1 // SPDX-License-Identifier: GPL-2.0-only
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>
9 #include <linux/init.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>
24 #include <asm/fixmap.h>
25 #include <asm/tlbflush.h>
26 #include <asm/sections.h>
29 #include <asm/ptdump.h>
32 #include "../kernel/head.h"
34 struct kernel_mapping kernel_map __ro_after_init;
35 EXPORT_SYMBOL(kernel_map);
36 #ifdef CONFIG_XIP_KERNEL
37 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
41 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39;
43 u64 satp_mode __ro_after_init = SATP_MODE_32;
45 EXPORT_SYMBOL(satp_mode);
47 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
48 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
49 EXPORT_SYMBOL(pgtable_l4_enabled);
50 EXPORT_SYMBOL(pgtable_l5_enabled);
52 phys_addr_t phys_ram_base __ro_after_init;
53 EXPORT_SYMBOL(phys_ram_base);
55 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
57 EXPORT_SYMBOL(empty_zero_page);
60 #define DTB_EARLY_BASE_VA PGDIR_SIZE
61 void *_dtb_early_va __initdata;
62 uintptr_t _dtb_early_pa __initdata;
64 static phys_addr_t dma32_phys_limit __initdata;
66 static void __init zone_sizes_init(void)
68 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
70 #ifdef CONFIG_ZONE_DMA32
71 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
73 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
75 free_area_init(max_zone_pfns);
78 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
79 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
81 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
85 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
87 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
91 static void __init print_vm_layout(void)
93 pr_notice("Virtual kernel memory layout:\n");
94 print_mlk("fixmap", (unsigned long)FIXADDR_START,
95 (unsigned long)FIXADDR_TOP);
96 print_mlm("pci io", (unsigned long)PCI_IO_START,
97 (unsigned long)PCI_IO_END);
98 print_mlm("vmemmap", (unsigned long)VMEMMAP_START,
99 (unsigned long)VMEMMAP_END);
100 print_mlm("vmalloc", (unsigned long)VMALLOC_START,
101 (unsigned long)VMALLOC_END);
102 print_mlm("lowmem", (unsigned long)PAGE_OFFSET,
103 (unsigned long)high_memory);
104 if (IS_ENABLED(CONFIG_64BIT)) {
106 print_mlm("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
109 print_mlm("kernel", (unsigned long)KERNEL_LINK_ADDR,
110 (unsigned long)ADDRESS_SPACE_END);
114 static void print_vm_layout(void) { }
115 #endif /* CONFIG_DEBUG_VM */
117 void __init mem_init(void)
119 #ifdef CONFIG_FLATMEM
121 #endif /* CONFIG_FLATMEM */
123 #ifdef CONFIG_SWIOTLB
124 if (swiotlb_force == SWIOTLB_FORCE ||
125 max_pfn > PFN_DOWN(dma32_phys_limit))
128 swiotlb_force = SWIOTLB_NO_FORCE;
135 /* Limit the memory size via mem. */
136 static phys_addr_t memory_limit;
138 static int __init early_mem(char *p)
145 size = memparse(p, &p) & PAGE_MASK;
146 memory_limit = min_t(u64, size, memory_limit);
148 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
152 early_param("mem", early_mem);
154 static void __init setup_bootmem(void)
156 phys_addr_t vmlinux_end = __pa_symbol(&_end);
157 phys_addr_t max_mapped_addr;
158 phys_addr_t phys_ram_end, vmlinux_start;
160 if (IS_ENABLED(CONFIG_XIP_KERNEL))
161 vmlinux_start = __pa_symbol(&_sdata);
163 vmlinux_start = __pa_symbol(&_start);
165 memblock_enforce_memory_limit(memory_limit);
168 * Make sure we align the reservation on PMD_SIZE since we will
169 * map the kernel in the linear mapping as read-only: we do not want
170 * any allocation to happen between _end and the next pmd aligned page.
172 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
173 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
175 * Reserve from the start of the kernel to the end of the kernel
177 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
179 phys_ram_end = memblock_end_of_DRAM();
180 if (!IS_ENABLED(CONFIG_XIP_KERNEL))
181 phys_ram_base = memblock_start_of_DRAM();
183 * memblock allocator is not aware of the fact that last 4K bytes of
184 * the addressable memory can not be mapped because of IS_ERR_VALUE
185 * macro. Make sure that last 4k bytes are not usable by memblock
186 * if end of dram is equal to maximum addressable memory. For 64-bit
187 * kernel, this problem can't happen here as the end of the virtual
188 * address space is occupied by the kernel mapping then this check must
189 * be done as soon as the kernel mapping base address is determined.
191 if (!IS_ENABLED(CONFIG_64BIT)) {
192 max_mapped_addr = __pa(~(ulong)0);
193 if (max_mapped_addr == (phys_ram_end - 1))
194 memblock_set_current_limit(max_mapped_addr - 4096);
197 min_low_pfn = PFN_UP(phys_ram_base);
198 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
199 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
201 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
202 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
204 reserve_initrd_mem();
206 * If DTB is built in, no need to reserve its memblock.
207 * Otherwise, do reserve it but avoid using
208 * early_init_fdt_reserve_self() since __pa() does
209 * not work for DTB pointers that are fixmap addresses
211 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) {
213 * In case the DTB is not located in a memory region we won't
214 * be able to locate it later on via the linear mapping and
215 * get a segfault when accessing it via __va(dtb_early_pa).
216 * To avoid this situation copy DTB to a memory region.
217 * Note that memblock_phys_alloc will also reserve DTB region.
219 if (!memblock_is_memory(dtb_early_pa)) {
220 size_t fdt_size = fdt_totalsize(dtb_early_va);
221 phys_addr_t new_dtb_early_pa = memblock_phys_alloc(fdt_size, PAGE_SIZE);
222 void *new_dtb_early_va = early_memremap(new_dtb_early_pa, fdt_size);
224 memcpy(new_dtb_early_va, dtb_early_va, fdt_size);
225 early_memunmap(new_dtb_early_va, fdt_size);
226 _dtb_early_pa = new_dtb_early_pa;
228 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
231 early_init_fdt_scan_reserved_mem();
232 dma_contiguous_reserve(dma32_phys_limit);
233 if (IS_ENABLED(CONFIG_64BIT))
234 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
235 memblock_allow_resize();
239 struct pt_alloc_ops pt_ops __initdata;
241 unsigned long riscv_pfn_base __ro_after_init;
242 EXPORT_SYMBOL(riscv_pfn_base);
244 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
245 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
246 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
248 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
249 static p4d_t __maybe_unused early_dtb_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
250 static pud_t __maybe_unused early_dtb_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
251 static pmd_t __maybe_unused early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
253 #ifdef CONFIG_XIP_KERNEL
254 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
255 #define riscv_pfn_base (*(unsigned long *)XIP_FIXUP(&riscv_pfn_base))
256 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
257 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
258 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
259 #endif /* CONFIG_XIP_KERNEL */
261 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
263 unsigned long addr = __fix_to_virt(idx);
266 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
268 ptep = &fixmap_pte[pte_index(addr)];
270 if (pgprot_val(prot))
271 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
273 pte_clear(&init_mm, addr, ptep);
274 local_flush_tlb_page(addr);
277 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
279 return (pte_t *)((uintptr_t)pa);
282 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
284 clear_fixmap(FIX_PTE);
285 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
288 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
290 return (pte_t *) __va(pa);
293 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
296 * We only create PMD or PGD early mappings so we
297 * should never reach here with MMU disabled.
302 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
304 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
307 static phys_addr_t __init alloc_pte_late(uintptr_t va)
311 vaddr = __get_free_page(GFP_KERNEL);
312 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)));
317 static void __init create_pte_mapping(pte_t *ptep,
318 uintptr_t va, phys_addr_t pa,
319 phys_addr_t sz, pgprot_t prot)
321 uintptr_t pte_idx = pte_index(va);
323 BUG_ON(sz != PAGE_SIZE);
325 if (pte_none(ptep[pte_idx]))
326 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
329 #ifndef __PAGETABLE_PMD_FOLDED
331 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
332 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
333 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
335 #ifdef CONFIG_XIP_KERNEL
336 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
337 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
338 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
339 #endif /* CONFIG_XIP_KERNEL */
341 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
342 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
343 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
345 #ifdef CONFIG_XIP_KERNEL
346 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
347 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d))
348 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d))
349 #endif /* CONFIG_XIP_KERNEL */
351 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
352 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
353 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
355 #ifdef CONFIG_XIP_KERNEL
356 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
357 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
358 #define early_pud ((pud_t *)XIP_FIXUP(early_pud))
359 #endif /* CONFIG_XIP_KERNEL */
361 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
363 /* Before MMU is enabled */
364 return (pmd_t *)((uintptr_t)pa);
367 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
369 clear_fixmap(FIX_PMD);
370 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
373 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
375 return (pmd_t *) __va(pa);
378 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
380 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
382 return (uintptr_t)early_pmd;
385 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
387 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
390 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
394 vaddr = __get_free_page(GFP_KERNEL);
395 BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page(vaddr)));
400 static void __init create_pmd_mapping(pmd_t *pmdp,
401 uintptr_t va, phys_addr_t pa,
402 phys_addr_t sz, pgprot_t prot)
405 phys_addr_t pte_phys;
406 uintptr_t pmd_idx = pmd_index(va);
408 if (sz == PMD_SIZE) {
409 if (pmd_none(pmdp[pmd_idx]))
410 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
414 if (pmd_none(pmdp[pmd_idx])) {
415 pte_phys = pt_ops.alloc_pte(va);
416 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
417 ptep = pt_ops.get_pte_virt(pte_phys);
418 memset(ptep, 0, PAGE_SIZE);
420 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
421 ptep = pt_ops.get_pte_virt(pte_phys);
424 create_pte_mapping(ptep, va, pa, sz, prot);
427 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
429 return (pud_t *)((uintptr_t)pa);
432 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
434 clear_fixmap(FIX_PUD);
435 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
438 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
440 return (pud_t *)__va(pa);
443 static phys_addr_t __init alloc_pud_early(uintptr_t va)
445 /* Only one PUD is available for early mapping */
446 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
448 return (uintptr_t)early_pud;
451 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
453 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
456 static phys_addr_t alloc_pud_late(uintptr_t va)
460 vaddr = __get_free_page(GFP_KERNEL);
465 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
467 return (p4d_t *)((uintptr_t)pa);
470 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
472 clear_fixmap(FIX_P4D);
473 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
476 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
478 return (p4d_t *)__va(pa);
481 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
483 /* Only one P4D is available for early mapping */
484 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
486 return (uintptr_t)early_p4d;
489 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
491 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
494 static phys_addr_t alloc_p4d_late(uintptr_t va)
498 vaddr = __get_free_page(GFP_KERNEL);
503 static void __init create_pud_mapping(pud_t *pudp,
504 uintptr_t va, phys_addr_t pa,
505 phys_addr_t sz, pgprot_t prot)
508 phys_addr_t next_phys;
509 uintptr_t pud_index = pud_index(va);
511 if (sz == PUD_SIZE) {
512 if (pud_val(pudp[pud_index]) == 0)
513 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
517 if (pud_val(pudp[pud_index]) == 0) {
518 next_phys = pt_ops.alloc_pmd(va);
519 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
520 nextp = pt_ops.get_pmd_virt(next_phys);
521 memset(nextp, 0, PAGE_SIZE);
523 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
524 nextp = pt_ops.get_pmd_virt(next_phys);
527 create_pmd_mapping(nextp, va, pa, sz, prot);
530 static void __init create_p4d_mapping(p4d_t *p4dp,
531 uintptr_t va, phys_addr_t pa,
532 phys_addr_t sz, pgprot_t prot)
535 phys_addr_t next_phys;
536 uintptr_t p4d_index = p4d_index(va);
538 if (sz == P4D_SIZE) {
539 if (p4d_val(p4dp[p4d_index]) == 0)
540 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
544 if (p4d_val(p4dp[p4d_index]) == 0) {
545 next_phys = pt_ops.alloc_pud(va);
546 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
547 nextp = pt_ops.get_pud_virt(next_phys);
548 memset(nextp, 0, PAGE_SIZE);
550 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
551 nextp = pt_ops.get_pud_virt(next_phys);
554 create_pud_mapping(nextp, va, pa, sz, prot);
557 #define pgd_next_t p4d_t
558 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \
559 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \
560 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
561 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \
562 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \
563 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
564 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
565 (pgtable_l5_enabled ? \
566 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
567 (pgtable_l4_enabled ? \
568 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \
569 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
570 #define fixmap_pgd_next (pgtable_l5_enabled ? \
571 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \
572 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
573 #define trampoline_pgd_next (pgtable_l5_enabled ? \
574 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \
575 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
576 #define early_dtb_pgd_next (pgtable_l5_enabled ? \
577 (uintptr_t)early_dtb_p4d : (pgtable_l4_enabled ? \
578 (uintptr_t)early_dtb_pud : (uintptr_t)early_dtb_pmd))
580 #define pgd_next_t pte_t
581 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
582 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
583 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
584 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
585 #define fixmap_pgd_next ((uintptr_t)fixmap_pte)
586 #define early_dtb_pgd_next ((uintptr_t)early_dtb_pmd)
587 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot)
588 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot)
589 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot)
590 #endif /* __PAGETABLE_PMD_FOLDED */
592 void __init create_pgd_mapping(pgd_t *pgdp,
593 uintptr_t va, phys_addr_t pa,
594 phys_addr_t sz, pgprot_t prot)
597 phys_addr_t next_phys;
598 uintptr_t pgd_idx = pgd_index(va);
600 if (sz == PGDIR_SIZE) {
601 if (pgd_val(pgdp[pgd_idx]) == 0)
602 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
606 if (pgd_val(pgdp[pgd_idx]) == 0) {
607 next_phys = alloc_pgd_next(va);
608 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
609 nextp = get_pgd_next_virt(next_phys);
610 memset(nextp, 0, PAGE_SIZE);
612 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
613 nextp = get_pgd_next_virt(next_phys);
616 create_pgd_next_mapping(nextp, va, pa, sz, prot);
619 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
621 /* Upgrade to PMD_SIZE mappings whenever possible */
622 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
628 #ifdef CONFIG_XIP_KERNEL
629 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
630 extern char _xiprom[], _exiprom[], __data_loc;
632 /* called from head.S with MMU off */
633 asmlinkage void __init __copy_data(void)
635 void *from = (void *)(&__data_loc);
636 void *to = (void *)CONFIG_PHYS_RAM_BASE;
637 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
639 memcpy(to, from, sz);
643 #ifdef CONFIG_STRICT_KERNEL_RWX
644 static __init pgprot_t pgprot_from_va(uintptr_t va)
646 if (is_va_kernel_text(va))
647 return PAGE_KERNEL_READ_EXEC;
650 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
651 * we must protect its linear mapping alias from being executed and
653 * And rodata section is marked readonly in mark_rodata_ro.
655 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
656 return PAGE_KERNEL_READ;
661 void mark_rodata_ro(void)
663 set_kernel_memory(__start_rodata, _data, set_memory_ro);
664 if (IS_ENABLED(CONFIG_64BIT))
665 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
671 static __init pgprot_t pgprot_from_va(uintptr_t va)
673 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
676 return PAGE_KERNEL_EXEC;
678 #endif /* CONFIG_STRICT_KERNEL_RWX */
681 static void __init disable_pgtable_l5(void)
683 pgtable_l5_enabled = false;
684 kernel_map.page_offset = PAGE_OFFSET_L4;
685 satp_mode = SATP_MODE_48;
688 static void __init disable_pgtable_l4(void)
690 pgtable_l4_enabled = false;
691 kernel_map.page_offset = PAGE_OFFSET_L3;
692 satp_mode = SATP_MODE_39;
696 * There is a simple way to determine if 4-level is supported by the
697 * underlying hardware: establish 1:1 mapping in 4-level page table mode
698 * then read SATP to see if the configuration was taken into account
699 * meaning sv48 is supported.
701 static __init void set_satp_mode(void)
703 u64 identity_satp, hw_satp;
704 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
705 bool check_l4 = false;
707 create_p4d_mapping(early_p4d,
708 set_satp_mode_pmd, (uintptr_t)early_pud,
709 P4D_SIZE, PAGE_TABLE);
710 create_pud_mapping(early_pud,
711 set_satp_mode_pmd, (uintptr_t)early_pmd,
712 PUD_SIZE, PAGE_TABLE);
713 /* Handle the case where set_satp_mode straddles 2 PMDs */
714 create_pmd_mapping(early_pmd,
715 set_satp_mode_pmd, set_satp_mode_pmd,
716 PMD_SIZE, PAGE_KERNEL_EXEC);
717 create_pmd_mapping(early_pmd,
718 set_satp_mode_pmd + PMD_SIZE,
719 set_satp_mode_pmd + PMD_SIZE,
720 PMD_SIZE, PAGE_KERNEL_EXEC);
722 create_pgd_mapping(early_pg_dir,
724 check_l4 ? (uintptr_t)early_pud : (uintptr_t)early_p4d,
725 PGDIR_SIZE, PAGE_TABLE);
727 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
729 local_flush_tlb_all();
730 csr_write(CSR_SATP, identity_satp);
731 hw_satp = csr_swap(CSR_SATP, 0ULL);
732 local_flush_tlb_all();
734 if (hw_satp != identity_satp) {
736 disable_pgtable_l5();
738 memset(early_pg_dir, 0, PAGE_SIZE);
741 disable_pgtable_l4();
744 memset(early_pg_dir, 0, PAGE_SIZE);
745 memset(early_p4d, 0, PAGE_SIZE);
746 memset(early_pud, 0, PAGE_SIZE);
747 memset(early_pmd, 0, PAGE_SIZE);
752 * setup_vm() is called from head.S with MMU-off.
754 * Following requirements should be honoured for setup_vm() to work
756 * 1) It should use PC-relative addressing for accessing kernel symbols.
757 * To achieve this we always use GCC cmodel=medany.
758 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
759 * so disable compiler instrumentation when FTRACE is enabled.
761 * Currently, the above requirements are honoured by using custom CFLAGS
762 * for init.o in mm/Makefile.
765 #ifndef __riscv_cmodel_medany
766 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
769 #ifdef CONFIG_XIP_KERNEL
770 static void __init create_kernel_page_table(pgd_t *pgdir,
771 __always_unused bool early)
773 uintptr_t va, end_va;
775 /* Map the flash resident part */
776 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
777 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
778 create_pgd_mapping(pgdir, va,
779 kernel_map.xiprom + (va - kernel_map.virt_addr),
780 PMD_SIZE, PAGE_KERNEL_EXEC);
782 /* Map the data in RAM */
783 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
784 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
785 create_pgd_mapping(pgdir, va,
786 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
787 PMD_SIZE, PAGE_KERNEL);
790 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
792 uintptr_t va, end_va;
794 end_va = kernel_map.virt_addr + kernel_map.size;
795 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
796 create_pgd_mapping(pgdir, va,
797 kernel_map.phys_addr + (va - kernel_map.virt_addr),
800 PAGE_KERNEL_EXEC : pgprot_from_va(va));
805 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
806 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
809 static void __init create_fdt_early_page_table(pgd_t *pgdir, uintptr_t dtb_pa)
811 #ifndef CONFIG_BUILTIN_DTB
812 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
814 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
815 IS_ENABLED(CONFIG_64BIT) ? early_dtb_pgd_next : pa,
817 IS_ENABLED(CONFIG_64BIT) ? PAGE_TABLE : PAGE_KERNEL);
819 if (pgtable_l5_enabled)
820 create_p4d_mapping(early_dtb_p4d, DTB_EARLY_BASE_VA,
821 (uintptr_t)early_dtb_pud, P4D_SIZE, PAGE_TABLE);
823 if (pgtable_l4_enabled)
824 create_pud_mapping(early_dtb_pud, DTB_EARLY_BASE_VA,
825 (uintptr_t)early_dtb_pmd, PUD_SIZE, PAGE_TABLE);
827 if (IS_ENABLED(CONFIG_64BIT)) {
828 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA,
829 pa, PMD_SIZE, PAGE_KERNEL);
830 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE,
831 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
834 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1));
837 * For 64-bit kernel, __va can't be used since it would return a linear
838 * mapping address whereas dtb_early_va will be used before
839 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
840 * kernel is mapped in the linear mapping, that makes no difference.
842 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
845 dtb_early_pa = dtb_pa;
849 * MMU is not enabled, the page tables are allocated directly using
850 * early_pmd/pud/p4d and the address returned is the physical one.
852 void __init pt_ops_set_early(void)
854 pt_ops.alloc_pte = alloc_pte_early;
855 pt_ops.get_pte_virt = get_pte_virt_early;
856 #ifndef __PAGETABLE_PMD_FOLDED
857 pt_ops.alloc_pmd = alloc_pmd_early;
858 pt_ops.get_pmd_virt = get_pmd_virt_early;
859 pt_ops.alloc_pud = alloc_pud_early;
860 pt_ops.get_pud_virt = get_pud_virt_early;
861 pt_ops.alloc_p4d = alloc_p4d_early;
862 pt_ops.get_p4d_virt = get_p4d_virt_early;
867 * MMU is enabled but page table setup is not complete yet.
868 * fixmap page table alloc functions must be used as a means to temporarily
869 * map the allocated physical pages since the linear mapping does not exist yet.
871 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
872 * but it will be used as described above.
874 void __init pt_ops_set_fixmap(void)
876 pt_ops.alloc_pte = kernel_mapping_pa_to_va((uintptr_t)alloc_pte_fixmap);
877 pt_ops.get_pte_virt = kernel_mapping_pa_to_va((uintptr_t)get_pte_virt_fixmap);
878 #ifndef __PAGETABLE_PMD_FOLDED
879 pt_ops.alloc_pmd = kernel_mapping_pa_to_va((uintptr_t)alloc_pmd_fixmap);
880 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va((uintptr_t)get_pmd_virt_fixmap);
881 pt_ops.alloc_pud = kernel_mapping_pa_to_va((uintptr_t)alloc_pud_fixmap);
882 pt_ops.get_pud_virt = kernel_mapping_pa_to_va((uintptr_t)get_pud_virt_fixmap);
883 pt_ops.alloc_p4d = kernel_mapping_pa_to_va((uintptr_t)alloc_p4d_fixmap);
884 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va((uintptr_t)get_p4d_virt_fixmap);
889 * MMU is enabled and page table setup is complete, so from now, we can use
890 * generic page allocation functions to setup page table.
892 void __init pt_ops_set_late(void)
894 pt_ops.alloc_pte = alloc_pte_late;
895 pt_ops.get_pte_virt = get_pte_virt_late;
896 #ifndef __PAGETABLE_PMD_FOLDED
897 pt_ops.alloc_pmd = alloc_pmd_late;
898 pt_ops.get_pmd_virt = get_pmd_virt_late;
899 pt_ops.alloc_pud = alloc_pud_late;
900 pt_ops.get_pud_virt = get_pud_virt_late;
901 pt_ops.alloc_p4d = alloc_p4d_late;
902 pt_ops.get_p4d_virt = get_p4d_virt_late;
906 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
908 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
910 kernel_map.virt_addr = KERNEL_LINK_ADDR;
911 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
913 #ifdef CONFIG_XIP_KERNEL
914 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
915 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
917 phys_ram_base = CONFIG_PHYS_RAM_BASE;
918 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
919 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
921 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
923 kernel_map.phys_addr = (uintptr_t)(&_start);
924 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
927 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
931 kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr;
932 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
934 riscv_pfn_base = PFN_DOWN(kernel_map.phys_addr);
937 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
938 * kernel, whereas for 64-bit kernel, the end of the virtual address
939 * space is occupied by the modules/BPF/kernel mappings which reduces
940 * the available size of the linear mapping.
942 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
944 /* Sanity check alignment and size */
945 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
946 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
950 * The last 4K bytes of the addressable memory can not be mapped because
951 * of IS_ERR_VALUE macro.
953 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
958 /* Setup early PGD for fixmap */
959 create_pgd_mapping(early_pg_dir, FIXADDR_START,
960 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
962 #ifndef __PAGETABLE_PMD_FOLDED
963 /* Setup fixmap P4D and PUD */
964 if (pgtable_l5_enabled)
965 create_p4d_mapping(fixmap_p4d, FIXADDR_START,
966 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
967 /* Setup fixmap PUD and PMD */
968 if (pgtable_l4_enabled)
969 create_pud_mapping(fixmap_pud, FIXADDR_START,
970 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
971 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
972 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
973 /* Setup trampoline PGD and PMD */
974 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
975 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
976 if (pgtable_l5_enabled)
977 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
978 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
979 if (pgtable_l4_enabled)
980 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
981 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
982 #ifdef CONFIG_XIP_KERNEL
983 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
984 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
986 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
987 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
990 /* Setup trampoline PGD */
991 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
992 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
996 * Setup early PGD covering entire kernel which will allow
997 * us to reach paging_init(). We map all memory banks later
998 * in setup_vm_final() below.
1000 create_kernel_page_table(early_pg_dir, true);
1002 /* Setup early mapping for FDT early scan */
1003 create_fdt_early_page_table(early_pg_dir, dtb_pa);
1006 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1007 * range can not span multiple pmds.
1009 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1010 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1012 #ifndef __PAGETABLE_PMD_FOLDED
1014 * Early ioremap fixmap is already created as it lies within first 2MB
1015 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1016 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1019 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1020 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1021 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1023 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1024 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1025 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1026 fix_to_virt(FIX_BTMAP_BEGIN));
1027 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1028 fix_to_virt(FIX_BTMAP_END));
1030 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1031 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1035 pt_ops_set_fixmap();
1038 static void __init setup_vm_final(void)
1040 uintptr_t va, map_size;
1041 phys_addr_t pa, start, end;
1044 /* Setup swapper PGD for fixmap */
1045 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1046 __pa_symbol(fixmap_pgd_next),
1047 PGDIR_SIZE, PAGE_TABLE);
1049 /* Map all memory banks in the linear mapping */
1050 for_each_mem_range(i, &start, &end) {
1053 if (start <= __pa(PAGE_OFFSET) &&
1054 __pa(PAGE_OFFSET) < end)
1055 start = __pa(PAGE_OFFSET);
1056 if (end >= __pa(PAGE_OFFSET) + memory_limit)
1057 end = __pa(PAGE_OFFSET) + memory_limit;
1059 map_size = best_map_size(start, end - start);
1060 for (pa = start; pa < end; pa += map_size) {
1061 va = (uintptr_t)__va(pa);
1063 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1064 pgprot_from_va(va));
1068 /* Map the kernel */
1069 if (IS_ENABLED(CONFIG_64BIT))
1070 create_kernel_page_table(swapper_pg_dir, false);
1073 kasan_swapper_init();
1076 /* Clear fixmap PTE and PMD mappings */
1077 clear_fixmap(FIX_PTE);
1078 clear_fixmap(FIX_PMD);
1079 clear_fixmap(FIX_PUD);
1080 clear_fixmap(FIX_P4D);
1082 /* Move to swapper page table */
1083 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1084 local_flush_tlb_all();
1089 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1091 dtb_early_va = (void *)dtb_pa;
1092 dtb_early_pa = dtb_pa;
1095 static inline void setup_vm_final(void)
1098 #endif /* CONFIG_MMU */
1101 * reserve_crashkernel() - reserves memory for crash kernel
1103 * This function reserves memory area given in "crashkernel=" kernel command
1104 * line parameter. The memory reserved is used by dump capture kernel when
1105 * primary kernel is crashing.
1107 static void __init reserve_crashkernel(void)
1109 unsigned long long crash_base = 0;
1110 unsigned long long crash_size = 0;
1111 unsigned long search_start = memblock_start_of_DRAM();
1112 unsigned long search_end = memblock_end_of_DRAM();
1116 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1119 * Don't reserve a region for a crash kernel on a crash kernel
1120 * since it doesn't make much sense and we have limited memory
1123 if (is_kdump_kernel()) {
1124 pr_info("crashkernel: ignoring reservation request\n");
1128 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
1129 &crash_size, &crash_base);
1130 if (ret || !crash_size)
1133 crash_size = PAGE_ALIGN(crash_size);
1136 search_start = crash_base;
1137 search_end = crash_base + crash_size;
1141 * Current riscv boot protocol requires 2MB alignment for
1142 * RV64 and 4MB alignment for RV32 (hugepage size)
1144 * Try to alloc from 32bit addressible physical memory so that
1145 * swiotlb can work on the crash kernel.
1147 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1149 min(search_end, (unsigned long) SZ_4G));
1150 if (crash_base == 0) {
1151 /* Try again without restricting region to 32bit addressible memory */
1152 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1153 search_start, search_end);
1154 if (crash_base == 0) {
1155 pr_warn("crashkernel: couldn't allocate %lldKB\n",
1161 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1162 crash_base, crash_base + crash_size, crash_size >> 20);
1164 crashk_res.start = crash_base;
1165 crashk_res.end = crash_base + crash_size - 1;
1168 void __init paging_init(void)
1174 void __init misc_mem_init(void)
1176 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1180 reserve_crashkernel();
1181 memblock_dump_all();
1184 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1185 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1186 struct vmem_altmap *altmap)
1188 return vmemmap_populate_basepages(start, end, node, NULL);