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>
23 #ifdef CONFIG_RELOCATABLE
24 #include <linux/elf.h>
26 #include <linux/kfence.h>
28 #include <asm/fixmap.h>
31 #include <asm/pgtable.h>
32 #include <asm/ptdump.h>
33 #include <asm/sections.h>
35 #include <asm/tlbflush.h>
37 #include "../kernel/head.h"
39 struct kernel_mapping kernel_map __ro_after_init;
40 EXPORT_SYMBOL(kernel_map);
41 #ifdef CONFIG_XIP_KERNEL
42 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
46 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39;
48 u64 satp_mode __ro_after_init = SATP_MODE_32;
50 EXPORT_SYMBOL(satp_mode);
52 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
53 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
54 EXPORT_SYMBOL(pgtable_l4_enabled);
55 EXPORT_SYMBOL(pgtable_l5_enabled);
57 phys_addr_t phys_ram_base __ro_after_init;
58 EXPORT_SYMBOL(phys_ram_base);
60 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
62 EXPORT_SYMBOL(empty_zero_page);
65 void *_dtb_early_va __initdata;
66 uintptr_t _dtb_early_pa __initdata;
68 static phys_addr_t dma32_phys_limit __initdata;
70 static void __init zone_sizes_init(void)
72 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
74 #ifdef CONFIG_ZONE_DMA32
75 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
77 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
79 free_area_init(max_zone_pfns);
82 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
84 #define LOG2_SZ_1K ilog2(SZ_1K)
85 #define LOG2_SZ_1M ilog2(SZ_1M)
86 #define LOG2_SZ_1G ilog2(SZ_1G)
87 #define LOG2_SZ_1T ilog2(SZ_1T)
89 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
91 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
92 (((t) - (b)) >> LOG2_SZ_1K));
95 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
97 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
98 (((t) - (b)) >> LOG2_SZ_1M));
101 static inline void print_mlg(char *name, unsigned long b, unsigned long t)
103 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t,
104 (((t) - (b)) >> LOG2_SZ_1G));
108 static inline void print_mlt(char *name, unsigned long b, unsigned long t)
110 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t,
111 (((t) - (b)) >> LOG2_SZ_1T));
114 #define print_mlt(n, b, t) do {} while (0)
117 static inline void print_ml(char *name, unsigned long b, unsigned long t)
119 unsigned long diff = t - b;
121 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10)
122 print_mlt(name, b, t);
123 else if ((diff >> LOG2_SZ_1G) >= 10)
124 print_mlg(name, b, t);
125 else if ((diff >> LOG2_SZ_1M) >= 10)
126 print_mlm(name, b, t);
128 print_mlk(name, b, t);
131 static void __init print_vm_layout(void)
133 pr_notice("Virtual kernel memory layout:\n");
134 print_ml("fixmap", (unsigned long)FIXADDR_START,
135 (unsigned long)FIXADDR_TOP);
136 print_ml("pci io", (unsigned long)PCI_IO_START,
137 (unsigned long)PCI_IO_END);
138 print_ml("vmemmap", (unsigned long)VMEMMAP_START,
139 (unsigned long)VMEMMAP_END);
140 print_ml("vmalloc", (unsigned long)VMALLOC_START,
141 (unsigned long)VMALLOC_END);
143 print_ml("modules", (unsigned long)MODULES_VADDR,
144 (unsigned long)MODULES_END);
146 print_ml("lowmem", (unsigned long)PAGE_OFFSET,
147 (unsigned long)high_memory);
148 if (IS_ENABLED(CONFIG_64BIT)) {
150 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
153 print_ml("kernel", (unsigned long)kernel_map.virt_addr,
154 (unsigned long)ADDRESS_SPACE_END);
158 static void print_vm_layout(void) { }
159 #endif /* CONFIG_DEBUG_VM */
161 void __init mem_init(void)
163 #ifdef CONFIG_FLATMEM
165 #endif /* CONFIG_FLATMEM */
167 swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE);
173 /* Limit the memory size via mem. */
174 static phys_addr_t memory_limit;
176 static int __init early_mem(char *p)
183 size = memparse(p, &p) & PAGE_MASK;
184 memory_limit = min_t(u64, size, memory_limit);
186 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
190 early_param("mem", early_mem);
192 static void __init setup_bootmem(void)
194 phys_addr_t vmlinux_end = __pa_symbol(&_end);
195 phys_addr_t max_mapped_addr;
196 phys_addr_t phys_ram_end, vmlinux_start;
198 if (IS_ENABLED(CONFIG_XIP_KERNEL))
199 vmlinux_start = __pa_symbol(&_sdata);
201 vmlinux_start = __pa_symbol(&_start);
203 memblock_enforce_memory_limit(memory_limit);
206 * Make sure we align the reservation on PMD_SIZE since we will
207 * map the kernel in the linear mapping as read-only: we do not want
208 * any allocation to happen between _end and the next pmd aligned page.
210 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
211 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
213 * Reserve from the start of the kernel to the end of the kernel
215 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
217 phys_ram_end = memblock_end_of_DRAM();
220 * Make sure we align the start of the memory on a PMD boundary so that
221 * at worst, we map the linear mapping with PMD mappings.
223 if (!IS_ENABLED(CONFIG_XIP_KERNEL))
224 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK;
227 * In 64-bit, any use of __va/__pa before this point is wrong as we
228 * did not know the start of DRAM before.
230 if (IS_ENABLED(CONFIG_64BIT))
231 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base;
234 * memblock allocator is not aware of the fact that last 4K bytes of
235 * the addressable memory can not be mapped because of IS_ERR_VALUE
236 * macro. Make sure that last 4k bytes are not usable by memblock
237 * if end of dram is equal to maximum addressable memory. For 64-bit
238 * kernel, this problem can't happen here as the end of the virtual
239 * address space is occupied by the kernel mapping then this check must
240 * be done as soon as the kernel mapping base address is determined.
242 if (!IS_ENABLED(CONFIG_64BIT)) {
243 max_mapped_addr = __pa(~(ulong)0);
244 if (max_mapped_addr == (phys_ram_end - 1))
245 memblock_set_current_limit(max_mapped_addr - 4096);
248 min_low_pfn = PFN_UP(phys_ram_base);
249 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
250 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
252 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
253 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
255 reserve_initrd_mem();
258 * No allocation should be done before reserving the memory as defined
259 * in the device tree, otherwise the allocation could end up in a
262 early_init_fdt_scan_reserved_mem();
265 * If DTB is built in, no need to reserve its memblock.
266 * Otherwise, do reserve it but avoid using
267 * early_init_fdt_reserve_self() since __pa() does
268 * not work for DTB pointers that are fixmap addresses
270 if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
271 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
273 dma_contiguous_reserve(dma32_phys_limit);
274 if (IS_ENABLED(CONFIG_64BIT))
275 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
279 struct pt_alloc_ops pt_ops __initdata;
281 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
282 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
283 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
285 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
287 #ifdef CONFIG_XIP_KERNEL
288 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
289 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
290 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
291 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
292 #endif /* CONFIG_XIP_KERNEL */
294 static const pgprot_t protection_map[16] = {
295 [VM_NONE] = PAGE_NONE,
296 [VM_READ] = PAGE_READ,
297 [VM_WRITE] = PAGE_COPY,
298 [VM_WRITE | VM_READ] = PAGE_COPY,
299 [VM_EXEC] = PAGE_EXEC,
300 [VM_EXEC | VM_READ] = PAGE_READ_EXEC,
301 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC,
302 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC,
303 [VM_SHARED] = PAGE_NONE,
304 [VM_SHARED | VM_READ] = PAGE_READ,
305 [VM_SHARED | VM_WRITE] = PAGE_SHARED,
306 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED,
307 [VM_SHARED | VM_EXEC] = PAGE_EXEC,
308 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC,
309 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC,
310 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC
312 DECLARE_VM_GET_PAGE_PROT
314 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
316 unsigned long addr = __fix_to_virt(idx);
319 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
321 ptep = &fixmap_pte[pte_index(addr)];
323 if (pgprot_val(prot))
324 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
326 pte_clear(&init_mm, addr, ptep);
327 local_flush_tlb_page(addr);
330 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
332 return (pte_t *)((uintptr_t)pa);
335 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
337 clear_fixmap(FIX_PTE);
338 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
341 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
343 return (pte_t *) __va(pa);
346 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
349 * We only create PMD or PGD early mappings so we
350 * should never reach here with MMU disabled.
355 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
357 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
360 static phys_addr_t __init alloc_pte_late(uintptr_t va)
364 vaddr = __get_free_page(GFP_KERNEL);
365 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page((void *)vaddr)));
370 static void __init create_pte_mapping(pte_t *ptep,
371 uintptr_t va, phys_addr_t pa,
372 phys_addr_t sz, pgprot_t prot)
374 uintptr_t pte_idx = pte_index(va);
376 BUG_ON(sz != PAGE_SIZE);
378 if (pte_none(ptep[pte_idx]))
379 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
382 #ifndef __PAGETABLE_PMD_FOLDED
384 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
385 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
386 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
388 #ifdef CONFIG_XIP_KERNEL
389 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
390 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
391 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
392 #endif /* CONFIG_XIP_KERNEL */
394 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
395 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
396 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
398 #ifdef CONFIG_XIP_KERNEL
399 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
400 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d))
401 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d))
402 #endif /* CONFIG_XIP_KERNEL */
404 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
405 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
406 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
408 #ifdef CONFIG_XIP_KERNEL
409 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
410 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
411 #define early_pud ((pud_t *)XIP_FIXUP(early_pud))
412 #endif /* CONFIG_XIP_KERNEL */
414 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
416 /* Before MMU is enabled */
417 return (pmd_t *)((uintptr_t)pa);
420 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
422 clear_fixmap(FIX_PMD);
423 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
426 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
428 return (pmd_t *) __va(pa);
431 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
433 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
435 return (uintptr_t)early_pmd;
438 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
440 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
443 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
447 vaddr = __get_free_page(GFP_KERNEL);
448 BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page((void *)vaddr)));
453 static void __init create_pmd_mapping(pmd_t *pmdp,
454 uintptr_t va, phys_addr_t pa,
455 phys_addr_t sz, pgprot_t prot)
458 phys_addr_t pte_phys;
459 uintptr_t pmd_idx = pmd_index(va);
461 if (sz == PMD_SIZE) {
462 if (pmd_none(pmdp[pmd_idx]))
463 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
467 if (pmd_none(pmdp[pmd_idx])) {
468 pte_phys = pt_ops.alloc_pte(va);
469 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
470 ptep = pt_ops.get_pte_virt(pte_phys);
471 memset(ptep, 0, PAGE_SIZE);
473 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
474 ptep = pt_ops.get_pte_virt(pte_phys);
477 create_pte_mapping(ptep, va, pa, sz, prot);
480 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
482 return (pud_t *)((uintptr_t)pa);
485 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
487 clear_fixmap(FIX_PUD);
488 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
491 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
493 return (pud_t *)__va(pa);
496 static phys_addr_t __init alloc_pud_early(uintptr_t va)
498 /* Only one PUD is available for early mapping */
499 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
501 return (uintptr_t)early_pud;
504 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
506 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
509 static phys_addr_t alloc_pud_late(uintptr_t va)
513 vaddr = __get_free_page(GFP_KERNEL);
518 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
520 return (p4d_t *)((uintptr_t)pa);
523 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
525 clear_fixmap(FIX_P4D);
526 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
529 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
531 return (p4d_t *)__va(pa);
534 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
536 /* Only one P4D is available for early mapping */
537 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
539 return (uintptr_t)early_p4d;
542 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
544 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
547 static phys_addr_t alloc_p4d_late(uintptr_t va)
551 vaddr = __get_free_page(GFP_KERNEL);
556 static void __init create_pud_mapping(pud_t *pudp,
557 uintptr_t va, phys_addr_t pa,
558 phys_addr_t sz, pgprot_t prot)
561 phys_addr_t next_phys;
562 uintptr_t pud_index = pud_index(va);
564 if (sz == PUD_SIZE) {
565 if (pud_val(pudp[pud_index]) == 0)
566 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
570 if (pud_val(pudp[pud_index]) == 0) {
571 next_phys = pt_ops.alloc_pmd(va);
572 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
573 nextp = pt_ops.get_pmd_virt(next_phys);
574 memset(nextp, 0, PAGE_SIZE);
576 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
577 nextp = pt_ops.get_pmd_virt(next_phys);
580 create_pmd_mapping(nextp, va, pa, sz, prot);
583 static void __init create_p4d_mapping(p4d_t *p4dp,
584 uintptr_t va, phys_addr_t pa,
585 phys_addr_t sz, pgprot_t prot)
588 phys_addr_t next_phys;
589 uintptr_t p4d_index = p4d_index(va);
591 if (sz == P4D_SIZE) {
592 if (p4d_val(p4dp[p4d_index]) == 0)
593 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
597 if (p4d_val(p4dp[p4d_index]) == 0) {
598 next_phys = pt_ops.alloc_pud(va);
599 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
600 nextp = pt_ops.get_pud_virt(next_phys);
601 memset(nextp, 0, PAGE_SIZE);
603 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
604 nextp = pt_ops.get_pud_virt(next_phys);
607 create_pud_mapping(nextp, va, pa, sz, prot);
610 #define pgd_next_t p4d_t
611 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \
612 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \
613 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
614 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \
615 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \
616 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
617 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
618 (pgtable_l5_enabled ? \
619 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
620 (pgtable_l4_enabled ? \
621 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \
622 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
623 #define fixmap_pgd_next (pgtable_l5_enabled ? \
624 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \
625 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
626 #define trampoline_pgd_next (pgtable_l5_enabled ? \
627 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \
628 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
630 #define pgd_next_t pte_t
631 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
632 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
633 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
634 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
635 #define fixmap_pgd_next ((uintptr_t)fixmap_pte)
636 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
637 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
638 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
639 #endif /* __PAGETABLE_PMD_FOLDED */
641 void __init create_pgd_mapping(pgd_t *pgdp,
642 uintptr_t va, phys_addr_t pa,
643 phys_addr_t sz, pgprot_t prot)
646 phys_addr_t next_phys;
647 uintptr_t pgd_idx = pgd_index(va);
649 if (sz == PGDIR_SIZE) {
650 if (pgd_val(pgdp[pgd_idx]) == 0)
651 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
655 if (pgd_val(pgdp[pgd_idx]) == 0) {
656 next_phys = alloc_pgd_next(va);
657 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
658 nextp = get_pgd_next_virt(next_phys);
659 memset(nextp, 0, PAGE_SIZE);
661 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
662 nextp = get_pgd_next_virt(next_phys);
665 create_pgd_next_mapping(nextp, va, pa, sz, prot);
668 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va,
671 if (!(pa & (PGDIR_SIZE - 1)) && !(va & (PGDIR_SIZE - 1)) && size >= PGDIR_SIZE)
674 if (!(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
677 if (!(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
680 if (!(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
686 #ifdef CONFIG_XIP_KERNEL
687 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
688 extern char _xiprom[], _exiprom[], __data_loc;
690 /* called from head.S with MMU off */
691 asmlinkage void __init __copy_data(void)
693 void *from = (void *)(&__data_loc);
694 void *to = (void *)CONFIG_PHYS_RAM_BASE;
695 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
697 memcpy(to, from, sz);
701 #ifdef CONFIG_STRICT_KERNEL_RWX
702 static __init pgprot_t pgprot_from_va(uintptr_t va)
704 if (is_va_kernel_text(va))
705 return PAGE_KERNEL_READ_EXEC;
708 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
709 * we must protect its linear mapping alias from being executed and
711 * And rodata section is marked readonly in mark_rodata_ro.
713 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
714 return PAGE_KERNEL_READ;
719 void mark_rodata_ro(void)
721 set_kernel_memory(__start_rodata, _data, set_memory_ro);
722 if (IS_ENABLED(CONFIG_64BIT))
723 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
729 static __init pgprot_t pgprot_from_va(uintptr_t va)
731 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
734 return PAGE_KERNEL_EXEC;
736 #endif /* CONFIG_STRICT_KERNEL_RWX */
738 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
739 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
741 static void __init disable_pgtable_l5(void)
743 pgtable_l5_enabled = false;
744 kernel_map.page_offset = PAGE_OFFSET_L4;
745 satp_mode = SATP_MODE_48;
748 static void __init disable_pgtable_l4(void)
750 pgtable_l4_enabled = false;
751 kernel_map.page_offset = PAGE_OFFSET_L3;
752 satp_mode = SATP_MODE_39;
755 static int __init print_no4lvl(char *p)
757 pr_info("Disabled 4-level and 5-level paging");
760 early_param("no4lvl", print_no4lvl);
762 static int __init print_no5lvl(char *p)
764 pr_info("Disabled 5-level paging");
767 early_param("no5lvl", print_no5lvl);
770 * There is a simple way to determine if 4-level is supported by the
771 * underlying hardware: establish 1:1 mapping in 4-level page table mode
772 * then read SATP to see if the configuration was taken into account
773 * meaning sv48 is supported.
775 static __init void set_satp_mode(uintptr_t dtb_pa)
777 u64 identity_satp, hw_satp;
778 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
779 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
781 if (satp_mode_cmdline == SATP_MODE_57) {
782 disable_pgtable_l5();
783 } else if (satp_mode_cmdline == SATP_MODE_48) {
784 disable_pgtable_l5();
785 disable_pgtable_l4();
789 create_p4d_mapping(early_p4d,
790 set_satp_mode_pmd, (uintptr_t)early_pud,
791 P4D_SIZE, PAGE_TABLE);
792 create_pud_mapping(early_pud,
793 set_satp_mode_pmd, (uintptr_t)early_pmd,
794 PUD_SIZE, PAGE_TABLE);
795 /* Handle the case where set_satp_mode straddles 2 PMDs */
796 create_pmd_mapping(early_pmd,
797 set_satp_mode_pmd, set_satp_mode_pmd,
798 PMD_SIZE, PAGE_KERNEL_EXEC);
799 create_pmd_mapping(early_pmd,
800 set_satp_mode_pmd + PMD_SIZE,
801 set_satp_mode_pmd + PMD_SIZE,
802 PMD_SIZE, PAGE_KERNEL_EXEC);
804 create_pgd_mapping(early_pg_dir,
807 (uintptr_t)early_p4d : (uintptr_t)early_pud,
808 PGDIR_SIZE, PAGE_TABLE);
810 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
812 local_flush_tlb_all();
813 csr_write(CSR_SATP, identity_satp);
814 hw_satp = csr_swap(CSR_SATP, 0ULL);
815 local_flush_tlb_all();
817 if (hw_satp != identity_satp) {
818 if (pgtable_l5_enabled) {
819 disable_pgtable_l5();
820 memset(early_pg_dir, 0, PAGE_SIZE);
823 disable_pgtable_l4();
826 memset(early_pg_dir, 0, PAGE_SIZE);
827 memset(early_p4d, 0, PAGE_SIZE);
828 memset(early_pud, 0, PAGE_SIZE);
829 memset(early_pmd, 0, PAGE_SIZE);
834 * setup_vm() is called from head.S with MMU-off.
836 * Following requirements should be honoured for setup_vm() to work
838 * 1) It should use PC-relative addressing for accessing kernel symbols.
839 * To achieve this we always use GCC cmodel=medany.
840 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
841 * so disable compiler instrumentation when FTRACE is enabled.
843 * Currently, the above requirements are honoured by using custom CFLAGS
844 * for init.o in mm/Makefile.
847 #ifndef __riscv_cmodel_medany
848 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
851 #ifdef CONFIG_RELOCATABLE
852 extern unsigned long __rela_dyn_start, __rela_dyn_end;
854 static void __init relocate_kernel(void)
856 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
858 * This holds the offset between the linked virtual address and the
859 * relocated virtual address.
861 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
863 * This holds the offset between kernel linked virtual address and
866 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
868 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
869 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
870 Elf64_Addr relocated_addr = rela->r_addend;
872 if (rela->r_info != R_RISCV_RELATIVE)
876 * Make sure to not relocate vdso symbols like rt_sigreturn
877 * which are linked from the address 0 in vmlinux since
878 * vdso symbol addresses are actually used as an offset from
879 * mm->context.vdso in VDSO_OFFSET macro.
881 if (relocated_addr >= KERNEL_LINK_ADDR)
882 relocated_addr += reloc_offset;
884 *(Elf64_Addr *)addr = relocated_addr;
887 #endif /* CONFIG_RELOCATABLE */
889 #ifdef CONFIG_XIP_KERNEL
890 static void __init create_kernel_page_table(pgd_t *pgdir,
891 __always_unused bool early)
893 uintptr_t va, end_va;
895 /* Map the flash resident part */
896 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
897 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
898 create_pgd_mapping(pgdir, va,
899 kernel_map.xiprom + (va - kernel_map.virt_addr),
900 PMD_SIZE, PAGE_KERNEL_EXEC);
902 /* Map the data in RAM */
903 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
904 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
905 create_pgd_mapping(pgdir, va,
906 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
907 PMD_SIZE, PAGE_KERNEL);
910 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
912 uintptr_t va, end_va;
914 end_va = kernel_map.virt_addr + kernel_map.size;
915 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
916 create_pgd_mapping(pgdir, va,
917 kernel_map.phys_addr + (va - kernel_map.virt_addr),
920 PAGE_KERNEL_EXEC : pgprot_from_va(va));
925 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
926 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
929 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
932 #ifndef CONFIG_BUILTIN_DTB
933 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
935 /* Make sure the fdt fixmap address is always aligned on PMD size */
936 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
938 /* In 32-bit only, the fdt lies in its own PGD */
939 if (!IS_ENABLED(CONFIG_64BIT)) {
940 create_pgd_mapping(early_pg_dir, fix_fdt_va,
941 pa, MAX_FDT_SIZE, PAGE_KERNEL);
943 create_pmd_mapping(fixmap_pmd, fix_fdt_va,
944 pa, PMD_SIZE, PAGE_KERNEL);
945 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
946 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
949 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
952 * For 64-bit kernel, __va can't be used since it would return a linear
953 * mapping address whereas dtb_early_va will be used before
954 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
955 * kernel is mapped in the linear mapping, that makes no difference.
957 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
960 dtb_early_pa = dtb_pa;
964 * MMU is not enabled, the page tables are allocated directly using
965 * early_pmd/pud/p4d and the address returned is the physical one.
967 static void __init pt_ops_set_early(void)
969 pt_ops.alloc_pte = alloc_pte_early;
970 pt_ops.get_pte_virt = get_pte_virt_early;
971 #ifndef __PAGETABLE_PMD_FOLDED
972 pt_ops.alloc_pmd = alloc_pmd_early;
973 pt_ops.get_pmd_virt = get_pmd_virt_early;
974 pt_ops.alloc_pud = alloc_pud_early;
975 pt_ops.get_pud_virt = get_pud_virt_early;
976 pt_ops.alloc_p4d = alloc_p4d_early;
977 pt_ops.get_p4d_virt = get_p4d_virt_early;
982 * MMU is enabled but page table setup is not complete yet.
983 * fixmap page table alloc functions must be used as a means to temporarily
984 * map the allocated physical pages since the linear mapping does not exist yet.
986 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
987 * but it will be used as described above.
989 static void __init pt_ops_set_fixmap(void)
991 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
992 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
993 #ifndef __PAGETABLE_PMD_FOLDED
994 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
995 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
996 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
997 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
998 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
999 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
1004 * MMU is enabled and page table setup is complete, so from now, we can use
1005 * generic page allocation functions to setup page table.
1007 static void __init pt_ops_set_late(void)
1009 pt_ops.alloc_pte = alloc_pte_late;
1010 pt_ops.get_pte_virt = get_pte_virt_late;
1011 #ifndef __PAGETABLE_PMD_FOLDED
1012 pt_ops.alloc_pmd = alloc_pmd_late;
1013 pt_ops.get_pmd_virt = get_pmd_virt_late;
1014 pt_ops.alloc_pud = alloc_pud_late;
1015 pt_ops.get_pud_virt = get_pud_virt_late;
1016 pt_ops.alloc_p4d = alloc_p4d_late;
1017 pt_ops.get_p4d_virt = get_p4d_virt_late;
1021 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1023 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1025 kernel_map.virt_addr = KERNEL_LINK_ADDR;
1026 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1028 #ifdef CONFIG_XIP_KERNEL
1029 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1030 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1032 phys_ram_base = CONFIG_PHYS_RAM_BASE;
1033 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1034 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
1036 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1038 kernel_map.phys_addr = (uintptr_t)(&_start);
1039 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1042 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1043 set_satp_mode(dtb_pa);
1047 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1048 * where we have the system memory layout: this allows us to align
1049 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1050 * for the linear mapping. This is only possible because the kernel
1051 * mapping lies outside the linear mapping.
1052 * In 32-bit however, as the kernel resides in the linear mapping,
1053 * setup_vm_final can not change the mapping established here,
1054 * otherwise the same kernel addresses would get mapped to different
1055 * physical addresses (if the start of dram is different from the
1056 * kernel physical address start).
1058 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1059 0UL : PAGE_OFFSET - kernel_map.phys_addr;
1060 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1063 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
1064 * kernel, whereas for 64-bit kernel, the end of the virtual address
1065 * space is occupied by the modules/BPF/kernel mappings which reduces
1066 * the available size of the linear mapping.
1068 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
1070 /* Sanity check alignment and size */
1071 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1072 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1076 * The last 4K bytes of the addressable memory can not be mapped because
1077 * of IS_ERR_VALUE macro.
1079 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1082 #ifdef CONFIG_RELOCATABLE
1084 * Early page table uses only one PUD, which makes it possible
1085 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1086 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1087 * since a part of the kernel would not get mapped.
1089 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1093 apply_early_boot_alternatives();
1096 /* Setup early PGD for fixmap */
1097 create_pgd_mapping(early_pg_dir, FIXADDR_START,
1098 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1100 #ifndef __PAGETABLE_PMD_FOLDED
1101 /* Setup fixmap P4D and PUD */
1102 if (pgtable_l5_enabled)
1103 create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1104 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1105 /* Setup fixmap PUD and PMD */
1106 if (pgtable_l4_enabled)
1107 create_pud_mapping(fixmap_pud, FIXADDR_START,
1108 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1109 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1110 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1111 /* Setup trampoline PGD and PMD */
1112 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1113 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1114 if (pgtable_l5_enabled)
1115 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1116 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1117 if (pgtable_l4_enabled)
1118 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1119 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1120 #ifdef CONFIG_XIP_KERNEL
1121 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1122 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1124 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1125 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1128 /* Setup trampoline PGD */
1129 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1130 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1134 * Setup early PGD covering entire kernel which will allow
1135 * us to reach paging_init(). We map all memory banks later
1136 * in setup_vm_final() below.
1138 create_kernel_page_table(early_pg_dir, true);
1140 /* Setup early mapping for FDT early scan */
1141 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1144 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1145 * range can not span multiple pmds.
1147 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1148 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1150 #ifndef __PAGETABLE_PMD_FOLDED
1152 * Early ioremap fixmap is already created as it lies within first 2MB
1153 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1154 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1157 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1158 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1159 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1161 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1162 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1163 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1164 fix_to_virt(FIX_BTMAP_BEGIN));
1165 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1166 fix_to_virt(FIX_BTMAP_END));
1168 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1169 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1173 pt_ops_set_fixmap();
1176 static void __init create_linear_mapping_range(phys_addr_t start,
1178 uintptr_t fixed_map_size)
1181 uintptr_t va, map_size;
1183 for (pa = start; pa < end; pa += map_size) {
1184 va = (uintptr_t)__va(pa);
1185 map_size = fixed_map_size ? fixed_map_size :
1186 best_map_size(pa, va, end - pa);
1188 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1189 pgprot_from_va(va));
1193 static void __init create_linear_mapping_page_table(void)
1195 phys_addr_t start, end;
1196 phys_addr_t kfence_pool __maybe_unused;
1199 #ifdef CONFIG_STRICT_KERNEL_RWX
1200 phys_addr_t ktext_start = __pa_symbol(_start);
1201 phys_addr_t ktext_size = __init_data_begin - _start;
1202 phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1203 phys_addr_t krodata_size = _data - __start_rodata;
1205 /* Isolate kernel text and rodata so they don't get mapped with a PUD */
1206 memblock_mark_nomap(ktext_start, ktext_size);
1207 memblock_mark_nomap(krodata_start, krodata_size);
1210 #ifdef CONFIG_KFENCE
1212 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1213 * before we setup the linear mapping so that we avoid using hugepages
1216 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1217 BUG_ON(!kfence_pool);
1219 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1220 __kfence_pool = __va(kfence_pool);
1223 /* Map all memory banks in the linear mapping */
1224 for_each_mem_range(i, &start, &end) {
1227 if (start <= __pa(PAGE_OFFSET) &&
1228 __pa(PAGE_OFFSET) < end)
1229 start = __pa(PAGE_OFFSET);
1230 if (end >= __pa(PAGE_OFFSET) + memory_limit)
1231 end = __pa(PAGE_OFFSET) + memory_limit;
1233 create_linear_mapping_range(start, end, 0);
1236 #ifdef CONFIG_STRICT_KERNEL_RWX
1237 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0);
1238 create_linear_mapping_range(krodata_start,
1239 krodata_start + krodata_size, 0);
1241 memblock_clear_nomap(ktext_start, ktext_size);
1242 memblock_clear_nomap(krodata_start, krodata_size);
1245 #ifdef CONFIG_KFENCE
1246 create_linear_mapping_range(kfence_pool,
1247 kfence_pool + KFENCE_POOL_SIZE,
1250 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1254 static void __init setup_vm_final(void)
1256 /* Setup swapper PGD for fixmap */
1257 #if !defined(CONFIG_64BIT)
1259 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1260 * directly in swapper_pg_dir in addition to the pgd entry that points
1263 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1265 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1267 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1268 __pa_symbol(fixmap_pgd_next),
1269 PGDIR_SIZE, PAGE_TABLE);
1271 /* Map the linear mapping */
1272 create_linear_mapping_page_table();
1274 /* Map the kernel */
1275 if (IS_ENABLED(CONFIG_64BIT))
1276 create_kernel_page_table(swapper_pg_dir, false);
1279 kasan_swapper_init();
1282 /* Clear fixmap PTE and PMD mappings */
1283 clear_fixmap(FIX_PTE);
1284 clear_fixmap(FIX_PMD);
1285 clear_fixmap(FIX_PUD);
1286 clear_fixmap(FIX_P4D);
1288 /* Move to swapper page table */
1289 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1290 local_flush_tlb_all();
1295 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1297 dtb_early_va = (void *)dtb_pa;
1298 dtb_early_pa = dtb_pa;
1301 static inline void setup_vm_final(void)
1304 #endif /* CONFIG_MMU */
1307 * reserve_crashkernel() - reserves memory for crash kernel
1309 * This function reserves memory area given in "crashkernel=" kernel command
1310 * line parameter. The memory reserved is used by dump capture kernel when
1311 * primary kernel is crashing.
1313 static void __init reserve_crashkernel(void)
1315 unsigned long long crash_base = 0;
1316 unsigned long long crash_size = 0;
1317 unsigned long search_start = memblock_start_of_DRAM();
1318 unsigned long search_end = memblock_end_of_DRAM();
1322 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1325 * Don't reserve a region for a crash kernel on a crash kernel
1326 * since it doesn't make much sense and we have limited memory
1329 if (is_kdump_kernel()) {
1330 pr_info("crashkernel: ignoring reservation request\n");
1334 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
1335 &crash_size, &crash_base);
1336 if (ret || !crash_size)
1339 crash_size = PAGE_ALIGN(crash_size);
1342 search_start = crash_base;
1343 search_end = crash_base + crash_size;
1347 * Current riscv boot protocol requires 2MB alignment for
1348 * RV64 and 4MB alignment for RV32 (hugepage size)
1350 * Try to alloc from 32bit addressible physical memory so that
1351 * swiotlb can work on the crash kernel.
1353 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1355 min(search_end, (unsigned long)(SZ_4G - 1)));
1356 if (crash_base == 0) {
1357 /* Try again without restricting region to 32bit addressible memory */
1358 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1359 search_start, search_end);
1360 if (crash_base == 0) {
1361 pr_warn("crashkernel: couldn't allocate %lldKB\n",
1367 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1368 crash_base, crash_base + crash_size, crash_size >> 20);
1370 crashk_res.start = crash_base;
1371 crashk_res.end = crash_base + crash_size - 1;
1374 void __init paging_init(void)
1379 /* Depend on that Linear Mapping is ready */
1380 memblock_allow_resize();
1383 void __init misc_mem_init(void)
1385 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1389 reserve_crashkernel();
1390 memblock_dump_all();
1393 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1394 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1395 struct vmem_altmap *altmap)
1397 return vmemmap_populate_basepages(start, end, node, NULL);
1401 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1403 * Pre-allocates page-table pages for a specific area in the kernel
1404 * page-table. Only the level which needs to be synchronized between
1405 * all page-tables is allocated because the synchronization can be
1408 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1414 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1415 pgd_t *pgd = pgd_offset_k(addr);
1421 p4d = p4d_alloc(&init_mm, pgd, addr);
1425 if (pgtable_l5_enabled)
1429 pud = pud_alloc(&init_mm, p4d, addr);
1433 if (pgtable_l4_enabled)
1437 pmd = pmd_alloc(&init_mm, pud, addr);
1445 * The pages have to be there now or they will be missing in
1446 * process page-tables later.
1448 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1451 void __init pgtable_cache_init(void)
1453 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1454 if (IS_ENABLED(CONFIG_MODULES))
1455 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");