mm/init.c (revision b348b5fe)
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 	struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
357 
358 	BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc));
359 	return __pa((pte_t *)ptdesc_address(ptdesc));
360 }
361 
362 static void __init create_pte_mapping(pte_t *ptep,
363 				      uintptr_t va, phys_addr_t pa,
364 				      phys_addr_t sz, pgprot_t prot)
365 {
366 	uintptr_t pte_idx = pte_index(va);
367 
368 	BUG_ON(sz != PAGE_SIZE);
369 
370 	if (pte_none(ptep[pte_idx]))
371 		ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
372 }
373 
374 #ifndef __PAGETABLE_PMD_FOLDED
375 
376 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
377 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
378 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
379 
380 #ifdef CONFIG_XIP_KERNEL
381 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
382 #define fixmap_pmd     ((pmd_t *)XIP_FIXUP(fixmap_pmd))
383 #define early_pmd      ((pmd_t *)XIP_FIXUP(early_pmd))
384 #endif /* CONFIG_XIP_KERNEL */
385 
386 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
387 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
388 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
389 
390 #ifdef CONFIG_XIP_KERNEL
391 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
392 #define fixmap_p4d     ((p4d_t *)XIP_FIXUP(fixmap_p4d))
393 #define early_p4d      ((p4d_t *)XIP_FIXUP(early_p4d))
394 #endif /* CONFIG_XIP_KERNEL */
395 
396 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
397 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
398 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
399 
400 #ifdef CONFIG_XIP_KERNEL
401 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
402 #define fixmap_pud     ((pud_t *)XIP_FIXUP(fixmap_pud))
403 #define early_pud      ((pud_t *)XIP_FIXUP(early_pud))
404 #endif /* CONFIG_XIP_KERNEL */
405 
406 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
407 {
408 	/* Before MMU is enabled */
409 	return (pmd_t *)((uintptr_t)pa);
410 }
411 
412 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
413 {
414 	clear_fixmap(FIX_PMD);
415 	return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
416 }
417 
418 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
419 {
420 	return (pmd_t *) __va(pa);
421 }
422 
423 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
424 {
425 	BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
426 
427 	return (uintptr_t)early_pmd;
428 }
429 
430 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
431 {
432 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
433 }
434 
435 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
436 {
437 	struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
438 
439 	BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc));
440 	return __pa((pmd_t *)ptdesc_address(ptdesc));
441 }
442 
443 static void __init create_pmd_mapping(pmd_t *pmdp,
444 				      uintptr_t va, phys_addr_t pa,
445 				      phys_addr_t sz, pgprot_t prot)
446 {
447 	pte_t *ptep;
448 	phys_addr_t pte_phys;
449 	uintptr_t pmd_idx = pmd_index(va);
450 
451 	if (sz == PMD_SIZE) {
452 		if (pmd_none(pmdp[pmd_idx]))
453 			pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
454 		return;
455 	}
456 
457 	if (pmd_none(pmdp[pmd_idx])) {
458 		pte_phys = pt_ops.alloc_pte(va);
459 		pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
460 		ptep = pt_ops.get_pte_virt(pte_phys);
461 		memset(ptep, 0, PAGE_SIZE);
462 	} else {
463 		pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
464 		ptep = pt_ops.get_pte_virt(pte_phys);
465 	}
466 
467 	create_pte_mapping(ptep, va, pa, sz, prot);
468 }
469 
470 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
471 {
472 	return (pud_t *)((uintptr_t)pa);
473 }
474 
475 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
476 {
477 	clear_fixmap(FIX_PUD);
478 	return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
479 }
480 
481 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
482 {
483 	return (pud_t *)__va(pa);
484 }
485 
486 static phys_addr_t __init alloc_pud_early(uintptr_t va)
487 {
488 	/* Only one PUD is available for early mapping */
489 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
490 
491 	return (uintptr_t)early_pud;
492 }
493 
494 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
495 {
496 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
497 }
498 
499 static phys_addr_t alloc_pud_late(uintptr_t va)
500 {
501 	unsigned long vaddr;
502 
503 	vaddr = __get_free_page(GFP_KERNEL);
504 	BUG_ON(!vaddr);
505 	return __pa(vaddr);
506 }
507 
508 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
509 {
510 	return (p4d_t *)((uintptr_t)pa);
511 }
512 
513 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
514 {
515 	clear_fixmap(FIX_P4D);
516 	return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
517 }
518 
519 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
520 {
521 	return (p4d_t *)__va(pa);
522 }
523 
524 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
525 {
526 	/* Only one P4D is available for early mapping */
527 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
528 
529 	return (uintptr_t)early_p4d;
530 }
531 
532 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
533 {
534 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
535 }
536 
537 static phys_addr_t alloc_p4d_late(uintptr_t va)
538 {
539 	unsigned long vaddr;
540 
541 	vaddr = __get_free_page(GFP_KERNEL);
542 	BUG_ON(!vaddr);
543 	return __pa(vaddr);
544 }
545 
546 static void __init create_pud_mapping(pud_t *pudp,
547 				      uintptr_t va, phys_addr_t pa,
548 				      phys_addr_t sz, pgprot_t prot)
549 {
550 	pmd_t *nextp;
551 	phys_addr_t next_phys;
552 	uintptr_t pud_index = pud_index(va);
553 
554 	if (sz == PUD_SIZE) {
555 		if (pud_val(pudp[pud_index]) == 0)
556 			pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
557 		return;
558 	}
559 
560 	if (pud_val(pudp[pud_index]) == 0) {
561 		next_phys = pt_ops.alloc_pmd(va);
562 		pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
563 		nextp = pt_ops.get_pmd_virt(next_phys);
564 		memset(nextp, 0, PAGE_SIZE);
565 	} else {
566 		next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
567 		nextp = pt_ops.get_pmd_virt(next_phys);
568 	}
569 
570 	create_pmd_mapping(nextp, va, pa, sz, prot);
571 }
572 
573 static void __init create_p4d_mapping(p4d_t *p4dp,
574 				      uintptr_t va, phys_addr_t pa,
575 				      phys_addr_t sz, pgprot_t prot)
576 {
577 	pud_t *nextp;
578 	phys_addr_t next_phys;
579 	uintptr_t p4d_index = p4d_index(va);
580 
581 	if (sz == P4D_SIZE) {
582 		if (p4d_val(p4dp[p4d_index]) == 0)
583 			p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
584 		return;
585 	}
586 
587 	if (p4d_val(p4dp[p4d_index]) == 0) {
588 		next_phys = pt_ops.alloc_pud(va);
589 		p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
590 		nextp = pt_ops.get_pud_virt(next_phys);
591 		memset(nextp, 0, PAGE_SIZE);
592 	} else {
593 		next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
594 		nextp = pt_ops.get_pud_virt(next_phys);
595 	}
596 
597 	create_pud_mapping(nextp, va, pa, sz, prot);
598 }
599 
600 #define pgd_next_t		p4d_t
601 #define alloc_pgd_next(__va)	(pgtable_l5_enabled ?			\
602 		pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ?		\
603 		pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
604 #define get_pgd_next_virt(__pa)	(pgtable_l5_enabled ?			\
605 		pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ?	\
606 		pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
607 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
608 				(pgtable_l5_enabled ?			\
609 		create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
610 				(pgtable_l4_enabled ?			\
611 		create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) :	\
612 		create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
613 #define fixmap_pgd_next		(pgtable_l5_enabled ?			\
614 		(uintptr_t)fixmap_p4d : (pgtable_l4_enabled ?		\
615 		(uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
616 #define trampoline_pgd_next	(pgtable_l5_enabled ?			\
617 		(uintptr_t)trampoline_p4d : (pgtable_l4_enabled ?	\
618 		(uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
619 #else
620 #define pgd_next_t		pte_t
621 #define alloc_pgd_next(__va)	pt_ops.alloc_pte(__va)
622 #define get_pgd_next_virt(__pa)	pt_ops.get_pte_virt(__pa)
623 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
624 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
625 #define fixmap_pgd_next		((uintptr_t)fixmap_pte)
626 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
627 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
628 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
629 #endif /* __PAGETABLE_PMD_FOLDED */
630 
631 void __init create_pgd_mapping(pgd_t *pgdp,
632 				      uintptr_t va, phys_addr_t pa,
633 				      phys_addr_t sz, pgprot_t prot)
634 {
635 	pgd_next_t *nextp;
636 	phys_addr_t next_phys;
637 	uintptr_t pgd_idx = pgd_index(va);
638 
639 	if (sz == PGDIR_SIZE) {
640 		if (pgd_val(pgdp[pgd_idx]) == 0)
641 			pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
642 		return;
643 	}
644 
645 	if (pgd_val(pgdp[pgd_idx]) == 0) {
646 		next_phys = alloc_pgd_next(va);
647 		pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
648 		nextp = get_pgd_next_virt(next_phys);
649 		memset(nextp, 0, PAGE_SIZE);
650 	} else {
651 		next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
652 		nextp = get_pgd_next_virt(next_phys);
653 	}
654 
655 	create_pgd_next_mapping(nextp, va, pa, sz, prot);
656 }
657 
658 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va,
659 				      phys_addr_t size)
660 {
661 	if (!(pa & (PGDIR_SIZE - 1)) && !(va & (PGDIR_SIZE - 1)) && size >= PGDIR_SIZE)
662 		return PGDIR_SIZE;
663 
664 	if (!(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
665 		return P4D_SIZE;
666 
667 	if (!(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
668 		return PUD_SIZE;
669 
670 	if (!(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
671 		return PMD_SIZE;
672 
673 	return PAGE_SIZE;
674 }
675 
676 #ifdef CONFIG_XIP_KERNEL
677 #define phys_ram_base  (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
678 extern char _xiprom[], _exiprom[], __data_loc;
679 
680 /* called from head.S with MMU off */
681 asmlinkage void __init __copy_data(void)
682 {
683 	void *from = (void *)(&__data_loc);
684 	void *to = (void *)CONFIG_PHYS_RAM_BASE;
685 	size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
686 
687 	memcpy(to, from, sz);
688 }
689 #endif
690 
691 #ifdef CONFIG_STRICT_KERNEL_RWX
692 static __init pgprot_t pgprot_from_va(uintptr_t va)
693 {
694 	if (is_va_kernel_text(va))
695 		return PAGE_KERNEL_READ_EXEC;
696 
697 	/*
698 	 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
699 	 * we must protect its linear mapping alias from being executed and
700 	 * written.
701 	 * And rodata section is marked readonly in mark_rodata_ro.
702 	 */
703 	if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
704 		return PAGE_KERNEL_READ;
705 
706 	return PAGE_KERNEL;
707 }
708 
709 void mark_rodata_ro(void)
710 {
711 	set_kernel_memory(__start_rodata, _data, set_memory_ro);
712 	if (IS_ENABLED(CONFIG_64BIT))
713 		set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
714 				  set_memory_ro);
715 
716 	debug_checkwx();
717 }
718 #else
719 static __init pgprot_t pgprot_from_va(uintptr_t va)
720 {
721 	if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
722 		return PAGE_KERNEL;
723 
724 	return PAGE_KERNEL_EXEC;
725 }
726 #endif /* CONFIG_STRICT_KERNEL_RWX */
727 
728 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
729 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
730 
731 static void __init disable_pgtable_l5(void)
732 {
733 	pgtable_l5_enabled = false;
734 	kernel_map.page_offset = PAGE_OFFSET_L4;
735 	satp_mode = SATP_MODE_48;
736 }
737 
738 static void __init disable_pgtable_l4(void)
739 {
740 	pgtable_l4_enabled = false;
741 	kernel_map.page_offset = PAGE_OFFSET_L3;
742 	satp_mode = SATP_MODE_39;
743 }
744 
745 static int __init print_no4lvl(char *p)
746 {
747 	pr_info("Disabled 4-level and 5-level paging");
748 	return 0;
749 }
750 early_param("no4lvl", print_no4lvl);
751 
752 static int __init print_no5lvl(char *p)
753 {
754 	pr_info("Disabled 5-level paging");
755 	return 0;
756 }
757 early_param("no5lvl", print_no5lvl);
758 
759 /*
760  * There is a simple way to determine if 4-level is supported by the
761  * underlying hardware: establish 1:1 mapping in 4-level page table mode
762  * then read SATP to see if the configuration was taken into account
763  * meaning sv48 is supported.
764  */
765 static __init void set_satp_mode(uintptr_t dtb_pa)
766 {
767 	u64 identity_satp, hw_satp;
768 	uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
769 	u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
770 
771 	if (satp_mode_cmdline == SATP_MODE_57) {
772 		disable_pgtable_l5();
773 	} else if (satp_mode_cmdline == SATP_MODE_48) {
774 		disable_pgtable_l5();
775 		disable_pgtable_l4();
776 		return;
777 	}
778 
779 	create_p4d_mapping(early_p4d,
780 			set_satp_mode_pmd, (uintptr_t)early_pud,
781 			P4D_SIZE, PAGE_TABLE);
782 	create_pud_mapping(early_pud,
783 			   set_satp_mode_pmd, (uintptr_t)early_pmd,
784 			   PUD_SIZE, PAGE_TABLE);
785 	/* Handle the case where set_satp_mode straddles 2 PMDs */
786 	create_pmd_mapping(early_pmd,
787 			   set_satp_mode_pmd, set_satp_mode_pmd,
788 			   PMD_SIZE, PAGE_KERNEL_EXEC);
789 	create_pmd_mapping(early_pmd,
790 			   set_satp_mode_pmd + PMD_SIZE,
791 			   set_satp_mode_pmd + PMD_SIZE,
792 			   PMD_SIZE, PAGE_KERNEL_EXEC);
793 retry:
794 	create_pgd_mapping(early_pg_dir,
795 			   set_satp_mode_pmd,
796 			   pgtable_l5_enabled ?
797 				(uintptr_t)early_p4d : (uintptr_t)early_pud,
798 			   PGDIR_SIZE, PAGE_TABLE);
799 
800 	identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
801 
802 	local_flush_tlb_all();
803 	csr_write(CSR_SATP, identity_satp);
804 	hw_satp = csr_swap(CSR_SATP, 0ULL);
805 	local_flush_tlb_all();
806 
807 	if (hw_satp != identity_satp) {
808 		if (pgtable_l5_enabled) {
809 			disable_pgtable_l5();
810 			memset(early_pg_dir, 0, PAGE_SIZE);
811 			goto retry;
812 		}
813 		disable_pgtable_l4();
814 	}
815 
816 	memset(early_pg_dir, 0, PAGE_SIZE);
817 	memset(early_p4d, 0, PAGE_SIZE);
818 	memset(early_pud, 0, PAGE_SIZE);
819 	memset(early_pmd, 0, PAGE_SIZE);
820 }
821 #endif
822 
823 /*
824  * setup_vm() is called from head.S with MMU-off.
825  *
826  * Following requirements should be honoured for setup_vm() to work
827  * correctly:
828  * 1) It should use PC-relative addressing for accessing kernel symbols.
829  *    To achieve this we always use GCC cmodel=medany.
830  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
831  *    so disable compiler instrumentation when FTRACE is enabled.
832  *
833  * Currently, the above requirements are honoured by using custom CFLAGS
834  * for init.o in mm/Makefile.
835  */
836 
837 #ifndef __riscv_cmodel_medany
838 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
839 #endif
840 
841 #ifdef CONFIG_RELOCATABLE
842 extern unsigned long __rela_dyn_start, __rela_dyn_end;
843 
844 static void __init relocate_kernel(void)
845 {
846 	Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
847 	/*
848 	 * This holds the offset between the linked virtual address and the
849 	 * relocated virtual address.
850 	 */
851 	uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
852 	/*
853 	 * This holds the offset between kernel linked virtual address and
854 	 * physical address.
855 	 */
856 	uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
857 
858 	for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
859 		Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
860 		Elf64_Addr relocated_addr = rela->r_addend;
861 
862 		if (rela->r_info != R_RISCV_RELATIVE)
863 			continue;
864 
865 		/*
866 		 * Make sure to not relocate vdso symbols like rt_sigreturn
867 		 * which are linked from the address 0 in vmlinux since
868 		 * vdso symbol addresses are actually used as an offset from
869 		 * mm->context.vdso in VDSO_OFFSET macro.
870 		 */
871 		if (relocated_addr >= KERNEL_LINK_ADDR)
872 			relocated_addr += reloc_offset;
873 
874 		*(Elf64_Addr *)addr = relocated_addr;
875 	}
876 }
877 #endif /* CONFIG_RELOCATABLE */
878 
879 #ifdef CONFIG_XIP_KERNEL
880 static void __init create_kernel_page_table(pgd_t *pgdir,
881 					    __always_unused bool early)
882 {
883 	uintptr_t va, end_va;
884 
885 	/* Map the flash resident part */
886 	end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
887 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
888 		create_pgd_mapping(pgdir, va,
889 				   kernel_map.xiprom + (va - kernel_map.virt_addr),
890 				   PMD_SIZE, PAGE_KERNEL_EXEC);
891 
892 	/* Map the data in RAM */
893 	end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
894 	for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
895 		create_pgd_mapping(pgdir, va,
896 				   kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
897 				   PMD_SIZE, PAGE_KERNEL);
898 }
899 #else
900 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
901 {
902 	uintptr_t va, end_va;
903 
904 	end_va = kernel_map.virt_addr + kernel_map.size;
905 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
906 		create_pgd_mapping(pgdir, va,
907 				   kernel_map.phys_addr + (va - kernel_map.virt_addr),
908 				   PMD_SIZE,
909 				   early ?
910 					PAGE_KERNEL_EXEC : pgprot_from_va(va));
911 }
912 #endif
913 
914 /*
915  * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
916  * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
917  * entry.
918  */
919 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
920 					       uintptr_t dtb_pa)
921 {
922 #ifndef CONFIG_BUILTIN_DTB
923 	uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
924 
925 	/* Make sure the fdt fixmap address is always aligned on PMD size */
926 	BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
927 
928 	/* In 32-bit only, the fdt lies in its own PGD */
929 	if (!IS_ENABLED(CONFIG_64BIT)) {
930 		create_pgd_mapping(early_pg_dir, fix_fdt_va,
931 				   pa, MAX_FDT_SIZE, PAGE_KERNEL);
932 	} else {
933 		create_pmd_mapping(fixmap_pmd, fix_fdt_va,
934 				   pa, PMD_SIZE, PAGE_KERNEL);
935 		create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
936 				   pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
937 	}
938 
939 	dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
940 #else
941 	/*
942 	 * For 64-bit kernel, __va can't be used since it would return a linear
943 	 * mapping address whereas dtb_early_va will be used before
944 	 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
945 	 * kernel is mapped in the linear mapping, that makes no difference.
946 	 */
947 	dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
948 #endif
949 
950 	dtb_early_pa = dtb_pa;
951 }
952 
953 /*
954  * MMU is not enabled, the page tables are allocated directly using
955  * early_pmd/pud/p4d and the address returned is the physical one.
956  */
957 static void __init pt_ops_set_early(void)
958 {
959 	pt_ops.alloc_pte = alloc_pte_early;
960 	pt_ops.get_pte_virt = get_pte_virt_early;
961 #ifndef __PAGETABLE_PMD_FOLDED
962 	pt_ops.alloc_pmd = alloc_pmd_early;
963 	pt_ops.get_pmd_virt = get_pmd_virt_early;
964 	pt_ops.alloc_pud = alloc_pud_early;
965 	pt_ops.get_pud_virt = get_pud_virt_early;
966 	pt_ops.alloc_p4d = alloc_p4d_early;
967 	pt_ops.get_p4d_virt = get_p4d_virt_early;
968 #endif
969 }
970 
971 /*
972  * MMU is enabled but page table setup is not complete yet.
973  * fixmap page table alloc functions must be used as a means to temporarily
974  * map the allocated physical pages since the linear mapping does not exist yet.
975  *
976  * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
977  * but it will be used as described above.
978  */
979 static void __init pt_ops_set_fixmap(void)
980 {
981 	pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
982 	pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
983 #ifndef __PAGETABLE_PMD_FOLDED
984 	pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
985 	pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
986 	pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
987 	pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
988 	pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
989 	pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
990 #endif
991 }
992 
993 /*
994  * MMU is enabled and page table setup is complete, so from now, we can use
995  * generic page allocation functions to setup page table.
996  */
997 static void __init pt_ops_set_late(void)
998 {
999 	pt_ops.alloc_pte = alloc_pte_late;
1000 	pt_ops.get_pte_virt = get_pte_virt_late;
1001 #ifndef __PAGETABLE_PMD_FOLDED
1002 	pt_ops.alloc_pmd = alloc_pmd_late;
1003 	pt_ops.get_pmd_virt = get_pmd_virt_late;
1004 	pt_ops.alloc_pud = alloc_pud_late;
1005 	pt_ops.get_pud_virt = get_pud_virt_late;
1006 	pt_ops.alloc_p4d = alloc_p4d_late;
1007 	pt_ops.get_p4d_virt = get_p4d_virt_late;
1008 #endif
1009 }
1010 
1011 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1012 {
1013 	pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1014 
1015 	kernel_map.virt_addr = KERNEL_LINK_ADDR;
1016 	kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1017 
1018 #ifdef CONFIG_XIP_KERNEL
1019 	kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1020 	kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1021 
1022 	phys_ram_base = CONFIG_PHYS_RAM_BASE;
1023 	kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1024 	kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
1025 
1026 	kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1027 #else
1028 	kernel_map.phys_addr = (uintptr_t)(&_start);
1029 	kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1030 #endif
1031 
1032 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1033 	set_satp_mode(dtb_pa);
1034 #endif
1035 
1036 	/*
1037 	 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1038 	 * where we have the system memory layout: this allows us to align
1039 	 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1040 	 * for the linear mapping. This is only possible because the kernel
1041 	 * mapping lies outside the linear mapping.
1042 	 * In 32-bit however, as the kernel resides in the linear mapping,
1043 	 * setup_vm_final can not change the mapping established here,
1044 	 * otherwise the same kernel addresses would get mapped to different
1045 	 * physical addresses (if the start of dram is different from the
1046 	 * kernel physical address start).
1047 	 */
1048 	kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1049 				0UL : PAGE_OFFSET - kernel_map.phys_addr;
1050 	kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1051 
1052 	/*
1053 	 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
1054 	 * kernel, whereas for 64-bit kernel, the end of the virtual address
1055 	 * space is occupied by the modules/BPF/kernel mappings which reduces
1056 	 * the available size of the linear mapping.
1057 	 */
1058 	memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
1059 
1060 	/* Sanity check alignment and size */
1061 	BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1062 	BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1063 
1064 #ifdef CONFIG_64BIT
1065 	/*
1066 	 * The last 4K bytes of the addressable memory can not be mapped because
1067 	 * of IS_ERR_VALUE macro.
1068 	 */
1069 	BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1070 #endif
1071 
1072 #ifdef CONFIG_RELOCATABLE
1073 	/*
1074 	 * Early page table uses only one PUD, which makes it possible
1075 	 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1076 	 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1077 	 * since a part of the kernel would not get mapped.
1078 	 */
1079 	BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1080 	relocate_kernel();
1081 #endif
1082 
1083 	apply_early_boot_alternatives();
1084 	pt_ops_set_early();
1085 
1086 	/* Setup early PGD for fixmap */
1087 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
1088 			   fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1089 
1090 #ifndef __PAGETABLE_PMD_FOLDED
1091 	/* Setup fixmap P4D and PUD */
1092 	if (pgtable_l5_enabled)
1093 		create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1094 				   (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1095 	/* Setup fixmap PUD and PMD */
1096 	if (pgtable_l4_enabled)
1097 		create_pud_mapping(fixmap_pud, FIXADDR_START,
1098 				   (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1099 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1100 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1101 	/* Setup trampoline PGD and PMD */
1102 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1103 			   trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1104 	if (pgtable_l5_enabled)
1105 		create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1106 				   (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1107 	if (pgtable_l4_enabled)
1108 		create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1109 				   (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1110 #ifdef CONFIG_XIP_KERNEL
1111 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1112 			   kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1113 #else
1114 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1115 			   kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1116 #endif
1117 #else
1118 	/* Setup trampoline PGD */
1119 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1120 			   kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1121 #endif
1122 
1123 	/*
1124 	 * Setup early PGD covering entire kernel which will allow
1125 	 * us to reach paging_init(). We map all memory banks later
1126 	 * in setup_vm_final() below.
1127 	 */
1128 	create_kernel_page_table(early_pg_dir, true);
1129 
1130 	/* Setup early mapping for FDT early scan */
1131 	create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1132 
1133 	/*
1134 	 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1135 	 * range can not span multiple pmds.
1136 	 */
1137 	BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1138 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1139 
1140 #ifndef __PAGETABLE_PMD_FOLDED
1141 	/*
1142 	 * Early ioremap fixmap is already created as it lies within first 2MB
1143 	 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1144 	 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1145 	 * the user if not.
1146 	 */
1147 	fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1148 	fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1149 	if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1150 		WARN_ON(1);
1151 		pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1152 			pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1153 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1154 			fix_to_virt(FIX_BTMAP_BEGIN));
1155 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
1156 			fix_to_virt(FIX_BTMAP_END));
1157 
1158 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
1159 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
1160 	}
1161 #endif
1162 
1163 	pt_ops_set_fixmap();
1164 }
1165 
1166 static void __init create_linear_mapping_range(phys_addr_t start,
1167 					       phys_addr_t end,
1168 					       uintptr_t fixed_map_size)
1169 {
1170 	phys_addr_t pa;
1171 	uintptr_t va, map_size;
1172 
1173 	for (pa = start; pa < end; pa += map_size) {
1174 		va = (uintptr_t)__va(pa);
1175 		map_size = fixed_map_size ? fixed_map_size :
1176 					    best_map_size(pa, va, end - pa);
1177 
1178 		create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1179 				   pgprot_from_va(va));
1180 	}
1181 }
1182 
1183 static void __init create_linear_mapping_page_table(void)
1184 {
1185 	phys_addr_t start, end;
1186 	phys_addr_t kfence_pool __maybe_unused;
1187 	u64 i;
1188 
1189 #ifdef CONFIG_STRICT_KERNEL_RWX
1190 	phys_addr_t ktext_start = __pa_symbol(_start);
1191 	phys_addr_t ktext_size = __init_data_begin - _start;
1192 	phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1193 	phys_addr_t krodata_size = _data - __start_rodata;
1194 
1195 	/* Isolate kernel text and rodata so they don't get mapped with a PUD */
1196 	memblock_mark_nomap(ktext_start,  ktext_size);
1197 	memblock_mark_nomap(krodata_start, krodata_size);
1198 #endif
1199 
1200 #ifdef CONFIG_KFENCE
1201 	/*
1202 	 *  kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1203 	 *  before we setup the linear mapping so that we avoid using hugepages
1204 	 *  for this region.
1205 	 */
1206 	kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1207 	BUG_ON(!kfence_pool);
1208 
1209 	memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1210 	__kfence_pool = __va(kfence_pool);
1211 #endif
1212 
1213 	/* Map all memory banks in the linear mapping */
1214 	for_each_mem_range(i, &start, &end) {
1215 		if (start >= end)
1216 			break;
1217 		if (start <= __pa(PAGE_OFFSET) &&
1218 		    __pa(PAGE_OFFSET) < end)
1219 			start = __pa(PAGE_OFFSET);
1220 		if (end >= __pa(PAGE_OFFSET) + memory_limit)
1221 			end = __pa(PAGE_OFFSET) + memory_limit;
1222 
1223 		create_linear_mapping_range(start, end, 0);
1224 	}
1225 
1226 #ifdef CONFIG_STRICT_KERNEL_RWX
1227 	create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0);
1228 	create_linear_mapping_range(krodata_start,
1229 				    krodata_start + krodata_size, 0);
1230 
1231 	memblock_clear_nomap(ktext_start,  ktext_size);
1232 	memblock_clear_nomap(krodata_start, krodata_size);
1233 #endif
1234 
1235 #ifdef CONFIG_KFENCE
1236 	create_linear_mapping_range(kfence_pool,
1237 				    kfence_pool + KFENCE_POOL_SIZE,
1238 				    PAGE_SIZE);
1239 
1240 	memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1241 #endif
1242 }
1243 
1244 static void __init setup_vm_final(void)
1245 {
1246 	/* Setup swapper PGD for fixmap */
1247 #if !defined(CONFIG_64BIT)
1248 	/*
1249 	 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1250 	 * directly in swapper_pg_dir in addition to the pgd entry that points
1251 	 * to fixmap_pte.
1252 	 */
1253 	unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1254 
1255 	set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1256 #endif
1257 	create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1258 			   __pa_symbol(fixmap_pgd_next),
1259 			   PGDIR_SIZE, PAGE_TABLE);
1260 
1261 	/* Map the linear mapping */
1262 	create_linear_mapping_page_table();
1263 
1264 	/* Map the kernel */
1265 	if (IS_ENABLED(CONFIG_64BIT))
1266 		create_kernel_page_table(swapper_pg_dir, false);
1267 
1268 #ifdef CONFIG_KASAN
1269 	kasan_swapper_init();
1270 #endif
1271 
1272 	/* Clear fixmap PTE and PMD mappings */
1273 	clear_fixmap(FIX_PTE);
1274 	clear_fixmap(FIX_PMD);
1275 	clear_fixmap(FIX_PUD);
1276 	clear_fixmap(FIX_P4D);
1277 
1278 	/* Move to swapper page table */
1279 	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1280 	local_flush_tlb_all();
1281 
1282 	pt_ops_set_late();
1283 }
1284 #else
1285 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1286 {
1287 	dtb_early_va = (void *)dtb_pa;
1288 	dtb_early_pa = dtb_pa;
1289 }
1290 
1291 static inline void setup_vm_final(void)
1292 {
1293 }
1294 #endif /* CONFIG_MMU */
1295 
1296 /*
1297  * reserve_crashkernel() - reserves memory for crash kernel
1298  *
1299  * This function reserves memory area given in "crashkernel=" kernel command
1300  * line parameter. The memory reserved is used by dump capture kernel when
1301  * primary kernel is crashing.
1302  */
1303 static void __init reserve_crashkernel(void)
1304 {
1305 	unsigned long long crash_base = 0;
1306 	unsigned long long crash_size = 0;
1307 	unsigned long search_start = memblock_start_of_DRAM();
1308 	unsigned long search_end = memblock_end_of_DRAM();
1309 
1310 	int ret = 0;
1311 
1312 	if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1313 		return;
1314 	/*
1315 	 * Don't reserve a region for a crash kernel on a crash kernel
1316 	 * since it doesn't make much sense and we have limited memory
1317 	 * resources.
1318 	 */
1319 	if (is_kdump_kernel()) {
1320 		pr_info("crashkernel: ignoring reservation request\n");
1321 		return;
1322 	}
1323 
1324 	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
1325 				&crash_size, &crash_base);
1326 	if (ret || !crash_size)
1327 		return;
1328 
1329 	crash_size = PAGE_ALIGN(crash_size);
1330 
1331 	if (crash_base) {
1332 		search_start = crash_base;
1333 		search_end = crash_base + crash_size;
1334 	}
1335 
1336 	/*
1337 	 * Current riscv boot protocol requires 2MB alignment for
1338 	 * RV64 and 4MB alignment for RV32 (hugepage size)
1339 	 *
1340 	 * Try to alloc from 32bit addressible physical memory so that
1341 	 * swiotlb can work on the crash kernel.
1342 	 */
1343 	crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1344 					       search_start,
1345 					       min(search_end, (unsigned long)(SZ_4G - 1)));
1346 	if (crash_base == 0) {
1347 		/* Try again without restricting region to 32bit addressible memory */
1348 		crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1349 						search_start, search_end);
1350 		if (crash_base == 0) {
1351 			pr_warn("crashkernel: couldn't allocate %lldKB\n",
1352 				crash_size >> 10);
1353 			return;
1354 		}
1355 	}
1356 
1357 	pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1358 		crash_base, crash_base + crash_size, crash_size >> 20);
1359 
1360 	crashk_res.start = crash_base;
1361 	crashk_res.end = crash_base + crash_size - 1;
1362 }
1363 
1364 void __init paging_init(void)
1365 {
1366 	setup_bootmem();
1367 	setup_vm_final();
1368 
1369 	/* Depend on that Linear Mapping is ready */
1370 	memblock_allow_resize();
1371 }
1372 
1373 void __init misc_mem_init(void)
1374 {
1375 	early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1376 	arch_numa_init();
1377 	sparse_init();
1378 	zone_sizes_init();
1379 	reserve_crashkernel();
1380 	memblock_dump_all();
1381 }
1382 
1383 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1384 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1385 			       struct vmem_altmap *altmap)
1386 {
1387 	return vmemmap_populate_basepages(start, end, node, NULL);
1388 }
1389 #endif
1390 
1391 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1392 /*
1393  * Pre-allocates page-table pages for a specific area in the kernel
1394  * page-table. Only the level which needs to be synchronized between
1395  * all page-tables is allocated because the synchronization can be
1396  * expensive.
1397  */
1398 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1399 					       const char *area)
1400 {
1401 	unsigned long addr;
1402 	const char *lvl;
1403 
1404 	for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1405 		pgd_t *pgd = pgd_offset_k(addr);
1406 		p4d_t *p4d;
1407 		pud_t *pud;
1408 		pmd_t *pmd;
1409 
1410 		lvl = "p4d";
1411 		p4d = p4d_alloc(&init_mm, pgd, addr);
1412 		if (!p4d)
1413 			goto failed;
1414 
1415 		if (pgtable_l5_enabled)
1416 			continue;
1417 
1418 		lvl = "pud";
1419 		pud = pud_alloc(&init_mm, p4d, addr);
1420 		if (!pud)
1421 			goto failed;
1422 
1423 		if (pgtable_l4_enabled)
1424 			continue;
1425 
1426 		lvl = "pmd";
1427 		pmd = pmd_alloc(&init_mm, pud, addr);
1428 		if (!pmd)
1429 			goto failed;
1430 	}
1431 	return;
1432 
1433 failed:
1434 	/*
1435 	 * The pages have to be there now or they will be missing in
1436 	 * process page-tables later.
1437 	 */
1438 	panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1439 }
1440 
1441 void __init pgtable_cache_init(void)
1442 {
1443 	preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1444 	if (IS_ENABLED(CONFIG_MODULES))
1445 		preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
1446 }
1447 #endif
1448