xref: /openbmc/linux/arch/riscv/mm/init.c (revision c2dea0bc53397166a77811d9b66928643a83d6f3)
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 
27 #include <asm/fixmap.h>
28 #include <asm/tlbflush.h>
29 #include <asm/sections.h>
30 #include <asm/soc.h>
31 #include <asm/io.h>
32 #include <asm/ptdump.h>
33 #include <asm/numa.h>
34 
35 #include "../kernel/head.h"
36 
37 struct kernel_mapping kernel_map __ro_after_init;
38 EXPORT_SYMBOL(kernel_map);
39 #ifdef CONFIG_XIP_KERNEL
40 #define kernel_map	(*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
41 #endif
42 
43 #ifdef CONFIG_64BIT
44 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39;
45 #else
46 u64 satp_mode __ro_after_init = SATP_MODE_32;
47 #endif
48 EXPORT_SYMBOL(satp_mode);
49 
50 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
51 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
52 EXPORT_SYMBOL(pgtable_l4_enabled);
53 EXPORT_SYMBOL(pgtable_l5_enabled);
54 
55 phys_addr_t phys_ram_base __ro_after_init;
56 EXPORT_SYMBOL(phys_ram_base);
57 
58 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
59 							__page_aligned_bss;
60 EXPORT_SYMBOL(empty_zero_page);
61 
62 extern char _start[];
63 #define DTB_EARLY_BASE_VA      PGDIR_SIZE
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 	 * memblock allocator is not aware of the fact that last 4K bytes of
221 	 * the addressable memory can not be mapped because of IS_ERR_VALUE
222 	 * macro. Make sure that last 4k bytes are not usable by memblock
223 	 * if end of dram is equal to maximum addressable memory.  For 64-bit
224 	 * kernel, this problem can't happen here as the end of the virtual
225 	 * address space is occupied by the kernel mapping then this check must
226 	 * be done as soon as the kernel mapping base address is determined.
227 	 */
228 	if (!IS_ENABLED(CONFIG_64BIT)) {
229 		max_mapped_addr = __pa(~(ulong)0);
230 		if (max_mapped_addr == (phys_ram_end - 1))
231 			memblock_set_current_limit(max_mapped_addr - 4096);
232 	}
233 
234 	min_low_pfn = PFN_UP(phys_ram_base);
235 	max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
236 	high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
237 
238 	dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
239 	set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
240 
241 	reserve_initrd_mem();
242 	/*
243 	 * If DTB is built in, no need to reserve its memblock.
244 	 * Otherwise, do reserve it but avoid using
245 	 * early_init_fdt_reserve_self() since __pa() does
246 	 * not work for DTB pointers that are fixmap addresses
247 	 */
248 	if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) {
249 		/*
250 		 * In case the DTB is not located in a memory region we won't
251 		 * be able to locate it later on via the linear mapping and
252 		 * get a segfault when accessing it via __va(dtb_early_pa).
253 		 * To avoid this situation copy DTB to a memory region.
254 		 * Note that memblock_phys_alloc will also reserve DTB region.
255 		 */
256 		if (!memblock_is_memory(dtb_early_pa)) {
257 			size_t fdt_size = fdt_totalsize(dtb_early_va);
258 			phys_addr_t new_dtb_early_pa = memblock_phys_alloc(fdt_size, PAGE_SIZE);
259 			void *new_dtb_early_va = early_memremap(new_dtb_early_pa, fdt_size);
260 
261 			memcpy(new_dtb_early_va, dtb_early_va, fdt_size);
262 			early_memunmap(new_dtb_early_va, fdt_size);
263 			_dtb_early_pa = new_dtb_early_pa;
264 		} else
265 			memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
266 	}
267 
268 	dma_contiguous_reserve(dma32_phys_limit);
269 	if (IS_ENABLED(CONFIG_64BIT))
270 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
271 	memblock_allow_resize();
272 }
273 
274 #ifdef CONFIG_MMU
275 struct pt_alloc_ops pt_ops __initdata;
276 
277 unsigned long riscv_pfn_base __ro_after_init;
278 EXPORT_SYMBOL(riscv_pfn_base);
279 
280 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
281 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
282 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
283 
284 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
285 static p4d_t __maybe_unused early_dtb_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
286 static pud_t __maybe_unused early_dtb_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
287 static pmd_t __maybe_unused early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
288 
289 #ifdef CONFIG_XIP_KERNEL
290 #define pt_ops			(*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
291 #define riscv_pfn_base         (*(unsigned long  *)XIP_FIXUP(&riscv_pfn_base))
292 #define trampoline_pg_dir      ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
293 #define fixmap_pte             ((pte_t *)XIP_FIXUP(fixmap_pte))
294 #define early_pg_dir           ((pgd_t *)XIP_FIXUP(early_pg_dir))
295 #endif /* CONFIG_XIP_KERNEL */
296 
297 static const pgprot_t protection_map[16] = {
298 	[VM_NONE]					= PAGE_NONE,
299 	[VM_READ]					= PAGE_READ,
300 	[VM_WRITE]					= PAGE_COPY,
301 	[VM_WRITE | VM_READ]				= PAGE_COPY,
302 	[VM_EXEC]					= PAGE_EXEC,
303 	[VM_EXEC | VM_READ]				= PAGE_READ_EXEC,
304 	[VM_EXEC | VM_WRITE]				= PAGE_COPY_EXEC,
305 	[VM_EXEC | VM_WRITE | VM_READ]			= PAGE_COPY_READ_EXEC,
306 	[VM_SHARED]					= PAGE_NONE,
307 	[VM_SHARED | VM_READ]				= PAGE_READ,
308 	[VM_SHARED | VM_WRITE]				= PAGE_SHARED,
309 	[VM_SHARED | VM_WRITE | VM_READ]		= PAGE_SHARED,
310 	[VM_SHARED | VM_EXEC]				= PAGE_EXEC,
311 	[VM_SHARED | VM_EXEC | VM_READ]			= PAGE_READ_EXEC,
312 	[VM_SHARED | VM_EXEC | VM_WRITE]		= PAGE_SHARED_EXEC,
313 	[VM_SHARED | VM_EXEC | VM_WRITE | VM_READ]	= PAGE_SHARED_EXEC
314 };
315 DECLARE_VM_GET_PAGE_PROT
316 
317 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
318 {
319 	unsigned long addr = __fix_to_virt(idx);
320 	pte_t *ptep;
321 
322 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
323 
324 	ptep = &fixmap_pte[pte_index(addr)];
325 
326 	if (pgprot_val(prot))
327 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
328 	else
329 		pte_clear(&init_mm, addr, ptep);
330 	local_flush_tlb_page(addr);
331 }
332 
333 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
334 {
335 	return (pte_t *)((uintptr_t)pa);
336 }
337 
338 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
339 {
340 	clear_fixmap(FIX_PTE);
341 	return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
342 }
343 
344 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
345 {
346 	return (pte_t *) __va(pa);
347 }
348 
349 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
350 {
351 	/*
352 	 * We only create PMD or PGD early mappings so we
353 	 * should never reach here with MMU disabled.
354 	 */
355 	BUG();
356 }
357 
358 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
359 {
360 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
361 }
362 
363 static phys_addr_t __init alloc_pte_late(uintptr_t va)
364 {
365 	unsigned long vaddr;
366 
367 	vaddr = __get_free_page(GFP_KERNEL);
368 	BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)));
369 
370 	return __pa(vaddr);
371 }
372 
373 static void __init create_pte_mapping(pte_t *ptep,
374 				      uintptr_t va, phys_addr_t pa,
375 				      phys_addr_t sz, pgprot_t prot)
376 {
377 	uintptr_t pte_idx = pte_index(va);
378 
379 	BUG_ON(sz != PAGE_SIZE);
380 
381 	if (pte_none(ptep[pte_idx]))
382 		ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
383 }
384 
385 #ifndef __PAGETABLE_PMD_FOLDED
386 
387 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
388 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
389 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
390 
391 #ifdef CONFIG_XIP_KERNEL
392 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
393 #define fixmap_pmd     ((pmd_t *)XIP_FIXUP(fixmap_pmd))
394 #define early_pmd      ((pmd_t *)XIP_FIXUP(early_pmd))
395 #endif /* CONFIG_XIP_KERNEL */
396 
397 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
398 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
399 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
400 
401 #ifdef CONFIG_XIP_KERNEL
402 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
403 #define fixmap_p4d     ((p4d_t *)XIP_FIXUP(fixmap_p4d))
404 #define early_p4d      ((p4d_t *)XIP_FIXUP(early_p4d))
405 #endif /* CONFIG_XIP_KERNEL */
406 
407 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
408 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
409 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
410 
411 #ifdef CONFIG_XIP_KERNEL
412 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
413 #define fixmap_pud     ((pud_t *)XIP_FIXUP(fixmap_pud))
414 #define early_pud      ((pud_t *)XIP_FIXUP(early_pud))
415 #endif /* CONFIG_XIP_KERNEL */
416 
417 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
418 {
419 	/* Before MMU is enabled */
420 	return (pmd_t *)((uintptr_t)pa);
421 }
422 
423 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
424 {
425 	clear_fixmap(FIX_PMD);
426 	return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
427 }
428 
429 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
430 {
431 	return (pmd_t *) __va(pa);
432 }
433 
434 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
435 {
436 	BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
437 
438 	return (uintptr_t)early_pmd;
439 }
440 
441 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
442 {
443 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
444 }
445 
446 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
447 {
448 	unsigned long vaddr;
449 
450 	vaddr = __get_free_page(GFP_KERNEL);
451 	BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page(vaddr)));
452 
453 	return __pa(vaddr);
454 }
455 
456 static void __init create_pmd_mapping(pmd_t *pmdp,
457 				      uintptr_t va, phys_addr_t pa,
458 				      phys_addr_t sz, pgprot_t prot)
459 {
460 	pte_t *ptep;
461 	phys_addr_t pte_phys;
462 	uintptr_t pmd_idx = pmd_index(va);
463 
464 	if (sz == PMD_SIZE) {
465 		if (pmd_none(pmdp[pmd_idx]))
466 			pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
467 		return;
468 	}
469 
470 	if (pmd_none(pmdp[pmd_idx])) {
471 		pte_phys = pt_ops.alloc_pte(va);
472 		pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
473 		ptep = pt_ops.get_pte_virt(pte_phys);
474 		memset(ptep, 0, PAGE_SIZE);
475 	} else {
476 		pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
477 		ptep = pt_ops.get_pte_virt(pte_phys);
478 	}
479 
480 	create_pte_mapping(ptep, va, pa, sz, prot);
481 }
482 
483 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
484 {
485 	return (pud_t *)((uintptr_t)pa);
486 }
487 
488 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
489 {
490 	clear_fixmap(FIX_PUD);
491 	return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
492 }
493 
494 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
495 {
496 	return (pud_t *)__va(pa);
497 }
498 
499 static phys_addr_t __init alloc_pud_early(uintptr_t va)
500 {
501 	/* Only one PUD is available for early mapping */
502 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
503 
504 	return (uintptr_t)early_pud;
505 }
506 
507 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
508 {
509 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
510 }
511 
512 static phys_addr_t alloc_pud_late(uintptr_t va)
513 {
514 	unsigned long vaddr;
515 
516 	vaddr = __get_free_page(GFP_KERNEL);
517 	BUG_ON(!vaddr);
518 	return __pa(vaddr);
519 }
520 
521 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
522 {
523 	return (p4d_t *)((uintptr_t)pa);
524 }
525 
526 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
527 {
528 	clear_fixmap(FIX_P4D);
529 	return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
530 }
531 
532 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
533 {
534 	return (p4d_t *)__va(pa);
535 }
536 
537 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
538 {
539 	/* Only one P4D is available for early mapping */
540 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
541 
542 	return (uintptr_t)early_p4d;
543 }
544 
545 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
546 {
547 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
548 }
549 
550 static phys_addr_t alloc_p4d_late(uintptr_t va)
551 {
552 	unsigned long vaddr;
553 
554 	vaddr = __get_free_page(GFP_KERNEL);
555 	BUG_ON(!vaddr);
556 	return __pa(vaddr);
557 }
558 
559 static void __init create_pud_mapping(pud_t *pudp,
560 				      uintptr_t va, phys_addr_t pa,
561 				      phys_addr_t sz, pgprot_t prot)
562 {
563 	pmd_t *nextp;
564 	phys_addr_t next_phys;
565 	uintptr_t pud_index = pud_index(va);
566 
567 	if (sz == PUD_SIZE) {
568 		if (pud_val(pudp[pud_index]) == 0)
569 			pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
570 		return;
571 	}
572 
573 	if (pud_val(pudp[pud_index]) == 0) {
574 		next_phys = pt_ops.alloc_pmd(va);
575 		pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
576 		nextp = pt_ops.get_pmd_virt(next_phys);
577 		memset(nextp, 0, PAGE_SIZE);
578 	} else {
579 		next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
580 		nextp = pt_ops.get_pmd_virt(next_phys);
581 	}
582 
583 	create_pmd_mapping(nextp, va, pa, sz, prot);
584 }
585 
586 static void __init create_p4d_mapping(p4d_t *p4dp,
587 				      uintptr_t va, phys_addr_t pa,
588 				      phys_addr_t sz, pgprot_t prot)
589 {
590 	pud_t *nextp;
591 	phys_addr_t next_phys;
592 	uintptr_t p4d_index = p4d_index(va);
593 
594 	if (sz == P4D_SIZE) {
595 		if (p4d_val(p4dp[p4d_index]) == 0)
596 			p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
597 		return;
598 	}
599 
600 	if (p4d_val(p4dp[p4d_index]) == 0) {
601 		next_phys = pt_ops.alloc_pud(va);
602 		p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
603 		nextp = pt_ops.get_pud_virt(next_phys);
604 		memset(nextp, 0, PAGE_SIZE);
605 	} else {
606 		next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
607 		nextp = pt_ops.get_pud_virt(next_phys);
608 	}
609 
610 	create_pud_mapping(nextp, va, pa, sz, prot);
611 }
612 
613 #define pgd_next_t		p4d_t
614 #define alloc_pgd_next(__va)	(pgtable_l5_enabled ?			\
615 		pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ?		\
616 		pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
617 #define get_pgd_next_virt(__pa)	(pgtable_l5_enabled ?			\
618 		pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ?	\
619 		pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
620 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
621 				(pgtable_l5_enabled ?			\
622 		create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
623 				(pgtable_l4_enabled ?			\
624 		create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) :	\
625 		create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
626 #define fixmap_pgd_next		(pgtable_l5_enabled ?			\
627 		(uintptr_t)fixmap_p4d : (pgtable_l4_enabled ?		\
628 		(uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
629 #define trampoline_pgd_next	(pgtable_l5_enabled ?			\
630 		(uintptr_t)trampoline_p4d : (pgtable_l4_enabled ?	\
631 		(uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
632 #define early_dtb_pgd_next	(pgtable_l5_enabled ?			\
633 		(uintptr_t)early_dtb_p4d : (pgtable_l4_enabled ?	\
634 		(uintptr_t)early_dtb_pud : (uintptr_t)early_dtb_pmd))
635 #else
636 #define pgd_next_t		pte_t
637 #define alloc_pgd_next(__va)	pt_ops.alloc_pte(__va)
638 #define get_pgd_next_virt(__pa)	pt_ops.get_pte_virt(__pa)
639 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
640 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
641 #define fixmap_pgd_next		((uintptr_t)fixmap_pte)
642 #define early_dtb_pgd_next	((uintptr_t)early_dtb_pmd)
643 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
644 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
645 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
646 #endif /* __PAGETABLE_PMD_FOLDED */
647 
648 void __init create_pgd_mapping(pgd_t *pgdp,
649 				      uintptr_t va, phys_addr_t pa,
650 				      phys_addr_t sz, pgprot_t prot)
651 {
652 	pgd_next_t *nextp;
653 	phys_addr_t next_phys;
654 	uintptr_t pgd_idx = pgd_index(va);
655 
656 	if (sz == PGDIR_SIZE) {
657 		if (pgd_val(pgdp[pgd_idx]) == 0)
658 			pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
659 		return;
660 	}
661 
662 	if (pgd_val(pgdp[pgd_idx]) == 0) {
663 		next_phys = alloc_pgd_next(va);
664 		pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
665 		nextp = get_pgd_next_virt(next_phys);
666 		memset(nextp, 0, PAGE_SIZE);
667 	} else {
668 		next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
669 		nextp = get_pgd_next_virt(next_phys);
670 	}
671 
672 	create_pgd_next_mapping(nextp, va, pa, sz, prot);
673 }
674 
675 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
676 {
677 	/* Upgrade to PMD_SIZE mappings whenever possible */
678 	base &= PMD_SIZE - 1;
679 	if (!base && size >= PMD_SIZE)
680 		return PMD_SIZE;
681 
682 	return PAGE_SIZE;
683 }
684 
685 #ifdef CONFIG_XIP_KERNEL
686 #define phys_ram_base  (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
687 extern char _xiprom[], _exiprom[], __data_loc;
688 
689 /* called from head.S with MMU off */
690 asmlinkage void __init __copy_data(void)
691 {
692 	void *from = (void *)(&__data_loc);
693 	void *to = (void *)CONFIG_PHYS_RAM_BASE;
694 	size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
695 
696 	memcpy(to, from, sz);
697 }
698 #endif
699 
700 #ifdef CONFIG_STRICT_KERNEL_RWX
701 static __init pgprot_t pgprot_from_va(uintptr_t va)
702 {
703 	if (is_va_kernel_text(va))
704 		return PAGE_KERNEL_READ_EXEC;
705 
706 	/*
707 	 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
708 	 * we must protect its linear mapping alias from being executed and
709 	 * written.
710 	 * And rodata section is marked readonly in mark_rodata_ro.
711 	 */
712 	if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
713 		return PAGE_KERNEL_READ;
714 
715 	return PAGE_KERNEL;
716 }
717 
718 void mark_rodata_ro(void)
719 {
720 	set_kernel_memory(__start_rodata, _data, set_memory_ro);
721 	if (IS_ENABLED(CONFIG_64BIT))
722 		set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
723 				  set_memory_ro);
724 
725 	debug_checkwx();
726 }
727 #else
728 static __init pgprot_t pgprot_from_va(uintptr_t va)
729 {
730 	if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
731 		return PAGE_KERNEL;
732 
733 	return PAGE_KERNEL_EXEC;
734 }
735 #endif /* CONFIG_STRICT_KERNEL_RWX */
736 
737 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
738 static void __init disable_pgtable_l5(void)
739 {
740 	pgtable_l5_enabled = false;
741 	kernel_map.page_offset = PAGE_OFFSET_L4;
742 	satp_mode = SATP_MODE_48;
743 }
744 
745 static void __init disable_pgtable_l4(void)
746 {
747 	pgtable_l4_enabled = false;
748 	kernel_map.page_offset = PAGE_OFFSET_L3;
749 	satp_mode = SATP_MODE_39;
750 }
751 
752 /*
753  * There is a simple way to determine if 4-level is supported by the
754  * underlying hardware: establish 1:1 mapping in 4-level page table mode
755  * then read SATP to see if the configuration was taken into account
756  * meaning sv48 is supported.
757  */
758 static __init void set_satp_mode(void)
759 {
760 	u64 identity_satp, hw_satp;
761 	uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
762 	bool check_l4 = false;
763 
764 	create_p4d_mapping(early_p4d,
765 			set_satp_mode_pmd, (uintptr_t)early_pud,
766 			P4D_SIZE, PAGE_TABLE);
767 	create_pud_mapping(early_pud,
768 			   set_satp_mode_pmd, (uintptr_t)early_pmd,
769 			   PUD_SIZE, PAGE_TABLE);
770 	/* Handle the case where set_satp_mode straddles 2 PMDs */
771 	create_pmd_mapping(early_pmd,
772 			   set_satp_mode_pmd, set_satp_mode_pmd,
773 			   PMD_SIZE, PAGE_KERNEL_EXEC);
774 	create_pmd_mapping(early_pmd,
775 			   set_satp_mode_pmd + PMD_SIZE,
776 			   set_satp_mode_pmd + PMD_SIZE,
777 			   PMD_SIZE, PAGE_KERNEL_EXEC);
778 retry:
779 	create_pgd_mapping(early_pg_dir,
780 			   set_satp_mode_pmd,
781 			   check_l4 ? (uintptr_t)early_pud : (uintptr_t)early_p4d,
782 			   PGDIR_SIZE, PAGE_TABLE);
783 
784 	identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
785 
786 	local_flush_tlb_all();
787 	csr_write(CSR_SATP, identity_satp);
788 	hw_satp = csr_swap(CSR_SATP, 0ULL);
789 	local_flush_tlb_all();
790 
791 	if (hw_satp != identity_satp) {
792 		if (!check_l4) {
793 			disable_pgtable_l5();
794 			check_l4 = true;
795 			memset(early_pg_dir, 0, PAGE_SIZE);
796 			goto retry;
797 		}
798 		disable_pgtable_l4();
799 	}
800 
801 	memset(early_pg_dir, 0, PAGE_SIZE);
802 	memset(early_p4d, 0, PAGE_SIZE);
803 	memset(early_pud, 0, PAGE_SIZE);
804 	memset(early_pmd, 0, PAGE_SIZE);
805 }
806 #endif
807 
808 /*
809  * setup_vm() is called from head.S with MMU-off.
810  *
811  * Following requirements should be honoured for setup_vm() to work
812  * correctly:
813  * 1) It should use PC-relative addressing for accessing kernel symbols.
814  *    To achieve this we always use GCC cmodel=medany.
815  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
816  *    so disable compiler instrumentation when FTRACE is enabled.
817  *
818  * Currently, the above requirements are honoured by using custom CFLAGS
819  * for init.o in mm/Makefile.
820  */
821 
822 #ifndef __riscv_cmodel_medany
823 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
824 #endif
825 
826 #ifdef CONFIG_RELOCATABLE
827 extern unsigned long __rela_dyn_start, __rela_dyn_end;
828 
829 static void __init relocate_kernel(void)
830 {
831 	Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
832 	/*
833 	 * This holds the offset between the linked virtual address and the
834 	 * relocated virtual address.
835 	 */
836 	uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
837 	/*
838 	 * This holds the offset between kernel linked virtual address and
839 	 * physical address.
840 	 */
841 	uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
842 
843 	for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
844 		Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
845 		Elf64_Addr relocated_addr = rela->r_addend;
846 
847 		if (rela->r_info != R_RISCV_RELATIVE)
848 			continue;
849 
850 		/*
851 		 * Make sure to not relocate vdso symbols like rt_sigreturn
852 		 * which are linked from the address 0 in vmlinux since
853 		 * vdso symbol addresses are actually used as an offset from
854 		 * mm->context.vdso in VDSO_OFFSET macro.
855 		 */
856 		if (relocated_addr >= KERNEL_LINK_ADDR)
857 			relocated_addr += reloc_offset;
858 
859 		*(Elf64_Addr *)addr = relocated_addr;
860 	}
861 }
862 #endif /* CONFIG_RELOCATABLE */
863 
864 #ifdef CONFIG_XIP_KERNEL
865 static void __init create_kernel_page_table(pgd_t *pgdir,
866 					    __always_unused bool early)
867 {
868 	uintptr_t va, end_va;
869 
870 	/* Map the flash resident part */
871 	end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
872 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
873 		create_pgd_mapping(pgdir, va,
874 				   kernel_map.xiprom + (va - kernel_map.virt_addr),
875 				   PMD_SIZE, PAGE_KERNEL_EXEC);
876 
877 	/* Map the data in RAM */
878 	end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
879 	for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
880 		create_pgd_mapping(pgdir, va,
881 				   kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
882 				   PMD_SIZE, PAGE_KERNEL);
883 }
884 #else
885 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
886 {
887 	uintptr_t va, end_va;
888 
889 	end_va = kernel_map.virt_addr + kernel_map.size;
890 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
891 		create_pgd_mapping(pgdir, va,
892 				   kernel_map.phys_addr + (va - kernel_map.virt_addr),
893 				   PMD_SIZE,
894 				   early ?
895 					PAGE_KERNEL_EXEC : pgprot_from_va(va));
896 }
897 #endif
898 
899 /*
900  * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
901  * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
902  * entry.
903  */
904 static void __init create_fdt_early_page_table(pgd_t *pgdir, uintptr_t dtb_pa)
905 {
906 #ifndef CONFIG_BUILTIN_DTB
907 	uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
908 
909 	create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
910 			   IS_ENABLED(CONFIG_64BIT) ? early_dtb_pgd_next : pa,
911 			   PGDIR_SIZE,
912 			   IS_ENABLED(CONFIG_64BIT) ? PAGE_TABLE : PAGE_KERNEL);
913 
914 	if (pgtable_l5_enabled)
915 		create_p4d_mapping(early_dtb_p4d, DTB_EARLY_BASE_VA,
916 				   (uintptr_t)early_dtb_pud, P4D_SIZE, PAGE_TABLE);
917 
918 	if (pgtable_l4_enabled)
919 		create_pud_mapping(early_dtb_pud, DTB_EARLY_BASE_VA,
920 				   (uintptr_t)early_dtb_pmd, PUD_SIZE, PAGE_TABLE);
921 
922 	if (IS_ENABLED(CONFIG_64BIT)) {
923 		create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA,
924 				   pa, PMD_SIZE, PAGE_KERNEL);
925 		create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE,
926 				   pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
927 	}
928 
929 	dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1));
930 #else
931 	/*
932 	 * For 64-bit kernel, __va can't be used since it would return a linear
933 	 * mapping address whereas dtb_early_va will be used before
934 	 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
935 	 * kernel is mapped in the linear mapping, that makes no difference.
936 	 */
937 	dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
938 #endif
939 
940 	dtb_early_pa = dtb_pa;
941 }
942 
943 /*
944  * MMU is not enabled, the page tables are allocated directly using
945  * early_pmd/pud/p4d and the address returned is the physical one.
946  */
947 static void __init pt_ops_set_early(void)
948 {
949 	pt_ops.alloc_pte = alloc_pte_early;
950 	pt_ops.get_pte_virt = get_pte_virt_early;
951 #ifndef __PAGETABLE_PMD_FOLDED
952 	pt_ops.alloc_pmd = alloc_pmd_early;
953 	pt_ops.get_pmd_virt = get_pmd_virt_early;
954 	pt_ops.alloc_pud = alloc_pud_early;
955 	pt_ops.get_pud_virt = get_pud_virt_early;
956 	pt_ops.alloc_p4d = alloc_p4d_early;
957 	pt_ops.get_p4d_virt = get_p4d_virt_early;
958 #endif
959 }
960 
961 /*
962  * MMU is enabled but page table setup is not complete yet.
963  * fixmap page table alloc functions must be used as a means to temporarily
964  * map the allocated physical pages since the linear mapping does not exist yet.
965  *
966  * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
967  * but it will be used as described above.
968  */
969 static void __init pt_ops_set_fixmap(void)
970 {
971 	pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
972 	pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
973 #ifndef __PAGETABLE_PMD_FOLDED
974 	pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
975 	pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
976 	pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
977 	pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
978 	pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
979 	pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
980 #endif
981 }
982 
983 /*
984  * MMU is enabled and page table setup is complete, so from now, we can use
985  * generic page allocation functions to setup page table.
986  */
987 static void __init pt_ops_set_late(void)
988 {
989 	pt_ops.alloc_pte = alloc_pte_late;
990 	pt_ops.get_pte_virt = get_pte_virt_late;
991 #ifndef __PAGETABLE_PMD_FOLDED
992 	pt_ops.alloc_pmd = alloc_pmd_late;
993 	pt_ops.get_pmd_virt = get_pmd_virt_late;
994 	pt_ops.alloc_pud = alloc_pud_late;
995 	pt_ops.get_pud_virt = get_pud_virt_late;
996 	pt_ops.alloc_p4d = alloc_p4d_late;
997 	pt_ops.get_p4d_virt = get_p4d_virt_late;
998 #endif
999 }
1000 
1001 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1002 {
1003 	pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1004 
1005 	kernel_map.virt_addr = KERNEL_LINK_ADDR;
1006 	kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1007 
1008 #ifdef CONFIG_XIP_KERNEL
1009 	kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1010 	kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1011 
1012 	phys_ram_base = CONFIG_PHYS_RAM_BASE;
1013 	kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1014 	kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
1015 
1016 	kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1017 #else
1018 	kernel_map.phys_addr = (uintptr_t)(&_start);
1019 	kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1020 #endif
1021 
1022 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1023 	set_satp_mode();
1024 #endif
1025 
1026 	kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr;
1027 	kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1028 
1029 	riscv_pfn_base = PFN_DOWN(kernel_map.phys_addr);
1030 
1031 	/*
1032 	 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
1033 	 * kernel, whereas for 64-bit kernel, the end of the virtual address
1034 	 * space is occupied by the modules/BPF/kernel mappings which reduces
1035 	 * the available size of the linear mapping.
1036 	 */
1037 	memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
1038 
1039 	/* Sanity check alignment and size */
1040 	BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1041 	BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1042 
1043 #ifdef CONFIG_64BIT
1044 	/*
1045 	 * The last 4K bytes of the addressable memory can not be mapped because
1046 	 * of IS_ERR_VALUE macro.
1047 	 */
1048 	BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1049 #endif
1050 
1051 #ifdef CONFIG_RELOCATABLE
1052 	/*
1053 	 * Early page table uses only one PUD, which makes it possible
1054 	 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1055 	 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1056 	 * since a part of the kernel would not get mapped.
1057 	 */
1058 	BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1059 	relocate_kernel();
1060 #endif
1061 
1062 	apply_early_boot_alternatives();
1063 	pt_ops_set_early();
1064 
1065 	/* Setup early PGD for fixmap */
1066 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
1067 			   fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1068 
1069 #ifndef __PAGETABLE_PMD_FOLDED
1070 	/* Setup fixmap P4D and PUD */
1071 	if (pgtable_l5_enabled)
1072 		create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1073 				   (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1074 	/* Setup fixmap PUD and PMD */
1075 	if (pgtable_l4_enabled)
1076 		create_pud_mapping(fixmap_pud, FIXADDR_START,
1077 				   (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1078 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1079 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1080 	/* Setup trampoline PGD and PMD */
1081 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1082 			   trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1083 	if (pgtable_l5_enabled)
1084 		create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1085 				   (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1086 	if (pgtable_l4_enabled)
1087 		create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1088 				   (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1089 #ifdef CONFIG_XIP_KERNEL
1090 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1091 			   kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1092 #else
1093 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1094 			   kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1095 #endif
1096 #else
1097 	/* Setup trampoline PGD */
1098 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1099 			   kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1100 #endif
1101 
1102 	/*
1103 	 * Setup early PGD covering entire kernel which will allow
1104 	 * us to reach paging_init(). We map all memory banks later
1105 	 * in setup_vm_final() below.
1106 	 */
1107 	create_kernel_page_table(early_pg_dir, true);
1108 
1109 	/* Setup early mapping for FDT early scan */
1110 	create_fdt_early_page_table(early_pg_dir, dtb_pa);
1111 
1112 	/*
1113 	 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1114 	 * range can not span multiple pmds.
1115 	 */
1116 	BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1117 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1118 
1119 #ifndef __PAGETABLE_PMD_FOLDED
1120 	/*
1121 	 * Early ioremap fixmap is already created as it lies within first 2MB
1122 	 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1123 	 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1124 	 * the user if not.
1125 	 */
1126 	fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1127 	fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1128 	if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1129 		WARN_ON(1);
1130 		pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1131 			pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1132 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1133 			fix_to_virt(FIX_BTMAP_BEGIN));
1134 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
1135 			fix_to_virt(FIX_BTMAP_END));
1136 
1137 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
1138 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
1139 	}
1140 #endif
1141 
1142 	pt_ops_set_fixmap();
1143 }
1144 
1145 static void __init setup_vm_final(void)
1146 {
1147 	uintptr_t va, map_size;
1148 	phys_addr_t pa, start, end;
1149 	u64 i;
1150 
1151 	/* Setup swapper PGD for fixmap */
1152 	create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1153 			   __pa_symbol(fixmap_pgd_next),
1154 			   PGDIR_SIZE, PAGE_TABLE);
1155 
1156 	/* Map all memory banks in the linear mapping */
1157 	for_each_mem_range(i, &start, &end) {
1158 		if (start >= end)
1159 			break;
1160 		if (start <= __pa(PAGE_OFFSET) &&
1161 		    __pa(PAGE_OFFSET) < end)
1162 			start = __pa(PAGE_OFFSET);
1163 		if (end >= __pa(PAGE_OFFSET) + memory_limit)
1164 			end = __pa(PAGE_OFFSET) + memory_limit;
1165 
1166 		for (pa = start; pa < end; pa += map_size) {
1167 			va = (uintptr_t)__va(pa);
1168 			map_size = best_map_size(pa, end - pa);
1169 
1170 			create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1171 					   pgprot_from_va(va));
1172 		}
1173 	}
1174 
1175 	/* Map the kernel */
1176 	if (IS_ENABLED(CONFIG_64BIT))
1177 		create_kernel_page_table(swapper_pg_dir, false);
1178 
1179 #ifdef CONFIG_KASAN
1180 	kasan_swapper_init();
1181 #endif
1182 
1183 	/* Clear fixmap PTE and PMD mappings */
1184 	clear_fixmap(FIX_PTE);
1185 	clear_fixmap(FIX_PMD);
1186 	clear_fixmap(FIX_PUD);
1187 	clear_fixmap(FIX_P4D);
1188 
1189 	/* Move to swapper page table */
1190 	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1191 	local_flush_tlb_all();
1192 
1193 	pt_ops_set_late();
1194 }
1195 #else
1196 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1197 {
1198 	dtb_early_va = (void *)dtb_pa;
1199 	dtb_early_pa = dtb_pa;
1200 }
1201 
1202 static inline void setup_vm_final(void)
1203 {
1204 }
1205 #endif /* CONFIG_MMU */
1206 
1207 /*
1208  * reserve_crashkernel() - reserves memory for crash kernel
1209  *
1210  * This function reserves memory area given in "crashkernel=" kernel command
1211  * line parameter. The memory reserved is used by dump capture kernel when
1212  * primary kernel is crashing.
1213  */
1214 static void __init reserve_crashkernel(void)
1215 {
1216 	unsigned long long crash_base = 0;
1217 	unsigned long long crash_size = 0;
1218 	unsigned long search_start = memblock_start_of_DRAM();
1219 	unsigned long search_end = memblock_end_of_DRAM();
1220 
1221 	int ret = 0;
1222 
1223 	if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1224 		return;
1225 	/*
1226 	 * Don't reserve a region for a crash kernel on a crash kernel
1227 	 * since it doesn't make much sense and we have limited memory
1228 	 * resources.
1229 	 */
1230 	if (is_kdump_kernel()) {
1231 		pr_info("crashkernel: ignoring reservation request\n");
1232 		return;
1233 	}
1234 
1235 	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
1236 				&crash_size, &crash_base);
1237 	if (ret || !crash_size)
1238 		return;
1239 
1240 	crash_size = PAGE_ALIGN(crash_size);
1241 
1242 	if (crash_base) {
1243 		search_start = crash_base;
1244 		search_end = crash_base + crash_size;
1245 	}
1246 
1247 	/*
1248 	 * Current riscv boot protocol requires 2MB alignment for
1249 	 * RV64 and 4MB alignment for RV32 (hugepage size)
1250 	 *
1251 	 * Try to alloc from 32bit addressible physical memory so that
1252 	 * swiotlb can work on the crash kernel.
1253 	 */
1254 	crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1255 					       search_start,
1256 					       min(search_end, (unsigned long) SZ_4G));
1257 	if (crash_base == 0) {
1258 		/* Try again without restricting region to 32bit addressible memory */
1259 		crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1260 						search_start, search_end);
1261 		if (crash_base == 0) {
1262 			pr_warn("crashkernel: couldn't allocate %lldKB\n",
1263 				crash_size >> 10);
1264 			return;
1265 		}
1266 	}
1267 
1268 	pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1269 		crash_base, crash_base + crash_size, crash_size >> 20);
1270 
1271 	crashk_res.start = crash_base;
1272 	crashk_res.end = crash_base + crash_size - 1;
1273 }
1274 
1275 void __init paging_init(void)
1276 {
1277 	setup_bootmem();
1278 	setup_vm_final();
1279 }
1280 
1281 void __init misc_mem_init(void)
1282 {
1283 	early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1284 	arch_numa_init();
1285 	sparse_init();
1286 	zone_sizes_init();
1287 	reserve_crashkernel();
1288 	memblock_dump_all();
1289 }
1290 
1291 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1292 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1293 			       struct vmem_altmap *altmap)
1294 {
1295 	return vmemmap_populate_basepages(start, end, node, NULL);
1296 }
1297 #endif
1298