Loading...
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 */
6
7#include <linux/init.h>
8#include <linux/mm.h>
9#include <linux/memblock.h>
10#include <linux/initrd.h>
11#include <linux/swap.h>
12#include <linux/sizes.h>
13#include <linux/of_fdt.h>
14#include <linux/libfdt.h>
15#include <linux/set_memory.h>
16
17#include <asm/fixmap.h>
18#include <asm/tlbflush.h>
19#include <asm/sections.h>
20#include <asm/soc.h>
21#include <asm/io.h>
22#include <asm/ptdump.h>
23
24#include "../kernel/head.h"
25
26unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
27 __page_aligned_bss;
28EXPORT_SYMBOL(empty_zero_page);
29
30extern char _start[];
31void *dtb_early_va;
32
33static void __init zone_sizes_init(void)
34{
35 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
36
37#ifdef CONFIG_ZONE_DMA32
38 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(min(4UL * SZ_1G,
39 (unsigned long) PFN_PHYS(max_low_pfn)));
40#endif
41 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
42
43 free_area_init(max_zone_pfns);
44}
45
46static void setup_zero_page(void)
47{
48 memset((void *)empty_zero_page, 0, PAGE_SIZE);
49}
50
51#if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
52static inline void print_mlk(char *name, unsigned long b, unsigned long t)
53{
54 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
55 (((t) - (b)) >> 10));
56}
57
58static inline void print_mlm(char *name, unsigned long b, unsigned long t)
59{
60 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
61 (((t) - (b)) >> 20));
62}
63
64static void print_vm_layout(void)
65{
66 pr_notice("Virtual kernel memory layout:\n");
67 print_mlk("fixmap", (unsigned long)FIXADDR_START,
68 (unsigned long)FIXADDR_TOP);
69 print_mlm("pci io", (unsigned long)PCI_IO_START,
70 (unsigned long)PCI_IO_END);
71 print_mlm("vmemmap", (unsigned long)VMEMMAP_START,
72 (unsigned long)VMEMMAP_END);
73 print_mlm("vmalloc", (unsigned long)VMALLOC_START,
74 (unsigned long)VMALLOC_END);
75 print_mlm("lowmem", (unsigned long)PAGE_OFFSET,
76 (unsigned long)high_memory);
77}
78#else
79static void print_vm_layout(void) { }
80#endif /* CONFIG_DEBUG_VM */
81
82void __init mem_init(void)
83{
84#ifdef CONFIG_FLATMEM
85 BUG_ON(!mem_map);
86#endif /* CONFIG_FLATMEM */
87
88 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
89 memblock_free_all();
90
91 mem_init_print_info(NULL);
92 print_vm_layout();
93}
94
95#ifdef CONFIG_BLK_DEV_INITRD
96static void __init setup_initrd(void)
97{
98 phys_addr_t start;
99 unsigned long size;
100
101 /* Ignore the virtul address computed during device tree parsing */
102 initrd_start = initrd_end = 0;
103
104 if (!phys_initrd_size)
105 return;
106 /*
107 * Round the memory region to page boundaries as per free_initrd_mem()
108 * This allows us to detect whether the pages overlapping the initrd
109 * are in use, but more importantly, reserves the entire set of pages
110 * as we don't want these pages allocated for other purposes.
111 */
112 start = round_down(phys_initrd_start, PAGE_SIZE);
113 size = phys_initrd_size + (phys_initrd_start - start);
114 size = round_up(size, PAGE_SIZE);
115
116 if (!memblock_is_region_memory(start, size)) {
117 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
118 (u64)start, size);
119 goto disable;
120 }
121
122 if (memblock_is_region_reserved(start, size)) {
123 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
124 (u64)start, size);
125 goto disable;
126 }
127
128 memblock_reserve(start, size);
129 /* Now convert initrd to virtual addresses */
130 initrd_start = (unsigned long)__va(phys_initrd_start);
131 initrd_end = initrd_start + phys_initrd_size;
132 initrd_below_start_ok = 1;
133
134 pr_info("Initial ramdisk at: 0x%p (%lu bytes)\n",
135 (void *)(initrd_start), size);
136 return;
137disable:
138 pr_cont(" - disabling initrd\n");
139 initrd_start = 0;
140 initrd_end = 0;
141}
142#endif /* CONFIG_BLK_DEV_INITRD */
143
144static phys_addr_t dtb_early_pa __initdata;
145
146void __init setup_bootmem(void)
147{
148 struct memblock_region *reg;
149 phys_addr_t mem_size = 0;
150 phys_addr_t total_mem = 0;
151 phys_addr_t mem_start, end = 0;
152 phys_addr_t vmlinux_end = __pa_symbol(&_end);
153 phys_addr_t vmlinux_start = __pa_symbol(&_start);
154
155 /* Find the memory region containing the kernel */
156 for_each_memblock(memory, reg) {
157 end = reg->base + reg->size;
158 if (!total_mem)
159 mem_start = reg->base;
160 if (reg->base <= vmlinux_start && vmlinux_end <= end)
161 BUG_ON(reg->size == 0);
162 total_mem = total_mem + reg->size;
163 }
164
165 /*
166 * Remove memblock from the end of usable area to the
167 * end of region
168 */
169 mem_size = min(total_mem, (phys_addr_t)-PAGE_OFFSET);
170 if (mem_start + mem_size < end)
171 memblock_remove(mem_start + mem_size,
172 end - mem_start - mem_size);
173
174 /* Reserve from the start of the kernel to the end of the kernel */
175 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
176
177 max_pfn = PFN_DOWN(memblock_end_of_DRAM());
178 max_low_pfn = max_pfn;
179 set_max_mapnr(max_low_pfn);
180
181#ifdef CONFIG_BLK_DEV_INITRD
182 setup_initrd();
183#endif /* CONFIG_BLK_DEV_INITRD */
184
185 /*
186 * Avoid using early_init_fdt_reserve_self() since __pa() does
187 * not work for DTB pointers that are fixmap addresses
188 */
189 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
190
191 early_init_fdt_scan_reserved_mem();
192 memblock_allow_resize();
193 memblock_dump_all();
194
195 for_each_memblock(memory, reg) {
196 unsigned long start_pfn = memblock_region_memory_base_pfn(reg);
197 unsigned long end_pfn = memblock_region_memory_end_pfn(reg);
198
199 memblock_set_node(PFN_PHYS(start_pfn),
200 PFN_PHYS(end_pfn - start_pfn),
201 &memblock.memory, 0);
202 }
203}
204
205#ifdef CONFIG_MMU
206unsigned long va_pa_offset;
207EXPORT_SYMBOL(va_pa_offset);
208unsigned long pfn_base;
209EXPORT_SYMBOL(pfn_base);
210
211pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
212pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
213pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
214static bool mmu_enabled;
215
216#define MAX_EARLY_MAPPING_SIZE SZ_128M
217
218pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
219
220void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
221{
222 unsigned long addr = __fix_to_virt(idx);
223 pte_t *ptep;
224
225 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
226
227 ptep = &fixmap_pte[pte_index(addr)];
228
229 if (pgprot_val(prot))
230 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
231 else
232 pte_clear(&init_mm, addr, ptep);
233 local_flush_tlb_page(addr);
234}
235
236static pte_t *__init get_pte_virt(phys_addr_t pa)
237{
238 if (mmu_enabled) {
239 clear_fixmap(FIX_PTE);
240 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
241 } else {
242 return (pte_t *)((uintptr_t)pa);
243 }
244}
245
246static phys_addr_t __init alloc_pte(uintptr_t va)
247{
248 /*
249 * We only create PMD or PGD early mappings so we
250 * should never reach here with MMU disabled.
251 */
252 BUG_ON(!mmu_enabled);
253
254 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
255}
256
257static void __init create_pte_mapping(pte_t *ptep,
258 uintptr_t va, phys_addr_t pa,
259 phys_addr_t sz, pgprot_t prot)
260{
261 uintptr_t pte_idx = pte_index(va);
262
263 BUG_ON(sz != PAGE_SIZE);
264
265 if (pte_none(ptep[pte_idx]))
266 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
267}
268
269#ifndef __PAGETABLE_PMD_FOLDED
270
271pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
272pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
273
274#if MAX_EARLY_MAPPING_SIZE < PGDIR_SIZE
275#define NUM_EARLY_PMDS 1UL
276#else
277#define NUM_EARLY_PMDS (1UL + MAX_EARLY_MAPPING_SIZE / PGDIR_SIZE)
278#endif
279pmd_t early_pmd[PTRS_PER_PMD * NUM_EARLY_PMDS] __initdata __aligned(PAGE_SIZE);
280
281static pmd_t *__init get_pmd_virt(phys_addr_t pa)
282{
283 if (mmu_enabled) {
284 clear_fixmap(FIX_PMD);
285 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
286 } else {
287 return (pmd_t *)((uintptr_t)pa);
288 }
289}
290
291static phys_addr_t __init alloc_pmd(uintptr_t va)
292{
293 uintptr_t pmd_num;
294
295 if (mmu_enabled)
296 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
297
298 pmd_num = (va - PAGE_OFFSET) >> PGDIR_SHIFT;
299 BUG_ON(pmd_num >= NUM_EARLY_PMDS);
300 return (uintptr_t)&early_pmd[pmd_num * PTRS_PER_PMD];
301}
302
303static void __init create_pmd_mapping(pmd_t *pmdp,
304 uintptr_t va, phys_addr_t pa,
305 phys_addr_t sz, pgprot_t prot)
306{
307 pte_t *ptep;
308 phys_addr_t pte_phys;
309 uintptr_t pmd_idx = pmd_index(va);
310
311 if (sz == PMD_SIZE) {
312 if (pmd_none(pmdp[pmd_idx]))
313 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
314 return;
315 }
316
317 if (pmd_none(pmdp[pmd_idx])) {
318 pte_phys = alloc_pte(va);
319 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
320 ptep = get_pte_virt(pte_phys);
321 memset(ptep, 0, PAGE_SIZE);
322 } else {
323 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
324 ptep = get_pte_virt(pte_phys);
325 }
326
327 create_pte_mapping(ptep, va, pa, sz, prot);
328}
329
330#define pgd_next_t pmd_t
331#define alloc_pgd_next(__va) alloc_pmd(__va)
332#define get_pgd_next_virt(__pa) get_pmd_virt(__pa)
333#define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
334 create_pmd_mapping(__nextp, __va, __pa, __sz, __prot)
335#define fixmap_pgd_next fixmap_pmd
336#else
337#define pgd_next_t pte_t
338#define alloc_pgd_next(__va) alloc_pte(__va)
339#define get_pgd_next_virt(__pa) get_pte_virt(__pa)
340#define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
341 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
342#define fixmap_pgd_next fixmap_pte
343#endif
344
345static void __init create_pgd_mapping(pgd_t *pgdp,
346 uintptr_t va, phys_addr_t pa,
347 phys_addr_t sz, pgprot_t prot)
348{
349 pgd_next_t *nextp;
350 phys_addr_t next_phys;
351 uintptr_t pgd_idx = pgd_index(va);
352
353 if (sz == PGDIR_SIZE) {
354 if (pgd_val(pgdp[pgd_idx]) == 0)
355 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
356 return;
357 }
358
359 if (pgd_val(pgdp[pgd_idx]) == 0) {
360 next_phys = alloc_pgd_next(va);
361 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
362 nextp = get_pgd_next_virt(next_phys);
363 memset(nextp, 0, PAGE_SIZE);
364 } else {
365 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
366 nextp = get_pgd_next_virt(next_phys);
367 }
368
369 create_pgd_next_mapping(nextp, va, pa, sz, prot);
370}
371
372static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
373{
374 /* Upgrade to PMD_SIZE mappings whenever possible */
375 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
376 return PAGE_SIZE;
377
378 return PMD_SIZE;
379}
380
381/*
382 * setup_vm() is called from head.S with MMU-off.
383 *
384 * Following requirements should be honoured for setup_vm() to work
385 * correctly:
386 * 1) It should use PC-relative addressing for accessing kernel symbols.
387 * To achieve this we always use GCC cmodel=medany.
388 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
389 * so disable compiler instrumentation when FTRACE is enabled.
390 *
391 * Currently, the above requirements are honoured by using custom CFLAGS
392 * for init.o in mm/Makefile.
393 */
394
395#ifndef __riscv_cmodel_medany
396#error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
397#endif
398
399asmlinkage void __init setup_vm(uintptr_t dtb_pa)
400{
401 uintptr_t va, end_va;
402 uintptr_t load_pa = (uintptr_t)(&_start);
403 uintptr_t load_sz = (uintptr_t)(&_end) - load_pa;
404 uintptr_t map_size = best_map_size(load_pa, MAX_EARLY_MAPPING_SIZE);
405
406 va_pa_offset = PAGE_OFFSET - load_pa;
407 pfn_base = PFN_DOWN(load_pa);
408
409 /*
410 * Enforce boot alignment requirements of RV32 and
411 * RV64 by only allowing PMD or PGD mappings.
412 */
413 BUG_ON(map_size == PAGE_SIZE);
414
415 /* Sanity check alignment and size */
416 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
417 BUG_ON((load_pa % map_size) != 0);
418 BUG_ON(load_sz > MAX_EARLY_MAPPING_SIZE);
419
420 /* Setup early PGD for fixmap */
421 create_pgd_mapping(early_pg_dir, FIXADDR_START,
422 (uintptr_t)fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
423
424#ifndef __PAGETABLE_PMD_FOLDED
425 /* Setup fixmap PMD */
426 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
427 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
428 /* Setup trampoline PGD and PMD */
429 create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
430 (uintptr_t)trampoline_pmd, PGDIR_SIZE, PAGE_TABLE);
431 create_pmd_mapping(trampoline_pmd, PAGE_OFFSET,
432 load_pa, PMD_SIZE, PAGE_KERNEL_EXEC);
433#else
434 /* Setup trampoline PGD */
435 create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
436 load_pa, PGDIR_SIZE, PAGE_KERNEL_EXEC);
437#endif
438
439 /*
440 * Setup early PGD covering entire kernel which will allows
441 * us to reach paging_init(). We map all memory banks later
442 * in setup_vm_final() below.
443 */
444 end_va = PAGE_OFFSET + load_sz;
445 for (va = PAGE_OFFSET; va < end_va; va += map_size)
446 create_pgd_mapping(early_pg_dir, va,
447 load_pa + (va - PAGE_OFFSET),
448 map_size, PAGE_KERNEL_EXEC);
449
450 /* Create fixed mapping for early FDT parsing */
451 end_va = __fix_to_virt(FIX_FDT) + FIX_FDT_SIZE;
452 for (va = __fix_to_virt(FIX_FDT); va < end_va; va += PAGE_SIZE)
453 create_pte_mapping(fixmap_pte, va,
454 dtb_pa + (va - __fix_to_virt(FIX_FDT)),
455 PAGE_SIZE, PAGE_KERNEL);
456
457 /* Save pointer to DTB for early FDT parsing */
458 dtb_early_va = (void *)fix_to_virt(FIX_FDT) + (dtb_pa & ~PAGE_MASK);
459 /* Save physical address for memblock reservation */
460 dtb_early_pa = dtb_pa;
461}
462
463static void __init setup_vm_final(void)
464{
465 uintptr_t va, map_size;
466 phys_addr_t pa, start, end;
467 struct memblock_region *reg;
468
469 /* Set mmu_enabled flag */
470 mmu_enabled = true;
471
472 /* Setup swapper PGD for fixmap */
473 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
474 __pa_symbol(fixmap_pgd_next),
475 PGDIR_SIZE, PAGE_TABLE);
476
477 /* Map all memory banks */
478 for_each_memblock(memory, reg) {
479 start = reg->base;
480 end = start + reg->size;
481
482 if (start >= end)
483 break;
484 if (memblock_is_nomap(reg))
485 continue;
486 if (start <= __pa(PAGE_OFFSET) &&
487 __pa(PAGE_OFFSET) < end)
488 start = __pa(PAGE_OFFSET);
489
490 map_size = best_map_size(start, end - start);
491 for (pa = start; pa < end; pa += map_size) {
492 va = (uintptr_t)__va(pa);
493 create_pgd_mapping(swapper_pg_dir, va, pa,
494 map_size, PAGE_KERNEL_EXEC);
495 }
496 }
497
498 /* Clear fixmap PTE and PMD mappings */
499 clear_fixmap(FIX_PTE);
500 clear_fixmap(FIX_PMD);
501
502 /* Move to swapper page table */
503 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE);
504 local_flush_tlb_all();
505}
506#else
507asmlinkage void __init setup_vm(uintptr_t dtb_pa)
508{
509#ifdef CONFIG_BUILTIN_DTB
510 dtb_early_va = soc_lookup_builtin_dtb();
511 if (!dtb_early_va) {
512 /* Fallback to first available DTS */
513 dtb_early_va = (void *) __dtb_start;
514 }
515#else
516 dtb_early_va = (void *)dtb_pa;
517#endif
518 dtb_early_pa = dtb_pa;
519}
520
521static inline void setup_vm_final(void)
522{
523}
524#endif /* CONFIG_MMU */
525
526#ifdef CONFIG_STRICT_KERNEL_RWX
527void mark_rodata_ro(void)
528{
529 unsigned long text_start = (unsigned long)_text;
530 unsigned long text_end = (unsigned long)_etext;
531 unsigned long rodata_start = (unsigned long)__start_rodata;
532 unsigned long data_start = (unsigned long)_data;
533 unsigned long max_low = (unsigned long)(__va(PFN_PHYS(max_low_pfn)));
534
535 set_memory_ro(text_start, (text_end - text_start) >> PAGE_SHIFT);
536 set_memory_ro(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
537 set_memory_nx(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
538 set_memory_nx(data_start, (max_low - data_start) >> PAGE_SHIFT);
539
540 debug_checkwx();
541}
542#endif
543
544static void __init resource_init(void)
545{
546 struct memblock_region *region;
547
548 for_each_memblock(memory, region) {
549 struct resource *res;
550
551 res = memblock_alloc(sizeof(struct resource), SMP_CACHE_BYTES);
552 if (!res)
553 panic("%s: Failed to allocate %zu bytes\n", __func__,
554 sizeof(struct resource));
555
556 if (memblock_is_nomap(region)) {
557 res->name = "reserved";
558 res->flags = IORESOURCE_MEM;
559 } else {
560 res->name = "System RAM";
561 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
562 }
563 res->start = __pfn_to_phys(memblock_region_memory_base_pfn(region));
564 res->end = __pfn_to_phys(memblock_region_memory_end_pfn(region)) - 1;
565
566 request_resource(&iomem_resource, res);
567 }
568}
569
570void __init paging_init(void)
571{
572 setup_vm_final();
573 sparse_init();
574 setup_zero_page();
575 zone_sizes_init();
576 resource_init();
577}
578
579#ifdef CONFIG_SPARSEMEM_VMEMMAP
580int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
581 struct vmem_altmap *altmap)
582{
583 return vmemmap_populate_basepages(start, end, node, NULL);
584}
585#endif
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
24#include <asm/fixmap.h>
25#include <asm/tlbflush.h>
26#include <asm/sections.h>
27#include <asm/soc.h>
28#include <asm/io.h>
29#include <asm/ptdump.h>
30#include <asm/numa.h>
31
32#include "../kernel/head.h"
33
34struct kernel_mapping kernel_map __ro_after_init;
35EXPORT_SYMBOL(kernel_map);
36#ifdef CONFIG_XIP_KERNEL
37#define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
38#endif
39
40#ifdef CONFIG_64BIT
41u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39;
42#else
43u64 satp_mode __ro_after_init = SATP_MODE_32;
44#endif
45EXPORT_SYMBOL(satp_mode);
46
47bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
48bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
49EXPORT_SYMBOL(pgtable_l4_enabled);
50EXPORT_SYMBOL(pgtable_l5_enabled);
51
52phys_addr_t phys_ram_base __ro_after_init;
53EXPORT_SYMBOL(phys_ram_base);
54
55unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
56 __page_aligned_bss;
57EXPORT_SYMBOL(empty_zero_page);
58
59extern char _start[];
60#define DTB_EARLY_BASE_VA PGDIR_SIZE
61void *_dtb_early_va __initdata;
62uintptr_t _dtb_early_pa __initdata;
63
64static phys_addr_t dma32_phys_limit __initdata;
65
66static void __init zone_sizes_init(void)
67{
68 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
69
70#ifdef CONFIG_ZONE_DMA32
71 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
72#endif
73 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
74
75 free_area_init(max_zone_pfns);
76}
77
78#if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
79
80#define LOG2_SZ_1K ilog2(SZ_1K)
81#define LOG2_SZ_1M ilog2(SZ_1M)
82#define LOG2_SZ_1G ilog2(SZ_1G)
83#define LOG2_SZ_1T ilog2(SZ_1T)
84
85static inline void print_mlk(char *name, unsigned long b, unsigned long t)
86{
87 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
88 (((t) - (b)) >> LOG2_SZ_1K));
89}
90
91static inline void print_mlm(char *name, unsigned long b, unsigned long t)
92{
93 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
94 (((t) - (b)) >> LOG2_SZ_1M));
95}
96
97static inline void print_mlg(char *name, unsigned long b, unsigned long t)
98{
99 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t,
100 (((t) - (b)) >> LOG2_SZ_1G));
101}
102
103#ifdef CONFIG_64BIT
104static inline void print_mlt(char *name, unsigned long b, unsigned long t)
105{
106 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t,
107 (((t) - (b)) >> LOG2_SZ_1T));
108}
109#else
110#define print_mlt(n, b, t) do {} while (0)
111#endif
112
113static inline void print_ml(char *name, unsigned long b, unsigned long t)
114{
115 unsigned long diff = t - b;
116
117 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10)
118 print_mlt(name, b, t);
119 else if ((diff >> LOG2_SZ_1G) >= 10)
120 print_mlg(name, b, t);
121 else if ((diff >> LOG2_SZ_1M) >= 10)
122 print_mlm(name, b, t);
123 else
124 print_mlk(name, b, t);
125}
126
127static void __init print_vm_layout(void)
128{
129 pr_notice("Virtual kernel memory layout:\n");
130 print_ml("fixmap", (unsigned long)FIXADDR_START,
131 (unsigned long)FIXADDR_TOP);
132 print_ml("pci io", (unsigned long)PCI_IO_START,
133 (unsigned long)PCI_IO_END);
134 print_ml("vmemmap", (unsigned long)VMEMMAP_START,
135 (unsigned long)VMEMMAP_END);
136 print_ml("vmalloc", (unsigned long)VMALLOC_START,
137 (unsigned long)VMALLOC_END);
138#ifdef CONFIG_64BIT
139 print_ml("modules", (unsigned long)MODULES_VADDR,
140 (unsigned long)MODULES_END);
141#endif
142 print_ml("lowmem", (unsigned long)PAGE_OFFSET,
143 (unsigned long)high_memory);
144 if (IS_ENABLED(CONFIG_64BIT)) {
145#ifdef CONFIG_KASAN
146 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
147#endif
148
149 print_ml("kernel", (unsigned long)KERNEL_LINK_ADDR,
150 (unsigned long)ADDRESS_SPACE_END);
151 }
152}
153#else
154static void print_vm_layout(void) { }
155#endif /* CONFIG_DEBUG_VM */
156
157void __init mem_init(void)
158{
159#ifdef CONFIG_FLATMEM
160 BUG_ON(!mem_map);
161#endif /* CONFIG_FLATMEM */
162
163 swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE);
164 memblock_free_all();
165
166 print_vm_layout();
167}
168
169/* Limit the memory size via mem. */
170static phys_addr_t memory_limit;
171
172static int __init early_mem(char *p)
173{
174 u64 size;
175
176 if (!p)
177 return 1;
178
179 size = memparse(p, &p) & PAGE_MASK;
180 memory_limit = min_t(u64, size, memory_limit);
181
182 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
183
184 return 0;
185}
186early_param("mem", early_mem);
187
188static void __init setup_bootmem(void)
189{
190 phys_addr_t vmlinux_end = __pa_symbol(&_end);
191 phys_addr_t max_mapped_addr;
192 phys_addr_t phys_ram_end, vmlinux_start;
193
194 if (IS_ENABLED(CONFIG_XIP_KERNEL))
195 vmlinux_start = __pa_symbol(&_sdata);
196 else
197 vmlinux_start = __pa_symbol(&_start);
198
199 memblock_enforce_memory_limit(memory_limit);
200
201 /*
202 * Make sure we align the reservation on PMD_SIZE since we will
203 * map the kernel in the linear mapping as read-only: we do not want
204 * any allocation to happen between _end and the next pmd aligned page.
205 */
206 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
207 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
208 /*
209 * Reserve from the start of the kernel to the end of the kernel
210 */
211 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
212
213 phys_ram_end = memblock_end_of_DRAM();
214 if (!IS_ENABLED(CONFIG_XIP_KERNEL))
215 phys_ram_base = memblock_start_of_DRAM();
216 /*
217 * memblock allocator is not aware of the fact that last 4K bytes of
218 * the addressable memory can not be mapped because of IS_ERR_VALUE
219 * macro. Make sure that last 4k bytes are not usable by memblock
220 * if end of dram is equal to maximum addressable memory. For 64-bit
221 * kernel, this problem can't happen here as the end of the virtual
222 * address space is occupied by the kernel mapping then this check must
223 * be done as soon as the kernel mapping base address is determined.
224 */
225 if (!IS_ENABLED(CONFIG_64BIT)) {
226 max_mapped_addr = __pa(~(ulong)0);
227 if (max_mapped_addr == (phys_ram_end - 1))
228 memblock_set_current_limit(max_mapped_addr - 4096);
229 }
230
231 min_low_pfn = PFN_UP(phys_ram_base);
232 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
233 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
234
235 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
236 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
237
238 reserve_initrd_mem();
239 /*
240 * If DTB is built in, no need to reserve its memblock.
241 * Otherwise, do reserve it but avoid using
242 * early_init_fdt_reserve_self() since __pa() does
243 * not work for DTB pointers that are fixmap addresses
244 */
245 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) {
246 /*
247 * In case the DTB is not located in a memory region we won't
248 * be able to locate it later on via the linear mapping and
249 * get a segfault when accessing it via __va(dtb_early_pa).
250 * To avoid this situation copy DTB to a memory region.
251 * Note that memblock_phys_alloc will also reserve DTB region.
252 */
253 if (!memblock_is_memory(dtb_early_pa)) {
254 size_t fdt_size = fdt_totalsize(dtb_early_va);
255 phys_addr_t new_dtb_early_pa = memblock_phys_alloc(fdt_size, PAGE_SIZE);
256 void *new_dtb_early_va = early_memremap(new_dtb_early_pa, fdt_size);
257
258 memcpy(new_dtb_early_va, dtb_early_va, fdt_size);
259 early_memunmap(new_dtb_early_va, fdt_size);
260 _dtb_early_pa = new_dtb_early_pa;
261 } else
262 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
263 }
264
265 dma_contiguous_reserve(dma32_phys_limit);
266 if (IS_ENABLED(CONFIG_64BIT))
267 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
268 memblock_allow_resize();
269}
270
271#ifdef CONFIG_MMU
272struct pt_alloc_ops pt_ops __initdata;
273
274unsigned long riscv_pfn_base __ro_after_init;
275EXPORT_SYMBOL(riscv_pfn_base);
276
277pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
278pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
279static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
280
281pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
282static p4d_t __maybe_unused early_dtb_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
283static pud_t __maybe_unused early_dtb_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
284static pmd_t __maybe_unused early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
285
286#ifdef CONFIG_XIP_KERNEL
287#define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
288#define riscv_pfn_base (*(unsigned long *)XIP_FIXUP(&riscv_pfn_base))
289#define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
290#define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
291#define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
292#endif /* CONFIG_XIP_KERNEL */
293
294static const pgprot_t protection_map[16] = {
295 [VM_NONE] = PAGE_NONE,
296 [VM_READ] = PAGE_READ,
297 [VM_WRITE] = PAGE_COPY,
298 [VM_WRITE | VM_READ] = PAGE_COPY,
299 [VM_EXEC] = PAGE_EXEC,
300 [VM_EXEC | VM_READ] = PAGE_READ_EXEC,
301 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC,
302 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_READ_EXEC,
303 [VM_SHARED] = PAGE_NONE,
304 [VM_SHARED | VM_READ] = PAGE_READ,
305 [VM_SHARED | VM_WRITE] = PAGE_SHARED,
306 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED,
307 [VM_SHARED | VM_EXEC] = PAGE_EXEC,
308 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC,
309 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC,
310 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC
311};
312DECLARE_VM_GET_PAGE_PROT
313
314void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
315{
316 unsigned long addr = __fix_to_virt(idx);
317 pte_t *ptep;
318
319 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
320
321 ptep = &fixmap_pte[pte_index(addr)];
322
323 if (pgprot_val(prot))
324 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
325 else
326 pte_clear(&init_mm, addr, ptep);
327 local_flush_tlb_page(addr);
328}
329
330static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
331{
332 return (pte_t *)((uintptr_t)pa);
333}
334
335static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
336{
337 clear_fixmap(FIX_PTE);
338 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
339}
340
341static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
342{
343 return (pte_t *) __va(pa);
344}
345
346static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
347{
348 /*
349 * We only create PMD or PGD early mappings so we
350 * should never reach here with MMU disabled.
351 */
352 BUG();
353}
354
355static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
356{
357 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
358}
359
360static phys_addr_t __init alloc_pte_late(uintptr_t va)
361{
362 unsigned long vaddr;
363
364 vaddr = __get_free_page(GFP_KERNEL);
365 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)));
366
367 return __pa(vaddr);
368}
369
370static void __init create_pte_mapping(pte_t *ptep,
371 uintptr_t va, phys_addr_t pa,
372 phys_addr_t sz, pgprot_t prot)
373{
374 uintptr_t pte_idx = pte_index(va);
375
376 BUG_ON(sz != PAGE_SIZE);
377
378 if (pte_none(ptep[pte_idx]))
379 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
380}
381
382#ifndef __PAGETABLE_PMD_FOLDED
383
384static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
385static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
386static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
387
388#ifdef CONFIG_XIP_KERNEL
389#define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
390#define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
391#define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
392#endif /* CONFIG_XIP_KERNEL */
393
394static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
395static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
396static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
397
398#ifdef CONFIG_XIP_KERNEL
399#define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
400#define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d))
401#define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d))
402#endif /* CONFIG_XIP_KERNEL */
403
404static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
405static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
406static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
407
408#ifdef CONFIG_XIP_KERNEL
409#define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
410#define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
411#define early_pud ((pud_t *)XIP_FIXUP(early_pud))
412#endif /* CONFIG_XIP_KERNEL */
413
414static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
415{
416 /* Before MMU is enabled */
417 return (pmd_t *)((uintptr_t)pa);
418}
419
420static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
421{
422 clear_fixmap(FIX_PMD);
423 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
424}
425
426static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
427{
428 return (pmd_t *) __va(pa);
429}
430
431static phys_addr_t __init alloc_pmd_early(uintptr_t va)
432{
433 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
434
435 return (uintptr_t)early_pmd;
436}
437
438static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
439{
440 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
441}
442
443static phys_addr_t __init alloc_pmd_late(uintptr_t va)
444{
445 unsigned long vaddr;
446
447 vaddr = __get_free_page(GFP_KERNEL);
448 BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page(vaddr)));
449
450 return __pa(vaddr);
451}
452
453static void __init create_pmd_mapping(pmd_t *pmdp,
454 uintptr_t va, phys_addr_t pa,
455 phys_addr_t sz, pgprot_t prot)
456{
457 pte_t *ptep;
458 phys_addr_t pte_phys;
459 uintptr_t pmd_idx = pmd_index(va);
460
461 if (sz == PMD_SIZE) {
462 if (pmd_none(pmdp[pmd_idx]))
463 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
464 return;
465 }
466
467 if (pmd_none(pmdp[pmd_idx])) {
468 pte_phys = pt_ops.alloc_pte(va);
469 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
470 ptep = pt_ops.get_pte_virt(pte_phys);
471 memset(ptep, 0, PAGE_SIZE);
472 } else {
473 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
474 ptep = pt_ops.get_pte_virt(pte_phys);
475 }
476
477 create_pte_mapping(ptep, va, pa, sz, prot);
478}
479
480static pud_t *__init get_pud_virt_early(phys_addr_t pa)
481{
482 return (pud_t *)((uintptr_t)pa);
483}
484
485static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
486{
487 clear_fixmap(FIX_PUD);
488 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
489}
490
491static pud_t *__init get_pud_virt_late(phys_addr_t pa)
492{
493 return (pud_t *)__va(pa);
494}
495
496static phys_addr_t __init alloc_pud_early(uintptr_t va)
497{
498 /* Only one PUD is available for early mapping */
499 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
500
501 return (uintptr_t)early_pud;
502}
503
504static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
505{
506 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
507}
508
509static phys_addr_t alloc_pud_late(uintptr_t va)
510{
511 unsigned long vaddr;
512
513 vaddr = __get_free_page(GFP_KERNEL);
514 BUG_ON(!vaddr);
515 return __pa(vaddr);
516}
517
518static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
519{
520 return (p4d_t *)((uintptr_t)pa);
521}
522
523static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
524{
525 clear_fixmap(FIX_P4D);
526 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
527}
528
529static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
530{
531 return (p4d_t *)__va(pa);
532}
533
534static phys_addr_t __init alloc_p4d_early(uintptr_t va)
535{
536 /* Only one P4D is available for early mapping */
537 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
538
539 return (uintptr_t)early_p4d;
540}
541
542static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
543{
544 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
545}
546
547static phys_addr_t alloc_p4d_late(uintptr_t va)
548{
549 unsigned long vaddr;
550
551 vaddr = __get_free_page(GFP_KERNEL);
552 BUG_ON(!vaddr);
553 return __pa(vaddr);
554}
555
556static void __init create_pud_mapping(pud_t *pudp,
557 uintptr_t va, phys_addr_t pa,
558 phys_addr_t sz, pgprot_t prot)
559{
560 pmd_t *nextp;
561 phys_addr_t next_phys;
562 uintptr_t pud_index = pud_index(va);
563
564 if (sz == PUD_SIZE) {
565 if (pud_val(pudp[pud_index]) == 0)
566 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
567 return;
568 }
569
570 if (pud_val(pudp[pud_index]) == 0) {
571 next_phys = pt_ops.alloc_pmd(va);
572 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
573 nextp = pt_ops.get_pmd_virt(next_phys);
574 memset(nextp, 0, PAGE_SIZE);
575 } else {
576 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
577 nextp = pt_ops.get_pmd_virt(next_phys);
578 }
579
580 create_pmd_mapping(nextp, va, pa, sz, prot);
581}
582
583static void __init create_p4d_mapping(p4d_t *p4dp,
584 uintptr_t va, phys_addr_t pa,
585 phys_addr_t sz, pgprot_t prot)
586{
587 pud_t *nextp;
588 phys_addr_t next_phys;
589 uintptr_t p4d_index = p4d_index(va);
590
591 if (sz == P4D_SIZE) {
592 if (p4d_val(p4dp[p4d_index]) == 0)
593 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
594 return;
595 }
596
597 if (p4d_val(p4dp[p4d_index]) == 0) {
598 next_phys = pt_ops.alloc_pud(va);
599 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
600 nextp = pt_ops.get_pud_virt(next_phys);
601 memset(nextp, 0, PAGE_SIZE);
602 } else {
603 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
604 nextp = pt_ops.get_pud_virt(next_phys);
605 }
606
607 create_pud_mapping(nextp, va, pa, sz, prot);
608}
609
610#define pgd_next_t p4d_t
611#define alloc_pgd_next(__va) (pgtable_l5_enabled ? \
612 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \
613 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
614#define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \
615 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \
616 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
617#define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
618 (pgtable_l5_enabled ? \
619 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
620 (pgtable_l4_enabled ? \
621 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \
622 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
623#define fixmap_pgd_next (pgtable_l5_enabled ? \
624 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \
625 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
626#define trampoline_pgd_next (pgtable_l5_enabled ? \
627 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \
628 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
629#define early_dtb_pgd_next (pgtable_l5_enabled ? \
630 (uintptr_t)early_dtb_p4d : (pgtable_l4_enabled ? \
631 (uintptr_t)early_dtb_pud : (uintptr_t)early_dtb_pmd))
632#else
633#define pgd_next_t pte_t
634#define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
635#define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
636#define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
637 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
638#define fixmap_pgd_next ((uintptr_t)fixmap_pte)
639#define early_dtb_pgd_next ((uintptr_t)early_dtb_pmd)
640#define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
641#define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
642#define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
643#endif /* __PAGETABLE_PMD_FOLDED */
644
645void __init create_pgd_mapping(pgd_t *pgdp,
646 uintptr_t va, phys_addr_t pa,
647 phys_addr_t sz, pgprot_t prot)
648{
649 pgd_next_t *nextp;
650 phys_addr_t next_phys;
651 uintptr_t pgd_idx = pgd_index(va);
652
653 if (sz == PGDIR_SIZE) {
654 if (pgd_val(pgdp[pgd_idx]) == 0)
655 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
656 return;
657 }
658
659 if (pgd_val(pgdp[pgd_idx]) == 0) {
660 next_phys = alloc_pgd_next(va);
661 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
662 nextp = get_pgd_next_virt(next_phys);
663 memset(nextp, 0, PAGE_SIZE);
664 } else {
665 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
666 nextp = get_pgd_next_virt(next_phys);
667 }
668
669 create_pgd_next_mapping(nextp, va, pa, sz, prot);
670}
671
672static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
673{
674 /* Upgrade to PMD_SIZE mappings whenever possible */
675 base &= PMD_SIZE - 1;
676 if (!base && size >= PMD_SIZE)
677 return PMD_SIZE;
678
679 return PAGE_SIZE;
680}
681
682#ifdef CONFIG_XIP_KERNEL
683#define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
684extern char _xiprom[], _exiprom[], __data_loc;
685
686/* called from head.S with MMU off */
687asmlinkage void __init __copy_data(void)
688{
689 void *from = (void *)(&__data_loc);
690 void *to = (void *)CONFIG_PHYS_RAM_BASE;
691 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
692
693 memcpy(to, from, sz);
694}
695#endif
696
697#ifdef CONFIG_STRICT_KERNEL_RWX
698static __init pgprot_t pgprot_from_va(uintptr_t va)
699{
700 if (is_va_kernel_text(va))
701 return PAGE_KERNEL_READ_EXEC;
702
703 /*
704 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
705 * we must protect its linear mapping alias from being executed and
706 * written.
707 * And rodata section is marked readonly in mark_rodata_ro.
708 */
709 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
710 return PAGE_KERNEL_READ;
711
712 return PAGE_KERNEL;
713}
714
715void mark_rodata_ro(void)
716{
717 set_kernel_memory(__start_rodata, _data, set_memory_ro);
718 if (IS_ENABLED(CONFIG_64BIT))
719 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
720 set_memory_ro);
721
722 debug_checkwx();
723}
724#else
725static __init pgprot_t pgprot_from_va(uintptr_t va)
726{
727 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
728 return PAGE_KERNEL;
729
730 return PAGE_KERNEL_EXEC;
731}
732#endif /* CONFIG_STRICT_KERNEL_RWX */
733
734#if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
735static void __init disable_pgtable_l5(void)
736{
737 pgtable_l5_enabled = false;
738 kernel_map.page_offset = PAGE_OFFSET_L4;
739 satp_mode = SATP_MODE_48;
740}
741
742static void __init disable_pgtable_l4(void)
743{
744 pgtable_l4_enabled = false;
745 kernel_map.page_offset = PAGE_OFFSET_L3;
746 satp_mode = SATP_MODE_39;
747}
748
749/*
750 * There is a simple way to determine if 4-level is supported by the
751 * underlying hardware: establish 1:1 mapping in 4-level page table mode
752 * then read SATP to see if the configuration was taken into account
753 * meaning sv48 is supported.
754 */
755static __init void set_satp_mode(void)
756{
757 u64 identity_satp, hw_satp;
758 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
759 bool check_l4 = false;
760
761 create_p4d_mapping(early_p4d,
762 set_satp_mode_pmd, (uintptr_t)early_pud,
763 P4D_SIZE, PAGE_TABLE);
764 create_pud_mapping(early_pud,
765 set_satp_mode_pmd, (uintptr_t)early_pmd,
766 PUD_SIZE, PAGE_TABLE);
767 /* Handle the case where set_satp_mode straddles 2 PMDs */
768 create_pmd_mapping(early_pmd,
769 set_satp_mode_pmd, set_satp_mode_pmd,
770 PMD_SIZE, PAGE_KERNEL_EXEC);
771 create_pmd_mapping(early_pmd,
772 set_satp_mode_pmd + PMD_SIZE,
773 set_satp_mode_pmd + PMD_SIZE,
774 PMD_SIZE, PAGE_KERNEL_EXEC);
775retry:
776 create_pgd_mapping(early_pg_dir,
777 set_satp_mode_pmd,
778 check_l4 ? (uintptr_t)early_pud : (uintptr_t)early_p4d,
779 PGDIR_SIZE, PAGE_TABLE);
780
781 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
782
783 local_flush_tlb_all();
784 csr_write(CSR_SATP, identity_satp);
785 hw_satp = csr_swap(CSR_SATP, 0ULL);
786 local_flush_tlb_all();
787
788 if (hw_satp != identity_satp) {
789 if (!check_l4) {
790 disable_pgtable_l5();
791 check_l4 = true;
792 memset(early_pg_dir, 0, PAGE_SIZE);
793 goto retry;
794 }
795 disable_pgtable_l4();
796 }
797
798 memset(early_pg_dir, 0, PAGE_SIZE);
799 memset(early_p4d, 0, PAGE_SIZE);
800 memset(early_pud, 0, PAGE_SIZE);
801 memset(early_pmd, 0, PAGE_SIZE);
802}
803#endif
804
805/*
806 * setup_vm() is called from head.S with MMU-off.
807 *
808 * Following requirements should be honoured for setup_vm() to work
809 * correctly:
810 * 1) It should use PC-relative addressing for accessing kernel symbols.
811 * To achieve this we always use GCC cmodel=medany.
812 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
813 * so disable compiler instrumentation when FTRACE is enabled.
814 *
815 * Currently, the above requirements are honoured by using custom CFLAGS
816 * for init.o in mm/Makefile.
817 */
818
819#ifndef __riscv_cmodel_medany
820#error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
821#endif
822
823#ifdef CONFIG_XIP_KERNEL
824static void __init create_kernel_page_table(pgd_t *pgdir,
825 __always_unused bool early)
826{
827 uintptr_t va, end_va;
828
829 /* Map the flash resident part */
830 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
831 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
832 create_pgd_mapping(pgdir, va,
833 kernel_map.xiprom + (va - kernel_map.virt_addr),
834 PMD_SIZE, PAGE_KERNEL_EXEC);
835
836 /* Map the data in RAM */
837 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
838 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
839 create_pgd_mapping(pgdir, va,
840 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
841 PMD_SIZE, PAGE_KERNEL);
842}
843#else
844static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
845{
846 uintptr_t va, end_va;
847
848 end_va = kernel_map.virt_addr + kernel_map.size;
849 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
850 create_pgd_mapping(pgdir, va,
851 kernel_map.phys_addr + (va - kernel_map.virt_addr),
852 PMD_SIZE,
853 early ?
854 PAGE_KERNEL_EXEC : pgprot_from_va(va));
855}
856#endif
857
858/*
859 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
860 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
861 * entry.
862 */
863static void __init create_fdt_early_page_table(pgd_t *pgdir, uintptr_t dtb_pa)
864{
865#ifndef CONFIG_BUILTIN_DTB
866 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
867
868 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
869 IS_ENABLED(CONFIG_64BIT) ? early_dtb_pgd_next : pa,
870 PGDIR_SIZE,
871 IS_ENABLED(CONFIG_64BIT) ? PAGE_TABLE : PAGE_KERNEL);
872
873 if (pgtable_l5_enabled)
874 create_p4d_mapping(early_dtb_p4d, DTB_EARLY_BASE_VA,
875 (uintptr_t)early_dtb_pud, P4D_SIZE, PAGE_TABLE);
876
877 if (pgtable_l4_enabled)
878 create_pud_mapping(early_dtb_pud, DTB_EARLY_BASE_VA,
879 (uintptr_t)early_dtb_pmd, PUD_SIZE, PAGE_TABLE);
880
881 if (IS_ENABLED(CONFIG_64BIT)) {
882 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA,
883 pa, PMD_SIZE, PAGE_KERNEL);
884 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE,
885 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
886 }
887
888 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1));
889#else
890 /*
891 * For 64-bit kernel, __va can't be used since it would return a linear
892 * mapping address whereas dtb_early_va will be used before
893 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
894 * kernel is mapped in the linear mapping, that makes no difference.
895 */
896 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
897#endif
898
899 dtb_early_pa = dtb_pa;
900}
901
902/*
903 * MMU is not enabled, the page tables are allocated directly using
904 * early_pmd/pud/p4d and the address returned is the physical one.
905 */
906static void __init pt_ops_set_early(void)
907{
908 pt_ops.alloc_pte = alloc_pte_early;
909 pt_ops.get_pte_virt = get_pte_virt_early;
910#ifndef __PAGETABLE_PMD_FOLDED
911 pt_ops.alloc_pmd = alloc_pmd_early;
912 pt_ops.get_pmd_virt = get_pmd_virt_early;
913 pt_ops.alloc_pud = alloc_pud_early;
914 pt_ops.get_pud_virt = get_pud_virt_early;
915 pt_ops.alloc_p4d = alloc_p4d_early;
916 pt_ops.get_p4d_virt = get_p4d_virt_early;
917#endif
918}
919
920/*
921 * MMU is enabled but page table setup is not complete yet.
922 * fixmap page table alloc functions must be used as a means to temporarily
923 * map the allocated physical pages since the linear mapping does not exist yet.
924 *
925 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
926 * but it will be used as described above.
927 */
928static void __init pt_ops_set_fixmap(void)
929{
930 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
931 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
932#ifndef __PAGETABLE_PMD_FOLDED
933 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
934 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
935 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
936 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
937 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
938 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
939#endif
940}
941
942/*
943 * MMU is enabled and page table setup is complete, so from now, we can use
944 * generic page allocation functions to setup page table.
945 */
946static void __init pt_ops_set_late(void)
947{
948 pt_ops.alloc_pte = alloc_pte_late;
949 pt_ops.get_pte_virt = get_pte_virt_late;
950#ifndef __PAGETABLE_PMD_FOLDED
951 pt_ops.alloc_pmd = alloc_pmd_late;
952 pt_ops.get_pmd_virt = get_pmd_virt_late;
953 pt_ops.alloc_pud = alloc_pud_late;
954 pt_ops.get_pud_virt = get_pud_virt_late;
955 pt_ops.alloc_p4d = alloc_p4d_late;
956 pt_ops.get_p4d_virt = get_p4d_virt_late;
957#endif
958}
959
960asmlinkage void __init setup_vm(uintptr_t dtb_pa)
961{
962 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
963
964 kernel_map.virt_addr = KERNEL_LINK_ADDR;
965 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
966
967#ifdef CONFIG_XIP_KERNEL
968 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
969 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
970
971 phys_ram_base = CONFIG_PHYS_RAM_BASE;
972 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
973 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
974
975 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
976#else
977 kernel_map.phys_addr = (uintptr_t)(&_start);
978 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
979#endif
980
981#if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
982 set_satp_mode();
983#endif
984
985 kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr;
986 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
987
988 riscv_pfn_base = PFN_DOWN(kernel_map.phys_addr);
989
990 /*
991 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
992 * kernel, whereas for 64-bit kernel, the end of the virtual address
993 * space is occupied by the modules/BPF/kernel mappings which reduces
994 * the available size of the linear mapping.
995 */
996 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
997
998 /* Sanity check alignment and size */
999 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1000 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1001
1002#ifdef CONFIG_64BIT
1003 /*
1004 * The last 4K bytes of the addressable memory can not be mapped because
1005 * of IS_ERR_VALUE macro.
1006 */
1007 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1008#endif
1009
1010 apply_early_boot_alternatives();
1011 pt_ops_set_early();
1012
1013 /* Setup early PGD for fixmap */
1014 create_pgd_mapping(early_pg_dir, FIXADDR_START,
1015 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1016
1017#ifndef __PAGETABLE_PMD_FOLDED
1018 /* Setup fixmap P4D and PUD */
1019 if (pgtable_l5_enabled)
1020 create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1021 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1022 /* Setup fixmap PUD and PMD */
1023 if (pgtable_l4_enabled)
1024 create_pud_mapping(fixmap_pud, FIXADDR_START,
1025 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1026 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1027 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1028 /* Setup trampoline PGD and PMD */
1029 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1030 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1031 if (pgtable_l5_enabled)
1032 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1033 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1034 if (pgtable_l4_enabled)
1035 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1036 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1037#ifdef CONFIG_XIP_KERNEL
1038 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1039 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1040#else
1041 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1042 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1043#endif
1044#else
1045 /* Setup trampoline PGD */
1046 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1047 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1048#endif
1049
1050 /*
1051 * Setup early PGD covering entire kernel which will allow
1052 * us to reach paging_init(). We map all memory banks later
1053 * in setup_vm_final() below.
1054 */
1055 create_kernel_page_table(early_pg_dir, true);
1056
1057 /* Setup early mapping for FDT early scan */
1058 create_fdt_early_page_table(early_pg_dir, dtb_pa);
1059
1060 /*
1061 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1062 * range can not span multiple pmds.
1063 */
1064 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1065 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1066
1067#ifndef __PAGETABLE_PMD_FOLDED
1068 /*
1069 * Early ioremap fixmap is already created as it lies within first 2MB
1070 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1071 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1072 * the user if not.
1073 */
1074 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1075 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1076 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1077 WARN_ON(1);
1078 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1079 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1080 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1081 fix_to_virt(FIX_BTMAP_BEGIN));
1082 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1083 fix_to_virt(FIX_BTMAP_END));
1084
1085 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1086 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1087 }
1088#endif
1089
1090 pt_ops_set_fixmap();
1091}
1092
1093static void __init setup_vm_final(void)
1094{
1095 uintptr_t va, map_size;
1096 phys_addr_t pa, start, end;
1097 u64 i;
1098
1099 /* Setup swapper PGD for fixmap */
1100 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1101 __pa_symbol(fixmap_pgd_next),
1102 PGDIR_SIZE, PAGE_TABLE);
1103
1104 /* Map all memory banks in the linear mapping */
1105 for_each_mem_range(i, &start, &end) {
1106 if (start >= end)
1107 break;
1108 if (start <= __pa(PAGE_OFFSET) &&
1109 __pa(PAGE_OFFSET) < end)
1110 start = __pa(PAGE_OFFSET);
1111 if (end >= __pa(PAGE_OFFSET) + memory_limit)
1112 end = __pa(PAGE_OFFSET) + memory_limit;
1113
1114 for (pa = start; pa < end; pa += map_size) {
1115 va = (uintptr_t)__va(pa);
1116 map_size = best_map_size(pa, end - pa);
1117
1118 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1119 pgprot_from_va(va));
1120 }
1121 }
1122
1123 /* Map the kernel */
1124 if (IS_ENABLED(CONFIG_64BIT))
1125 create_kernel_page_table(swapper_pg_dir, false);
1126
1127#ifdef CONFIG_KASAN
1128 kasan_swapper_init();
1129#endif
1130
1131 /* Clear fixmap PTE and PMD mappings */
1132 clear_fixmap(FIX_PTE);
1133 clear_fixmap(FIX_PMD);
1134 clear_fixmap(FIX_PUD);
1135 clear_fixmap(FIX_P4D);
1136
1137 /* Move to swapper page table */
1138 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1139 local_flush_tlb_all();
1140
1141 pt_ops_set_late();
1142}
1143#else
1144asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1145{
1146 dtb_early_va = (void *)dtb_pa;
1147 dtb_early_pa = dtb_pa;
1148}
1149
1150static inline void setup_vm_final(void)
1151{
1152}
1153#endif /* CONFIG_MMU */
1154
1155/*
1156 * reserve_crashkernel() - reserves memory for crash kernel
1157 *
1158 * This function reserves memory area given in "crashkernel=" kernel command
1159 * line parameter. The memory reserved is used by dump capture kernel when
1160 * primary kernel is crashing.
1161 */
1162static void __init reserve_crashkernel(void)
1163{
1164 unsigned long long crash_base = 0;
1165 unsigned long long crash_size = 0;
1166 unsigned long search_start = memblock_start_of_DRAM();
1167 unsigned long search_end = memblock_end_of_DRAM();
1168
1169 int ret = 0;
1170
1171 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1172 return;
1173 /*
1174 * Don't reserve a region for a crash kernel on a crash kernel
1175 * since it doesn't make much sense and we have limited memory
1176 * resources.
1177 */
1178 if (is_kdump_kernel()) {
1179 pr_info("crashkernel: ignoring reservation request\n");
1180 return;
1181 }
1182
1183 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
1184 &crash_size, &crash_base);
1185 if (ret || !crash_size)
1186 return;
1187
1188 crash_size = PAGE_ALIGN(crash_size);
1189
1190 if (crash_base) {
1191 search_start = crash_base;
1192 search_end = crash_base + crash_size;
1193 }
1194
1195 /*
1196 * Current riscv boot protocol requires 2MB alignment for
1197 * RV64 and 4MB alignment for RV32 (hugepage size)
1198 *
1199 * Try to alloc from 32bit addressible physical memory so that
1200 * swiotlb can work on the crash kernel.
1201 */
1202 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1203 search_start,
1204 min(search_end, (unsigned long) SZ_4G));
1205 if (crash_base == 0) {
1206 /* Try again without restricting region to 32bit addressible memory */
1207 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1208 search_start, search_end);
1209 if (crash_base == 0) {
1210 pr_warn("crashkernel: couldn't allocate %lldKB\n",
1211 crash_size >> 10);
1212 return;
1213 }
1214 }
1215
1216 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1217 crash_base, crash_base + crash_size, crash_size >> 20);
1218
1219 crashk_res.start = crash_base;
1220 crashk_res.end = crash_base + crash_size - 1;
1221}
1222
1223void __init paging_init(void)
1224{
1225 setup_bootmem();
1226 setup_vm_final();
1227}
1228
1229void __init misc_mem_init(void)
1230{
1231 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1232 arch_numa_init();
1233 sparse_init();
1234 zone_sizes_init();
1235 reserve_crashkernel();
1236 memblock_dump_all();
1237}
1238
1239#ifdef CONFIG_SPARSEMEM_VMEMMAP
1240int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1241 struct vmem_altmap *altmap)
1242{
1243 return vmemmap_populate_basepages(start, end, node, NULL);
1244}
1245#endif