xref: /openbmc/linux/arch/riscv/mm/init.c (revision 35267cea)
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 for 32-bit kernel,
131  * whereas for 64-bit kernel, the end of the virtual address space is occupied
132  * by the modules/BPF/kernel mappings which reduces the available size of the
133  * linear mapping.
134  * Limit the memory size via mem.
135  */
136 #ifdef CONFIG_64BIT
137 static phys_addr_t memory_limit = -PAGE_OFFSET - SZ_4G;
138 #else
139 static phys_addr_t memory_limit = -PAGE_OFFSET;
140 #endif
141 
142 static int __init early_mem(char *p)
143 {
144 	u64 size;
145 
146 	if (!p)
147 		return 1;
148 
149 	size = memparse(p, &p) & PAGE_MASK;
150 	memory_limit = min_t(u64, size, memory_limit);
151 
152 	pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
153 
154 	return 0;
155 }
156 early_param("mem", early_mem);
157 
158 static void __init setup_bootmem(void)
159 {
160 	phys_addr_t vmlinux_end = __pa_symbol(&_end);
161 	phys_addr_t vmlinux_start = __pa_symbol(&_start);
162 	phys_addr_t __maybe_unused max_mapped_addr;
163 	phys_addr_t dram_end;
164 
165 #ifdef CONFIG_XIP_KERNEL
166 	vmlinux_start = __pa_symbol(&_sdata);
167 #endif
168 
169 	memblock_enforce_memory_limit(memory_limit);
170 
171 	/*
172 	 * Reserve from the start of the kernel to the end of the kernel
173 	 */
174 #if defined(CONFIG_64BIT) && defined(CONFIG_STRICT_KERNEL_RWX)
175 	/*
176 	 * Make sure we align the reservation on PMD_SIZE since we will
177 	 * map the kernel in the linear mapping as read-only: we do not want
178 	 * any allocation to happen between _end and the next pmd aligned page.
179 	 */
180 	vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
181 #endif
182 	memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
183 
184 	dram_end = memblock_end_of_DRAM();
185 
186 #ifndef CONFIG_64BIT
187 	/*
188 	 * memblock allocator is not aware of the fact that last 4K bytes of
189 	 * the addressable memory can not be mapped because of IS_ERR_VALUE
190 	 * macro. Make sure that last 4k bytes are not usable by memblock
191 	 * if end of dram is equal to maximum addressable memory.  For 64-bit
192 	 * kernel, this problem can't happen here as the end of the virtual
193 	 * address space is occupied by the kernel mapping then this check must
194 	 * be done in create_kernel_page_table.
195 	 */
196 	max_mapped_addr = __pa(~(ulong)0);
197 	if (max_mapped_addr == (dram_end - 1))
198 		memblock_set_current_limit(max_mapped_addr - 4096);
199 #endif
200 
201 	min_low_pfn = PFN_UP(memblock_start_of_DRAM());
202 	max_low_pfn = max_pfn = PFN_DOWN(dram_end);
203 
204 	dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
205 	set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
206 
207 	reserve_initrd_mem();
208 	/*
209 	 * If DTB is built in, no need to reserve its memblock.
210 	 * Otherwise, do reserve it but avoid using
211 	 * early_init_fdt_reserve_self() since __pa() does
212 	 * not work for DTB pointers that are fixmap addresses
213 	 */
214 	if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
215 		memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
216 
217 	early_init_fdt_scan_reserved_mem();
218 	dma_contiguous_reserve(dma32_phys_limit);
219 	memblock_allow_resize();
220 }
221 
222 #ifdef CONFIG_MMU
223 static struct pt_alloc_ops _pt_ops __initdata;
224 
225 #ifdef CONFIG_XIP_KERNEL
226 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&_pt_ops))
227 #else
228 #define pt_ops _pt_ops
229 #endif
230 
231 unsigned long pfn_base __ro_after_init;
232 EXPORT_SYMBOL(pfn_base);
233 
234 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
235 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
236 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
237 
238 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
239 
240 #ifdef CONFIG_XIP_KERNEL
241 #define trampoline_pg_dir      ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
242 #define fixmap_pte             ((pte_t *)XIP_FIXUP(fixmap_pte))
243 #define early_pg_dir           ((pgd_t *)XIP_FIXUP(early_pg_dir))
244 #endif /* CONFIG_XIP_KERNEL */
245 
246 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
247 {
248 	unsigned long addr = __fix_to_virt(idx);
249 	pte_t *ptep;
250 
251 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
252 
253 	ptep = &fixmap_pte[pte_index(addr)];
254 
255 	if (pgprot_val(prot))
256 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
257 	else
258 		pte_clear(&init_mm, addr, ptep);
259 	local_flush_tlb_page(addr);
260 }
261 
262 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
263 {
264 	return (pte_t *)((uintptr_t)pa);
265 }
266 
267 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
268 {
269 	clear_fixmap(FIX_PTE);
270 	return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
271 }
272 
273 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
274 {
275 	return (pte_t *) __va(pa);
276 }
277 
278 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
279 {
280 	/*
281 	 * We only create PMD or PGD early mappings so we
282 	 * should never reach here with MMU disabled.
283 	 */
284 	BUG();
285 }
286 
287 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
288 {
289 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
290 }
291 
292 static phys_addr_t __init alloc_pte_late(uintptr_t va)
293 {
294 	unsigned long vaddr;
295 
296 	vaddr = __get_free_page(GFP_KERNEL);
297 	BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)));
298 
299 	return __pa(vaddr);
300 }
301 
302 static void __init create_pte_mapping(pte_t *ptep,
303 				      uintptr_t va, phys_addr_t pa,
304 				      phys_addr_t sz, pgprot_t prot)
305 {
306 	uintptr_t pte_idx = pte_index(va);
307 
308 	BUG_ON(sz != PAGE_SIZE);
309 
310 	if (pte_none(ptep[pte_idx]))
311 		ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
312 }
313 
314 #ifndef __PAGETABLE_PMD_FOLDED
315 
316 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
317 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
318 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
319 static pmd_t early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
320 
321 #ifdef CONFIG_XIP_KERNEL
322 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
323 #define fixmap_pmd     ((pmd_t *)XIP_FIXUP(fixmap_pmd))
324 #define early_pmd      ((pmd_t *)XIP_FIXUP(early_pmd))
325 #endif /* CONFIG_XIP_KERNEL */
326 
327 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
328 {
329 	/* Before MMU is enabled */
330 	return (pmd_t *)((uintptr_t)pa);
331 }
332 
333 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
334 {
335 	clear_fixmap(FIX_PMD);
336 	return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
337 }
338 
339 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
340 {
341 	return (pmd_t *) __va(pa);
342 }
343 
344 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
345 {
346 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
347 
348 	return (uintptr_t)early_pmd;
349 }
350 
351 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
352 {
353 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
354 }
355 
356 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
357 {
358 	unsigned long vaddr;
359 
360 	vaddr = __get_free_page(GFP_KERNEL);
361 	BUG_ON(!vaddr);
362 	return __pa(vaddr);
363 }
364 
365 static void __init create_pmd_mapping(pmd_t *pmdp,
366 				      uintptr_t va, phys_addr_t pa,
367 				      phys_addr_t sz, pgprot_t prot)
368 {
369 	pte_t *ptep;
370 	phys_addr_t pte_phys;
371 	uintptr_t pmd_idx = pmd_index(va);
372 
373 	if (sz == PMD_SIZE) {
374 		if (pmd_none(pmdp[pmd_idx]))
375 			pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
376 		return;
377 	}
378 
379 	if (pmd_none(pmdp[pmd_idx])) {
380 		pte_phys = pt_ops.alloc_pte(va);
381 		pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
382 		ptep = pt_ops.get_pte_virt(pte_phys);
383 		memset(ptep, 0, PAGE_SIZE);
384 	} else {
385 		pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
386 		ptep = pt_ops.get_pte_virt(pte_phys);
387 	}
388 
389 	create_pte_mapping(ptep, va, pa, sz, prot);
390 }
391 
392 #define pgd_next_t		pmd_t
393 #define alloc_pgd_next(__va)	pt_ops.alloc_pmd(__va)
394 #define get_pgd_next_virt(__pa)	pt_ops.get_pmd_virt(__pa)
395 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
396 	create_pmd_mapping(__nextp, __va, __pa, __sz, __prot)
397 #define fixmap_pgd_next		fixmap_pmd
398 #else
399 #define pgd_next_t		pte_t
400 #define alloc_pgd_next(__va)	pt_ops.alloc_pte(__va)
401 #define get_pgd_next_virt(__pa)	pt_ops.get_pte_virt(__pa)
402 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
403 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
404 #define fixmap_pgd_next		fixmap_pte
405 #endif
406 
407 void __init create_pgd_mapping(pgd_t *pgdp,
408 				      uintptr_t va, phys_addr_t pa,
409 				      phys_addr_t sz, pgprot_t prot)
410 {
411 	pgd_next_t *nextp;
412 	phys_addr_t next_phys;
413 	uintptr_t pgd_idx = pgd_index(va);
414 
415 	if (sz == PGDIR_SIZE) {
416 		if (pgd_val(pgdp[pgd_idx]) == 0)
417 			pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
418 		return;
419 	}
420 
421 	if (pgd_val(pgdp[pgd_idx]) == 0) {
422 		next_phys = alloc_pgd_next(va);
423 		pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
424 		nextp = get_pgd_next_virt(next_phys);
425 		memset(nextp, 0, PAGE_SIZE);
426 	} else {
427 		next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
428 		nextp = get_pgd_next_virt(next_phys);
429 	}
430 
431 	create_pgd_next_mapping(nextp, va, pa, sz, prot);
432 }
433 
434 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
435 {
436 	/* Upgrade to PMD_SIZE mappings whenever possible */
437 	if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
438 		return PAGE_SIZE;
439 
440 	return PMD_SIZE;
441 }
442 
443 #ifdef CONFIG_XIP_KERNEL
444 /* called from head.S with MMU off */
445 asmlinkage void __init __copy_data(void)
446 {
447 	void *from = (void *)(&_sdata);
448 	void *end = (void *)(&_end);
449 	void *to = (void *)CONFIG_PHYS_RAM_BASE;
450 	size_t sz = (size_t)(end - from + 1);
451 
452 	memcpy(to, from, sz);
453 }
454 #endif
455 
456 #ifdef CONFIG_STRICT_KERNEL_RWX
457 static __init pgprot_t pgprot_from_va(uintptr_t va)
458 {
459 	if (is_va_kernel_text(va))
460 		return PAGE_KERNEL_READ_EXEC;
461 
462 	/*
463 	 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
464 	 * we must protect its linear mapping alias from being executed and
465 	 * written.
466 	 * And rodata section is marked readonly in mark_rodata_ro.
467 	 */
468 	if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
469 		return PAGE_KERNEL_READ;
470 
471 	return PAGE_KERNEL;
472 }
473 
474 void mark_rodata_ro(void)
475 {
476 	set_kernel_memory(__start_rodata, _data, set_memory_ro);
477 	if (IS_ENABLED(CONFIG_64BIT))
478 		set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
479 				  set_memory_ro);
480 
481 	debug_checkwx();
482 }
483 #else
484 static __init pgprot_t pgprot_from_va(uintptr_t va)
485 {
486 	if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
487 		return PAGE_KERNEL;
488 
489 	return PAGE_KERNEL_EXEC;
490 }
491 #endif /* CONFIG_STRICT_KERNEL_RWX */
492 
493 /*
494  * setup_vm() is called from head.S with MMU-off.
495  *
496  * Following requirements should be honoured for setup_vm() to work
497  * correctly:
498  * 1) It should use PC-relative addressing for accessing kernel symbols.
499  *    To achieve this we always use GCC cmodel=medany.
500  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
501  *    so disable compiler instrumentation when FTRACE is enabled.
502  *
503  * Currently, the above requirements are honoured by using custom CFLAGS
504  * for init.o in mm/Makefile.
505  */
506 
507 #ifndef __riscv_cmodel_medany
508 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
509 #endif
510 
511 #ifdef CONFIG_XIP_KERNEL
512 static void __init create_kernel_page_table(pgd_t *pgdir, uintptr_t map_size,
513 					    __always_unused bool early)
514 {
515 	uintptr_t va, end_va;
516 
517 	/* Map the flash resident part */
518 	end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
519 	for (va = kernel_map.virt_addr; va < end_va; va += map_size)
520 		create_pgd_mapping(pgdir, va,
521 				   kernel_map.xiprom + (va - kernel_map.virt_addr),
522 				   map_size, PAGE_KERNEL_EXEC);
523 
524 	/* Map the data in RAM */
525 	end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
526 	for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += map_size)
527 		create_pgd_mapping(pgdir, va,
528 				   kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
529 				   map_size, PAGE_KERNEL);
530 }
531 #else
532 static void __init create_kernel_page_table(pgd_t *pgdir, uintptr_t map_size,
533 					    bool early)
534 {
535 	uintptr_t va, end_va;
536 
537 	end_va = kernel_map.virt_addr + kernel_map.size;
538 	for (va = kernel_map.virt_addr; va < end_va; va += map_size)
539 		create_pgd_mapping(pgdir, va,
540 				   kernel_map.phys_addr + (va - kernel_map.virt_addr),
541 				   map_size,
542 				   early ?
543 					PAGE_KERNEL_EXEC : pgprot_from_va(va));
544 }
545 #endif
546 
547 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
548 {
549 	uintptr_t __maybe_unused pa;
550 	uintptr_t map_size;
551 #ifndef __PAGETABLE_PMD_FOLDED
552 	pmd_t fix_bmap_spmd, fix_bmap_epmd;
553 #endif
554 
555 	kernel_map.virt_addr = KERNEL_LINK_ADDR;
556 
557 #ifdef CONFIG_XIP_KERNEL
558 	kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
559 	kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
560 
561 	kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
562 	kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
563 
564 	kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
565 #else
566 	kernel_map.phys_addr = (uintptr_t)(&_start);
567 	kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
568 #endif
569 
570 	kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr;
571 #ifdef CONFIG_64BIT
572 	kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
573 #endif
574 
575 	pfn_base = PFN_DOWN(kernel_map.phys_addr);
576 
577 	/*
578 	 * Enforce boot alignment requirements of RV32 and
579 	 * RV64 by only allowing PMD or PGD mappings.
580 	 */
581 	map_size = PMD_SIZE;
582 
583 	/* Sanity check alignment and size */
584 	BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
585 	BUG_ON((kernel_map.phys_addr % map_size) != 0);
586 
587 #ifdef CONFIG_64BIT
588 	/*
589 	 * The last 4K bytes of the addressable memory can not be mapped because
590 	 * of IS_ERR_VALUE macro.
591 	 */
592 	BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
593 #endif
594 
595 	pt_ops.alloc_pte = alloc_pte_early;
596 	pt_ops.get_pte_virt = get_pte_virt_early;
597 #ifndef __PAGETABLE_PMD_FOLDED
598 	pt_ops.alloc_pmd = alloc_pmd_early;
599 	pt_ops.get_pmd_virt = get_pmd_virt_early;
600 #endif
601 	/* Setup early PGD for fixmap */
602 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
603 			   (uintptr_t)fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
604 
605 #ifndef __PAGETABLE_PMD_FOLDED
606 	/* Setup fixmap PMD */
607 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
608 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
609 	/* Setup trampoline PGD and PMD */
610 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
611 			   (uintptr_t)trampoline_pmd, PGDIR_SIZE, PAGE_TABLE);
612 #ifdef CONFIG_XIP_KERNEL
613 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
614 			   kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
615 #else
616 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
617 			   kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
618 #endif
619 #else
620 	/* Setup trampoline PGD */
621 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
622 			   kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
623 #endif
624 
625 	/*
626 	 * Setup early PGD covering entire kernel which will allow
627 	 * us to reach paging_init(). We map all memory banks later
628 	 * in setup_vm_final() below.
629 	 */
630 	create_kernel_page_table(early_pg_dir, map_size, true);
631 
632 #ifndef __PAGETABLE_PMD_FOLDED
633 	/* Setup early PMD for DTB */
634 	create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
635 			   (uintptr_t)early_dtb_pmd, PGDIR_SIZE, PAGE_TABLE);
636 #ifndef CONFIG_BUILTIN_DTB
637 	/* Create two consecutive PMD mappings for FDT early scan */
638 	pa = dtb_pa & ~(PMD_SIZE - 1);
639 	create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA,
640 			   pa, PMD_SIZE, PAGE_KERNEL);
641 	create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE,
642 			   pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
643 	dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1));
644 #else /* CONFIG_BUILTIN_DTB */
645 #ifdef CONFIG_64BIT
646 	/*
647 	 * __va can't be used since it would return a linear mapping address
648 	 * whereas dtb_early_va will be used before setup_vm_final installs
649 	 * the linear mapping.
650 	 */
651 	dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
652 #else
653 	dtb_early_va = __va(dtb_pa);
654 #endif /* CONFIG_64BIT */
655 #endif /* CONFIG_BUILTIN_DTB */
656 #else
657 #ifndef CONFIG_BUILTIN_DTB
658 	/* Create two consecutive PGD mappings for FDT early scan */
659 	pa = dtb_pa & ~(PGDIR_SIZE - 1);
660 	create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
661 			   pa, PGDIR_SIZE, PAGE_KERNEL);
662 	create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA + PGDIR_SIZE,
663 			   pa + PGDIR_SIZE, PGDIR_SIZE, PAGE_KERNEL);
664 	dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PGDIR_SIZE - 1));
665 #else /* CONFIG_BUILTIN_DTB */
666 #ifdef CONFIG_64BIT
667 	dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
668 #else
669 	dtb_early_va = __va(dtb_pa);
670 #endif /* CONFIG_64BIT */
671 #endif /* CONFIG_BUILTIN_DTB */
672 #endif
673 	dtb_early_pa = dtb_pa;
674 
675 	/*
676 	 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
677 	 * range can not span multiple pmds.
678 	 */
679 	BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
680 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
681 
682 #ifndef __PAGETABLE_PMD_FOLDED
683 	/*
684 	 * Early ioremap fixmap is already created as it lies within first 2MB
685 	 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
686 	 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
687 	 * the user if not.
688 	 */
689 	fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
690 	fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
691 	if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
692 		WARN_ON(1);
693 		pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
694 			pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
695 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
696 			fix_to_virt(FIX_BTMAP_BEGIN));
697 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
698 			fix_to_virt(FIX_BTMAP_END));
699 
700 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
701 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
702 	}
703 #endif
704 }
705 
706 static void __init setup_vm_final(void)
707 {
708 	uintptr_t va, map_size;
709 	phys_addr_t pa, start, end;
710 	u64 i;
711 
712 	/**
713 	 * MMU is enabled at this point. But page table setup is not complete yet.
714 	 * fixmap page table alloc functions should be used at this point
715 	 */
716 	pt_ops.alloc_pte = alloc_pte_fixmap;
717 	pt_ops.get_pte_virt = get_pte_virt_fixmap;
718 #ifndef __PAGETABLE_PMD_FOLDED
719 	pt_ops.alloc_pmd = alloc_pmd_fixmap;
720 	pt_ops.get_pmd_virt = get_pmd_virt_fixmap;
721 #endif
722 	/* Setup swapper PGD for fixmap */
723 	create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
724 			   __pa_symbol(fixmap_pgd_next),
725 			   PGDIR_SIZE, PAGE_TABLE);
726 
727 	/* Map all memory banks in the linear mapping */
728 	for_each_mem_range(i, &start, &end) {
729 		if (start >= end)
730 			break;
731 		if (start <= __pa(PAGE_OFFSET) &&
732 		    __pa(PAGE_OFFSET) < end)
733 			start = __pa(PAGE_OFFSET);
734 		if (end >= __pa(PAGE_OFFSET) + memory_limit)
735 			end = __pa(PAGE_OFFSET) + memory_limit;
736 
737 		map_size = best_map_size(start, end - start);
738 		for (pa = start; pa < end; pa += map_size) {
739 			va = (uintptr_t)__va(pa);
740 
741 			create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
742 					   pgprot_from_va(va));
743 		}
744 	}
745 
746 #ifdef CONFIG_64BIT
747 	/* Map the kernel */
748 	create_kernel_page_table(swapper_pg_dir, PMD_SIZE, false);
749 #endif
750 
751 	/* Clear fixmap PTE and PMD mappings */
752 	clear_fixmap(FIX_PTE);
753 	clear_fixmap(FIX_PMD);
754 
755 	/* Move to swapper page table */
756 	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE);
757 	local_flush_tlb_all();
758 
759 	/* generic page allocation functions must be used to setup page table */
760 	pt_ops.alloc_pte = alloc_pte_late;
761 	pt_ops.get_pte_virt = get_pte_virt_late;
762 #ifndef __PAGETABLE_PMD_FOLDED
763 	pt_ops.alloc_pmd = alloc_pmd_late;
764 	pt_ops.get_pmd_virt = get_pmd_virt_late;
765 #endif
766 }
767 #else
768 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
769 {
770 	dtb_early_va = (void *)dtb_pa;
771 	dtb_early_pa = dtb_pa;
772 }
773 
774 static inline void setup_vm_final(void)
775 {
776 }
777 #endif /* CONFIG_MMU */
778 
779 #ifdef CONFIG_KEXEC_CORE
780 /*
781  * reserve_crashkernel() - reserves memory for crash kernel
782  *
783  * This function reserves memory area given in "crashkernel=" kernel command
784  * line parameter. The memory reserved is used by dump capture kernel when
785  * primary kernel is crashing.
786  */
787 static void __init reserve_crashkernel(void)
788 {
789 	unsigned long long crash_base = 0;
790 	unsigned long long crash_size = 0;
791 	unsigned long search_start = memblock_start_of_DRAM();
792 	unsigned long search_end = memblock_end_of_DRAM();
793 
794 	int ret = 0;
795 
796 	/*
797 	 * Don't reserve a region for a crash kernel on a crash kernel
798 	 * since it doesn't make much sense and we have limited memory
799 	 * resources.
800 	 */
801 #ifdef CONFIG_CRASH_DUMP
802 	if (is_kdump_kernel()) {
803 		pr_info("crashkernel: ignoring reservation request\n");
804 		return;
805 	}
806 #endif
807 
808 	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
809 				&crash_size, &crash_base);
810 	if (ret || !crash_size)
811 		return;
812 
813 	crash_size = PAGE_ALIGN(crash_size);
814 
815 	if (crash_base == 0) {
816 		/*
817 		 * Current riscv boot protocol requires 2MB alignment for
818 		 * RV64 and 4MB alignment for RV32 (hugepage size)
819 		 */
820 		crash_base = memblock_find_in_range(search_start, search_end,
821 						    crash_size, PMD_SIZE);
822 
823 		if (crash_base == 0) {
824 			pr_warn("crashkernel: couldn't allocate %lldKB\n",
825 				crash_size >> 10);
826 			return;
827 		}
828 	} else {
829 		/* User specifies base address explicitly. */
830 		if (!memblock_is_region_memory(crash_base, crash_size)) {
831 			pr_warn("crashkernel: requested region is not memory\n");
832 			return;
833 		}
834 
835 		if (memblock_is_region_reserved(crash_base, crash_size)) {
836 			pr_warn("crashkernel: requested region is reserved\n");
837 			return;
838 		}
839 
840 
841 		if (!IS_ALIGNED(crash_base, PMD_SIZE)) {
842 			pr_warn("crashkernel: requested region is misaligned\n");
843 			return;
844 		}
845 	}
846 	memblock_reserve(crash_base, crash_size);
847 
848 	pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
849 		crash_base, crash_base + crash_size, crash_size >> 20);
850 
851 	crashk_res.start = crash_base;
852 	crashk_res.end = crash_base + crash_size - 1;
853 }
854 #endif /* CONFIG_KEXEC_CORE */
855 
856 #ifdef CONFIG_CRASH_DUMP
857 /*
858  * We keep track of the ELF core header of the crashed
859  * kernel with a reserved-memory region with compatible
860  * string "linux,elfcorehdr". Here we register a callback
861  * to populate elfcorehdr_addr/size when this region is
862  * present. Note that this region will be marked as
863  * reserved once we call early_init_fdt_scan_reserved_mem()
864  * later on.
865  */
866 static int __init elfcore_hdr_setup(struct reserved_mem *rmem)
867 {
868 	elfcorehdr_addr = rmem->base;
869 	elfcorehdr_size = rmem->size;
870 	return 0;
871 }
872 
873 RESERVEDMEM_OF_DECLARE(elfcorehdr, "linux,elfcorehdr", elfcore_hdr_setup);
874 #endif
875 
876 void __init paging_init(void)
877 {
878 	setup_bootmem();
879 	setup_vm_final();
880 }
881 
882 void __init misc_mem_init(void)
883 {
884 	early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
885 	arch_numa_init();
886 	sparse_init();
887 	zone_sizes_init();
888 #ifdef CONFIG_KEXEC_CORE
889 	reserve_crashkernel();
890 #endif
891 	memblock_dump_all();
892 }
893 
894 #ifdef CONFIG_SPARSEMEM_VMEMMAP
895 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
896 			       struct vmem_altmap *altmap)
897 {
898 	return vmemmap_populate_basepages(start, end, node, NULL);
899 }
900 #endif
901