xref: /openbmc/linux/arch/riscv/mm/init.c (revision e855afd715656a9f25cf62fa68d99c33213b83b7)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012 Regents of the University of California
4  * Copyright (C) 2019 Western Digital Corporation or its affiliates.
5  * Copyright (C) 2020 FORTH-ICS/CARV
6  *  Nick Kossifidis <mick@ics.forth.gr>
7  */
8 
9 #include <linux/init.h>
10 #include <linux/mm.h>
11 #include <linux/memblock.h>
12 #include <linux/initrd.h>
13 #include <linux/swap.h>
14 #include <linux/swiotlb.h>
15 #include <linux/sizes.h>
16 #include <linux/of_fdt.h>
17 #include <linux/of_reserved_mem.h>
18 #include <linux/libfdt.h>
19 #include <linux/set_memory.h>
20 #include <linux/dma-map-ops.h>
21 #include <linux/crash_dump.h>
22 #include <linux/hugetlb.h>
23 #ifdef CONFIG_RELOCATABLE
24 #include <linux/elf.h>
25 #endif
26 #include <linux/kfence.h>
27 
28 #include <asm/fixmap.h>
29 #include <asm/io.h>
30 #include <asm/numa.h>
31 #include <asm/pgtable.h>
32 #include <asm/ptdump.h>
33 #include <asm/sections.h>
34 #include <asm/soc.h>
35 #include <asm/tlbflush.h>
36 
37 #include "../kernel/head.h"
38 
39 struct kernel_mapping kernel_map __ro_after_init;
40 EXPORT_SYMBOL(kernel_map);
41 #ifdef CONFIG_XIP_KERNEL
42 #define kernel_map	(*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
43 #endif
44 
45 #ifdef CONFIG_64BIT
46 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39;
47 #else
48 u64 satp_mode __ro_after_init = SATP_MODE_32;
49 #endif
50 EXPORT_SYMBOL(satp_mode);
51 
52 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
53 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
54 EXPORT_SYMBOL(pgtable_l4_enabled);
55 EXPORT_SYMBOL(pgtable_l5_enabled);
56 
57 phys_addr_t phys_ram_base __ro_after_init;
58 EXPORT_SYMBOL(phys_ram_base);
59 
60 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
61 							__page_aligned_bss;
62 EXPORT_SYMBOL(empty_zero_page);
63 
64 extern char _start[];
65 void *_dtb_early_va __initdata;
66 uintptr_t _dtb_early_pa __initdata;
67 
68 static phys_addr_t dma32_phys_limit __initdata;
69 
70 static void __init zone_sizes_init(void)
71 {
72 	unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
73 
74 #ifdef CONFIG_ZONE_DMA32
75 	max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
76 #endif
77 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
78 
79 	free_area_init(max_zone_pfns);
80 }
81 
82 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
83 
84 #define LOG2_SZ_1K  ilog2(SZ_1K)
85 #define LOG2_SZ_1M  ilog2(SZ_1M)
86 #define LOG2_SZ_1G  ilog2(SZ_1G)
87 #define LOG2_SZ_1T  ilog2(SZ_1T)
88 
89 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
90 {
91 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld kB)\n", name, b, t,
92 		  (((t) - (b)) >> LOG2_SZ_1K));
93 }
94 
95 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
96 {
97 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld MB)\n", name, b, t,
98 		  (((t) - (b)) >> LOG2_SZ_1M));
99 }
100 
101 static inline void print_mlg(char *name, unsigned long b, unsigned long t)
102 {
103 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld GB)\n", name, b, t,
104 		   (((t) - (b)) >> LOG2_SZ_1G));
105 }
106 
107 #ifdef CONFIG_64BIT
108 static inline void print_mlt(char *name, unsigned long b, unsigned long t)
109 {
110 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld TB)\n", name, b, t,
111 		   (((t) - (b)) >> LOG2_SZ_1T));
112 }
113 #else
114 #define print_mlt(n, b, t) do {} while (0)
115 #endif
116 
117 static inline void print_ml(char *name, unsigned long b, unsigned long t)
118 {
119 	unsigned long diff = t - b;
120 
121 	if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10)
122 		print_mlt(name, b, t);
123 	else if ((diff >> LOG2_SZ_1G) >= 10)
124 		print_mlg(name, b, t);
125 	else if ((diff >> LOG2_SZ_1M) >= 10)
126 		print_mlm(name, b, t);
127 	else
128 		print_mlk(name, b, t);
129 }
130 
131 static void __init print_vm_layout(void)
132 {
133 	pr_notice("Virtual kernel memory layout:\n");
134 	print_ml("fixmap", (unsigned long)FIXADDR_START,
135 		(unsigned long)FIXADDR_TOP);
136 	print_ml("pci io", (unsigned long)PCI_IO_START,
137 		(unsigned long)PCI_IO_END);
138 	print_ml("vmemmap", (unsigned long)VMEMMAP_START,
139 		(unsigned long)VMEMMAP_END);
140 	print_ml("vmalloc", (unsigned long)VMALLOC_START,
141 		(unsigned long)VMALLOC_END);
142 #ifdef CONFIG_64BIT
143 	print_ml("modules", (unsigned long)MODULES_VADDR,
144 		(unsigned long)MODULES_END);
145 #endif
146 	print_ml("lowmem", (unsigned long)PAGE_OFFSET,
147 		(unsigned long)high_memory);
148 	if (IS_ENABLED(CONFIG_64BIT)) {
149 #ifdef CONFIG_KASAN
150 		print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
151 #endif
152 
153 		print_ml("kernel", (unsigned long)kernel_map.virt_addr,
154 			 (unsigned long)ADDRESS_SPACE_END);
155 	}
156 }
157 #else
158 static void print_vm_layout(void) { }
159 #endif /* CONFIG_DEBUG_VM */
160 
161 void __init mem_init(void)
162 {
163 #ifdef CONFIG_FLATMEM
164 	BUG_ON(!mem_map);
165 #endif /* CONFIG_FLATMEM */
166 
167 	swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE);
168 	memblock_free_all();
169 
170 	print_vm_layout();
171 }
172 
173 /* Limit the memory size via mem. */
174 static phys_addr_t memory_limit;
175 
176 static int __init early_mem(char *p)
177 {
178 	u64 size;
179 
180 	if (!p)
181 		return 1;
182 
183 	size = memparse(p, &p) & PAGE_MASK;
184 	memory_limit = min_t(u64, size, memory_limit);
185 
186 	pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
187 
188 	return 0;
189 }
190 early_param("mem", early_mem);
191 
192 static void __init setup_bootmem(void)
193 {
194 	phys_addr_t vmlinux_end = __pa_symbol(&_end);
195 	phys_addr_t max_mapped_addr;
196 	phys_addr_t phys_ram_end, vmlinux_start;
197 
198 	if (IS_ENABLED(CONFIG_XIP_KERNEL))
199 		vmlinux_start = __pa_symbol(&_sdata);
200 	else
201 		vmlinux_start = __pa_symbol(&_start);
202 
203 	memblock_enforce_memory_limit(memory_limit);
204 
205 	/*
206 	 * Make sure we align the reservation on PMD_SIZE since we will
207 	 * map the kernel in the linear mapping as read-only: we do not want
208 	 * any allocation to happen between _end and the next pmd aligned page.
209 	 */
210 	if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
211 		vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
212 	/*
213 	 * Reserve from the start of the kernel to the end of the kernel
214 	 */
215 	memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
216 
217 	phys_ram_end = memblock_end_of_DRAM();
218 
219 	/*
220 	 * Make sure we align the start of the memory on a PMD boundary so that
221 	 * at worst, we map the linear mapping with PMD mappings.
222 	 */
223 	if (!IS_ENABLED(CONFIG_XIP_KERNEL))
224 		phys_ram_base = memblock_start_of_DRAM() & PMD_MASK;
225 
226 	/*
227 	 * In 64-bit, any use of __va/__pa before this point is wrong as we
228 	 * did not know the start of DRAM before.
229 	 */
230 	if (IS_ENABLED(CONFIG_64BIT))
231 		kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base;
232 
233 	/*
234 	 * memblock allocator is not aware of the fact that last 4K bytes of
235 	 * the addressable memory can not be mapped because of IS_ERR_VALUE
236 	 * macro. Make sure that last 4k bytes are not usable by memblock
237 	 * if end of dram is equal to maximum addressable memory.  For 64-bit
238 	 * kernel, this problem can't happen here as the end of the virtual
239 	 * address space is occupied by the kernel mapping then this check must
240 	 * be done as soon as the kernel mapping base address is determined.
241 	 */
242 	if (!IS_ENABLED(CONFIG_64BIT)) {
243 		max_mapped_addr = __pa(~(ulong)0);
244 		if (max_mapped_addr == (phys_ram_end - 1))
245 			memblock_set_current_limit(max_mapped_addr - 4096);
246 	}
247 
248 	min_low_pfn = PFN_UP(phys_ram_base);
249 	max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
250 	high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
251 
252 	dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
253 	set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
254 
255 	reserve_initrd_mem();
256 
257 	/*
258 	 * No allocation should be done before reserving the memory as defined
259 	 * in the device tree, otherwise the allocation could end up in a
260 	 * reserved region.
261 	 */
262 	early_init_fdt_scan_reserved_mem();
263 
264 	/*
265 	 * If DTB is built in, no need to reserve its memblock.
266 	 * Otherwise, do reserve it but avoid using
267 	 * early_init_fdt_reserve_self() since __pa() does
268 	 * not work for DTB pointers that are fixmap addresses
269 	 */
270 	if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
271 		memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
272 
273 	dma_contiguous_reserve(dma32_phys_limit);
274 	if (IS_ENABLED(CONFIG_64BIT))
275 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
276 }
277 
278 #ifdef CONFIG_MMU
279 struct pt_alloc_ops pt_ops __initdata;
280 
281 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
282 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
283 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
284 
285 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
286 
287 #ifdef CONFIG_XIP_KERNEL
288 #define pt_ops			(*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
289 #define trampoline_pg_dir      ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
290 #define fixmap_pte             ((pte_t *)XIP_FIXUP(fixmap_pte))
291 #define early_pg_dir           ((pgd_t *)XIP_FIXUP(early_pg_dir))
292 #endif /* CONFIG_XIP_KERNEL */
293 
294 static const pgprot_t protection_map[16] = {
295 	[VM_NONE]					= PAGE_NONE,
296 	[VM_READ]					= PAGE_READ,
297 	[VM_WRITE]					= PAGE_COPY,
298 	[VM_WRITE | VM_READ]				= PAGE_COPY,
299 	[VM_EXEC]					= PAGE_EXEC,
300 	[VM_EXEC | VM_READ]				= PAGE_READ_EXEC,
301 	[VM_EXEC | VM_WRITE]				= PAGE_COPY_EXEC,
302 	[VM_EXEC | VM_WRITE | VM_READ]			= PAGE_COPY_EXEC,
303 	[VM_SHARED]					= PAGE_NONE,
304 	[VM_SHARED | VM_READ]				= PAGE_READ,
305 	[VM_SHARED | VM_WRITE]				= PAGE_SHARED,
306 	[VM_SHARED | VM_WRITE | VM_READ]		= PAGE_SHARED,
307 	[VM_SHARED | VM_EXEC]				= PAGE_EXEC,
308 	[VM_SHARED | VM_EXEC | VM_READ]			= PAGE_READ_EXEC,
309 	[VM_SHARED | VM_EXEC | VM_WRITE]		= PAGE_SHARED_EXEC,
310 	[VM_SHARED | VM_EXEC | VM_WRITE | VM_READ]	= PAGE_SHARED_EXEC
311 };
312 DECLARE_VM_GET_PAGE_PROT
313 
314 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
315 {
316 	unsigned long addr = __fix_to_virt(idx);
317 	pte_t *ptep;
318 
319 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
320 
321 	ptep = &fixmap_pte[pte_index(addr)];
322 
323 	if (pgprot_val(prot))
324 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
325 	else
326 		pte_clear(&init_mm, addr, ptep);
327 	local_flush_tlb_page(addr);
328 }
329 
330 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
331 {
332 	return (pte_t *)((uintptr_t)pa);
333 }
334 
335 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
336 {
337 	clear_fixmap(FIX_PTE);
338 	return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
339 }
340 
341 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
342 {
343 	return (pte_t *) __va(pa);
344 }
345 
346 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
347 {
348 	/*
349 	 * We only create PMD or PGD early mappings so we
350 	 * should never reach here with MMU disabled.
351 	 */
352 	BUG();
353 }
354 
355 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
356 {
357 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
358 }
359 
360 static phys_addr_t __init alloc_pte_late(uintptr_t va)
361 {
362 	unsigned long vaddr;
363 
364 	vaddr = __get_free_page(GFP_KERNEL);
365 	BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page((void *)vaddr)));
366 
367 	return __pa(vaddr);
368 }
369 
370 static void __init create_pte_mapping(pte_t *ptep,
371 				      uintptr_t va, phys_addr_t pa,
372 				      phys_addr_t sz, pgprot_t prot)
373 {
374 	uintptr_t pte_idx = pte_index(va);
375 
376 	BUG_ON(sz != PAGE_SIZE);
377 
378 	if (pte_none(ptep[pte_idx]))
379 		ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
380 }
381 
382 #ifndef __PAGETABLE_PMD_FOLDED
383 
384 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
385 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
386 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
387 
388 #ifdef CONFIG_XIP_KERNEL
389 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
390 #define fixmap_pmd     ((pmd_t *)XIP_FIXUP(fixmap_pmd))
391 #define early_pmd      ((pmd_t *)XIP_FIXUP(early_pmd))
392 #endif /* CONFIG_XIP_KERNEL */
393 
394 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
395 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
396 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
397 
398 #ifdef CONFIG_XIP_KERNEL
399 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
400 #define fixmap_p4d     ((p4d_t *)XIP_FIXUP(fixmap_p4d))
401 #define early_p4d      ((p4d_t *)XIP_FIXUP(early_p4d))
402 #endif /* CONFIG_XIP_KERNEL */
403 
404 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
405 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
406 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
407 
408 #ifdef CONFIG_XIP_KERNEL
409 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
410 #define fixmap_pud     ((pud_t *)XIP_FIXUP(fixmap_pud))
411 #define early_pud      ((pud_t *)XIP_FIXUP(early_pud))
412 #endif /* CONFIG_XIP_KERNEL */
413 
414 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
415 {
416 	/* Before MMU is enabled */
417 	return (pmd_t *)((uintptr_t)pa);
418 }
419 
420 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
421 {
422 	clear_fixmap(FIX_PMD);
423 	return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
424 }
425 
426 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
427 {
428 	return (pmd_t *) __va(pa);
429 }
430 
431 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
432 {
433 	BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
434 
435 	return (uintptr_t)early_pmd;
436 }
437 
438 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
439 {
440 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
441 }
442 
443 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
444 {
445 	unsigned long vaddr;
446 
447 	vaddr = __get_free_page(GFP_KERNEL);
448 	BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page((void *)vaddr)));
449 
450 	return __pa(vaddr);
451 }
452 
453 static void __init create_pmd_mapping(pmd_t *pmdp,
454 				      uintptr_t va, phys_addr_t pa,
455 				      phys_addr_t sz, pgprot_t prot)
456 {
457 	pte_t *ptep;
458 	phys_addr_t pte_phys;
459 	uintptr_t pmd_idx = pmd_index(va);
460 
461 	if (sz == PMD_SIZE) {
462 		if (pmd_none(pmdp[pmd_idx]))
463 			pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
464 		return;
465 	}
466 
467 	if (pmd_none(pmdp[pmd_idx])) {
468 		pte_phys = pt_ops.alloc_pte(va);
469 		pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
470 		ptep = pt_ops.get_pte_virt(pte_phys);
471 		memset(ptep, 0, PAGE_SIZE);
472 	} else {
473 		pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
474 		ptep = pt_ops.get_pte_virt(pte_phys);
475 	}
476 
477 	create_pte_mapping(ptep, va, pa, sz, prot);
478 }
479 
480 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
481 {
482 	return (pud_t *)((uintptr_t)pa);
483 }
484 
485 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
486 {
487 	clear_fixmap(FIX_PUD);
488 	return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
489 }
490 
491 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
492 {
493 	return (pud_t *)__va(pa);
494 }
495 
496 static phys_addr_t __init alloc_pud_early(uintptr_t va)
497 {
498 	/* Only one PUD is available for early mapping */
499 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
500 
501 	return (uintptr_t)early_pud;
502 }
503 
504 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
505 {
506 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
507 }
508 
509 static phys_addr_t alloc_pud_late(uintptr_t va)
510 {
511 	unsigned long vaddr;
512 
513 	vaddr = __get_free_page(GFP_KERNEL);
514 	BUG_ON(!vaddr);
515 	return __pa(vaddr);
516 }
517 
518 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
519 {
520 	return (p4d_t *)((uintptr_t)pa);
521 }
522 
523 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
524 {
525 	clear_fixmap(FIX_P4D);
526 	return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
527 }
528 
529 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
530 {
531 	return (p4d_t *)__va(pa);
532 }
533 
534 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
535 {
536 	/* Only one P4D is available for early mapping */
537 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
538 
539 	return (uintptr_t)early_p4d;
540 }
541 
542 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
543 {
544 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
545 }
546 
547 static phys_addr_t alloc_p4d_late(uintptr_t va)
548 {
549 	unsigned long vaddr;
550 
551 	vaddr = __get_free_page(GFP_KERNEL);
552 	BUG_ON(!vaddr);
553 	return __pa(vaddr);
554 }
555 
556 static void __init create_pud_mapping(pud_t *pudp,
557 				      uintptr_t va, phys_addr_t pa,
558 				      phys_addr_t sz, pgprot_t prot)
559 {
560 	pmd_t *nextp;
561 	phys_addr_t next_phys;
562 	uintptr_t pud_index = pud_index(va);
563 
564 	if (sz == PUD_SIZE) {
565 		if (pud_val(pudp[pud_index]) == 0)
566 			pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
567 		return;
568 	}
569 
570 	if (pud_val(pudp[pud_index]) == 0) {
571 		next_phys = pt_ops.alloc_pmd(va);
572 		pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
573 		nextp = pt_ops.get_pmd_virt(next_phys);
574 		memset(nextp, 0, PAGE_SIZE);
575 	} else {
576 		next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
577 		nextp = pt_ops.get_pmd_virt(next_phys);
578 	}
579 
580 	create_pmd_mapping(nextp, va, pa, sz, prot);
581 }
582 
583 static void __init create_p4d_mapping(p4d_t *p4dp,
584 				      uintptr_t va, phys_addr_t pa,
585 				      phys_addr_t sz, pgprot_t prot)
586 {
587 	pud_t *nextp;
588 	phys_addr_t next_phys;
589 	uintptr_t p4d_index = p4d_index(va);
590 
591 	if (sz == P4D_SIZE) {
592 		if (p4d_val(p4dp[p4d_index]) == 0)
593 			p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
594 		return;
595 	}
596 
597 	if (p4d_val(p4dp[p4d_index]) == 0) {
598 		next_phys = pt_ops.alloc_pud(va);
599 		p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
600 		nextp = pt_ops.get_pud_virt(next_phys);
601 		memset(nextp, 0, PAGE_SIZE);
602 	} else {
603 		next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
604 		nextp = pt_ops.get_pud_virt(next_phys);
605 	}
606 
607 	create_pud_mapping(nextp, va, pa, sz, prot);
608 }
609 
610 #define pgd_next_t		p4d_t
611 #define alloc_pgd_next(__va)	(pgtable_l5_enabled ?			\
612 		pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ?		\
613 		pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
614 #define get_pgd_next_virt(__pa)	(pgtable_l5_enabled ?			\
615 		pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ?	\
616 		pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
617 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
618 				(pgtable_l5_enabled ?			\
619 		create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
620 				(pgtable_l4_enabled ?			\
621 		create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) :	\
622 		create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
623 #define fixmap_pgd_next		(pgtable_l5_enabled ?			\
624 		(uintptr_t)fixmap_p4d : (pgtable_l4_enabled ?		\
625 		(uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
626 #define trampoline_pgd_next	(pgtable_l5_enabled ?			\
627 		(uintptr_t)trampoline_p4d : (pgtable_l4_enabled ?	\
628 		(uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
629 #else
630 #define pgd_next_t		pte_t
631 #define alloc_pgd_next(__va)	pt_ops.alloc_pte(__va)
632 #define get_pgd_next_virt(__pa)	pt_ops.get_pte_virt(__pa)
633 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
634 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
635 #define fixmap_pgd_next		((uintptr_t)fixmap_pte)
636 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
637 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
638 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
639 #endif /* __PAGETABLE_PMD_FOLDED */
640 
641 void __init create_pgd_mapping(pgd_t *pgdp,
642 				      uintptr_t va, phys_addr_t pa,
643 				      phys_addr_t sz, pgprot_t prot)
644 {
645 	pgd_next_t *nextp;
646 	phys_addr_t next_phys;
647 	uintptr_t pgd_idx = pgd_index(va);
648 
649 	if (sz == PGDIR_SIZE) {
650 		if (pgd_val(pgdp[pgd_idx]) == 0)
651 			pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
652 		return;
653 	}
654 
655 	if (pgd_val(pgdp[pgd_idx]) == 0) {
656 		next_phys = alloc_pgd_next(va);
657 		pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
658 		nextp = get_pgd_next_virt(next_phys);
659 		memset(nextp, 0, PAGE_SIZE);
660 	} else {
661 		next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
662 		nextp = get_pgd_next_virt(next_phys);
663 	}
664 
665 	create_pgd_next_mapping(nextp, va, pa, sz, prot);
666 }
667 
668 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va,
669 				      phys_addr_t size)
670 {
671 	if (!(pa & (PGDIR_SIZE - 1)) && !(va & (PGDIR_SIZE - 1)) && size >= PGDIR_SIZE)
672 		return PGDIR_SIZE;
673 
674 	if (!(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
675 		return P4D_SIZE;
676 
677 	if (!(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
678 		return PUD_SIZE;
679 
680 	if (!(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
681 		return PMD_SIZE;
682 
683 	return PAGE_SIZE;
684 }
685 
686 #ifdef CONFIG_XIP_KERNEL
687 #define phys_ram_base  (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
688 extern char _xiprom[], _exiprom[], __data_loc;
689 
690 /* called from head.S with MMU off */
691 asmlinkage void __init __copy_data(void)
692 {
693 	void *from = (void *)(&__data_loc);
694 	void *to = (void *)CONFIG_PHYS_RAM_BASE;
695 	size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
696 
697 	memcpy(to, from, sz);
698 }
699 #endif
700 
701 #ifdef CONFIG_STRICT_KERNEL_RWX
702 static __init pgprot_t pgprot_from_va(uintptr_t va)
703 {
704 	if (is_va_kernel_text(va))
705 		return PAGE_KERNEL_READ_EXEC;
706 
707 	/*
708 	 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
709 	 * we must protect its linear mapping alias from being executed and
710 	 * written.
711 	 * And rodata section is marked readonly in mark_rodata_ro.
712 	 */
713 	if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
714 		return PAGE_KERNEL_READ;
715 
716 	return PAGE_KERNEL;
717 }
718 
719 void mark_rodata_ro(void)
720 {
721 	set_kernel_memory(__start_rodata, _data, set_memory_ro);
722 	if (IS_ENABLED(CONFIG_64BIT))
723 		set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
724 				  set_memory_ro);
725 
726 	debug_checkwx();
727 }
728 #else
729 static __init pgprot_t pgprot_from_va(uintptr_t va)
730 {
731 	if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
732 		return PAGE_KERNEL;
733 
734 	return PAGE_KERNEL_EXEC;
735 }
736 #endif /* CONFIG_STRICT_KERNEL_RWX */
737 
738 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
739 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
740 
741 static void __init disable_pgtable_l5(void)
742 {
743 	pgtable_l5_enabled = false;
744 	kernel_map.page_offset = PAGE_OFFSET_L4;
745 	satp_mode = SATP_MODE_48;
746 }
747 
748 static void __init disable_pgtable_l4(void)
749 {
750 	pgtable_l4_enabled = false;
751 	kernel_map.page_offset = PAGE_OFFSET_L3;
752 	satp_mode = SATP_MODE_39;
753 }
754 
755 static int __init print_no4lvl(char *p)
756 {
757 	pr_info("Disabled 4-level and 5-level paging");
758 	return 0;
759 }
760 early_param("no4lvl", print_no4lvl);
761 
762 static int __init print_no5lvl(char *p)
763 {
764 	pr_info("Disabled 5-level paging");
765 	return 0;
766 }
767 early_param("no5lvl", print_no5lvl);
768 
769 /*
770  * There is a simple way to determine if 4-level is supported by the
771  * underlying hardware: establish 1:1 mapping in 4-level page table mode
772  * then read SATP to see if the configuration was taken into account
773  * meaning sv48 is supported.
774  */
775 static __init void set_satp_mode(uintptr_t dtb_pa)
776 {
777 	u64 identity_satp, hw_satp;
778 	uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
779 	u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
780 
781 	if (satp_mode_cmdline == SATP_MODE_57) {
782 		disable_pgtable_l5();
783 	} else if (satp_mode_cmdline == SATP_MODE_48) {
784 		disable_pgtable_l5();
785 		disable_pgtable_l4();
786 		return;
787 	}
788 
789 	create_p4d_mapping(early_p4d,
790 			set_satp_mode_pmd, (uintptr_t)early_pud,
791 			P4D_SIZE, PAGE_TABLE);
792 	create_pud_mapping(early_pud,
793 			   set_satp_mode_pmd, (uintptr_t)early_pmd,
794 			   PUD_SIZE, PAGE_TABLE);
795 	/* Handle the case where set_satp_mode straddles 2 PMDs */
796 	create_pmd_mapping(early_pmd,
797 			   set_satp_mode_pmd, set_satp_mode_pmd,
798 			   PMD_SIZE, PAGE_KERNEL_EXEC);
799 	create_pmd_mapping(early_pmd,
800 			   set_satp_mode_pmd + PMD_SIZE,
801 			   set_satp_mode_pmd + PMD_SIZE,
802 			   PMD_SIZE, PAGE_KERNEL_EXEC);
803 retry:
804 	create_pgd_mapping(early_pg_dir,
805 			   set_satp_mode_pmd,
806 			   pgtable_l5_enabled ?
807 				(uintptr_t)early_p4d : (uintptr_t)early_pud,
808 			   PGDIR_SIZE, PAGE_TABLE);
809 
810 	identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
811 
812 	local_flush_tlb_all();
813 	csr_write(CSR_SATP, identity_satp);
814 	hw_satp = csr_swap(CSR_SATP, 0ULL);
815 	local_flush_tlb_all();
816 
817 	if (hw_satp != identity_satp) {
818 		if (pgtable_l5_enabled) {
819 			disable_pgtable_l5();
820 			memset(early_pg_dir, 0, PAGE_SIZE);
821 			goto retry;
822 		}
823 		disable_pgtable_l4();
824 	}
825 
826 	memset(early_pg_dir, 0, PAGE_SIZE);
827 	memset(early_p4d, 0, PAGE_SIZE);
828 	memset(early_pud, 0, PAGE_SIZE);
829 	memset(early_pmd, 0, PAGE_SIZE);
830 }
831 #endif
832 
833 /*
834  * setup_vm() is called from head.S with MMU-off.
835  *
836  * Following requirements should be honoured for setup_vm() to work
837  * correctly:
838  * 1) It should use PC-relative addressing for accessing kernel symbols.
839  *    To achieve this we always use GCC cmodel=medany.
840  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
841  *    so disable compiler instrumentation when FTRACE is enabled.
842  *
843  * Currently, the above requirements are honoured by using custom CFLAGS
844  * for init.o in mm/Makefile.
845  */
846 
847 #ifndef __riscv_cmodel_medany
848 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
849 #endif
850 
851 #ifdef CONFIG_RELOCATABLE
852 extern unsigned long __rela_dyn_start, __rela_dyn_end;
853 
854 static void __init relocate_kernel(void)
855 {
856 	Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
857 	/*
858 	 * This holds the offset between the linked virtual address and the
859 	 * relocated virtual address.
860 	 */
861 	uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
862 	/*
863 	 * This holds the offset between kernel linked virtual address and
864 	 * physical address.
865 	 */
866 	uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
867 
868 	for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
869 		Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
870 		Elf64_Addr relocated_addr = rela->r_addend;
871 
872 		if (rela->r_info != R_RISCV_RELATIVE)
873 			continue;
874 
875 		/*
876 		 * Make sure to not relocate vdso symbols like rt_sigreturn
877 		 * which are linked from the address 0 in vmlinux since
878 		 * vdso symbol addresses are actually used as an offset from
879 		 * mm->context.vdso in VDSO_OFFSET macro.
880 		 */
881 		if (relocated_addr >= KERNEL_LINK_ADDR)
882 			relocated_addr += reloc_offset;
883 
884 		*(Elf64_Addr *)addr = relocated_addr;
885 	}
886 }
887 #endif /* CONFIG_RELOCATABLE */
888 
889 #ifdef CONFIG_XIP_KERNEL
890 static void __init create_kernel_page_table(pgd_t *pgdir,
891 					    __always_unused bool early)
892 {
893 	uintptr_t va, end_va;
894 
895 	/* Map the flash resident part */
896 	end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
897 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
898 		create_pgd_mapping(pgdir, va,
899 				   kernel_map.xiprom + (va - kernel_map.virt_addr),
900 				   PMD_SIZE, PAGE_KERNEL_EXEC);
901 
902 	/* Map the data in RAM */
903 	end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
904 	for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
905 		create_pgd_mapping(pgdir, va,
906 				   kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
907 				   PMD_SIZE, PAGE_KERNEL);
908 }
909 #else
910 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
911 {
912 	uintptr_t va, end_va;
913 
914 	end_va = kernel_map.virt_addr + kernel_map.size;
915 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
916 		create_pgd_mapping(pgdir, va,
917 				   kernel_map.phys_addr + (va - kernel_map.virt_addr),
918 				   PMD_SIZE,
919 				   early ?
920 					PAGE_KERNEL_EXEC : pgprot_from_va(va));
921 }
922 #endif
923 
924 /*
925  * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
926  * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
927  * entry.
928  */
929 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
930 					       uintptr_t dtb_pa)
931 {
932 #ifndef CONFIG_BUILTIN_DTB
933 	uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
934 
935 	/* Make sure the fdt fixmap address is always aligned on PMD size */
936 	BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
937 
938 	/* In 32-bit only, the fdt lies in its own PGD */
939 	if (!IS_ENABLED(CONFIG_64BIT)) {
940 		create_pgd_mapping(early_pg_dir, fix_fdt_va,
941 				   pa, MAX_FDT_SIZE, PAGE_KERNEL);
942 	} else {
943 		create_pmd_mapping(fixmap_pmd, fix_fdt_va,
944 				   pa, PMD_SIZE, PAGE_KERNEL);
945 		create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
946 				   pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
947 	}
948 
949 	dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
950 #else
951 	/*
952 	 * For 64-bit kernel, __va can't be used since it would return a linear
953 	 * mapping address whereas dtb_early_va will be used before
954 	 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
955 	 * kernel is mapped in the linear mapping, that makes no difference.
956 	 */
957 	dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
958 #endif
959 
960 	dtb_early_pa = dtb_pa;
961 }
962 
963 /*
964  * MMU is not enabled, the page tables are allocated directly using
965  * early_pmd/pud/p4d and the address returned is the physical one.
966  */
967 static void __init pt_ops_set_early(void)
968 {
969 	pt_ops.alloc_pte = alloc_pte_early;
970 	pt_ops.get_pte_virt = get_pte_virt_early;
971 #ifndef __PAGETABLE_PMD_FOLDED
972 	pt_ops.alloc_pmd = alloc_pmd_early;
973 	pt_ops.get_pmd_virt = get_pmd_virt_early;
974 	pt_ops.alloc_pud = alloc_pud_early;
975 	pt_ops.get_pud_virt = get_pud_virt_early;
976 	pt_ops.alloc_p4d = alloc_p4d_early;
977 	pt_ops.get_p4d_virt = get_p4d_virt_early;
978 #endif
979 }
980 
981 /*
982  * MMU is enabled but page table setup is not complete yet.
983  * fixmap page table alloc functions must be used as a means to temporarily
984  * map the allocated physical pages since the linear mapping does not exist yet.
985  *
986  * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
987  * but it will be used as described above.
988  */
989 static void __init pt_ops_set_fixmap(void)
990 {
991 	pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
992 	pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
993 #ifndef __PAGETABLE_PMD_FOLDED
994 	pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
995 	pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
996 	pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
997 	pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
998 	pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
999 	pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
1000 #endif
1001 }
1002 
1003 /*
1004  * MMU is enabled and page table setup is complete, so from now, we can use
1005  * generic page allocation functions to setup page table.
1006  */
1007 static void __init pt_ops_set_late(void)
1008 {
1009 	pt_ops.alloc_pte = alloc_pte_late;
1010 	pt_ops.get_pte_virt = get_pte_virt_late;
1011 #ifndef __PAGETABLE_PMD_FOLDED
1012 	pt_ops.alloc_pmd = alloc_pmd_late;
1013 	pt_ops.get_pmd_virt = get_pmd_virt_late;
1014 	pt_ops.alloc_pud = alloc_pud_late;
1015 	pt_ops.get_pud_virt = get_pud_virt_late;
1016 	pt_ops.alloc_p4d = alloc_p4d_late;
1017 	pt_ops.get_p4d_virt = get_p4d_virt_late;
1018 #endif
1019 }
1020 
1021 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1022 {
1023 	pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1024 
1025 	kernel_map.virt_addr = KERNEL_LINK_ADDR;
1026 	kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1027 
1028 #ifdef CONFIG_XIP_KERNEL
1029 	kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1030 	kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1031 
1032 	phys_ram_base = CONFIG_PHYS_RAM_BASE;
1033 	kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1034 	kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
1035 
1036 	kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1037 #else
1038 	kernel_map.phys_addr = (uintptr_t)(&_start);
1039 	kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1040 #endif
1041 
1042 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1043 	set_satp_mode(dtb_pa);
1044 #endif
1045 
1046 	/*
1047 	 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1048 	 * where we have the system memory layout: this allows us to align
1049 	 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1050 	 * for the linear mapping. This is only possible because the kernel
1051 	 * mapping lies outside the linear mapping.
1052 	 * In 32-bit however, as the kernel resides in the linear mapping,
1053 	 * setup_vm_final can not change the mapping established here,
1054 	 * otherwise the same kernel addresses would get mapped to different
1055 	 * physical addresses (if the start of dram is different from the
1056 	 * kernel physical address start).
1057 	 */
1058 	kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1059 				0UL : PAGE_OFFSET - kernel_map.phys_addr;
1060 	kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1061 
1062 	/*
1063 	 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
1064 	 * kernel, whereas for 64-bit kernel, the end of the virtual address
1065 	 * space is occupied by the modules/BPF/kernel mappings which reduces
1066 	 * the available size of the linear mapping.
1067 	 */
1068 	memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
1069 
1070 	/* Sanity check alignment and size */
1071 	BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1072 	BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1073 
1074 #ifdef CONFIG_64BIT
1075 	/*
1076 	 * The last 4K bytes of the addressable memory can not be mapped because
1077 	 * of IS_ERR_VALUE macro.
1078 	 */
1079 	BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1080 #endif
1081 
1082 #ifdef CONFIG_RELOCATABLE
1083 	/*
1084 	 * Early page table uses only one PUD, which makes it possible
1085 	 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1086 	 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1087 	 * since a part of the kernel would not get mapped.
1088 	 */
1089 	BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1090 	relocate_kernel();
1091 #endif
1092 
1093 	apply_early_boot_alternatives();
1094 	pt_ops_set_early();
1095 
1096 	/* Setup early PGD for fixmap */
1097 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
1098 			   fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1099 
1100 #ifndef __PAGETABLE_PMD_FOLDED
1101 	/* Setup fixmap P4D and PUD */
1102 	if (pgtable_l5_enabled)
1103 		create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1104 				   (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1105 	/* Setup fixmap PUD and PMD */
1106 	if (pgtable_l4_enabled)
1107 		create_pud_mapping(fixmap_pud, FIXADDR_START,
1108 				   (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1109 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1110 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1111 	/* Setup trampoline PGD and PMD */
1112 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1113 			   trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1114 	if (pgtable_l5_enabled)
1115 		create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1116 				   (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1117 	if (pgtable_l4_enabled)
1118 		create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1119 				   (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1120 #ifdef CONFIG_XIP_KERNEL
1121 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1122 			   kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1123 #else
1124 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1125 			   kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1126 #endif
1127 #else
1128 	/* Setup trampoline PGD */
1129 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1130 			   kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1131 #endif
1132 
1133 	/*
1134 	 * Setup early PGD covering entire kernel which will allow
1135 	 * us to reach paging_init(). We map all memory banks later
1136 	 * in setup_vm_final() below.
1137 	 */
1138 	create_kernel_page_table(early_pg_dir, true);
1139 
1140 	/* Setup early mapping for FDT early scan */
1141 	create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1142 
1143 	/*
1144 	 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1145 	 * range can not span multiple pmds.
1146 	 */
1147 	BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1148 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1149 
1150 #ifndef __PAGETABLE_PMD_FOLDED
1151 	/*
1152 	 * Early ioremap fixmap is already created as it lies within first 2MB
1153 	 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1154 	 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1155 	 * the user if not.
1156 	 */
1157 	fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1158 	fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1159 	if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1160 		WARN_ON(1);
1161 		pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1162 			pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1163 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1164 			fix_to_virt(FIX_BTMAP_BEGIN));
1165 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
1166 			fix_to_virt(FIX_BTMAP_END));
1167 
1168 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
1169 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
1170 	}
1171 #endif
1172 
1173 	pt_ops_set_fixmap();
1174 }
1175 
1176 static void __init create_linear_mapping_range(phys_addr_t start,
1177 					       phys_addr_t end,
1178 					       uintptr_t fixed_map_size)
1179 {
1180 	phys_addr_t pa;
1181 	uintptr_t va, map_size;
1182 
1183 	for (pa = start; pa < end; pa += map_size) {
1184 		va = (uintptr_t)__va(pa);
1185 		map_size = fixed_map_size ? fixed_map_size :
1186 					    best_map_size(pa, va, end - pa);
1187 
1188 		create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1189 				   pgprot_from_va(va));
1190 	}
1191 }
1192 
1193 static void __init create_linear_mapping_page_table(void)
1194 {
1195 	phys_addr_t start, end;
1196 	phys_addr_t kfence_pool __maybe_unused;
1197 	u64 i;
1198 
1199 #ifdef CONFIG_STRICT_KERNEL_RWX
1200 	phys_addr_t ktext_start = __pa_symbol(_start);
1201 	phys_addr_t ktext_size = __init_data_begin - _start;
1202 	phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1203 	phys_addr_t krodata_size = _data - __start_rodata;
1204 
1205 	/* Isolate kernel text and rodata so they don't get mapped with a PUD */
1206 	memblock_mark_nomap(ktext_start,  ktext_size);
1207 	memblock_mark_nomap(krodata_start, krodata_size);
1208 #endif
1209 
1210 #ifdef CONFIG_KFENCE
1211 	/*
1212 	 *  kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1213 	 *  before we setup the linear mapping so that we avoid using hugepages
1214 	 *  for this region.
1215 	 */
1216 	kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1217 	BUG_ON(!kfence_pool);
1218 
1219 	memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1220 	__kfence_pool = __va(kfence_pool);
1221 #endif
1222 
1223 	/* Map all memory banks in the linear mapping */
1224 	for_each_mem_range(i, &start, &end) {
1225 		if (start >= end)
1226 			break;
1227 		if (start <= __pa(PAGE_OFFSET) &&
1228 		    __pa(PAGE_OFFSET) < end)
1229 			start = __pa(PAGE_OFFSET);
1230 		if (end >= __pa(PAGE_OFFSET) + memory_limit)
1231 			end = __pa(PAGE_OFFSET) + memory_limit;
1232 
1233 		create_linear_mapping_range(start, end, 0);
1234 	}
1235 
1236 #ifdef CONFIG_STRICT_KERNEL_RWX
1237 	create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0);
1238 	create_linear_mapping_range(krodata_start,
1239 				    krodata_start + krodata_size, 0);
1240 
1241 	memblock_clear_nomap(ktext_start,  ktext_size);
1242 	memblock_clear_nomap(krodata_start, krodata_size);
1243 #endif
1244 
1245 #ifdef CONFIG_KFENCE
1246 	create_linear_mapping_range(kfence_pool,
1247 				    kfence_pool + KFENCE_POOL_SIZE,
1248 				    PAGE_SIZE);
1249 
1250 	memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1251 #endif
1252 }
1253 
1254 static void __init setup_vm_final(void)
1255 {
1256 	/* Setup swapper PGD for fixmap */
1257 #if !defined(CONFIG_64BIT)
1258 	/*
1259 	 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1260 	 * directly in swapper_pg_dir in addition to the pgd entry that points
1261 	 * to fixmap_pte.
1262 	 */
1263 	unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1264 
1265 	set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1266 #endif
1267 	create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1268 			   __pa_symbol(fixmap_pgd_next),
1269 			   PGDIR_SIZE, PAGE_TABLE);
1270 
1271 	/* Map the linear mapping */
1272 	create_linear_mapping_page_table();
1273 
1274 	/* Map the kernel */
1275 	if (IS_ENABLED(CONFIG_64BIT))
1276 		create_kernel_page_table(swapper_pg_dir, false);
1277 
1278 #ifdef CONFIG_KASAN
1279 	kasan_swapper_init();
1280 #endif
1281 
1282 	/* Clear fixmap PTE and PMD mappings */
1283 	clear_fixmap(FIX_PTE);
1284 	clear_fixmap(FIX_PMD);
1285 	clear_fixmap(FIX_PUD);
1286 	clear_fixmap(FIX_P4D);
1287 
1288 	/* Move to swapper page table */
1289 	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1290 	local_flush_tlb_all();
1291 
1292 	pt_ops_set_late();
1293 }
1294 #else
1295 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1296 {
1297 	dtb_early_va = (void *)dtb_pa;
1298 	dtb_early_pa = dtb_pa;
1299 }
1300 
1301 static inline void setup_vm_final(void)
1302 {
1303 }
1304 #endif /* CONFIG_MMU */
1305 
1306 /*
1307  * reserve_crashkernel() - reserves memory for crash kernel
1308  *
1309  * This function reserves memory area given in "crashkernel=" kernel command
1310  * line parameter. The memory reserved is used by dump capture kernel when
1311  * primary kernel is crashing.
1312  */
1313 static void __init reserve_crashkernel(void)
1314 {
1315 	unsigned long long crash_base = 0;
1316 	unsigned long long crash_size = 0;
1317 	unsigned long search_start = memblock_start_of_DRAM();
1318 	unsigned long search_end = memblock_end_of_DRAM();
1319 
1320 	int ret = 0;
1321 
1322 	if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1323 		return;
1324 	/*
1325 	 * Don't reserve a region for a crash kernel on a crash kernel
1326 	 * since it doesn't make much sense and we have limited memory
1327 	 * resources.
1328 	 */
1329 	if (is_kdump_kernel()) {
1330 		pr_info("crashkernel: ignoring reservation request\n");
1331 		return;
1332 	}
1333 
1334 	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
1335 				&crash_size, &crash_base);
1336 	if (ret || !crash_size)
1337 		return;
1338 
1339 	crash_size = PAGE_ALIGN(crash_size);
1340 
1341 	if (crash_base) {
1342 		search_start = crash_base;
1343 		search_end = crash_base + crash_size;
1344 	}
1345 
1346 	/*
1347 	 * Current riscv boot protocol requires 2MB alignment for
1348 	 * RV64 and 4MB alignment for RV32 (hugepage size)
1349 	 *
1350 	 * Try to alloc from 32bit addressible physical memory so that
1351 	 * swiotlb can work on the crash kernel.
1352 	 */
1353 	crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1354 					       search_start,
1355 					       min(search_end, (unsigned long)(SZ_4G - 1)));
1356 	if (crash_base == 0) {
1357 		/* Try again without restricting region to 32bit addressible memory */
1358 		crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1359 						search_start, search_end);
1360 		if (crash_base == 0) {
1361 			pr_warn("crashkernel: couldn't allocate %lldKB\n",
1362 				crash_size >> 10);
1363 			return;
1364 		}
1365 	}
1366 
1367 	pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1368 		crash_base, crash_base + crash_size, crash_size >> 20);
1369 
1370 	crashk_res.start = crash_base;
1371 	crashk_res.end = crash_base + crash_size - 1;
1372 }
1373 
1374 void __init paging_init(void)
1375 {
1376 	setup_bootmem();
1377 	setup_vm_final();
1378 
1379 	/* Depend on that Linear Mapping is ready */
1380 	memblock_allow_resize();
1381 }
1382 
1383 void __init misc_mem_init(void)
1384 {
1385 	early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1386 	arch_numa_init();
1387 	sparse_init();
1388 	zone_sizes_init();
1389 	reserve_crashkernel();
1390 	memblock_dump_all();
1391 }
1392 
1393 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1394 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1395 			       struct vmem_altmap *altmap)
1396 {
1397 	return vmemmap_populate_basepages(start, end, node, NULL);
1398 }
1399 #endif
1400 
1401 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1402 /*
1403  * Pre-allocates page-table pages for a specific area in the kernel
1404  * page-table. Only the level which needs to be synchronized between
1405  * all page-tables is allocated because the synchronization can be
1406  * expensive.
1407  */
1408 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1409 					       const char *area)
1410 {
1411 	unsigned long addr;
1412 	const char *lvl;
1413 
1414 	for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1415 		pgd_t *pgd = pgd_offset_k(addr);
1416 		p4d_t *p4d;
1417 		pud_t *pud;
1418 		pmd_t *pmd;
1419 
1420 		lvl = "p4d";
1421 		p4d = p4d_alloc(&init_mm, pgd, addr);
1422 		if (!p4d)
1423 			goto failed;
1424 
1425 		if (pgtable_l5_enabled)
1426 			continue;
1427 
1428 		lvl = "pud";
1429 		pud = pud_alloc(&init_mm, p4d, addr);
1430 		if (!pud)
1431 			goto failed;
1432 
1433 		if (pgtable_l4_enabled)
1434 			continue;
1435 
1436 		lvl = "pmd";
1437 		pmd = pmd_alloc(&init_mm, pud, addr);
1438 		if (!pmd)
1439 			goto failed;
1440 	}
1441 	return;
1442 
1443 failed:
1444 	/*
1445 	 * The pages have to be there now or they will be missing in
1446 	 * process page-tables later.
1447 	 */
1448 	panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1449 }
1450 
1451 void __init pgtable_cache_init(void)
1452 {
1453 	preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1454 	if (IS_ENABLED(CONFIG_MODULES))
1455 		preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
1456 }
1457 #endif
1458