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
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 swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE);
129 /* Limit the memory size via mem. */
130 static phys_addr_t memory_limit;
132 static int __init early_mem(char *p)
139 size = memparse(p, &p) & PAGE_MASK;
140 memory_limit = min_t(u64, size, memory_limit);
142 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
146 early_param("mem", early_mem);
148 static void __init setup_bootmem(void)
150 phys_addr_t vmlinux_end = __pa_symbol(&_end);
151 phys_addr_t max_mapped_addr;
152 phys_addr_t phys_ram_end, vmlinux_start;
154 if (IS_ENABLED(CONFIG_XIP_KERNEL))
155 vmlinux_start = __pa_symbol(&_sdata);
157 vmlinux_start = __pa_symbol(&_start);
159 memblock_enforce_memory_limit(memory_limit);
162 * Make sure we align the reservation on PMD_SIZE since we will
163 * map the kernel in the linear mapping as read-only: we do not want
164 * any allocation to happen between _end and the next pmd aligned page.
166 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
167 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
169 * Reserve from the start of the kernel to the end of the kernel
171 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
173 phys_ram_end = memblock_end_of_DRAM();
174 if (!IS_ENABLED(CONFIG_XIP_KERNEL))
175 phys_ram_base = memblock_start_of_DRAM();
177 * memblock allocator is not aware of the fact that last 4K bytes of
178 * the addressable memory can not be mapped because of IS_ERR_VALUE
179 * macro. Make sure that last 4k bytes are not usable by memblock
180 * if end of dram is equal to maximum addressable memory. For 64-bit
181 * kernel, this problem can't happen here as the end of the virtual
182 * address space is occupied by the kernel mapping then this check must
183 * be done as soon as the kernel mapping base address is determined.
185 if (!IS_ENABLED(CONFIG_64BIT)) {
186 max_mapped_addr = __pa(~(ulong)0);
187 if (max_mapped_addr == (phys_ram_end - 1))
188 memblock_set_current_limit(max_mapped_addr - 4096);
191 min_low_pfn = PFN_UP(phys_ram_base);
192 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
193 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
195 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
196 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
198 reserve_initrd_mem();
200 * If DTB is built in, no need to reserve its memblock.
201 * Otherwise, do reserve it but avoid using
202 * early_init_fdt_reserve_self() since __pa() does
203 * not work for DTB pointers that are fixmap addresses
205 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) {
207 * In case the DTB is not located in a memory region we won't
208 * be able to locate it later on via the linear mapping and
209 * get a segfault when accessing it via __va(dtb_early_pa).
210 * To avoid this situation copy DTB to a memory region.
211 * Note that memblock_phys_alloc will also reserve DTB region.
213 if (!memblock_is_memory(dtb_early_pa)) {
214 size_t fdt_size = fdt_totalsize(dtb_early_va);
215 phys_addr_t new_dtb_early_pa = memblock_phys_alloc(fdt_size, PAGE_SIZE);
216 void *new_dtb_early_va = early_memremap(new_dtb_early_pa, fdt_size);
218 memcpy(new_dtb_early_va, dtb_early_va, fdt_size);
219 early_memunmap(new_dtb_early_va, fdt_size);
220 _dtb_early_pa = new_dtb_early_pa;
222 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
225 early_init_fdt_scan_reserved_mem();
226 dma_contiguous_reserve(dma32_phys_limit);
227 if (IS_ENABLED(CONFIG_64BIT))
228 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
229 memblock_allow_resize();
233 struct pt_alloc_ops pt_ops __initdata;
235 unsigned long riscv_pfn_base __ro_after_init;
236 EXPORT_SYMBOL(riscv_pfn_base);
238 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
239 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
240 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
242 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
243 static p4d_t __maybe_unused early_dtb_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
244 static pud_t __maybe_unused early_dtb_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
245 static pmd_t __maybe_unused early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
247 #ifdef CONFIG_XIP_KERNEL
248 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
249 #define riscv_pfn_base (*(unsigned long *)XIP_FIXUP(&riscv_pfn_base))
250 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
251 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
252 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
253 #endif /* CONFIG_XIP_KERNEL */
255 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
257 unsigned long addr = __fix_to_virt(idx);
260 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
262 ptep = &fixmap_pte[pte_index(addr)];
264 if (pgprot_val(prot))
265 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
267 pte_clear(&init_mm, addr, ptep);
268 local_flush_tlb_page(addr);
271 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
273 return (pte_t *)((uintptr_t)pa);
276 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
278 clear_fixmap(FIX_PTE);
279 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
282 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
284 return (pte_t *) __va(pa);
287 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
290 * We only create PMD or PGD early mappings so we
291 * should never reach here with MMU disabled.
296 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
298 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
301 static phys_addr_t __init alloc_pte_late(uintptr_t va)
305 vaddr = __get_free_page(GFP_KERNEL);
306 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)));
311 static void __init create_pte_mapping(pte_t *ptep,
312 uintptr_t va, phys_addr_t pa,
313 phys_addr_t sz, pgprot_t prot)
315 uintptr_t pte_idx = pte_index(va);
317 BUG_ON(sz != PAGE_SIZE);
319 if (pte_none(ptep[pte_idx]))
320 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
323 #ifndef __PAGETABLE_PMD_FOLDED
325 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
326 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
327 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
329 #ifdef CONFIG_XIP_KERNEL
330 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
331 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
332 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
333 #endif /* CONFIG_XIP_KERNEL */
335 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
336 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
337 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
339 #ifdef CONFIG_XIP_KERNEL
340 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
341 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d))
342 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d))
343 #endif /* CONFIG_XIP_KERNEL */
345 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
346 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
347 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
349 #ifdef CONFIG_XIP_KERNEL
350 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
351 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
352 #define early_pud ((pud_t *)XIP_FIXUP(early_pud))
353 #endif /* CONFIG_XIP_KERNEL */
355 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
357 /* Before MMU is enabled */
358 return (pmd_t *)((uintptr_t)pa);
361 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
363 clear_fixmap(FIX_PMD);
364 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
367 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
369 return (pmd_t *) __va(pa);
372 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
374 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
376 return (uintptr_t)early_pmd;
379 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
381 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
384 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
388 vaddr = __get_free_page(GFP_KERNEL);
389 BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page(vaddr)));
394 static void __init create_pmd_mapping(pmd_t *pmdp,
395 uintptr_t va, phys_addr_t pa,
396 phys_addr_t sz, pgprot_t prot)
399 phys_addr_t pte_phys;
400 uintptr_t pmd_idx = pmd_index(va);
402 if (sz == PMD_SIZE) {
403 if (pmd_none(pmdp[pmd_idx]))
404 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
408 if (pmd_none(pmdp[pmd_idx])) {
409 pte_phys = pt_ops.alloc_pte(va);
410 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
411 ptep = pt_ops.get_pte_virt(pte_phys);
412 memset(ptep, 0, PAGE_SIZE);
414 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
415 ptep = pt_ops.get_pte_virt(pte_phys);
418 create_pte_mapping(ptep, va, pa, sz, prot);
421 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
423 return (pud_t *)((uintptr_t)pa);
426 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
428 clear_fixmap(FIX_PUD);
429 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
432 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
434 return (pud_t *)__va(pa);
437 static phys_addr_t __init alloc_pud_early(uintptr_t va)
439 /* Only one PUD is available for early mapping */
440 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
442 return (uintptr_t)early_pud;
445 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
447 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
450 static phys_addr_t alloc_pud_late(uintptr_t va)
454 vaddr = __get_free_page(GFP_KERNEL);
459 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
461 return (p4d_t *)((uintptr_t)pa);
464 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
466 clear_fixmap(FIX_P4D);
467 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
470 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
472 return (p4d_t *)__va(pa);
475 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
477 /* Only one P4D is available for early mapping */
478 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
480 return (uintptr_t)early_p4d;
483 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
485 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
488 static phys_addr_t alloc_p4d_late(uintptr_t va)
492 vaddr = __get_free_page(GFP_KERNEL);
497 static void __init create_pud_mapping(pud_t *pudp,
498 uintptr_t va, phys_addr_t pa,
499 phys_addr_t sz, pgprot_t prot)
502 phys_addr_t next_phys;
503 uintptr_t pud_index = pud_index(va);
505 if (sz == PUD_SIZE) {
506 if (pud_val(pudp[pud_index]) == 0)
507 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
511 if (pud_val(pudp[pud_index]) == 0) {
512 next_phys = pt_ops.alloc_pmd(va);
513 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
514 nextp = pt_ops.get_pmd_virt(next_phys);
515 memset(nextp, 0, PAGE_SIZE);
517 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
518 nextp = pt_ops.get_pmd_virt(next_phys);
521 create_pmd_mapping(nextp, va, pa, sz, prot);
524 static void __init create_p4d_mapping(p4d_t *p4dp,
525 uintptr_t va, phys_addr_t pa,
526 phys_addr_t sz, pgprot_t prot)
529 phys_addr_t next_phys;
530 uintptr_t p4d_index = p4d_index(va);
532 if (sz == P4D_SIZE) {
533 if (p4d_val(p4dp[p4d_index]) == 0)
534 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
538 if (p4d_val(p4dp[p4d_index]) == 0) {
539 next_phys = pt_ops.alloc_pud(va);
540 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
541 nextp = pt_ops.get_pud_virt(next_phys);
542 memset(nextp, 0, PAGE_SIZE);
544 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
545 nextp = pt_ops.get_pud_virt(next_phys);
548 create_pud_mapping(nextp, va, pa, sz, prot);
551 #define pgd_next_t p4d_t
552 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \
553 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \
554 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
555 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \
556 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \
557 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
558 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
559 (pgtable_l5_enabled ? \
560 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
561 (pgtable_l4_enabled ? \
562 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \
563 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
564 #define fixmap_pgd_next (pgtable_l5_enabled ? \
565 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \
566 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
567 #define trampoline_pgd_next (pgtable_l5_enabled ? \
568 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \
569 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
570 #define early_dtb_pgd_next (pgtable_l5_enabled ? \
571 (uintptr_t)early_dtb_p4d : (pgtable_l4_enabled ? \
572 (uintptr_t)early_dtb_pud : (uintptr_t)early_dtb_pmd))
574 #define pgd_next_t pte_t
575 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
576 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
577 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
578 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
579 #define fixmap_pgd_next ((uintptr_t)fixmap_pte)
580 #define early_dtb_pgd_next ((uintptr_t)early_dtb_pmd)
581 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
582 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
583 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
584 #endif /* __PAGETABLE_PMD_FOLDED */
586 void __init create_pgd_mapping(pgd_t *pgdp,
587 uintptr_t va, phys_addr_t pa,
588 phys_addr_t sz, pgprot_t prot)
591 phys_addr_t next_phys;
592 uintptr_t pgd_idx = pgd_index(va);
594 if (sz == PGDIR_SIZE) {
595 if (pgd_val(pgdp[pgd_idx]) == 0)
596 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
600 if (pgd_val(pgdp[pgd_idx]) == 0) {
601 next_phys = alloc_pgd_next(va);
602 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
603 nextp = get_pgd_next_virt(next_phys);
604 memset(nextp, 0, PAGE_SIZE);
606 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
607 nextp = get_pgd_next_virt(next_phys);
610 create_pgd_next_mapping(nextp, va, pa, sz, prot);
613 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
615 /* Upgrade to PMD_SIZE mappings whenever possible */
616 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
622 #ifdef CONFIG_XIP_KERNEL
623 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
624 extern char _xiprom[], _exiprom[], __data_loc;
626 /* called from head.S with MMU off */
627 asmlinkage void __init __copy_data(void)
629 void *from = (void *)(&__data_loc);
630 void *to = (void *)CONFIG_PHYS_RAM_BASE;
631 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
633 memcpy(to, from, sz);
637 #ifdef CONFIG_STRICT_KERNEL_RWX
638 static __init pgprot_t pgprot_from_va(uintptr_t va)
640 if (is_va_kernel_text(va))
641 return PAGE_KERNEL_READ_EXEC;
644 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
645 * we must protect its linear mapping alias from being executed and
647 * And rodata section is marked readonly in mark_rodata_ro.
649 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
650 return PAGE_KERNEL_READ;
655 void mark_rodata_ro(void)
657 set_kernel_memory(__start_rodata, _data, set_memory_ro);
658 if (IS_ENABLED(CONFIG_64BIT))
659 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
665 static __init pgprot_t pgprot_from_va(uintptr_t va)
667 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
670 return PAGE_KERNEL_EXEC;
672 #endif /* CONFIG_STRICT_KERNEL_RWX */
674 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
675 static void __init disable_pgtable_l5(void)
677 pgtable_l5_enabled = false;
678 kernel_map.page_offset = PAGE_OFFSET_L4;
679 satp_mode = SATP_MODE_48;
682 static void __init disable_pgtable_l4(void)
684 pgtable_l4_enabled = false;
685 kernel_map.page_offset = PAGE_OFFSET_L3;
686 satp_mode = SATP_MODE_39;
690 * There is a simple way to determine if 4-level is supported by the
691 * underlying hardware: establish 1:1 mapping in 4-level page table mode
692 * then read SATP to see if the configuration was taken into account
693 * meaning sv48 is supported.
695 static __init void set_satp_mode(void)
697 u64 identity_satp, hw_satp;
698 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
699 bool check_l4 = false;
701 create_p4d_mapping(early_p4d,
702 set_satp_mode_pmd, (uintptr_t)early_pud,
703 P4D_SIZE, PAGE_TABLE);
704 create_pud_mapping(early_pud,
705 set_satp_mode_pmd, (uintptr_t)early_pmd,
706 PUD_SIZE, PAGE_TABLE);
707 /* Handle the case where set_satp_mode straddles 2 PMDs */
708 create_pmd_mapping(early_pmd,
709 set_satp_mode_pmd, set_satp_mode_pmd,
710 PMD_SIZE, PAGE_KERNEL_EXEC);
711 create_pmd_mapping(early_pmd,
712 set_satp_mode_pmd + PMD_SIZE,
713 set_satp_mode_pmd + PMD_SIZE,
714 PMD_SIZE, PAGE_KERNEL_EXEC);
716 create_pgd_mapping(early_pg_dir,
718 check_l4 ? (uintptr_t)early_pud : (uintptr_t)early_p4d,
719 PGDIR_SIZE, PAGE_TABLE);
721 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
723 local_flush_tlb_all();
724 csr_write(CSR_SATP, identity_satp);
725 hw_satp = csr_swap(CSR_SATP, 0ULL);
726 local_flush_tlb_all();
728 if (hw_satp != identity_satp) {
730 disable_pgtable_l5();
732 memset(early_pg_dir, 0, PAGE_SIZE);
735 disable_pgtable_l4();
738 memset(early_pg_dir, 0, PAGE_SIZE);
739 memset(early_p4d, 0, PAGE_SIZE);
740 memset(early_pud, 0, PAGE_SIZE);
741 memset(early_pmd, 0, PAGE_SIZE);
746 * setup_vm() is called from head.S with MMU-off.
748 * Following requirements should be honoured for setup_vm() to work
750 * 1) It should use PC-relative addressing for accessing kernel symbols.
751 * To achieve this we always use GCC cmodel=medany.
752 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
753 * so disable compiler instrumentation when FTRACE is enabled.
755 * Currently, the above requirements are honoured by using custom CFLAGS
756 * for init.o in mm/Makefile.
759 #ifndef __riscv_cmodel_medany
760 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
763 #ifdef CONFIG_XIP_KERNEL
764 static void __init create_kernel_page_table(pgd_t *pgdir,
765 __always_unused bool early)
767 uintptr_t va, end_va;
769 /* Map the flash resident part */
770 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
771 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
772 create_pgd_mapping(pgdir, va,
773 kernel_map.xiprom + (va - kernel_map.virt_addr),
774 PMD_SIZE, PAGE_KERNEL_EXEC);
776 /* Map the data in RAM */
777 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
778 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
779 create_pgd_mapping(pgdir, va,
780 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
781 PMD_SIZE, PAGE_KERNEL);
784 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
786 uintptr_t va, end_va;
788 end_va = kernel_map.virt_addr + kernel_map.size;
789 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
790 create_pgd_mapping(pgdir, va,
791 kernel_map.phys_addr + (va - kernel_map.virt_addr),
794 PAGE_KERNEL_EXEC : pgprot_from_va(va));
799 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
800 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
803 static void __init create_fdt_early_page_table(pgd_t *pgdir, uintptr_t dtb_pa)
805 #ifndef CONFIG_BUILTIN_DTB
806 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
808 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
809 IS_ENABLED(CONFIG_64BIT) ? early_dtb_pgd_next : pa,
811 IS_ENABLED(CONFIG_64BIT) ? PAGE_TABLE : PAGE_KERNEL);
813 if (pgtable_l5_enabled)
814 create_p4d_mapping(early_dtb_p4d, DTB_EARLY_BASE_VA,
815 (uintptr_t)early_dtb_pud, P4D_SIZE, PAGE_TABLE);
817 if (pgtable_l4_enabled)
818 create_pud_mapping(early_dtb_pud, DTB_EARLY_BASE_VA,
819 (uintptr_t)early_dtb_pmd, PUD_SIZE, PAGE_TABLE);
821 if (IS_ENABLED(CONFIG_64BIT)) {
822 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA,
823 pa, PMD_SIZE, PAGE_KERNEL);
824 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE,
825 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
828 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1));
831 * For 64-bit kernel, __va can't be used since it would return a linear
832 * mapping address whereas dtb_early_va will be used before
833 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
834 * kernel is mapped in the linear mapping, that makes no difference.
836 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
839 dtb_early_pa = dtb_pa;
843 * MMU is not enabled, the page tables are allocated directly using
844 * early_pmd/pud/p4d and the address returned is the physical one.
846 void __init pt_ops_set_early(void)
848 pt_ops.alloc_pte = alloc_pte_early;
849 pt_ops.get_pte_virt = get_pte_virt_early;
850 #ifndef __PAGETABLE_PMD_FOLDED
851 pt_ops.alloc_pmd = alloc_pmd_early;
852 pt_ops.get_pmd_virt = get_pmd_virt_early;
853 pt_ops.alloc_pud = alloc_pud_early;
854 pt_ops.get_pud_virt = get_pud_virt_early;
855 pt_ops.alloc_p4d = alloc_p4d_early;
856 pt_ops.get_p4d_virt = get_p4d_virt_early;
861 * MMU is enabled but page table setup is not complete yet.
862 * fixmap page table alloc functions must be used as a means to temporarily
863 * map the allocated physical pages since the linear mapping does not exist yet.
865 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
866 * but it will be used as described above.
868 void __init pt_ops_set_fixmap(void)
870 pt_ops.alloc_pte = kernel_mapping_pa_to_va((uintptr_t)alloc_pte_fixmap);
871 pt_ops.get_pte_virt = kernel_mapping_pa_to_va((uintptr_t)get_pte_virt_fixmap);
872 #ifndef __PAGETABLE_PMD_FOLDED
873 pt_ops.alloc_pmd = kernel_mapping_pa_to_va((uintptr_t)alloc_pmd_fixmap);
874 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va((uintptr_t)get_pmd_virt_fixmap);
875 pt_ops.alloc_pud = kernel_mapping_pa_to_va((uintptr_t)alloc_pud_fixmap);
876 pt_ops.get_pud_virt = kernel_mapping_pa_to_va((uintptr_t)get_pud_virt_fixmap);
877 pt_ops.alloc_p4d = kernel_mapping_pa_to_va((uintptr_t)alloc_p4d_fixmap);
878 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va((uintptr_t)get_p4d_virt_fixmap);
883 * MMU is enabled and page table setup is complete, so from now, we can use
884 * generic page allocation functions to setup page table.
886 void __init pt_ops_set_late(void)
888 pt_ops.alloc_pte = alloc_pte_late;
889 pt_ops.get_pte_virt = get_pte_virt_late;
890 #ifndef __PAGETABLE_PMD_FOLDED
891 pt_ops.alloc_pmd = alloc_pmd_late;
892 pt_ops.get_pmd_virt = get_pmd_virt_late;
893 pt_ops.alloc_pud = alloc_pud_late;
894 pt_ops.get_pud_virt = get_pud_virt_late;
895 pt_ops.alloc_p4d = alloc_p4d_late;
896 pt_ops.get_p4d_virt = get_p4d_virt_late;
900 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
902 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
904 kernel_map.virt_addr = KERNEL_LINK_ADDR;
905 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
907 #ifdef CONFIG_XIP_KERNEL
908 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
909 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
911 phys_ram_base = CONFIG_PHYS_RAM_BASE;
912 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
913 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
915 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
917 kernel_map.phys_addr = (uintptr_t)(&_start);
918 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
921 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
925 kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr;
926 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
928 riscv_pfn_base = PFN_DOWN(kernel_map.phys_addr);
931 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
932 * kernel, whereas for 64-bit kernel, the end of the virtual address
933 * space is occupied by the modules/BPF/kernel mappings which reduces
934 * the available size of the linear mapping.
936 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
938 /* Sanity check alignment and size */
939 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
940 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
944 * The last 4K bytes of the addressable memory can not be mapped because
945 * of IS_ERR_VALUE macro.
947 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
950 apply_early_boot_alternatives();
953 /* Setup early PGD for fixmap */
954 create_pgd_mapping(early_pg_dir, FIXADDR_START,
955 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
957 #ifndef __PAGETABLE_PMD_FOLDED
958 /* Setup fixmap P4D and PUD */
959 if (pgtable_l5_enabled)
960 create_p4d_mapping(fixmap_p4d, FIXADDR_START,
961 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
962 /* Setup fixmap PUD and PMD */
963 if (pgtable_l4_enabled)
964 create_pud_mapping(fixmap_pud, FIXADDR_START,
965 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
966 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
967 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
968 /* Setup trampoline PGD and PMD */
969 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
970 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
971 if (pgtable_l5_enabled)
972 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
973 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
974 if (pgtable_l4_enabled)
975 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
976 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
977 #ifdef CONFIG_XIP_KERNEL
978 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
979 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
981 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
982 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
985 /* Setup trampoline PGD */
986 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
987 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
991 * Setup early PGD covering entire kernel which will allow
992 * us to reach paging_init(). We map all memory banks later
993 * in setup_vm_final() below.
995 create_kernel_page_table(early_pg_dir, true);
997 /* Setup early mapping for FDT early scan */
998 create_fdt_early_page_table(early_pg_dir, dtb_pa);
1001 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1002 * range can not span multiple pmds.
1004 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1005 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1007 #ifndef __PAGETABLE_PMD_FOLDED
1009 * Early ioremap fixmap is already created as it lies within first 2MB
1010 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1011 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1014 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1015 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1016 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1018 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1019 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1020 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1021 fix_to_virt(FIX_BTMAP_BEGIN));
1022 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1023 fix_to_virt(FIX_BTMAP_END));
1025 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1026 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1030 pt_ops_set_fixmap();
1033 static void __init setup_vm_final(void)
1035 uintptr_t va, map_size;
1036 phys_addr_t pa, start, end;
1039 /* Setup swapper PGD for fixmap */
1040 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1041 __pa_symbol(fixmap_pgd_next),
1042 PGDIR_SIZE, PAGE_TABLE);
1044 /* Map all memory banks in the linear mapping */
1045 for_each_mem_range(i, &start, &end) {
1048 if (start <= __pa(PAGE_OFFSET) &&
1049 __pa(PAGE_OFFSET) < end)
1050 start = __pa(PAGE_OFFSET);
1051 if (end >= __pa(PAGE_OFFSET) + memory_limit)
1052 end = __pa(PAGE_OFFSET) + memory_limit;
1054 map_size = best_map_size(start, end - start);
1055 for (pa = start; pa < end; pa += map_size) {
1056 va = (uintptr_t)__va(pa);
1058 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1059 pgprot_from_va(va));
1063 /* Map the kernel */
1064 if (IS_ENABLED(CONFIG_64BIT))
1065 create_kernel_page_table(swapper_pg_dir, false);
1068 kasan_swapper_init();
1071 /* Clear fixmap PTE and PMD mappings */
1072 clear_fixmap(FIX_PTE);
1073 clear_fixmap(FIX_PMD);
1074 clear_fixmap(FIX_PUD);
1075 clear_fixmap(FIX_P4D);
1077 /* Move to swapper page table */
1078 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1079 local_flush_tlb_all();
1084 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1086 dtb_early_va = (void *)dtb_pa;
1087 dtb_early_pa = dtb_pa;
1090 static inline void setup_vm_final(void)
1093 #endif /* CONFIG_MMU */
1096 * reserve_crashkernel() - reserves memory for crash kernel
1098 * This function reserves memory area given in "crashkernel=" kernel command
1099 * line parameter. The memory reserved is used by dump capture kernel when
1100 * primary kernel is crashing.
1102 static void __init reserve_crashkernel(void)
1104 unsigned long long crash_base = 0;
1105 unsigned long long crash_size = 0;
1106 unsigned long search_start = memblock_start_of_DRAM();
1107 unsigned long search_end = memblock_end_of_DRAM();
1111 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1114 * Don't reserve a region for a crash kernel on a crash kernel
1115 * since it doesn't make much sense and we have limited memory
1118 if (is_kdump_kernel()) {
1119 pr_info("crashkernel: ignoring reservation request\n");
1123 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
1124 &crash_size, &crash_base);
1125 if (ret || !crash_size)
1128 crash_size = PAGE_ALIGN(crash_size);
1131 search_start = crash_base;
1132 search_end = crash_base + crash_size;
1136 * Current riscv boot protocol requires 2MB alignment for
1137 * RV64 and 4MB alignment for RV32 (hugepage size)
1139 * Try to alloc from 32bit addressible physical memory so that
1140 * swiotlb can work on the crash kernel.
1142 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1144 min(search_end, (unsigned long) SZ_4G));
1145 if (crash_base == 0) {
1146 /* Try again without restricting region to 32bit addressible memory */
1147 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1148 search_start, search_end);
1149 if (crash_base == 0) {
1150 pr_warn("crashkernel: couldn't allocate %lldKB\n",
1156 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1157 crash_base, crash_base + crash_size, crash_size >> 20);
1159 crashk_res.start = crash_base;
1160 crashk_res.end = crash_base + crash_size - 1;
1163 void __init paging_init(void)
1169 void __init misc_mem_init(void)
1171 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1175 reserve_crashkernel();
1176 memblock_dump_all();
1179 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1180 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1181 struct vmem_altmap *altmap)
1183 return vmemmap_populate_basepages(start, end, node, NULL);