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);
53 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
54 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
55 EXPORT_SYMBOL(pgtable_l4_enabled);
56 EXPORT_SYMBOL(pgtable_l5_enabled);
59 phys_addr_t phys_ram_base __ro_after_init;
60 EXPORT_SYMBOL(phys_ram_base);
62 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
64 EXPORT_SYMBOL(empty_zero_page);
67 void *_dtb_early_va __initdata;
68 uintptr_t _dtb_early_pa __initdata;
70 phys_addr_t dma32_phys_limit __initdata;
72 static void __init zone_sizes_init(void)
74 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
76 #ifdef CONFIG_ZONE_DMA32
77 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
79 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
81 free_area_init(max_zone_pfns);
84 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
86 #define LOG2_SZ_1K ilog2(SZ_1K)
87 #define LOG2_SZ_1M ilog2(SZ_1M)
88 #define LOG2_SZ_1G ilog2(SZ_1G)
89 #define LOG2_SZ_1T ilog2(SZ_1T)
91 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
93 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
94 (((t) - (b)) >> LOG2_SZ_1K));
97 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
99 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
100 (((t) - (b)) >> LOG2_SZ_1M));
103 static inline void print_mlg(char *name, unsigned long b, unsigned long t)
105 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t,
106 (((t) - (b)) >> LOG2_SZ_1G));
110 static inline void print_mlt(char *name, unsigned long b, unsigned long t)
112 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t,
113 (((t) - (b)) >> LOG2_SZ_1T));
116 #define print_mlt(n, b, t) do {} while (0)
119 static inline void print_ml(char *name, unsigned long b, unsigned long t)
121 unsigned long diff = t - b;
123 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10)
124 print_mlt(name, b, t);
125 else if ((diff >> LOG2_SZ_1G) >= 10)
126 print_mlg(name, b, t);
127 else if ((diff >> LOG2_SZ_1M) >= 10)
128 print_mlm(name, b, t);
130 print_mlk(name, b, t);
133 static void __init print_vm_layout(void)
135 pr_notice("Virtual kernel memory layout:\n");
136 print_ml("fixmap", (unsigned long)FIXADDR_START,
137 (unsigned long)FIXADDR_TOP);
138 print_ml("pci io", (unsigned long)PCI_IO_START,
139 (unsigned long)PCI_IO_END);
140 print_ml("vmemmap", (unsigned long)VMEMMAP_START,
141 (unsigned long)VMEMMAP_END);
142 print_ml("vmalloc", (unsigned long)VMALLOC_START,
143 (unsigned long)VMALLOC_END);
145 print_ml("modules", (unsigned long)MODULES_VADDR,
146 (unsigned long)MODULES_END);
148 print_ml("lowmem", (unsigned long)PAGE_OFFSET,
149 (unsigned long)high_memory);
150 if (IS_ENABLED(CONFIG_64BIT)) {
152 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
155 print_ml("kernel", (unsigned long)kernel_map.virt_addr,
156 (unsigned long)ADDRESS_SPACE_END);
160 static void print_vm_layout(void) { }
161 #endif /* CONFIG_DEBUG_VM */
163 void __init mem_init(void)
165 #ifdef CONFIG_FLATMEM
167 #endif /* CONFIG_FLATMEM */
169 swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE);
175 /* Limit the memory size via mem. */
176 static phys_addr_t memory_limit;
177 #ifdef CONFIG_XIP_KERNEL
178 #define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit))
179 #endif /* CONFIG_XIP_KERNEL */
181 static int __init early_mem(char *p)
188 size = memparse(p, &p) & PAGE_MASK;
189 memory_limit = min_t(u64, size, memory_limit);
191 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
195 early_param("mem", early_mem);
197 static void __init setup_bootmem(void)
199 phys_addr_t vmlinux_end = __pa_symbol(&_end);
200 phys_addr_t max_mapped_addr;
201 phys_addr_t phys_ram_end, vmlinux_start;
203 if (IS_ENABLED(CONFIG_XIP_KERNEL))
204 vmlinux_start = __pa_symbol(&_sdata);
206 vmlinux_start = __pa_symbol(&_start);
208 memblock_enforce_memory_limit(memory_limit);
211 * Make sure we align the reservation on PMD_SIZE since we will
212 * map the kernel in the linear mapping as read-only: we do not want
213 * any allocation to happen between _end and the next pmd aligned page.
215 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
216 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
218 * Reserve from the start of the kernel to the end of the kernel
220 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
222 phys_ram_end = memblock_end_of_DRAM();
225 * Make sure we align the start of the memory on a PMD boundary so that
226 * at worst, we map the linear mapping with PMD mappings.
228 if (!IS_ENABLED(CONFIG_XIP_KERNEL))
229 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK;
232 * In 64-bit, any use of __va/__pa before this point is wrong as we
233 * did not know the start of DRAM before.
235 if (IS_ENABLED(CONFIG_64BIT))
236 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base;
239 * memblock allocator is not aware of the fact that last 4K bytes of
240 * the addressable memory can not be mapped because of IS_ERR_VALUE
241 * macro. Make sure that last 4k bytes are not usable by memblock
242 * if end of dram is equal to maximum addressable memory. For 64-bit
243 * kernel, this problem can't happen here as the end of the virtual
244 * address space is occupied by the kernel mapping then this check must
245 * be done as soon as the kernel mapping base address is determined.
247 if (!IS_ENABLED(CONFIG_64BIT)) {
248 max_mapped_addr = __pa(~(ulong)0);
249 if (max_mapped_addr == (phys_ram_end - 1))
250 memblock_set_current_limit(max_mapped_addr - 4096);
253 min_low_pfn = PFN_UP(phys_ram_base);
254 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
255 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
257 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
258 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
260 reserve_initrd_mem();
263 * No allocation should be done before reserving the memory as defined
264 * in the device tree, otherwise the allocation could end up in a
267 early_init_fdt_scan_reserved_mem();
270 * If DTB is built in, no need to reserve its memblock.
271 * Otherwise, do reserve it but avoid using
272 * early_init_fdt_reserve_self() since __pa() does
273 * not work for DTB pointers that are fixmap addresses
275 if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
276 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
278 dma_contiguous_reserve(dma32_phys_limit);
279 if (IS_ENABLED(CONFIG_64BIT))
280 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
284 struct pt_alloc_ops pt_ops __initdata;
286 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
287 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
288 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
290 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
292 #ifdef CONFIG_XIP_KERNEL
293 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
294 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
295 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
296 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
297 #endif /* CONFIG_XIP_KERNEL */
299 static const pgprot_t protection_map[16] = {
300 [VM_NONE] = PAGE_NONE,
301 [VM_READ] = PAGE_READ,
302 [VM_WRITE] = PAGE_COPY,
303 [VM_WRITE | VM_READ] = PAGE_COPY,
304 [VM_EXEC] = PAGE_EXEC,
305 [VM_EXEC | VM_READ] = PAGE_READ_EXEC,
306 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC,
307 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC,
308 [VM_SHARED] = PAGE_NONE,
309 [VM_SHARED | VM_READ] = PAGE_READ,
310 [VM_SHARED | VM_WRITE] = PAGE_SHARED,
311 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED,
312 [VM_SHARED | VM_EXEC] = PAGE_EXEC,
313 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC,
314 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC,
315 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC
317 DECLARE_VM_GET_PAGE_PROT
319 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
321 unsigned long addr = __fix_to_virt(idx);
324 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
326 ptep = &fixmap_pte[pte_index(addr)];
328 if (pgprot_val(prot))
329 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
331 pte_clear(&init_mm, addr, ptep);
332 local_flush_tlb_page(addr);
335 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
337 return (pte_t *)((uintptr_t)pa);
340 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
342 clear_fixmap(FIX_PTE);
343 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
346 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
348 return (pte_t *) __va(pa);
351 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
354 * We only create PMD or PGD early mappings so we
355 * should never reach here with MMU disabled.
360 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
362 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
365 static phys_addr_t __init alloc_pte_late(uintptr_t va)
367 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
369 BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc));
370 return __pa((pte_t *)ptdesc_address(ptdesc));
373 static void __init create_pte_mapping(pte_t *ptep,
374 uintptr_t va, phys_addr_t pa,
375 phys_addr_t sz, pgprot_t prot)
377 uintptr_t pte_idx = pte_index(va);
379 BUG_ON(sz != PAGE_SIZE);
381 if (pte_none(ptep[pte_idx]))
382 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
385 #ifndef __PAGETABLE_PMD_FOLDED
387 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
388 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
389 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
391 #ifdef CONFIG_XIP_KERNEL
392 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
393 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
394 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
395 #endif /* CONFIG_XIP_KERNEL */
397 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
398 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
399 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
401 #ifdef CONFIG_XIP_KERNEL
402 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
403 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d))
404 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d))
405 #endif /* CONFIG_XIP_KERNEL */
407 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
408 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
409 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
411 #ifdef CONFIG_XIP_KERNEL
412 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
413 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
414 #define early_pud ((pud_t *)XIP_FIXUP(early_pud))
415 #endif /* CONFIG_XIP_KERNEL */
417 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
419 /* Before MMU is enabled */
420 return (pmd_t *)((uintptr_t)pa);
423 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
425 clear_fixmap(FIX_PMD);
426 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
429 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
431 return (pmd_t *) __va(pa);
434 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
436 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
438 return (uintptr_t)early_pmd;
441 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
443 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
446 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
448 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
450 BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc));
451 return __pa((pmd_t *)ptdesc_address(ptdesc));
454 static void __init create_pmd_mapping(pmd_t *pmdp,
455 uintptr_t va, phys_addr_t pa,
456 phys_addr_t sz, pgprot_t prot)
459 phys_addr_t pte_phys;
460 uintptr_t pmd_idx = pmd_index(va);
462 if (sz == PMD_SIZE) {
463 if (pmd_none(pmdp[pmd_idx]))
464 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
468 if (pmd_none(pmdp[pmd_idx])) {
469 pte_phys = pt_ops.alloc_pte(va);
470 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
471 ptep = pt_ops.get_pte_virt(pte_phys);
472 memset(ptep, 0, PAGE_SIZE);
474 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
475 ptep = pt_ops.get_pte_virt(pte_phys);
478 create_pte_mapping(ptep, va, pa, sz, prot);
481 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
483 return (pud_t *)((uintptr_t)pa);
486 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
488 clear_fixmap(FIX_PUD);
489 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
492 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
494 return (pud_t *)__va(pa);
497 static phys_addr_t __init alloc_pud_early(uintptr_t va)
499 /* Only one PUD is available for early mapping */
500 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
502 return (uintptr_t)early_pud;
505 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
507 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
510 static phys_addr_t alloc_pud_late(uintptr_t va)
514 vaddr = __get_free_page(GFP_KERNEL);
519 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
521 return (p4d_t *)((uintptr_t)pa);
524 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
526 clear_fixmap(FIX_P4D);
527 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
530 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
532 return (p4d_t *)__va(pa);
535 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
537 /* Only one P4D is available for early mapping */
538 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
540 return (uintptr_t)early_p4d;
543 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
545 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
548 static phys_addr_t alloc_p4d_late(uintptr_t va)
552 vaddr = __get_free_page(GFP_KERNEL);
557 static void __init create_pud_mapping(pud_t *pudp,
558 uintptr_t va, phys_addr_t pa,
559 phys_addr_t sz, pgprot_t prot)
562 phys_addr_t next_phys;
563 uintptr_t pud_index = pud_index(va);
565 if (sz == PUD_SIZE) {
566 if (pud_val(pudp[pud_index]) == 0)
567 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
571 if (pud_val(pudp[pud_index]) == 0) {
572 next_phys = pt_ops.alloc_pmd(va);
573 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
574 nextp = pt_ops.get_pmd_virt(next_phys);
575 memset(nextp, 0, PAGE_SIZE);
577 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
578 nextp = pt_ops.get_pmd_virt(next_phys);
581 create_pmd_mapping(nextp, va, pa, sz, prot);
584 static void __init create_p4d_mapping(p4d_t *p4dp,
585 uintptr_t va, phys_addr_t pa,
586 phys_addr_t sz, pgprot_t prot)
589 phys_addr_t next_phys;
590 uintptr_t p4d_index = p4d_index(va);
592 if (sz == P4D_SIZE) {
593 if (p4d_val(p4dp[p4d_index]) == 0)
594 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
598 if (p4d_val(p4dp[p4d_index]) == 0) {
599 next_phys = pt_ops.alloc_pud(va);
600 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
601 nextp = pt_ops.get_pud_virt(next_phys);
602 memset(nextp, 0, PAGE_SIZE);
604 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
605 nextp = pt_ops.get_pud_virt(next_phys);
608 create_pud_mapping(nextp, va, pa, sz, prot);
611 #define pgd_next_t p4d_t
612 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \
613 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \
614 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
615 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \
616 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \
617 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
618 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
619 (pgtable_l5_enabled ? \
620 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
621 (pgtable_l4_enabled ? \
622 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \
623 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
624 #define fixmap_pgd_next (pgtable_l5_enabled ? \
625 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \
626 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
627 #define trampoline_pgd_next (pgtable_l5_enabled ? \
628 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \
629 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
631 #define pgd_next_t pte_t
632 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
633 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
634 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
635 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
636 #define fixmap_pgd_next ((uintptr_t)fixmap_pte)
637 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
638 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
639 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
640 #endif /* __PAGETABLE_PMD_FOLDED */
642 void __init create_pgd_mapping(pgd_t *pgdp,
643 uintptr_t va, phys_addr_t pa,
644 phys_addr_t sz, pgprot_t prot)
647 phys_addr_t next_phys;
648 uintptr_t pgd_idx = pgd_index(va);
650 if (sz == PGDIR_SIZE) {
651 if (pgd_val(pgdp[pgd_idx]) == 0)
652 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
656 if (pgd_val(pgdp[pgd_idx]) == 0) {
657 next_phys = alloc_pgd_next(va);
658 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
659 nextp = get_pgd_next_virt(next_phys);
660 memset(nextp, 0, PAGE_SIZE);
662 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
663 nextp = get_pgd_next_virt(next_phys);
666 create_pgd_next_mapping(nextp, va, pa, sz, prot);
669 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va,
672 if (pgtable_l5_enabled &&
673 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
676 if (pgtable_l4_enabled &&
677 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
680 if (IS_ENABLED(CONFIG_64BIT) &&
681 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
687 #ifdef CONFIG_XIP_KERNEL
688 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
689 extern char _xiprom[], _exiprom[], __data_loc;
691 /* called from head.S with MMU off */
692 asmlinkage void __init __copy_data(void)
694 void *from = (void *)(&__data_loc);
695 void *to = (void *)CONFIG_PHYS_RAM_BASE;
696 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
698 memcpy(to, from, sz);
702 #ifdef CONFIG_STRICT_KERNEL_RWX
703 static __init pgprot_t pgprot_from_va(uintptr_t va)
705 if (is_va_kernel_text(va))
706 return PAGE_KERNEL_READ_EXEC;
709 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
710 * we must protect its linear mapping alias from being executed and
712 * And rodata section is marked readonly in mark_rodata_ro.
714 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
715 return PAGE_KERNEL_READ;
720 void mark_rodata_ro(void)
722 set_kernel_memory(__start_rodata, _data, set_memory_ro);
723 if (IS_ENABLED(CONFIG_64BIT))
724 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
730 static __init pgprot_t pgprot_from_va(uintptr_t va)
732 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
735 return PAGE_KERNEL_EXEC;
737 #endif /* CONFIG_STRICT_KERNEL_RWX */
739 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
740 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
742 static void __init disable_pgtable_l5(void)
744 pgtable_l5_enabled = false;
745 kernel_map.page_offset = PAGE_OFFSET_L4;
746 satp_mode = SATP_MODE_48;
749 static void __init disable_pgtable_l4(void)
751 pgtable_l4_enabled = false;
752 kernel_map.page_offset = PAGE_OFFSET_L3;
753 satp_mode = SATP_MODE_39;
756 static int __init print_no4lvl(char *p)
758 pr_info("Disabled 4-level and 5-level paging");
761 early_param("no4lvl", print_no4lvl);
763 static int __init print_no5lvl(char *p)
765 pr_info("Disabled 5-level paging");
768 early_param("no5lvl", print_no5lvl);
771 * There is a simple way to determine if 4-level is supported by the
772 * underlying hardware: establish 1:1 mapping in 4-level page table mode
773 * then read SATP to see if the configuration was taken into account
774 * meaning sv48 is supported.
776 static __init void set_satp_mode(uintptr_t dtb_pa)
778 u64 identity_satp, hw_satp;
779 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
780 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
782 if (satp_mode_cmdline == SATP_MODE_57) {
783 disable_pgtable_l5();
784 } else if (satp_mode_cmdline == SATP_MODE_48) {
785 disable_pgtable_l5();
786 disable_pgtable_l4();
790 create_p4d_mapping(early_p4d,
791 set_satp_mode_pmd, (uintptr_t)early_pud,
792 P4D_SIZE, PAGE_TABLE);
793 create_pud_mapping(early_pud,
794 set_satp_mode_pmd, (uintptr_t)early_pmd,
795 PUD_SIZE, PAGE_TABLE);
796 /* Handle the case where set_satp_mode straddles 2 PMDs */
797 create_pmd_mapping(early_pmd,
798 set_satp_mode_pmd, set_satp_mode_pmd,
799 PMD_SIZE, PAGE_KERNEL_EXEC);
800 create_pmd_mapping(early_pmd,
801 set_satp_mode_pmd + PMD_SIZE,
802 set_satp_mode_pmd + PMD_SIZE,
803 PMD_SIZE, PAGE_KERNEL_EXEC);
805 create_pgd_mapping(early_pg_dir,
808 (uintptr_t)early_p4d : (uintptr_t)early_pud,
809 PGDIR_SIZE, PAGE_TABLE);
811 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
813 local_flush_tlb_all();
814 csr_write(CSR_SATP, identity_satp);
815 hw_satp = csr_swap(CSR_SATP, 0ULL);
816 local_flush_tlb_all();
818 if (hw_satp != identity_satp) {
819 if (pgtable_l5_enabled) {
820 disable_pgtable_l5();
821 memset(early_pg_dir, 0, PAGE_SIZE);
824 disable_pgtable_l4();
827 memset(early_pg_dir, 0, PAGE_SIZE);
828 memset(early_p4d, 0, PAGE_SIZE);
829 memset(early_pud, 0, PAGE_SIZE);
830 memset(early_pmd, 0, PAGE_SIZE);
835 * setup_vm() is called from head.S with MMU-off.
837 * Following requirements should be honoured for setup_vm() to work
839 * 1) It should use PC-relative addressing for accessing kernel symbols.
840 * To achieve this we always use GCC cmodel=medany.
841 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
842 * so disable compiler instrumentation when FTRACE is enabled.
844 * Currently, the above requirements are honoured by using custom CFLAGS
845 * for init.o in mm/Makefile.
848 #ifndef __riscv_cmodel_medany
849 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
852 #ifdef CONFIG_RELOCATABLE
853 extern unsigned long __rela_dyn_start, __rela_dyn_end;
855 static void __init relocate_kernel(void)
857 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
859 * This holds the offset between the linked virtual address and the
860 * relocated virtual address.
862 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
864 * This holds the offset between kernel linked virtual address and
867 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
869 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
870 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
871 Elf64_Addr relocated_addr = rela->r_addend;
873 if (rela->r_info != R_RISCV_RELATIVE)
877 * Make sure to not relocate vdso symbols like rt_sigreturn
878 * which are linked from the address 0 in vmlinux since
879 * vdso symbol addresses are actually used as an offset from
880 * mm->context.vdso in VDSO_OFFSET macro.
882 if (relocated_addr >= KERNEL_LINK_ADDR)
883 relocated_addr += reloc_offset;
885 *(Elf64_Addr *)addr = relocated_addr;
888 #endif /* CONFIG_RELOCATABLE */
890 #ifdef CONFIG_XIP_KERNEL
891 static void __init create_kernel_page_table(pgd_t *pgdir,
892 __always_unused bool early)
894 uintptr_t va, end_va;
896 /* Map the flash resident part */
897 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
898 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
899 create_pgd_mapping(pgdir, va,
900 kernel_map.xiprom + (va - kernel_map.virt_addr),
901 PMD_SIZE, PAGE_KERNEL_EXEC);
903 /* Map the data in RAM */
904 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
905 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
906 create_pgd_mapping(pgdir, va,
907 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
908 PMD_SIZE, PAGE_KERNEL);
911 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
913 uintptr_t va, end_va;
915 end_va = kernel_map.virt_addr + kernel_map.size;
916 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
917 create_pgd_mapping(pgdir, va,
918 kernel_map.phys_addr + (va - kernel_map.virt_addr),
921 PAGE_KERNEL_EXEC : pgprot_from_va(va));
926 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
927 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
930 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
933 #ifndef CONFIG_BUILTIN_DTB
934 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
936 /* Make sure the fdt fixmap address is always aligned on PMD size */
937 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
939 /* In 32-bit only, the fdt lies in its own PGD */
940 if (!IS_ENABLED(CONFIG_64BIT)) {
941 create_pgd_mapping(early_pg_dir, fix_fdt_va,
942 pa, MAX_FDT_SIZE, PAGE_KERNEL);
944 create_pmd_mapping(fixmap_pmd, fix_fdt_va,
945 pa, PMD_SIZE, PAGE_KERNEL);
946 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
947 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
950 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
953 * For 64-bit kernel, __va can't be used since it would return a linear
954 * mapping address whereas dtb_early_va will be used before
955 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
956 * kernel is mapped in the linear mapping, that makes no difference.
958 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa);
961 dtb_early_pa = dtb_pa;
965 * MMU is not enabled, the page tables are allocated directly using
966 * early_pmd/pud/p4d and the address returned is the physical one.
968 static void __init pt_ops_set_early(void)
970 pt_ops.alloc_pte = alloc_pte_early;
971 pt_ops.get_pte_virt = get_pte_virt_early;
972 #ifndef __PAGETABLE_PMD_FOLDED
973 pt_ops.alloc_pmd = alloc_pmd_early;
974 pt_ops.get_pmd_virt = get_pmd_virt_early;
975 pt_ops.alloc_pud = alloc_pud_early;
976 pt_ops.get_pud_virt = get_pud_virt_early;
977 pt_ops.alloc_p4d = alloc_p4d_early;
978 pt_ops.get_p4d_virt = get_p4d_virt_early;
983 * MMU is enabled but page table setup is not complete yet.
984 * fixmap page table alloc functions must be used as a means to temporarily
985 * map the allocated physical pages since the linear mapping does not exist yet.
987 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
988 * but it will be used as described above.
990 static void __init pt_ops_set_fixmap(void)
992 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
993 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
994 #ifndef __PAGETABLE_PMD_FOLDED
995 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
996 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
997 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
998 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
999 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
1000 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
1005 * MMU is enabled and page table setup is complete, so from now, we can use
1006 * generic page allocation functions to setup page table.
1008 static void __init pt_ops_set_late(void)
1010 pt_ops.alloc_pte = alloc_pte_late;
1011 pt_ops.get_pte_virt = get_pte_virt_late;
1012 #ifndef __PAGETABLE_PMD_FOLDED
1013 pt_ops.alloc_pmd = alloc_pmd_late;
1014 pt_ops.get_pmd_virt = get_pmd_virt_late;
1015 pt_ops.alloc_pud = alloc_pud_late;
1016 pt_ops.get_pud_virt = get_pud_virt_late;
1017 pt_ops.alloc_p4d = alloc_p4d_late;
1018 pt_ops.get_p4d_virt = get_p4d_virt_late;
1022 #ifdef CONFIG_RANDOMIZE_BASE
1023 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa);
1024 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa);
1026 static int __init print_nokaslr(char *p)
1028 pr_info("Disabled KASLR");
1031 early_param("nokaslr", print_nokaslr);
1033 unsigned long kaslr_offset(void)
1035 return kernel_map.virt_offset;
1039 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1041 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1043 #ifdef CONFIG_RANDOMIZE_BASE
1044 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) {
1045 u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa);
1046 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
1050 * Compute the number of positions available: we are limited
1051 * by the early page table that only has one PUD and we must
1052 * be aligned on PMD_SIZE.
1054 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE;
1056 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE;
1060 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset;
1062 #ifdef CONFIG_XIP_KERNEL
1064 kernel_map.page_offset = PAGE_OFFSET_L3;
1066 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1068 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1069 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1071 phys_ram_base = CONFIG_PHYS_RAM_BASE;
1072 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1073 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
1075 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1077 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1078 kernel_map.phys_addr = (uintptr_t)(&_start);
1079 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1082 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1083 set_satp_mode(dtb_pa);
1087 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1088 * where we have the system memory layout: this allows us to align
1089 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1090 * for the linear mapping. This is only possible because the kernel
1091 * mapping lies outside the linear mapping.
1092 * In 32-bit however, as the kernel resides in the linear mapping,
1093 * setup_vm_final can not change the mapping established here,
1094 * otherwise the same kernel addresses would get mapped to different
1095 * physical addresses (if the start of dram is different from the
1096 * kernel physical address start).
1098 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1099 0UL : PAGE_OFFSET - kernel_map.phys_addr;
1100 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1103 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
1104 * kernel, whereas for 64-bit kernel, the end of the virtual address
1105 * space is occupied by the modules/BPF/kernel mappings which reduces
1106 * the available size of the linear mapping.
1108 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
1110 /* Sanity check alignment and size */
1111 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1112 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1116 * The last 4K bytes of the addressable memory can not be mapped because
1117 * of IS_ERR_VALUE macro.
1119 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1122 #ifdef CONFIG_RELOCATABLE
1124 * Early page table uses only one PUD, which makes it possible
1125 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1126 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1127 * since a part of the kernel would not get mapped.
1129 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1133 apply_early_boot_alternatives();
1136 /* Setup early PGD for fixmap */
1137 create_pgd_mapping(early_pg_dir, FIXADDR_START,
1138 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1140 #ifndef __PAGETABLE_PMD_FOLDED
1141 /* Setup fixmap P4D and PUD */
1142 if (pgtable_l5_enabled)
1143 create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1144 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1145 /* Setup fixmap PUD and PMD */
1146 if (pgtable_l4_enabled)
1147 create_pud_mapping(fixmap_pud, FIXADDR_START,
1148 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1149 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1150 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1151 /* Setup trampoline PGD and PMD */
1152 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1153 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1154 if (pgtable_l5_enabled)
1155 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1156 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1157 if (pgtable_l4_enabled)
1158 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1159 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1160 #ifdef CONFIG_XIP_KERNEL
1161 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1162 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1164 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1165 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1168 /* Setup trampoline PGD */
1169 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1170 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1174 * Setup early PGD covering entire kernel which will allow
1175 * us to reach paging_init(). We map all memory banks later
1176 * in setup_vm_final() below.
1178 create_kernel_page_table(early_pg_dir, true);
1180 /* Setup early mapping for FDT early scan */
1181 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1184 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1185 * range can not span multiple pmds.
1187 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1188 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1190 #ifndef __PAGETABLE_PMD_FOLDED
1192 * Early ioremap fixmap is already created as it lies within first 2MB
1193 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1194 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1197 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1198 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1199 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1201 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1202 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1203 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1204 fix_to_virt(FIX_BTMAP_BEGIN));
1205 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1206 fix_to_virt(FIX_BTMAP_END));
1208 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1209 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1213 pt_ops_set_fixmap();
1216 static void __init create_linear_mapping_range(phys_addr_t start,
1218 uintptr_t fixed_map_size)
1221 uintptr_t va, map_size;
1223 for (pa = start; pa < end; pa += map_size) {
1224 va = (uintptr_t)__va(pa);
1225 map_size = fixed_map_size ? fixed_map_size :
1226 best_map_size(pa, va, end - pa);
1228 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1229 pgprot_from_va(va));
1233 static void __init create_linear_mapping_page_table(void)
1235 phys_addr_t start, end;
1236 phys_addr_t kfence_pool __maybe_unused;
1239 #ifdef CONFIG_STRICT_KERNEL_RWX
1240 phys_addr_t ktext_start = __pa_symbol(_start);
1241 phys_addr_t ktext_size = __init_data_begin - _start;
1242 phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1243 phys_addr_t krodata_size = _data - __start_rodata;
1245 /* Isolate kernel text and rodata so they don't get mapped with a PUD */
1246 memblock_mark_nomap(ktext_start, ktext_size);
1247 memblock_mark_nomap(krodata_start, krodata_size);
1250 #ifdef CONFIG_KFENCE
1252 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1253 * before we setup the linear mapping so that we avoid using hugepages
1256 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1257 BUG_ON(!kfence_pool);
1259 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1260 __kfence_pool = __va(kfence_pool);
1263 /* Map all memory banks in the linear mapping */
1264 for_each_mem_range(i, &start, &end) {
1267 if (start <= __pa(PAGE_OFFSET) &&
1268 __pa(PAGE_OFFSET) < end)
1269 start = __pa(PAGE_OFFSET);
1270 if (end >= __pa(PAGE_OFFSET) + memory_limit)
1271 end = __pa(PAGE_OFFSET) + memory_limit;
1273 create_linear_mapping_range(start, end, 0);
1276 #ifdef CONFIG_STRICT_KERNEL_RWX
1277 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0);
1278 create_linear_mapping_range(krodata_start,
1279 krodata_start + krodata_size, 0);
1281 memblock_clear_nomap(ktext_start, ktext_size);
1282 memblock_clear_nomap(krodata_start, krodata_size);
1285 #ifdef CONFIG_KFENCE
1286 create_linear_mapping_range(kfence_pool,
1287 kfence_pool + KFENCE_POOL_SIZE,
1290 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1294 static void __init setup_vm_final(void)
1296 /* Setup swapper PGD for fixmap */
1297 #if !defined(CONFIG_64BIT)
1299 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1300 * directly in swapper_pg_dir in addition to the pgd entry that points
1303 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1305 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1307 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1308 __pa_symbol(fixmap_pgd_next),
1309 PGDIR_SIZE, PAGE_TABLE);
1311 /* Map the linear mapping */
1312 create_linear_mapping_page_table();
1314 /* Map the kernel */
1315 if (IS_ENABLED(CONFIG_64BIT))
1316 create_kernel_page_table(swapper_pg_dir, false);
1319 kasan_swapper_init();
1322 /* Clear fixmap PTE and PMD mappings */
1323 clear_fixmap(FIX_PTE);
1324 clear_fixmap(FIX_PMD);
1325 clear_fixmap(FIX_PUD);
1326 clear_fixmap(FIX_P4D);
1328 /* Move to swapper page table */
1329 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1330 local_flush_tlb_all();
1335 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1337 dtb_early_va = (void *)dtb_pa;
1338 dtb_early_pa = dtb_pa;
1341 static inline void setup_vm_final(void)
1344 #endif /* CONFIG_MMU */
1347 * reserve_crashkernel() - reserves memory for crash kernel
1349 * This function reserves memory area given in "crashkernel=" kernel command
1350 * line parameter. The memory reserved is used by dump capture kernel when
1351 * primary kernel is crashing.
1353 static void __init arch_reserve_crashkernel(void)
1355 unsigned long long low_size = 0;
1356 unsigned long long crash_base, crash_size;
1357 char *cmdline = boot_command_line;
1361 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1364 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(),
1365 &crash_size, &crash_base,
1370 reserve_crashkernel_generic(cmdline, crash_size, crash_base,
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);
1388 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1389 /* The entire VMEMMAP region has been populated. Flush TLB for this region */
1390 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END);
1393 arch_reserve_crashkernel();
1394 memblock_dump_all();
1397 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1398 void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node,
1399 unsigned long addr, unsigned long next)
1401 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL);
1404 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node,
1405 unsigned long addr, unsigned long next)
1407 vmemmap_verify((pte_t *)pmdp, node, addr, next);
1411 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1412 struct vmem_altmap *altmap)
1415 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we
1416 * can't use hugepage mappings for 2-level page table because in case of
1417 * memory hotplug, we are not able to update all the page tables with
1420 return vmemmap_populate_hugepages(start, end, node, NULL);
1424 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1426 * Pre-allocates page-table pages for a specific area in the kernel
1427 * page-table. Only the level which needs to be synchronized between
1428 * all page-tables is allocated because the synchronization can be
1431 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1437 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1438 pgd_t *pgd = pgd_offset_k(addr);
1444 p4d = p4d_alloc(&init_mm, pgd, addr);
1448 if (pgtable_l5_enabled)
1452 pud = pud_alloc(&init_mm, p4d, addr);
1456 if (pgtable_l4_enabled)
1460 pmd = pmd_alloc(&init_mm, pud, addr);
1468 * The pages have to be there now or they will be missing in
1469 * process page-tables later.
1471 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1474 void __init pgtable_cache_init(void)
1476 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1477 if (IS_ENABLED(CONFIG_MODULES))
1478 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");