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))
212 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
214 early_init_fdt_scan_reserved_mem();
215 dma_contiguous_reserve(dma32_phys_limit);
216 if (IS_ENABLED(CONFIG_64BIT))
217 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
218 memblock_allow_resize();
222 struct pt_alloc_ops pt_ops __initdata;
224 unsigned long riscv_pfn_base __ro_after_init;
225 EXPORT_SYMBOL(riscv_pfn_base);
227 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
228 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
229 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
231 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
232 static p4d_t __maybe_unused early_dtb_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
233 static pud_t __maybe_unused early_dtb_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
234 static pmd_t __maybe_unused early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
236 #ifdef CONFIG_XIP_KERNEL
237 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
238 #define riscv_pfn_base (*(unsigned long *)XIP_FIXUP(&riscv_pfn_base))
239 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
240 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
241 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
242 #endif /* CONFIG_XIP_KERNEL */
244 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
246 unsigned long addr = __fix_to_virt(idx);
249 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
251 ptep = &fixmap_pte[pte_index(addr)];
253 if (pgprot_val(prot))
254 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
256 pte_clear(&init_mm, addr, ptep);
257 local_flush_tlb_page(addr);
260 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
262 return (pte_t *)((uintptr_t)pa);
265 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
267 clear_fixmap(FIX_PTE);
268 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
271 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
273 return (pte_t *) __va(pa);
276 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
279 * We only create PMD or PGD early mappings so we
280 * should never reach here with MMU disabled.
285 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
287 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
290 static phys_addr_t __init alloc_pte_late(uintptr_t va)
294 vaddr = __get_free_page(GFP_KERNEL);
295 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)));
300 static void __init create_pte_mapping(pte_t *ptep,
301 uintptr_t va, phys_addr_t pa,
302 phys_addr_t sz, pgprot_t prot)
304 uintptr_t pte_idx = pte_index(va);
306 BUG_ON(sz != PAGE_SIZE);
308 if (pte_none(ptep[pte_idx]))
309 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
312 #ifndef __PAGETABLE_PMD_FOLDED
314 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
315 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
316 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
318 #ifdef CONFIG_XIP_KERNEL
319 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
320 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
321 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
322 #endif /* CONFIG_XIP_KERNEL */
324 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
325 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
326 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
328 #ifdef CONFIG_XIP_KERNEL
329 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
330 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d))
331 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d))
332 #endif /* CONFIG_XIP_KERNEL */
334 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
335 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
336 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
338 #ifdef CONFIG_XIP_KERNEL
339 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
340 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
341 #define early_pud ((pud_t *)XIP_FIXUP(early_pud))
342 #endif /* CONFIG_XIP_KERNEL */
344 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
346 /* Before MMU is enabled */
347 return (pmd_t *)((uintptr_t)pa);
350 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
352 clear_fixmap(FIX_PMD);
353 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
356 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
358 return (pmd_t *) __va(pa);
361 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
363 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
365 return (uintptr_t)early_pmd;
368 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
370 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
373 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
377 vaddr = __get_free_page(GFP_KERNEL);
378 BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page(vaddr)));
383 static void __init create_pmd_mapping(pmd_t *pmdp,
384 uintptr_t va, phys_addr_t pa,
385 phys_addr_t sz, pgprot_t prot)
388 phys_addr_t pte_phys;
389 uintptr_t pmd_idx = pmd_index(va);
391 if (sz == PMD_SIZE) {
392 if (pmd_none(pmdp[pmd_idx]))
393 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
397 if (pmd_none(pmdp[pmd_idx])) {
398 pte_phys = pt_ops.alloc_pte(va);
399 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
400 ptep = pt_ops.get_pte_virt(pte_phys);
401 memset(ptep, 0, PAGE_SIZE);
403 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
404 ptep = pt_ops.get_pte_virt(pte_phys);
407 create_pte_mapping(ptep, va, pa, sz, prot);
410 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
412 return (pud_t *)((uintptr_t)pa);
415 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
417 clear_fixmap(FIX_PUD);
418 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
421 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
423 return (pud_t *)__va(pa);
426 static phys_addr_t __init alloc_pud_early(uintptr_t va)
428 /* Only one PUD is available for early mapping */
429 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
431 return (uintptr_t)early_pud;
434 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
436 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
439 static phys_addr_t alloc_pud_late(uintptr_t va)
443 vaddr = __get_free_page(GFP_KERNEL);
448 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
450 return (p4d_t *)((uintptr_t)pa);
453 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
455 clear_fixmap(FIX_P4D);
456 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
459 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
461 return (p4d_t *)__va(pa);
464 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
466 /* Only one P4D is available for early mapping */
467 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
469 return (uintptr_t)early_p4d;
472 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
474 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
477 static phys_addr_t alloc_p4d_late(uintptr_t va)
481 vaddr = __get_free_page(GFP_KERNEL);
486 static void __init create_pud_mapping(pud_t *pudp,
487 uintptr_t va, phys_addr_t pa,
488 phys_addr_t sz, pgprot_t prot)
491 phys_addr_t next_phys;
492 uintptr_t pud_index = pud_index(va);
494 if (sz == PUD_SIZE) {
495 if (pud_val(pudp[pud_index]) == 0)
496 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
500 if (pud_val(pudp[pud_index]) == 0) {
501 next_phys = pt_ops.alloc_pmd(va);
502 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
503 nextp = pt_ops.get_pmd_virt(next_phys);
504 memset(nextp, 0, PAGE_SIZE);
506 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
507 nextp = pt_ops.get_pmd_virt(next_phys);
510 create_pmd_mapping(nextp, va, pa, sz, prot);
513 static void __init create_p4d_mapping(p4d_t *p4dp,
514 uintptr_t va, phys_addr_t pa,
515 phys_addr_t sz, pgprot_t prot)
518 phys_addr_t next_phys;
519 uintptr_t p4d_index = p4d_index(va);
521 if (sz == P4D_SIZE) {
522 if (p4d_val(p4dp[p4d_index]) == 0)
523 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
527 if (p4d_val(p4dp[p4d_index]) == 0) {
528 next_phys = pt_ops.alloc_pud(va);
529 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
530 nextp = pt_ops.get_pud_virt(next_phys);
531 memset(nextp, 0, PAGE_SIZE);
533 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
534 nextp = pt_ops.get_pud_virt(next_phys);
537 create_pud_mapping(nextp, va, pa, sz, prot);
540 #define pgd_next_t p4d_t
541 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \
542 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \
543 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
544 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \
545 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \
546 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
547 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
548 (pgtable_l5_enabled ? \
549 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
550 (pgtable_l4_enabled ? \
551 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \
552 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
553 #define fixmap_pgd_next (pgtable_l5_enabled ? \
554 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \
555 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
556 #define trampoline_pgd_next (pgtable_l5_enabled ? \
557 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \
558 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
559 #define early_dtb_pgd_next (pgtable_l5_enabled ? \
560 (uintptr_t)early_dtb_p4d : (pgtable_l4_enabled ? \
561 (uintptr_t)early_dtb_pud : (uintptr_t)early_dtb_pmd))
563 #define pgd_next_t pte_t
564 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
565 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
566 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
567 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
568 #define fixmap_pgd_next ((uintptr_t)fixmap_pte)
569 #define early_dtb_pgd_next ((uintptr_t)early_dtb_pmd)
570 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot)
571 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot)
572 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot)
573 #endif /* __PAGETABLE_PMD_FOLDED */
575 void __init create_pgd_mapping(pgd_t *pgdp,
576 uintptr_t va, phys_addr_t pa,
577 phys_addr_t sz, pgprot_t prot)
580 phys_addr_t next_phys;
581 uintptr_t pgd_idx = pgd_index(va);
583 if (sz == PGDIR_SIZE) {
584 if (pgd_val(pgdp[pgd_idx]) == 0)
585 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
589 if (pgd_val(pgdp[pgd_idx]) == 0) {
590 next_phys = alloc_pgd_next(va);
591 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
592 nextp = get_pgd_next_virt(next_phys);
593 memset(nextp, 0, PAGE_SIZE);
595 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
596 nextp = get_pgd_next_virt(next_phys);
599 create_pgd_next_mapping(nextp, va, pa, sz, prot);
602 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
604 /* Upgrade to PMD_SIZE mappings whenever possible */
605 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
611 #ifdef CONFIG_XIP_KERNEL
612 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
613 extern char _xiprom[], _exiprom[], __data_loc;
615 /* called from head.S with MMU off */
616 asmlinkage void __init __copy_data(void)
618 void *from = (void *)(&__data_loc);
619 void *to = (void *)CONFIG_PHYS_RAM_BASE;
620 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
622 memcpy(to, from, sz);
626 #ifdef CONFIG_STRICT_KERNEL_RWX
627 static __init pgprot_t pgprot_from_va(uintptr_t va)
629 if (is_va_kernel_text(va))
630 return PAGE_KERNEL_READ_EXEC;
633 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
634 * we must protect its linear mapping alias from being executed and
636 * And rodata section is marked readonly in mark_rodata_ro.
638 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
639 return PAGE_KERNEL_READ;
644 void mark_rodata_ro(void)
646 set_kernel_memory(__start_rodata, _data, set_memory_ro);
647 if (IS_ENABLED(CONFIG_64BIT))
648 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
654 static __init pgprot_t pgprot_from_va(uintptr_t va)
656 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
659 return PAGE_KERNEL_EXEC;
661 #endif /* CONFIG_STRICT_KERNEL_RWX */
664 static void __init disable_pgtable_l5(void)
666 pgtable_l5_enabled = false;
667 kernel_map.page_offset = PAGE_OFFSET_L4;
668 satp_mode = SATP_MODE_48;
671 static void __init disable_pgtable_l4(void)
673 pgtable_l4_enabled = false;
674 kernel_map.page_offset = PAGE_OFFSET_L3;
675 satp_mode = SATP_MODE_39;
679 * There is a simple way to determine if 4-level is supported by the
680 * underlying hardware: establish 1:1 mapping in 4-level page table mode
681 * then read SATP to see if the configuration was taken into account
682 * meaning sv48 is supported.
684 static __init void set_satp_mode(void)
686 u64 identity_satp, hw_satp;
687 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
688 bool check_l4 = false;
690 create_p4d_mapping(early_p4d,
691 set_satp_mode_pmd, (uintptr_t)early_pud,
692 P4D_SIZE, PAGE_TABLE);
693 create_pud_mapping(early_pud,
694 set_satp_mode_pmd, (uintptr_t)early_pmd,
695 PUD_SIZE, PAGE_TABLE);
696 /* Handle the case where set_satp_mode straddles 2 PMDs */
697 create_pmd_mapping(early_pmd,
698 set_satp_mode_pmd, set_satp_mode_pmd,
699 PMD_SIZE, PAGE_KERNEL_EXEC);
700 create_pmd_mapping(early_pmd,
701 set_satp_mode_pmd + PMD_SIZE,
702 set_satp_mode_pmd + PMD_SIZE,
703 PMD_SIZE, PAGE_KERNEL_EXEC);
705 create_pgd_mapping(early_pg_dir,
707 check_l4 ? (uintptr_t)early_pud : (uintptr_t)early_p4d,
708 PGDIR_SIZE, PAGE_TABLE);
710 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
712 local_flush_tlb_all();
713 csr_write(CSR_SATP, identity_satp);
714 hw_satp = csr_swap(CSR_SATP, 0ULL);
715 local_flush_tlb_all();
717 if (hw_satp != identity_satp) {
719 disable_pgtable_l5();
723 disable_pgtable_l4();
726 memset(early_pg_dir, 0, PAGE_SIZE);
727 memset(early_p4d, 0, PAGE_SIZE);
728 memset(early_pud, 0, PAGE_SIZE);
729 memset(early_pmd, 0, PAGE_SIZE);
734 * setup_vm() is called from head.S with MMU-off.
736 * Following requirements should be honoured for setup_vm() to work
738 * 1) It should use PC-relative addressing for accessing kernel symbols.
739 * To achieve this we always use GCC cmodel=medany.
740 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
741 * so disable compiler instrumentation when FTRACE is enabled.
743 * Currently, the above requirements are honoured by using custom CFLAGS
744 * for init.o in mm/Makefile.
747 #ifndef __riscv_cmodel_medany
748 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
751 #ifdef CONFIG_XIP_KERNEL
752 static void __init create_kernel_page_table(pgd_t *pgdir,
753 __always_unused bool early)
755 uintptr_t va, end_va;
757 /* Map the flash resident part */
758 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
759 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
760 create_pgd_mapping(pgdir, va,
761 kernel_map.xiprom + (va - kernel_map.virt_addr),
762 PMD_SIZE, PAGE_KERNEL_EXEC);
764 /* Map the data in RAM */
765 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
766 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
767 create_pgd_mapping(pgdir, va,
768 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
769 PMD_SIZE, PAGE_KERNEL);
772 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
774 uintptr_t va, end_va;
776 end_va = kernel_map.virt_addr + kernel_map.size;
777 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
778 create_pgd_mapping(pgdir, va,
779 kernel_map.phys_addr + (va - kernel_map.virt_addr),
782 PAGE_KERNEL_EXEC : pgprot_from_va(va));
787 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
788 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
791 static void __init create_fdt_early_page_table(pgd_t *pgdir, uintptr_t dtb_pa)
793 #ifndef CONFIG_BUILTIN_DTB
794 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
796 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
797 IS_ENABLED(CONFIG_64BIT) ? early_dtb_pgd_next : pa,
799 IS_ENABLED(CONFIG_64BIT) ? PAGE_TABLE : PAGE_KERNEL);
801 if (pgtable_l5_enabled)
802 create_p4d_mapping(early_dtb_p4d, DTB_EARLY_BASE_VA,
803 (uintptr_t)early_dtb_pud, P4D_SIZE, PAGE_TABLE);
805 if (pgtable_l4_enabled)
806 create_pud_mapping(early_dtb_pud, DTB_EARLY_BASE_VA,
807 (uintptr_t)early_dtb_pmd, PUD_SIZE, PAGE_TABLE);
809 if (IS_ENABLED(CONFIG_64BIT)) {
810 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA,
811 pa, PMD_SIZE, PAGE_KERNEL);
812 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE,
813 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
816 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1));
819 * For 64-bit kernel, __va can't be used since it would return a linear
820 * mapping address whereas dtb_early_va will be used before
821 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
822 * kernel is mapped in the linear mapping, that makes no difference.
824 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
827 dtb_early_pa = dtb_pa;
831 * MMU is not enabled, the page tables are allocated directly using
832 * early_pmd/pud/p4d and the address returned is the physical one.
834 void __init pt_ops_set_early(void)
836 pt_ops.alloc_pte = alloc_pte_early;
837 pt_ops.get_pte_virt = get_pte_virt_early;
838 #ifndef __PAGETABLE_PMD_FOLDED
839 pt_ops.alloc_pmd = alloc_pmd_early;
840 pt_ops.get_pmd_virt = get_pmd_virt_early;
841 pt_ops.alloc_pud = alloc_pud_early;
842 pt_ops.get_pud_virt = get_pud_virt_early;
843 pt_ops.alloc_p4d = alloc_p4d_early;
844 pt_ops.get_p4d_virt = get_p4d_virt_early;
849 * MMU is enabled but page table setup is not complete yet.
850 * fixmap page table alloc functions must be used as a means to temporarily
851 * map the allocated physical pages since the linear mapping does not exist yet.
853 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
854 * but it will be used as described above.
856 void __init pt_ops_set_fixmap(void)
858 pt_ops.alloc_pte = kernel_mapping_pa_to_va((uintptr_t)alloc_pte_fixmap);
859 pt_ops.get_pte_virt = kernel_mapping_pa_to_va((uintptr_t)get_pte_virt_fixmap);
860 #ifndef __PAGETABLE_PMD_FOLDED
861 pt_ops.alloc_pmd = kernel_mapping_pa_to_va((uintptr_t)alloc_pmd_fixmap);
862 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va((uintptr_t)get_pmd_virt_fixmap);
863 pt_ops.alloc_pud = kernel_mapping_pa_to_va((uintptr_t)alloc_pud_fixmap);
864 pt_ops.get_pud_virt = kernel_mapping_pa_to_va((uintptr_t)get_pud_virt_fixmap);
865 pt_ops.alloc_p4d = kernel_mapping_pa_to_va((uintptr_t)alloc_p4d_fixmap);
866 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va((uintptr_t)get_p4d_virt_fixmap);
871 * MMU is enabled and page table setup is complete, so from now, we can use
872 * generic page allocation functions to setup page table.
874 void __init pt_ops_set_late(void)
876 pt_ops.alloc_pte = alloc_pte_late;
877 pt_ops.get_pte_virt = get_pte_virt_late;
878 #ifndef __PAGETABLE_PMD_FOLDED
879 pt_ops.alloc_pmd = alloc_pmd_late;
880 pt_ops.get_pmd_virt = get_pmd_virt_late;
881 pt_ops.alloc_pud = alloc_pud_late;
882 pt_ops.get_pud_virt = get_pud_virt_late;
883 pt_ops.alloc_p4d = alloc_p4d_late;
884 pt_ops.get_p4d_virt = get_p4d_virt_late;
888 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
890 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
892 kernel_map.virt_addr = KERNEL_LINK_ADDR;
893 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
895 #ifdef CONFIG_XIP_KERNEL
896 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
897 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
899 phys_ram_base = CONFIG_PHYS_RAM_BASE;
900 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
901 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
903 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
905 kernel_map.phys_addr = (uintptr_t)(&_start);
906 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
909 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
913 kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr;
914 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
916 riscv_pfn_base = PFN_DOWN(kernel_map.phys_addr);
919 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
920 * kernel, whereas for 64-bit kernel, the end of the virtual address
921 * space is occupied by the modules/BPF/kernel mappings which reduces
922 * the available size of the linear mapping.
924 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
926 /* Sanity check alignment and size */
927 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
928 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
932 * The last 4K bytes of the addressable memory can not be mapped because
933 * of IS_ERR_VALUE macro.
935 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
940 /* Setup early PGD for fixmap */
941 create_pgd_mapping(early_pg_dir, FIXADDR_START,
942 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
944 #ifndef __PAGETABLE_PMD_FOLDED
945 /* Setup fixmap P4D and PUD */
946 if (pgtable_l5_enabled)
947 create_p4d_mapping(fixmap_p4d, FIXADDR_START,
948 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
949 /* Setup fixmap PUD and PMD */
950 if (pgtable_l4_enabled)
951 create_pud_mapping(fixmap_pud, FIXADDR_START,
952 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
953 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
954 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
955 /* Setup trampoline PGD and PMD */
956 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
957 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
958 if (pgtable_l5_enabled)
959 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
960 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
961 if (pgtable_l4_enabled)
962 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
963 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
964 #ifdef CONFIG_XIP_KERNEL
965 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
966 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
968 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
969 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
972 /* Setup trampoline PGD */
973 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
974 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
978 * Setup early PGD covering entire kernel which will allow
979 * us to reach paging_init(). We map all memory banks later
980 * in setup_vm_final() below.
982 create_kernel_page_table(early_pg_dir, true);
984 /* Setup early mapping for FDT early scan */
985 create_fdt_early_page_table(early_pg_dir, dtb_pa);
988 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
989 * range can not span multiple pmds.
991 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
992 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
994 #ifndef __PAGETABLE_PMD_FOLDED
996 * Early ioremap fixmap is already created as it lies within first 2MB
997 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
998 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1001 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1002 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1003 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1005 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1006 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1007 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1008 fix_to_virt(FIX_BTMAP_BEGIN));
1009 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1010 fix_to_virt(FIX_BTMAP_END));
1012 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1013 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1017 pt_ops_set_fixmap();
1020 static void __init setup_vm_final(void)
1022 uintptr_t va, map_size;
1023 phys_addr_t pa, start, end;
1026 /* Setup swapper PGD for fixmap */
1027 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1028 __pa_symbol(fixmap_pgd_next),
1029 PGDIR_SIZE, PAGE_TABLE);
1031 /* Map all memory banks in the linear mapping */
1032 for_each_mem_range(i, &start, &end) {
1035 if (start <= __pa(PAGE_OFFSET) &&
1036 __pa(PAGE_OFFSET) < end)
1037 start = __pa(PAGE_OFFSET);
1038 if (end >= __pa(PAGE_OFFSET) + memory_limit)
1039 end = __pa(PAGE_OFFSET) + memory_limit;
1041 map_size = best_map_size(start, end - start);
1042 for (pa = start; pa < end; pa += map_size) {
1043 va = (uintptr_t)__va(pa);
1045 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1046 pgprot_from_va(va));
1050 /* Map the kernel */
1051 if (IS_ENABLED(CONFIG_64BIT))
1052 create_kernel_page_table(swapper_pg_dir, false);
1055 kasan_swapper_init();
1058 /* Clear fixmap PTE and PMD mappings */
1059 clear_fixmap(FIX_PTE);
1060 clear_fixmap(FIX_PMD);
1061 clear_fixmap(FIX_PUD);
1062 clear_fixmap(FIX_P4D);
1064 /* Move to swapper page table */
1065 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1066 local_flush_tlb_all();
1071 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1073 dtb_early_va = (void *)dtb_pa;
1074 dtb_early_pa = dtb_pa;
1077 static inline void setup_vm_final(void)
1080 #endif /* CONFIG_MMU */
1083 * reserve_crashkernel() - reserves memory for crash kernel
1085 * This function reserves memory area given in "crashkernel=" kernel command
1086 * line parameter. The memory reserved is used by dump capture kernel when
1087 * primary kernel is crashing.
1089 static void __init reserve_crashkernel(void)
1091 unsigned long long crash_base = 0;
1092 unsigned long long crash_size = 0;
1093 unsigned long search_start = memblock_start_of_DRAM();
1094 unsigned long search_end = memblock_end_of_DRAM();
1098 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1101 * Don't reserve a region for a crash kernel on a crash kernel
1102 * since it doesn't make much sense and we have limited memory
1105 if (is_kdump_kernel()) {
1106 pr_info("crashkernel: ignoring reservation request\n");
1110 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
1111 &crash_size, &crash_base);
1112 if (ret || !crash_size)
1115 crash_size = PAGE_ALIGN(crash_size);
1118 search_start = crash_base;
1119 search_end = crash_base + crash_size;
1123 * Current riscv boot protocol requires 2MB alignment for
1124 * RV64 and 4MB alignment for RV32 (hugepage size)
1126 * Try to alloc from 32bit addressible physical memory so that
1127 * swiotlb can work on the crash kernel.
1129 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1131 min(search_end, (unsigned long) SZ_4G));
1132 if (crash_base == 0) {
1133 /* Try again without restricting region to 32bit addressible memory */
1134 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1135 search_start, search_end);
1136 if (crash_base == 0) {
1137 pr_warn("crashkernel: couldn't allocate %lldKB\n",
1143 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1144 crash_base, crash_base + crash_size, crash_size >> 20);
1146 crashk_res.start = crash_base;
1147 crashk_res.end = crash_base + crash_size - 1;
1150 void __init paging_init(void)
1156 void __init misc_mem_init(void)
1158 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1162 reserve_crashkernel();
1163 memblock_dump_all();
1166 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1167 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1168 struct vmem_altmap *altmap)
1170 return vmemmap_populate_basepages(start, end, node, NULL);