xref: /openbmc/linux/arch/riscv/mm/init.c (revision 8dda2eac)
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 
23 #include <asm/fixmap.h>
24 #include <asm/tlbflush.h>
25 #include <asm/sections.h>
26 #include <asm/soc.h>
27 #include <asm/io.h>
28 #include <asm/ptdump.h>
29 #include <asm/numa.h>
30 
31 #include "../kernel/head.h"
32 
33 struct kernel_mapping kernel_map __ro_after_init;
34 EXPORT_SYMBOL(kernel_map);
35 #ifdef CONFIG_XIP_KERNEL
36 #define kernel_map	(*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
37 #endif
38 
39 #ifdef CONFIG_XIP_KERNEL
40 extern char _xiprom[], _exiprom[];
41 #endif
42 
43 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
44 							__page_aligned_bss;
45 EXPORT_SYMBOL(empty_zero_page);
46 
47 extern char _start[];
48 #define DTB_EARLY_BASE_VA      PGDIR_SIZE
49 void *_dtb_early_va __initdata;
50 uintptr_t _dtb_early_pa __initdata;
51 
52 struct pt_alloc_ops {
53 	pte_t *(*get_pte_virt)(phys_addr_t pa);
54 	phys_addr_t (*alloc_pte)(uintptr_t va);
55 #ifndef __PAGETABLE_PMD_FOLDED
56 	pmd_t *(*get_pmd_virt)(phys_addr_t pa);
57 	phys_addr_t (*alloc_pmd)(uintptr_t va);
58 #endif
59 };
60 
61 static phys_addr_t dma32_phys_limit __initdata;
62 
63 static void __init zone_sizes_init(void)
64 {
65 	unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
66 
67 #ifdef CONFIG_ZONE_DMA32
68 	max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
69 #endif
70 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
71 
72 	free_area_init(max_zone_pfns);
73 }
74 
75 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
76 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
77 {
78 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld kB)\n", name, b, t,
79 		  (((t) - (b)) >> 10));
80 }
81 
82 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
83 {
84 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld MB)\n", name, b, t,
85 		  (((t) - (b)) >> 20));
86 }
87 
88 static void __init print_vm_layout(void)
89 {
90 	pr_notice("Virtual kernel memory layout:\n");
91 	print_mlk("fixmap", (unsigned long)FIXADDR_START,
92 		  (unsigned long)FIXADDR_TOP);
93 	print_mlm("pci io", (unsigned long)PCI_IO_START,
94 		  (unsigned long)PCI_IO_END);
95 	print_mlm("vmemmap", (unsigned long)VMEMMAP_START,
96 		  (unsigned long)VMEMMAP_END);
97 	print_mlm("vmalloc", (unsigned long)VMALLOC_START,
98 		  (unsigned long)VMALLOC_END);
99 	print_mlm("lowmem", (unsigned long)PAGE_OFFSET,
100 		  (unsigned long)high_memory);
101 #ifdef CONFIG_64BIT
102 	print_mlm("kernel", (unsigned long)KERNEL_LINK_ADDR,
103 		  (unsigned long)ADDRESS_SPACE_END);
104 #endif
105 }
106 #else
107 static void print_vm_layout(void) { }
108 #endif /* CONFIG_DEBUG_VM */
109 
110 void __init mem_init(void)
111 {
112 #ifdef CONFIG_FLATMEM
113 	BUG_ON(!mem_map);
114 #endif /* CONFIG_FLATMEM */
115 
116 #ifdef CONFIG_SWIOTLB
117 	if (swiotlb_force == SWIOTLB_FORCE ||
118 	    max_pfn > PFN_DOWN(dma32_phys_limit))
119 		swiotlb_init(1);
120 	else
121 		swiotlb_force = SWIOTLB_NO_FORCE;
122 #endif
123 	high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
124 	memblock_free_all();
125 
126 	print_vm_layout();
127 }
128 
129 /*
130  * The default maximal physical memory size is -PAGE_OFFSET,
131  * limit the memory size via mem.
132  */
133 static phys_addr_t memory_limit = -PAGE_OFFSET;
134 
135 static int __init early_mem(char *p)
136 {
137 	u64 size;
138 
139 	if (!p)
140 		return 1;
141 
142 	size = memparse(p, &p) & PAGE_MASK;
143 	memory_limit = min_t(u64, size, memory_limit);
144 
145 	pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
146 
147 	return 0;
148 }
149 early_param("mem", early_mem);
150 
151 static void __init setup_bootmem(void)
152 {
153 	phys_addr_t vmlinux_end = __pa_symbol(&_end);
154 	phys_addr_t vmlinux_start = __pa_symbol(&_start);
155 	phys_addr_t max_mapped_addr = __pa(~(ulong)0);
156 	phys_addr_t dram_end;
157 
158 #ifdef CONFIG_XIP_KERNEL
159 	vmlinux_start = __pa_symbol(&_sdata);
160 #endif
161 
162 	memblock_enforce_memory_limit(memory_limit);
163 
164 	/*
165 	 * Reserve from the start of the kernel to the end of the kernel
166 	 */
167 #if defined(CONFIG_64BIT) && defined(CONFIG_STRICT_KERNEL_RWX)
168 	/*
169 	 * Make sure we align the reservation on PMD_SIZE since we will
170 	 * map the kernel in the linear mapping as read-only: we do not want
171 	 * any allocation to happen between _end and the next pmd aligned page.
172 	 */
173 	vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
174 #endif
175 	memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
176 
177 	dram_end = memblock_end_of_DRAM();
178 	/*
179 	 * memblock allocator is not aware of the fact that last 4K bytes of
180 	 * the addressable memory can not be mapped because of IS_ERR_VALUE
181 	 * macro. Make sure that last 4k bytes are not usable by memblock
182 	 * if end of dram is equal to maximum addressable memory.
183 	 */
184 	if (max_mapped_addr == (dram_end - 1))
185 		memblock_set_current_limit(max_mapped_addr - 4096);
186 
187 	min_low_pfn = PFN_UP(memblock_start_of_DRAM());
188 	max_low_pfn = max_pfn = PFN_DOWN(dram_end);
189 
190 	dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
191 	set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
192 
193 	reserve_initrd_mem();
194 	/*
195 	 * If DTB is built in, no need to reserve its memblock.
196 	 * Otherwise, do reserve it but avoid using
197 	 * early_init_fdt_reserve_self() since __pa() does
198 	 * not work for DTB pointers that are fixmap addresses
199 	 */
200 	if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
201 		memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
202 
203 	early_init_fdt_scan_reserved_mem();
204 	dma_contiguous_reserve(dma32_phys_limit);
205 	memblock_allow_resize();
206 }
207 
208 #ifdef CONFIG_MMU
209 static struct pt_alloc_ops _pt_ops __initdata;
210 
211 #ifdef CONFIG_XIP_KERNEL
212 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&_pt_ops))
213 #else
214 #define pt_ops _pt_ops
215 #endif
216 
217 unsigned long pfn_base __ro_after_init;
218 EXPORT_SYMBOL(pfn_base);
219 
220 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
221 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
222 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
223 
224 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
225 
226 #ifdef CONFIG_XIP_KERNEL
227 #define trampoline_pg_dir      ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
228 #define fixmap_pte             ((pte_t *)XIP_FIXUP(fixmap_pte))
229 #define early_pg_dir           ((pgd_t *)XIP_FIXUP(early_pg_dir))
230 #endif /* CONFIG_XIP_KERNEL */
231 
232 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
233 {
234 	unsigned long addr = __fix_to_virt(idx);
235 	pte_t *ptep;
236 
237 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
238 
239 	ptep = &fixmap_pte[pte_index(addr)];
240 
241 	if (pgprot_val(prot))
242 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
243 	else
244 		pte_clear(&init_mm, addr, ptep);
245 	local_flush_tlb_page(addr);
246 }
247 
248 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
249 {
250 	return (pte_t *)((uintptr_t)pa);
251 }
252 
253 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
254 {
255 	clear_fixmap(FIX_PTE);
256 	return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
257 }
258 
259 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
260 {
261 	return (pte_t *) __va(pa);
262 }
263 
264 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
265 {
266 	/*
267 	 * We only create PMD or PGD early mappings so we
268 	 * should never reach here with MMU disabled.
269 	 */
270 	BUG();
271 }
272 
273 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
274 {
275 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
276 }
277 
278 static phys_addr_t __init alloc_pte_late(uintptr_t va)
279 {
280 	unsigned long vaddr;
281 
282 	vaddr = __get_free_page(GFP_KERNEL);
283 	BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)));
284 
285 	return __pa(vaddr);
286 }
287 
288 static void __init create_pte_mapping(pte_t *ptep,
289 				      uintptr_t va, phys_addr_t pa,
290 				      phys_addr_t sz, pgprot_t prot)
291 {
292 	uintptr_t pte_idx = pte_index(va);
293 
294 	BUG_ON(sz != PAGE_SIZE);
295 
296 	if (pte_none(ptep[pte_idx]))
297 		ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
298 }
299 
300 #ifndef __PAGETABLE_PMD_FOLDED
301 
302 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
303 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
304 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
305 static pmd_t early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
306 
307 #ifdef CONFIG_XIP_KERNEL
308 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
309 #define fixmap_pmd     ((pmd_t *)XIP_FIXUP(fixmap_pmd))
310 #define early_pmd      ((pmd_t *)XIP_FIXUP(early_pmd))
311 #endif /* CONFIG_XIP_KERNEL */
312 
313 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
314 {
315 	/* Before MMU is enabled */
316 	return (pmd_t *)((uintptr_t)pa);
317 }
318 
319 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
320 {
321 	clear_fixmap(FIX_PMD);
322 	return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
323 }
324 
325 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
326 {
327 	return (pmd_t *) __va(pa);
328 }
329 
330 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
331 {
332 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
333 
334 	return (uintptr_t)early_pmd;
335 }
336 
337 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
338 {
339 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
340 }
341 
342 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
343 {
344 	unsigned long vaddr;
345 
346 	vaddr = __get_free_page(GFP_KERNEL);
347 	BUG_ON(!vaddr);
348 	return __pa(vaddr);
349 }
350 
351 static void __init create_pmd_mapping(pmd_t *pmdp,
352 				      uintptr_t va, phys_addr_t pa,
353 				      phys_addr_t sz, pgprot_t prot)
354 {
355 	pte_t *ptep;
356 	phys_addr_t pte_phys;
357 	uintptr_t pmd_idx = pmd_index(va);
358 
359 	if (sz == PMD_SIZE) {
360 		if (pmd_none(pmdp[pmd_idx]))
361 			pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
362 		return;
363 	}
364 
365 	if (pmd_none(pmdp[pmd_idx])) {
366 		pte_phys = pt_ops.alloc_pte(va);
367 		pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
368 		ptep = pt_ops.get_pte_virt(pte_phys);
369 		memset(ptep, 0, PAGE_SIZE);
370 	} else {
371 		pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
372 		ptep = pt_ops.get_pte_virt(pte_phys);
373 	}
374 
375 	create_pte_mapping(ptep, va, pa, sz, prot);
376 }
377 
378 #define pgd_next_t		pmd_t
379 #define alloc_pgd_next(__va)	pt_ops.alloc_pmd(__va)
380 #define get_pgd_next_virt(__pa)	pt_ops.get_pmd_virt(__pa)
381 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
382 	create_pmd_mapping(__nextp, __va, __pa, __sz, __prot)
383 #define fixmap_pgd_next		fixmap_pmd
384 #else
385 #define pgd_next_t		pte_t
386 #define alloc_pgd_next(__va)	pt_ops.alloc_pte(__va)
387 #define get_pgd_next_virt(__pa)	pt_ops.get_pte_virt(__pa)
388 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
389 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
390 #define fixmap_pgd_next		fixmap_pte
391 #endif
392 
393 void __init create_pgd_mapping(pgd_t *pgdp,
394 				      uintptr_t va, phys_addr_t pa,
395 				      phys_addr_t sz, pgprot_t prot)
396 {
397 	pgd_next_t *nextp;
398 	phys_addr_t next_phys;
399 	uintptr_t pgd_idx = pgd_index(va);
400 
401 	if (sz == PGDIR_SIZE) {
402 		if (pgd_val(pgdp[pgd_idx]) == 0)
403 			pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
404 		return;
405 	}
406 
407 	if (pgd_val(pgdp[pgd_idx]) == 0) {
408 		next_phys = alloc_pgd_next(va);
409 		pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
410 		nextp = get_pgd_next_virt(next_phys);
411 		memset(nextp, 0, PAGE_SIZE);
412 	} else {
413 		next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
414 		nextp = get_pgd_next_virt(next_phys);
415 	}
416 
417 	create_pgd_next_mapping(nextp, va, pa, sz, prot);
418 }
419 
420 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
421 {
422 	/* Upgrade to PMD_SIZE mappings whenever possible */
423 	if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
424 		return PAGE_SIZE;
425 
426 	return PMD_SIZE;
427 }
428 
429 #ifdef CONFIG_XIP_KERNEL
430 /* called from head.S with MMU off */
431 asmlinkage void __init __copy_data(void)
432 {
433 	void *from = (void *)(&_sdata);
434 	void *end = (void *)(&_end);
435 	void *to = (void *)CONFIG_PHYS_RAM_BASE;
436 	size_t sz = (size_t)(end - from + 1);
437 
438 	memcpy(to, from, sz);
439 }
440 #endif
441 
442 #ifdef CONFIG_STRICT_KERNEL_RWX
443 static __init pgprot_t pgprot_from_va(uintptr_t va)
444 {
445 	if (is_va_kernel_text(va))
446 		return PAGE_KERNEL_READ_EXEC;
447 
448 	/*
449 	 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
450 	 * we must protect its linear mapping alias from being executed and
451 	 * written.
452 	 * And rodata section is marked readonly in mark_rodata_ro.
453 	 */
454 	if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
455 		return PAGE_KERNEL_READ;
456 
457 	return PAGE_KERNEL;
458 }
459 
460 void mark_rodata_ro(void)
461 {
462 	set_kernel_memory(__start_rodata, _data, set_memory_ro);
463 	if (IS_ENABLED(CONFIG_64BIT))
464 		set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
465 				  set_memory_ro);
466 
467 	debug_checkwx();
468 }
469 #else
470 static __init pgprot_t pgprot_from_va(uintptr_t va)
471 {
472 	if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
473 		return PAGE_KERNEL;
474 
475 	return PAGE_KERNEL_EXEC;
476 }
477 #endif /* CONFIG_STRICT_KERNEL_RWX */
478 
479 /*
480  * setup_vm() is called from head.S with MMU-off.
481  *
482  * Following requirements should be honoured for setup_vm() to work
483  * correctly:
484  * 1) It should use PC-relative addressing for accessing kernel symbols.
485  *    To achieve this we always use GCC cmodel=medany.
486  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
487  *    so disable compiler instrumentation when FTRACE is enabled.
488  *
489  * Currently, the above requirements are honoured by using custom CFLAGS
490  * for init.o in mm/Makefile.
491  */
492 
493 #ifndef __riscv_cmodel_medany
494 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
495 #endif
496 
497 #ifdef CONFIG_XIP_KERNEL
498 static void __init create_kernel_page_table(pgd_t *pgdir, uintptr_t map_size,
499 					    __always_unused bool early)
500 {
501 	uintptr_t va, end_va;
502 
503 	/* Map the flash resident part */
504 	end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
505 	for (va = kernel_map.virt_addr; va < end_va; va += map_size)
506 		create_pgd_mapping(pgdir, va,
507 				   kernel_map.xiprom + (va - kernel_map.virt_addr),
508 				   map_size, PAGE_KERNEL_EXEC);
509 
510 	/* Map the data in RAM */
511 	end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
512 	for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += map_size)
513 		create_pgd_mapping(pgdir, va,
514 				   kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
515 				   map_size, PAGE_KERNEL);
516 }
517 #else
518 static void __init create_kernel_page_table(pgd_t *pgdir, uintptr_t map_size,
519 					    bool early)
520 {
521 	uintptr_t va, end_va;
522 
523 	end_va = kernel_map.virt_addr + kernel_map.size;
524 	for (va = kernel_map.virt_addr; va < end_va; va += map_size)
525 		create_pgd_mapping(pgdir, va,
526 				   kernel_map.phys_addr + (va - kernel_map.virt_addr),
527 				   map_size,
528 				   early ?
529 					PAGE_KERNEL_EXEC : pgprot_from_va(va));
530 }
531 #endif
532 
533 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
534 {
535 	uintptr_t __maybe_unused pa;
536 	uintptr_t map_size;
537 #ifndef __PAGETABLE_PMD_FOLDED
538 	pmd_t fix_bmap_spmd, fix_bmap_epmd;
539 #endif
540 
541 	kernel_map.virt_addr = KERNEL_LINK_ADDR;
542 
543 #ifdef CONFIG_XIP_KERNEL
544 	kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
545 	kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
546 
547 	kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
548 	kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
549 
550 	kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
551 #else
552 	kernel_map.phys_addr = (uintptr_t)(&_start);
553 	kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
554 #endif
555 
556 	kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr;
557 #ifdef CONFIG_64BIT
558 	kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
559 #endif
560 
561 	pfn_base = PFN_DOWN(kernel_map.phys_addr);
562 
563 	/*
564 	 * Enforce boot alignment requirements of RV32 and
565 	 * RV64 by only allowing PMD or PGD mappings.
566 	 */
567 	map_size = PMD_SIZE;
568 
569 	/* Sanity check alignment and size */
570 	BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
571 	BUG_ON((kernel_map.phys_addr % map_size) != 0);
572 
573 	pt_ops.alloc_pte = alloc_pte_early;
574 	pt_ops.get_pte_virt = get_pte_virt_early;
575 #ifndef __PAGETABLE_PMD_FOLDED
576 	pt_ops.alloc_pmd = alloc_pmd_early;
577 	pt_ops.get_pmd_virt = get_pmd_virt_early;
578 #endif
579 	/* Setup early PGD for fixmap */
580 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
581 			   (uintptr_t)fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
582 
583 #ifndef __PAGETABLE_PMD_FOLDED
584 	/* Setup fixmap PMD */
585 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
586 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
587 	/* Setup trampoline PGD and PMD */
588 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
589 			   (uintptr_t)trampoline_pmd, PGDIR_SIZE, PAGE_TABLE);
590 #ifdef CONFIG_XIP_KERNEL
591 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
592 			   kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
593 #else
594 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
595 			   kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
596 #endif
597 #else
598 	/* Setup trampoline PGD */
599 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
600 			   kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
601 #endif
602 
603 	/*
604 	 * Setup early PGD covering entire kernel which will allow
605 	 * us to reach paging_init(). We map all memory banks later
606 	 * in setup_vm_final() below.
607 	 */
608 	create_kernel_page_table(early_pg_dir, map_size, true);
609 
610 #ifndef __PAGETABLE_PMD_FOLDED
611 	/* Setup early PMD for DTB */
612 	create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
613 			   (uintptr_t)early_dtb_pmd, PGDIR_SIZE, PAGE_TABLE);
614 #ifndef CONFIG_BUILTIN_DTB
615 	/* Create two consecutive PMD mappings for FDT early scan */
616 	pa = dtb_pa & ~(PMD_SIZE - 1);
617 	create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA,
618 			   pa, PMD_SIZE, PAGE_KERNEL);
619 	create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE,
620 			   pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
621 	dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1));
622 #else /* CONFIG_BUILTIN_DTB */
623 #ifdef CONFIG_64BIT
624 	/*
625 	 * __va can't be used since it would return a linear mapping address
626 	 * whereas dtb_early_va will be used before setup_vm_final installs
627 	 * the linear mapping.
628 	 */
629 	dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
630 #else
631 	dtb_early_va = __va(dtb_pa);
632 #endif /* CONFIG_64BIT */
633 #endif /* CONFIG_BUILTIN_DTB */
634 #else
635 #ifndef CONFIG_BUILTIN_DTB
636 	/* Create two consecutive PGD mappings for FDT early scan */
637 	pa = dtb_pa & ~(PGDIR_SIZE - 1);
638 	create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
639 			   pa, PGDIR_SIZE, PAGE_KERNEL);
640 	create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA + PGDIR_SIZE,
641 			   pa + PGDIR_SIZE, PGDIR_SIZE, PAGE_KERNEL);
642 	dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PGDIR_SIZE - 1));
643 #else /* CONFIG_BUILTIN_DTB */
644 #ifdef CONFIG_64BIT
645 	dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
646 #else
647 	dtb_early_va = __va(dtb_pa);
648 #endif /* CONFIG_64BIT */
649 #endif /* CONFIG_BUILTIN_DTB */
650 #endif
651 	dtb_early_pa = dtb_pa;
652 
653 	/*
654 	 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
655 	 * range can not span multiple pmds.
656 	 */
657 	BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
658 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
659 
660 #ifndef __PAGETABLE_PMD_FOLDED
661 	/*
662 	 * Early ioremap fixmap is already created as it lies within first 2MB
663 	 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
664 	 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
665 	 * the user if not.
666 	 */
667 	fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
668 	fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
669 	if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
670 		WARN_ON(1);
671 		pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
672 			pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
673 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
674 			fix_to_virt(FIX_BTMAP_BEGIN));
675 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
676 			fix_to_virt(FIX_BTMAP_END));
677 
678 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
679 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
680 	}
681 #endif
682 }
683 
684 static void __init setup_vm_final(void)
685 {
686 	uintptr_t va, map_size;
687 	phys_addr_t pa, start, end;
688 	u64 i;
689 
690 	/**
691 	 * MMU is enabled at this point. But page table setup is not complete yet.
692 	 * fixmap page table alloc functions should be used at this point
693 	 */
694 	pt_ops.alloc_pte = alloc_pte_fixmap;
695 	pt_ops.get_pte_virt = get_pte_virt_fixmap;
696 #ifndef __PAGETABLE_PMD_FOLDED
697 	pt_ops.alloc_pmd = alloc_pmd_fixmap;
698 	pt_ops.get_pmd_virt = get_pmd_virt_fixmap;
699 #endif
700 	/* Setup swapper PGD for fixmap */
701 	create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
702 			   __pa_symbol(fixmap_pgd_next),
703 			   PGDIR_SIZE, PAGE_TABLE);
704 
705 	/* Map all memory banks in the linear mapping */
706 	for_each_mem_range(i, &start, &end) {
707 		if (start >= end)
708 			break;
709 		if (start <= __pa(PAGE_OFFSET) &&
710 		    __pa(PAGE_OFFSET) < end)
711 			start = __pa(PAGE_OFFSET);
712 
713 		map_size = best_map_size(start, end - start);
714 		for (pa = start; pa < end; pa += map_size) {
715 			va = (uintptr_t)__va(pa);
716 
717 			create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
718 					   pgprot_from_va(va));
719 		}
720 	}
721 
722 #ifdef CONFIG_64BIT
723 	/* Map the kernel */
724 	create_kernel_page_table(swapper_pg_dir, PMD_SIZE, false);
725 #endif
726 
727 	/* Clear fixmap PTE and PMD mappings */
728 	clear_fixmap(FIX_PTE);
729 	clear_fixmap(FIX_PMD);
730 
731 	/* Move to swapper page table */
732 	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE);
733 	local_flush_tlb_all();
734 
735 	/* generic page allocation functions must be used to setup page table */
736 	pt_ops.alloc_pte = alloc_pte_late;
737 	pt_ops.get_pte_virt = get_pte_virt_late;
738 #ifndef __PAGETABLE_PMD_FOLDED
739 	pt_ops.alloc_pmd = alloc_pmd_late;
740 	pt_ops.get_pmd_virt = get_pmd_virt_late;
741 #endif
742 }
743 #else
744 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
745 {
746 	dtb_early_va = (void *)dtb_pa;
747 	dtb_early_pa = dtb_pa;
748 }
749 
750 static inline void setup_vm_final(void)
751 {
752 }
753 #endif /* CONFIG_MMU */
754 
755 #ifdef CONFIG_KEXEC_CORE
756 /*
757  * reserve_crashkernel() - reserves memory for crash kernel
758  *
759  * This function reserves memory area given in "crashkernel=" kernel command
760  * line parameter. The memory reserved is used by dump capture kernel when
761  * primary kernel is crashing.
762  */
763 static void __init reserve_crashkernel(void)
764 {
765 	unsigned long long crash_base = 0;
766 	unsigned long long crash_size = 0;
767 	unsigned long search_start = memblock_start_of_DRAM();
768 	unsigned long search_end = memblock_end_of_DRAM();
769 
770 	int ret = 0;
771 
772 	/*
773 	 * Don't reserve a region for a crash kernel on a crash kernel
774 	 * since it doesn't make much sense and we have limited memory
775 	 * resources.
776 	 */
777 #ifdef CONFIG_CRASH_DUMP
778 	if (is_kdump_kernel()) {
779 		pr_info("crashkernel: ignoring reservation request\n");
780 		return;
781 	}
782 #endif
783 
784 	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
785 				&crash_size, &crash_base);
786 	if (ret || !crash_size)
787 		return;
788 
789 	crash_size = PAGE_ALIGN(crash_size);
790 
791 	if (crash_base == 0) {
792 		/*
793 		 * Current riscv boot protocol requires 2MB alignment for
794 		 * RV64 and 4MB alignment for RV32 (hugepage size)
795 		 */
796 		crash_base = memblock_find_in_range(search_start, search_end,
797 						    crash_size, PMD_SIZE);
798 
799 		if (crash_base == 0) {
800 			pr_warn("crashkernel: couldn't allocate %lldKB\n",
801 				crash_size >> 10);
802 			return;
803 		}
804 	} else {
805 		/* User specifies base address explicitly. */
806 		if (!memblock_is_region_memory(crash_base, crash_size)) {
807 			pr_warn("crashkernel: requested region is not memory\n");
808 			return;
809 		}
810 
811 		if (memblock_is_region_reserved(crash_base, crash_size)) {
812 			pr_warn("crashkernel: requested region is reserved\n");
813 			return;
814 		}
815 
816 
817 		if (!IS_ALIGNED(crash_base, PMD_SIZE)) {
818 			pr_warn("crashkernel: requested region is misaligned\n");
819 			return;
820 		}
821 	}
822 	memblock_reserve(crash_base, crash_size);
823 
824 	pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
825 		crash_base, crash_base + crash_size, crash_size >> 20);
826 
827 	crashk_res.start = crash_base;
828 	crashk_res.end = crash_base + crash_size - 1;
829 }
830 #endif /* CONFIG_KEXEC_CORE */
831 
832 #ifdef CONFIG_CRASH_DUMP
833 /*
834  * We keep track of the ELF core header of the crashed
835  * kernel with a reserved-memory region with compatible
836  * string "linux,elfcorehdr". Here we register a callback
837  * to populate elfcorehdr_addr/size when this region is
838  * present. Note that this region will be marked as
839  * reserved once we call early_init_fdt_scan_reserved_mem()
840  * later on.
841  */
842 static int __init elfcore_hdr_setup(struct reserved_mem *rmem)
843 {
844 	elfcorehdr_addr = rmem->base;
845 	elfcorehdr_size = rmem->size;
846 	return 0;
847 }
848 
849 RESERVEDMEM_OF_DECLARE(elfcorehdr, "linux,elfcorehdr", elfcore_hdr_setup);
850 #endif
851 
852 void __init paging_init(void)
853 {
854 	setup_bootmem();
855 	setup_vm_final();
856 }
857 
858 void __init misc_mem_init(void)
859 {
860 	early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
861 	arch_numa_init();
862 	sparse_init();
863 	zone_sizes_init();
864 #ifdef CONFIG_KEXEC_CORE
865 	reserve_crashkernel();
866 #endif
867 	memblock_dump_all();
868 }
869 
870 #ifdef CONFIG_SPARSEMEM_VMEMMAP
871 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
872 			       struct vmem_altmap *altmap)
873 {
874 	return vmemmap_populate_basepages(start, end, node, NULL);
875 }
876 #endif
877