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>
23 #ifdef CONFIG_RELOCATABLE
24 #include <linux/elf.h>
26 #include <linux/kfence.h>
27 #include <linux/execmem.h>
29 #include <asm/fixmap.h>
32 #include <asm/pgtable.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 __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL);
54 bool pgtable_l5_enabled __ro_after_init = !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 bool swiotlb = max_pfn > PFN_DOWN(dma32_phys_limit);
166 #ifdef CONFIG_FLATMEM
168 #endif /* CONFIG_FLATMEM */
170 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && !swiotlb &&
171 dma_cache_alignment != 1) {
173 * If no bouncing needed for ZONE_DMA, allocate 1MB swiotlb
174 * buffer per 1GB of RAM for kmalloc() bouncing on
175 * non-coherent platforms.
178 DIV_ROUND_UP(memblock_phys_mem_size(), 1024);
179 swiotlb_adjust_size(min(swiotlb_size_or_default(), size));
183 swiotlb_init(swiotlb, SWIOTLB_VERBOSE);
189 /* Limit the memory size via mem. */
190 static phys_addr_t memory_limit;
191 #ifdef CONFIG_XIP_KERNEL
192 #define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit))
193 #endif /* CONFIG_XIP_KERNEL */
195 static int __init early_mem(char *p)
202 size = memparse(p, &p) & PAGE_MASK;
203 memory_limit = min_t(u64, size, memory_limit);
205 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
209 early_param("mem", early_mem);
211 static void __init setup_bootmem(void)
213 phys_addr_t vmlinux_end = __pa_symbol(&_end);
214 phys_addr_t max_mapped_addr;
215 phys_addr_t phys_ram_end, vmlinux_start;
217 if (IS_ENABLED(CONFIG_XIP_KERNEL))
218 vmlinux_start = __pa_symbol(&_sdata);
220 vmlinux_start = __pa_symbol(&_start);
222 memblock_enforce_memory_limit(memory_limit);
225 * Make sure we align the reservation on PMD_SIZE since we will
226 * map the kernel in the linear mapping as read-only: we do not want
227 * any allocation to happen between _end and the next pmd aligned page.
229 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
230 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
232 * Reserve from the start of the kernel to the end of the kernel
234 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
236 phys_ram_end = memblock_end_of_DRAM();
239 * Make sure we align the start of the memory on a PMD boundary so that
240 * at worst, we map the linear mapping with PMD mappings.
242 if (!IS_ENABLED(CONFIG_XIP_KERNEL))
243 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK;
246 * In 64-bit, any use of __va/__pa before this point is wrong as we
247 * did not know the start of DRAM before.
249 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU))
250 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base;
253 * memblock allocator is not aware of the fact that last 4K bytes of
254 * the addressable memory can not be mapped because of IS_ERR_VALUE
255 * macro. Make sure that last 4k bytes are not usable by memblock
256 * if end of dram is equal to maximum addressable memory. For 64-bit
257 * kernel, this problem can't happen here as the end of the virtual
258 * address space is occupied by the kernel mapping then this check must
259 * be done as soon as the kernel mapping base address is determined.
261 if (!IS_ENABLED(CONFIG_64BIT)) {
262 max_mapped_addr = __pa(~(ulong)0);
263 if (max_mapped_addr == (phys_ram_end - 1))
264 memblock_set_current_limit(max_mapped_addr - 4096);
267 min_low_pfn = PFN_UP(phys_ram_base);
268 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
269 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
271 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
272 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
274 reserve_initrd_mem();
277 * No allocation should be done before reserving the memory as defined
278 * in the device tree, otherwise the allocation could end up in a
281 early_init_fdt_scan_reserved_mem();
284 * If DTB is built in, no need to reserve its memblock.
285 * Otherwise, do reserve it but avoid using
286 * early_init_fdt_reserve_self() since __pa() does
287 * not work for DTB pointers that are fixmap addresses
289 if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
290 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
292 dma_contiguous_reserve(dma32_phys_limit);
293 if (IS_ENABLED(CONFIG_64BIT))
294 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
298 struct pt_alloc_ops pt_ops __initdata;
300 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
301 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
302 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
304 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
306 #ifdef CONFIG_XIP_KERNEL
307 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
308 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
309 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
310 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
311 #endif /* CONFIG_XIP_KERNEL */
313 static const pgprot_t protection_map[16] = {
314 [VM_NONE] = PAGE_NONE,
315 [VM_READ] = PAGE_READ,
316 [VM_WRITE] = PAGE_COPY,
317 [VM_WRITE | VM_READ] = PAGE_COPY,
318 [VM_EXEC] = PAGE_EXEC,
319 [VM_EXEC | VM_READ] = PAGE_READ_EXEC,
320 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC,
321 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC,
322 [VM_SHARED] = PAGE_NONE,
323 [VM_SHARED | VM_READ] = PAGE_READ,
324 [VM_SHARED | VM_WRITE] = PAGE_SHARED,
325 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED,
326 [VM_SHARED | VM_EXEC] = PAGE_EXEC,
327 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC,
328 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC,
329 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC
331 DECLARE_VM_GET_PAGE_PROT
333 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
335 unsigned long addr = __fix_to_virt(idx);
338 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
340 ptep = &fixmap_pte[pte_index(addr)];
342 if (pgprot_val(prot))
343 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
345 pte_clear(&init_mm, addr, ptep);
346 local_flush_tlb_page(addr);
349 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
351 return (pte_t *)((uintptr_t)pa);
354 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
356 clear_fixmap(FIX_PTE);
357 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
360 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
362 return (pte_t *) __va(pa);
365 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
368 * We only create PMD or PGD early mappings so we
369 * should never reach here with MMU disabled.
374 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
376 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
379 static phys_addr_t __init alloc_pte_late(uintptr_t va)
381 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
383 BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc));
384 return __pa((pte_t *)ptdesc_address(ptdesc));
387 static void __init create_pte_mapping(pte_t *ptep,
388 uintptr_t va, phys_addr_t pa,
389 phys_addr_t sz, pgprot_t prot)
391 uintptr_t pte_idx = pte_index(va);
393 BUG_ON(sz != PAGE_SIZE);
395 if (pte_none(ptep[pte_idx]))
396 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
399 #ifndef __PAGETABLE_PMD_FOLDED
401 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
402 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
403 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
405 #ifdef CONFIG_XIP_KERNEL
406 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
407 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
408 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
409 #endif /* CONFIG_XIP_KERNEL */
411 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
412 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
413 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
415 #ifdef CONFIG_XIP_KERNEL
416 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
417 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d))
418 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d))
419 #endif /* CONFIG_XIP_KERNEL */
421 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
422 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
423 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
425 #ifdef CONFIG_XIP_KERNEL
426 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
427 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
428 #define early_pud ((pud_t *)XIP_FIXUP(early_pud))
429 #endif /* CONFIG_XIP_KERNEL */
431 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
433 /* Before MMU is enabled */
434 return (pmd_t *)((uintptr_t)pa);
437 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
439 clear_fixmap(FIX_PMD);
440 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
443 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
445 return (pmd_t *) __va(pa);
448 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
450 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
452 return (uintptr_t)early_pmd;
455 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
457 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
460 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
462 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
464 BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc));
465 return __pa((pmd_t *)ptdesc_address(ptdesc));
468 static void __init create_pmd_mapping(pmd_t *pmdp,
469 uintptr_t va, phys_addr_t pa,
470 phys_addr_t sz, pgprot_t prot)
473 phys_addr_t pte_phys;
474 uintptr_t pmd_idx = pmd_index(va);
476 if (sz == PMD_SIZE) {
477 if (pmd_none(pmdp[pmd_idx]))
478 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
482 if (pmd_none(pmdp[pmd_idx])) {
483 pte_phys = pt_ops.alloc_pte(va);
484 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
485 ptep = pt_ops.get_pte_virt(pte_phys);
486 memset(ptep, 0, PAGE_SIZE);
488 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
489 ptep = pt_ops.get_pte_virt(pte_phys);
492 create_pte_mapping(ptep, va, pa, sz, prot);
495 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
497 return (pud_t *)((uintptr_t)pa);
500 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
502 clear_fixmap(FIX_PUD);
503 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
506 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
508 return (pud_t *)__va(pa);
511 static phys_addr_t __init alloc_pud_early(uintptr_t va)
513 /* Only one PUD is available for early mapping */
514 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
516 return (uintptr_t)early_pud;
519 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
521 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
524 static phys_addr_t alloc_pud_late(uintptr_t va)
528 vaddr = __get_free_page(GFP_KERNEL);
533 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
535 return (p4d_t *)((uintptr_t)pa);
538 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
540 clear_fixmap(FIX_P4D);
541 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
544 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
546 return (p4d_t *)__va(pa);
549 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
551 /* Only one P4D is available for early mapping */
552 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
554 return (uintptr_t)early_p4d;
557 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
559 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
562 static phys_addr_t alloc_p4d_late(uintptr_t va)
566 vaddr = __get_free_page(GFP_KERNEL);
571 static void __init create_pud_mapping(pud_t *pudp,
572 uintptr_t va, phys_addr_t pa,
573 phys_addr_t sz, pgprot_t prot)
576 phys_addr_t next_phys;
577 uintptr_t pud_index = pud_index(va);
579 if (sz == PUD_SIZE) {
580 if (pud_val(pudp[pud_index]) == 0)
581 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
585 if (pud_val(pudp[pud_index]) == 0) {
586 next_phys = pt_ops.alloc_pmd(va);
587 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
588 nextp = pt_ops.get_pmd_virt(next_phys);
589 memset(nextp, 0, PAGE_SIZE);
591 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
592 nextp = pt_ops.get_pmd_virt(next_phys);
595 create_pmd_mapping(nextp, va, pa, sz, prot);
598 static void __init create_p4d_mapping(p4d_t *p4dp,
599 uintptr_t va, phys_addr_t pa,
600 phys_addr_t sz, pgprot_t prot)
603 phys_addr_t next_phys;
604 uintptr_t p4d_index = p4d_index(va);
606 if (sz == P4D_SIZE) {
607 if (p4d_val(p4dp[p4d_index]) == 0)
608 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
612 if (p4d_val(p4dp[p4d_index]) == 0) {
613 next_phys = pt_ops.alloc_pud(va);
614 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
615 nextp = pt_ops.get_pud_virt(next_phys);
616 memset(nextp, 0, PAGE_SIZE);
618 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
619 nextp = pt_ops.get_pud_virt(next_phys);
622 create_pud_mapping(nextp, va, pa, sz, prot);
625 #define pgd_next_t p4d_t
626 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \
627 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \
628 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
629 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \
630 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \
631 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
632 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
633 (pgtable_l5_enabled ? \
634 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
635 (pgtable_l4_enabled ? \
636 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \
637 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
638 #define fixmap_pgd_next (pgtable_l5_enabled ? \
639 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \
640 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
641 #define trampoline_pgd_next (pgtable_l5_enabled ? \
642 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \
643 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
645 #define pgd_next_t pte_t
646 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
647 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
648 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
649 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
650 #define fixmap_pgd_next ((uintptr_t)fixmap_pte)
651 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
652 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
653 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
654 #endif /* __PAGETABLE_PMD_FOLDED */
656 void __init create_pgd_mapping(pgd_t *pgdp,
657 uintptr_t va, phys_addr_t pa,
658 phys_addr_t sz, pgprot_t prot)
661 phys_addr_t next_phys;
662 uintptr_t pgd_idx = pgd_index(va);
664 if (sz == PGDIR_SIZE) {
665 if (pgd_val(pgdp[pgd_idx]) == 0)
666 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
670 if (pgd_val(pgdp[pgd_idx]) == 0) {
671 next_phys = alloc_pgd_next(va);
672 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
673 nextp = get_pgd_next_virt(next_phys);
674 memset(nextp, 0, PAGE_SIZE);
676 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
677 nextp = get_pgd_next_virt(next_phys);
680 create_pgd_next_mapping(nextp, va, pa, sz, prot);
683 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va,
686 if (debug_pagealloc_enabled())
689 if (pgtable_l5_enabled &&
690 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
693 if (pgtable_l4_enabled &&
694 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
697 if (IS_ENABLED(CONFIG_64BIT) &&
698 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
704 #ifdef CONFIG_XIP_KERNEL
705 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
706 extern char _xiprom[], _exiprom[], __data_loc;
708 /* called from head.S with MMU off */
709 asmlinkage void __init __copy_data(void)
711 void *from = (void *)(&__data_loc);
712 void *to = (void *)CONFIG_PHYS_RAM_BASE;
713 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
715 memcpy(to, from, sz);
719 #ifdef CONFIG_STRICT_KERNEL_RWX
720 static __init pgprot_t pgprot_from_va(uintptr_t va)
722 if (is_va_kernel_text(va))
723 return PAGE_KERNEL_READ_EXEC;
726 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
727 * we must protect its linear mapping alias from being executed and
729 * And rodata section is marked readonly in mark_rodata_ro.
731 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
732 return PAGE_KERNEL_READ;
737 void mark_rodata_ro(void)
739 set_kernel_memory(__start_rodata, _data, set_memory_ro);
740 if (IS_ENABLED(CONFIG_64BIT))
741 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
745 static __init pgprot_t pgprot_from_va(uintptr_t va)
747 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
750 return PAGE_KERNEL_EXEC;
752 #endif /* CONFIG_STRICT_KERNEL_RWX */
754 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
755 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
757 static void __init disable_pgtable_l5(void)
759 pgtable_l5_enabled = false;
760 kernel_map.page_offset = PAGE_OFFSET_L4;
761 satp_mode = SATP_MODE_48;
764 static void __init disable_pgtable_l4(void)
766 pgtable_l4_enabled = false;
767 kernel_map.page_offset = PAGE_OFFSET_L3;
768 satp_mode = SATP_MODE_39;
771 static int __init print_no4lvl(char *p)
773 pr_info("Disabled 4-level and 5-level paging");
776 early_param("no4lvl", print_no4lvl);
778 static int __init print_no5lvl(char *p)
780 pr_info("Disabled 5-level paging");
783 early_param("no5lvl", print_no5lvl);
785 static void __init set_mmap_rnd_bits_max(void)
787 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3;
791 * There is a simple way to determine if 4-level is supported by the
792 * underlying hardware: establish 1:1 mapping in 4-level page table mode
793 * then read SATP to see if the configuration was taken into account
794 * meaning sv48 is supported.
796 static __init void set_satp_mode(uintptr_t dtb_pa)
798 u64 identity_satp, hw_satp;
799 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
800 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
802 if (satp_mode_cmdline == SATP_MODE_57) {
803 disable_pgtable_l5();
804 } else if (satp_mode_cmdline == SATP_MODE_48) {
805 disable_pgtable_l5();
806 disable_pgtable_l4();
810 create_p4d_mapping(early_p4d,
811 set_satp_mode_pmd, (uintptr_t)early_pud,
812 P4D_SIZE, PAGE_TABLE);
813 create_pud_mapping(early_pud,
814 set_satp_mode_pmd, (uintptr_t)early_pmd,
815 PUD_SIZE, PAGE_TABLE);
816 /* Handle the case where set_satp_mode straddles 2 PMDs */
817 create_pmd_mapping(early_pmd,
818 set_satp_mode_pmd, set_satp_mode_pmd,
819 PMD_SIZE, PAGE_KERNEL_EXEC);
820 create_pmd_mapping(early_pmd,
821 set_satp_mode_pmd + PMD_SIZE,
822 set_satp_mode_pmd + PMD_SIZE,
823 PMD_SIZE, PAGE_KERNEL_EXEC);
825 create_pgd_mapping(early_pg_dir,
828 (uintptr_t)early_p4d : (uintptr_t)early_pud,
829 PGDIR_SIZE, PAGE_TABLE);
831 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
833 local_flush_tlb_all();
834 csr_write(CSR_SATP, identity_satp);
835 hw_satp = csr_swap(CSR_SATP, 0ULL);
836 local_flush_tlb_all();
838 if (hw_satp != identity_satp) {
839 if (pgtable_l5_enabled) {
840 disable_pgtable_l5();
841 memset(early_pg_dir, 0, PAGE_SIZE);
844 disable_pgtable_l4();
847 memset(early_pg_dir, 0, PAGE_SIZE);
848 memset(early_p4d, 0, PAGE_SIZE);
849 memset(early_pud, 0, PAGE_SIZE);
850 memset(early_pmd, 0, PAGE_SIZE);
855 * setup_vm() is called from head.S with MMU-off.
857 * Following requirements should be honoured for setup_vm() to work
859 * 1) It should use PC-relative addressing for accessing kernel symbols.
860 * To achieve this we always use GCC cmodel=medany.
861 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
862 * so disable compiler instrumentation when FTRACE is enabled.
864 * Currently, the above requirements are honoured by using custom CFLAGS
865 * for init.o in mm/Makefile.
868 #ifndef __riscv_cmodel_medany
869 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
872 #ifdef CONFIG_RELOCATABLE
873 extern unsigned long __rela_dyn_start, __rela_dyn_end;
875 static void __init relocate_kernel(void)
877 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
879 * This holds the offset between the linked virtual address and the
880 * relocated virtual address.
882 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
884 * This holds the offset between kernel linked virtual address and
887 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
889 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
890 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
891 Elf64_Addr relocated_addr = rela->r_addend;
893 if (rela->r_info != R_RISCV_RELATIVE)
897 * Make sure to not relocate vdso symbols like rt_sigreturn
898 * which are linked from the address 0 in vmlinux since
899 * vdso symbol addresses are actually used as an offset from
900 * mm->context.vdso in VDSO_OFFSET macro.
902 if (relocated_addr >= KERNEL_LINK_ADDR)
903 relocated_addr += reloc_offset;
905 *(Elf64_Addr *)addr = relocated_addr;
908 #endif /* CONFIG_RELOCATABLE */
910 #ifdef CONFIG_XIP_KERNEL
911 static void __init create_kernel_page_table(pgd_t *pgdir,
912 __always_unused bool early)
914 uintptr_t va, end_va;
916 /* Map the flash resident part */
917 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
918 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
919 create_pgd_mapping(pgdir, va,
920 kernel_map.xiprom + (va - kernel_map.virt_addr),
921 PMD_SIZE, PAGE_KERNEL_EXEC);
923 /* Map the data in RAM */
924 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
925 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
926 create_pgd_mapping(pgdir, va,
927 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
928 PMD_SIZE, PAGE_KERNEL);
931 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
933 uintptr_t va, end_va;
935 end_va = kernel_map.virt_addr + kernel_map.size;
936 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
937 create_pgd_mapping(pgdir, va,
938 kernel_map.phys_addr + (va - kernel_map.virt_addr),
941 PAGE_KERNEL_EXEC : pgprot_from_va(va));
946 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
947 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
950 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
953 #ifndef CONFIG_BUILTIN_DTB
954 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
956 /* Make sure the fdt fixmap address is always aligned on PMD size */
957 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
959 /* In 32-bit only, the fdt lies in its own PGD */
960 if (!IS_ENABLED(CONFIG_64BIT)) {
961 create_pgd_mapping(early_pg_dir, fix_fdt_va,
962 pa, MAX_FDT_SIZE, PAGE_KERNEL);
964 create_pmd_mapping(fixmap_pmd, fix_fdt_va,
965 pa, PMD_SIZE, PAGE_KERNEL);
966 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
967 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
970 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
973 * For 64-bit kernel, __va can't be used since it would return a linear
974 * mapping address whereas dtb_early_va will be used before
975 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
976 * kernel is mapped in the linear mapping, that makes no difference.
978 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa);
981 dtb_early_pa = dtb_pa;
985 * MMU is not enabled, the page tables are allocated directly using
986 * early_pmd/pud/p4d and the address returned is the physical one.
988 static void __init pt_ops_set_early(void)
990 pt_ops.alloc_pte = alloc_pte_early;
991 pt_ops.get_pte_virt = get_pte_virt_early;
992 #ifndef __PAGETABLE_PMD_FOLDED
993 pt_ops.alloc_pmd = alloc_pmd_early;
994 pt_ops.get_pmd_virt = get_pmd_virt_early;
995 pt_ops.alloc_pud = alloc_pud_early;
996 pt_ops.get_pud_virt = get_pud_virt_early;
997 pt_ops.alloc_p4d = alloc_p4d_early;
998 pt_ops.get_p4d_virt = get_p4d_virt_early;
1003 * MMU is enabled but page table setup is not complete yet.
1004 * fixmap page table alloc functions must be used as a means to temporarily
1005 * map the allocated physical pages since the linear mapping does not exist yet.
1007 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
1008 * but it will be used as described above.
1010 static void __init pt_ops_set_fixmap(void)
1012 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
1013 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
1014 #ifndef __PAGETABLE_PMD_FOLDED
1015 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
1016 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
1017 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
1018 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
1019 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
1020 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
1025 * MMU is enabled and page table setup is complete, so from now, we can use
1026 * generic page allocation functions to setup page table.
1028 static void __init pt_ops_set_late(void)
1030 pt_ops.alloc_pte = alloc_pte_late;
1031 pt_ops.get_pte_virt = get_pte_virt_late;
1032 #ifndef __PAGETABLE_PMD_FOLDED
1033 pt_ops.alloc_pmd = alloc_pmd_late;
1034 pt_ops.get_pmd_virt = get_pmd_virt_late;
1035 pt_ops.alloc_pud = alloc_pud_late;
1036 pt_ops.get_pud_virt = get_pud_virt_late;
1037 pt_ops.alloc_p4d = alloc_p4d_late;
1038 pt_ops.get_p4d_virt = get_p4d_virt_late;
1042 #ifdef CONFIG_RANDOMIZE_BASE
1043 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa);
1044 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa);
1046 static int __init print_nokaslr(char *p)
1048 pr_info("Disabled KASLR");
1051 early_param("nokaslr", print_nokaslr);
1053 unsigned long kaslr_offset(void)
1055 return kernel_map.virt_offset;
1059 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1061 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1063 #ifdef CONFIG_RANDOMIZE_BASE
1064 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) {
1065 u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa);
1066 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
1070 * Compute the number of positions available: we are limited
1071 * by the early page table that only has one PUD and we must
1072 * be aligned on PMD_SIZE.
1074 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE;
1076 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE;
1080 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset;
1082 #ifdef CONFIG_XIP_KERNEL
1084 kernel_map.page_offset = PAGE_OFFSET_L3;
1086 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1088 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1089 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1091 phys_ram_base = CONFIG_PHYS_RAM_BASE;
1092 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1093 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
1095 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1097 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1098 kernel_map.phys_addr = (uintptr_t)(&_start);
1099 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1102 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1103 set_satp_mode(dtb_pa);
1104 set_mmap_rnd_bits_max();
1108 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1109 * where we have the system memory layout: this allows us to align
1110 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1111 * for the linear mapping. This is only possible because the kernel
1112 * mapping lies outside the linear mapping.
1113 * In 32-bit however, as the kernel resides in the linear mapping,
1114 * setup_vm_final can not change the mapping established here,
1115 * otherwise the same kernel addresses would get mapped to different
1116 * physical addresses (if the start of dram is different from the
1117 * kernel physical address start).
1119 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1120 0UL : PAGE_OFFSET - kernel_map.phys_addr;
1121 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1124 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
1125 * kernel, whereas for 64-bit kernel, the end of the virtual address
1126 * space is occupied by the modules/BPF/kernel mappings which reduces
1127 * the available size of the linear mapping.
1129 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
1131 /* Sanity check alignment and size */
1132 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1133 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1137 * The last 4K bytes of the addressable memory can not be mapped because
1138 * of IS_ERR_VALUE macro.
1140 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1143 #ifdef CONFIG_RELOCATABLE
1145 * Early page table uses only one PUD, which makes it possible
1146 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1147 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1148 * since a part of the kernel would not get mapped.
1150 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1154 apply_early_boot_alternatives();
1157 /* Setup early PGD for fixmap */
1158 create_pgd_mapping(early_pg_dir, FIXADDR_START,
1159 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1161 #ifndef __PAGETABLE_PMD_FOLDED
1162 /* Setup fixmap P4D and PUD */
1163 if (pgtable_l5_enabled)
1164 create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1165 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1166 /* Setup fixmap PUD and PMD */
1167 if (pgtable_l4_enabled)
1168 create_pud_mapping(fixmap_pud, FIXADDR_START,
1169 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1170 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1171 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1172 /* Setup trampoline PGD and PMD */
1173 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1174 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1175 if (pgtable_l5_enabled)
1176 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1177 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1178 if (pgtable_l4_enabled)
1179 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1180 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1181 #ifdef CONFIG_XIP_KERNEL
1182 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1183 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1185 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1186 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1189 /* Setup trampoline PGD */
1190 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1191 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1195 * Setup early PGD covering entire kernel which will allow
1196 * us to reach paging_init(). We map all memory banks later
1197 * in setup_vm_final() below.
1199 create_kernel_page_table(early_pg_dir, true);
1201 /* Setup early mapping for FDT early scan */
1202 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1205 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1206 * range can not span multiple pmds.
1208 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1209 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1211 #ifndef __PAGETABLE_PMD_FOLDED
1213 * Early ioremap fixmap is already created as it lies within first 2MB
1214 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1215 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1218 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1219 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1220 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1222 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1223 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1224 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1225 fix_to_virt(FIX_BTMAP_BEGIN));
1226 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1227 fix_to_virt(FIX_BTMAP_END));
1229 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1230 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1234 pt_ops_set_fixmap();
1237 static void __init create_linear_mapping_range(phys_addr_t start,
1239 uintptr_t fixed_map_size)
1242 uintptr_t va, map_size;
1244 for (pa = start; pa < end; pa += map_size) {
1245 va = (uintptr_t)__va(pa);
1246 map_size = fixed_map_size ? fixed_map_size :
1247 best_map_size(pa, va, end - pa);
1249 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1250 pgprot_from_va(va));
1254 static void __init create_linear_mapping_page_table(void)
1256 phys_addr_t start, end;
1257 phys_addr_t kfence_pool __maybe_unused;
1260 #ifdef CONFIG_STRICT_KERNEL_RWX
1261 phys_addr_t ktext_start = __pa_symbol(_start);
1262 phys_addr_t ktext_size = __init_data_begin - _start;
1263 phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1264 phys_addr_t krodata_size = _data - __start_rodata;
1266 /* Isolate kernel text and rodata so they don't get mapped with a PUD */
1267 memblock_mark_nomap(ktext_start, ktext_size);
1268 memblock_mark_nomap(krodata_start, krodata_size);
1271 #ifdef CONFIG_KFENCE
1273 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1274 * before we setup the linear mapping so that we avoid using hugepages
1277 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1278 BUG_ON(!kfence_pool);
1280 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1281 __kfence_pool = __va(kfence_pool);
1284 /* Map all memory banks in the linear mapping */
1285 for_each_mem_range(i, &start, &end) {
1288 if (start <= __pa(PAGE_OFFSET) &&
1289 __pa(PAGE_OFFSET) < end)
1290 start = __pa(PAGE_OFFSET);
1291 if (end >= __pa(PAGE_OFFSET) + memory_limit)
1292 end = __pa(PAGE_OFFSET) + memory_limit;
1294 create_linear_mapping_range(start, end, 0);
1297 #ifdef CONFIG_STRICT_KERNEL_RWX
1298 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0);
1299 create_linear_mapping_range(krodata_start,
1300 krodata_start + krodata_size, 0);
1302 memblock_clear_nomap(ktext_start, ktext_size);
1303 memblock_clear_nomap(krodata_start, krodata_size);
1306 #ifdef CONFIG_KFENCE
1307 create_linear_mapping_range(kfence_pool,
1308 kfence_pool + KFENCE_POOL_SIZE,
1311 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1315 static void __init setup_vm_final(void)
1317 /* Setup swapper PGD for fixmap */
1318 #if !defined(CONFIG_64BIT)
1320 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1321 * directly in swapper_pg_dir in addition to the pgd entry that points
1324 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1326 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1328 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1329 __pa_symbol(fixmap_pgd_next),
1330 PGDIR_SIZE, PAGE_TABLE);
1332 /* Map the linear mapping */
1333 create_linear_mapping_page_table();
1335 /* Map the kernel */
1336 if (IS_ENABLED(CONFIG_64BIT))
1337 create_kernel_page_table(swapper_pg_dir, false);
1340 kasan_swapper_init();
1343 /* Clear fixmap PTE and PMD mappings */
1344 clear_fixmap(FIX_PTE);
1345 clear_fixmap(FIX_PMD);
1346 clear_fixmap(FIX_PUD);
1347 clear_fixmap(FIX_P4D);
1349 /* Move to swapper page table */
1350 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1351 local_flush_tlb_all();
1356 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1358 dtb_early_va = (void *)dtb_pa;
1359 dtb_early_pa = dtb_pa;
1362 static inline void setup_vm_final(void)
1365 #endif /* CONFIG_MMU */
1368 * reserve_crashkernel() - reserves memory for crash kernel
1370 * This function reserves memory area given in "crashkernel=" kernel command
1371 * line parameter. The memory reserved is used by dump capture kernel when
1372 * primary kernel is crashing.
1374 static void __init arch_reserve_crashkernel(void)
1376 unsigned long long low_size = 0;
1377 unsigned long long crash_base, crash_size;
1378 char *cmdline = boot_command_line;
1382 if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
1385 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(),
1386 &crash_size, &crash_base,
1391 reserve_crashkernel_generic(cmdline, crash_size, crash_base,
1395 void __init paging_init(void)
1400 /* Depend on that Linear Mapping is ready */
1401 memblock_allow_resize();
1404 void __init misc_mem_init(void)
1406 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1409 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1410 /* The entire VMEMMAP region has been populated. Flush TLB for this region */
1411 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END);
1414 arch_reserve_crashkernel();
1415 memblock_dump_all();
1418 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1419 void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node,
1420 unsigned long addr, unsigned long next)
1422 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL);
1425 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node,
1426 unsigned long addr, unsigned long next)
1428 vmemmap_verify((pte_t *)pmdp, node, addr, next);
1432 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1433 struct vmem_altmap *altmap)
1436 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we
1437 * can't use hugepage mappings for 2-level page table because in case of
1438 * memory hotplug, we are not able to update all the page tables with
1441 return vmemmap_populate_hugepages(start, end, node, NULL);
1445 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1447 * Pre-allocates page-table pages for a specific area in the kernel
1448 * page-table. Only the level which needs to be synchronized between
1449 * all page-tables is allocated because the synchronization can be
1452 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1458 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1459 pgd_t *pgd = pgd_offset_k(addr);
1465 p4d = p4d_alloc(&init_mm, pgd, addr);
1469 if (pgtable_l5_enabled)
1473 pud = pud_alloc(&init_mm, p4d, addr);
1477 if (pgtable_l4_enabled)
1481 pmd = pmd_alloc(&init_mm, pud, addr);
1489 * The pages have to be there now or they will be missing in
1490 * process page-tables later.
1492 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1495 void __init pgtable_cache_init(void)
1497 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1498 if (IS_ENABLED(CONFIG_MODULES))
1499 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
1503 #ifdef CONFIG_EXECMEM
1505 static struct execmem_info execmem_info __ro_after_init;
1507 struct execmem_info __init *execmem_arch_setup(void)
1509 execmem_info = (struct execmem_info){
1511 [EXECMEM_DEFAULT] = {
1512 .start = MODULES_VADDR,
1514 .pgprot = PAGE_KERNEL,
1517 [EXECMEM_KPROBES] = {
1518 .start = VMALLOC_START,
1520 .pgprot = PAGE_KERNEL_READ_EXEC,
1524 .start = BPF_JIT_REGION_START,
1525 .end = BPF_JIT_REGION_END,
1526 .pgprot = PAGE_KERNEL,
1527 .alignment = PAGE_SIZE,
1532 return &execmem_info;
1534 #endif /* CONFIG_MMU */
1535 #endif /* CONFIG_EXECMEM */