1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2012 Regents of the University of California
4 * Copyright (C) 2019 Western Digital Corporation or its affiliates.
5 * Copyright (C) 2020 FORTH-ICS/CARV
6 * Nick Kossifidis <mick@ics.forth.gr>
9 #include <linux/init.h>
11 #include <linux/memblock.h>
12 #include <linux/initrd.h>
13 #include <linux/swap.h>
14 #include <linux/swiotlb.h>
15 #include <linux/sizes.h>
16 #include <linux/of_fdt.h>
17 #include <linux/of_reserved_mem.h>
18 #include <linux/libfdt.h>
19 #include <linux/set_memory.h>
20 #include <linux/dma-map-ops.h>
21 #include <linux/crash_dump.h>
22 #include <linux/hugetlb.h>
24 #include <asm/fixmap.h>
25 #include <asm/tlbflush.h>
26 #include <asm/sections.h>
29 #include <asm/ptdump.h>
32 #include "../kernel/head.h"
34 struct kernel_mapping kernel_map __ro_after_init;
35 EXPORT_SYMBOL(kernel_map);
36 #ifdef CONFIG_XIP_KERNEL
37 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
41 u64 satp_mode = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_48 : SATP_MODE_39;
43 u64 satp_mode = SATP_MODE_32;
45 EXPORT_SYMBOL(satp_mode);
47 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
48 EXPORT_SYMBOL(pgtable_l4_enabled);
50 phys_addr_t phys_ram_base __ro_after_init;
51 EXPORT_SYMBOL(phys_ram_base);
53 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
55 EXPORT_SYMBOL(empty_zero_page);
58 #define DTB_EARLY_BASE_VA PGDIR_SIZE
59 void *_dtb_early_va __initdata;
60 uintptr_t _dtb_early_pa __initdata;
62 static phys_addr_t dma32_phys_limit __initdata;
64 static void __init zone_sizes_init(void)
66 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
68 #ifdef CONFIG_ZONE_DMA32
69 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
71 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
73 free_area_init(max_zone_pfns);
76 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
77 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
79 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
83 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
85 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
89 static void __init print_vm_layout(void)
91 pr_notice("Virtual kernel memory layout:\n");
92 print_mlk("fixmap", (unsigned long)FIXADDR_START,
93 (unsigned long)FIXADDR_TOP);
94 print_mlm("pci io", (unsigned long)PCI_IO_START,
95 (unsigned long)PCI_IO_END);
96 print_mlm("vmemmap", (unsigned long)VMEMMAP_START,
97 (unsigned long)VMEMMAP_END);
98 print_mlm("vmalloc", (unsigned long)VMALLOC_START,
99 (unsigned long)VMALLOC_END);
100 print_mlm("lowmem", (unsigned long)PAGE_OFFSET,
101 (unsigned long)high_memory);
102 if (IS_ENABLED(CONFIG_64BIT)) {
104 print_mlm("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
107 print_mlm("kernel", (unsigned long)KERNEL_LINK_ADDR,
108 (unsigned long)ADDRESS_SPACE_END);
112 static void print_vm_layout(void) { }
113 #endif /* CONFIG_DEBUG_VM */
115 void __init mem_init(void)
117 #ifdef CONFIG_FLATMEM
119 #endif /* CONFIG_FLATMEM */
121 #ifdef CONFIG_SWIOTLB
122 if (swiotlb_force == SWIOTLB_FORCE ||
123 max_pfn > PFN_DOWN(dma32_phys_limit))
126 swiotlb_force = SWIOTLB_NO_FORCE;
133 /* Limit the memory size via mem. */
134 static phys_addr_t memory_limit;
136 static int __init early_mem(char *p)
143 size = memparse(p, &p) & PAGE_MASK;
144 memory_limit = min_t(u64, size, memory_limit);
146 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
150 early_param("mem", early_mem);
152 static void __init setup_bootmem(void)
154 phys_addr_t vmlinux_end = __pa_symbol(&_end);
155 phys_addr_t max_mapped_addr;
156 phys_addr_t phys_ram_end, vmlinux_start;
158 if (IS_ENABLED(CONFIG_XIP_KERNEL))
159 vmlinux_start = __pa_symbol(&_sdata);
161 vmlinux_start = __pa_symbol(&_start);
163 memblock_enforce_memory_limit(memory_limit);
166 * Make sure we align the reservation on PMD_SIZE since we will
167 * map the kernel in the linear mapping as read-only: we do not want
168 * any allocation to happen between _end and the next pmd aligned page.
170 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
171 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
173 * Reserve from the start of the kernel to the end of the kernel
175 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
177 phys_ram_end = memblock_end_of_DRAM();
178 if (!IS_ENABLED(CONFIG_XIP_KERNEL))
179 phys_ram_base = memblock_start_of_DRAM();
181 * memblock allocator is not aware of the fact that last 4K bytes of
182 * the addressable memory can not be mapped because of IS_ERR_VALUE
183 * macro. Make sure that last 4k bytes are not usable by memblock
184 * if end of dram is equal to maximum addressable memory. For 64-bit
185 * kernel, this problem can't happen here as the end of the virtual
186 * address space is occupied by the kernel mapping then this check must
187 * be done as soon as the kernel mapping base address is determined.
189 if (!IS_ENABLED(CONFIG_64BIT)) {
190 max_mapped_addr = __pa(~(ulong)0);
191 if (max_mapped_addr == (phys_ram_end - 1))
192 memblock_set_current_limit(max_mapped_addr - 4096);
195 min_low_pfn = PFN_UP(phys_ram_base);
196 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
197 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
199 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
200 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
202 reserve_initrd_mem();
204 * If DTB is built in, no need to reserve its memblock.
205 * Otherwise, do reserve it but avoid using
206 * early_init_fdt_reserve_self() since __pa() does
207 * not work for DTB pointers that are fixmap addresses
209 if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
210 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
212 early_init_fdt_scan_reserved_mem();
213 dma_contiguous_reserve(dma32_phys_limit);
214 if (IS_ENABLED(CONFIG_64BIT))
215 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
216 memblock_allow_resize();
220 struct pt_alloc_ops pt_ops __initdata;
222 unsigned long riscv_pfn_base __ro_after_init;
223 EXPORT_SYMBOL(riscv_pfn_base);
225 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
226 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
227 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
229 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
230 static pud_t __maybe_unused early_dtb_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
231 static pmd_t __maybe_unused early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
233 #ifdef CONFIG_XIP_KERNEL
234 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
235 #define riscv_pfn_base (*(unsigned long *)XIP_FIXUP(&riscv_pfn_base))
236 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
237 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
238 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
239 #endif /* CONFIG_XIP_KERNEL */
241 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
243 unsigned long addr = __fix_to_virt(idx);
246 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
248 ptep = &fixmap_pte[pte_index(addr)];
250 if (pgprot_val(prot))
251 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
253 pte_clear(&init_mm, addr, ptep);
254 local_flush_tlb_page(addr);
257 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
259 return (pte_t *)((uintptr_t)pa);
262 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
264 clear_fixmap(FIX_PTE);
265 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
268 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
270 return (pte_t *) __va(pa);
273 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
276 * We only create PMD or PGD early mappings so we
277 * should never reach here with MMU disabled.
282 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
284 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
287 static phys_addr_t __init alloc_pte_late(uintptr_t va)
291 vaddr = __get_free_page(GFP_KERNEL);
292 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)));
297 static void __init create_pte_mapping(pte_t *ptep,
298 uintptr_t va, phys_addr_t pa,
299 phys_addr_t sz, pgprot_t prot)
301 uintptr_t pte_idx = pte_index(va);
303 BUG_ON(sz != PAGE_SIZE);
305 if (pte_none(ptep[pte_idx]))
306 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
309 #ifndef __PAGETABLE_PMD_FOLDED
311 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
312 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
313 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
315 #ifdef CONFIG_XIP_KERNEL
316 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
317 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
318 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
319 #endif /* CONFIG_XIP_KERNEL */
321 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
322 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
323 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
325 #ifdef CONFIG_XIP_KERNEL
326 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
327 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
328 #define early_pud ((pud_t *)XIP_FIXUP(early_pud))
329 #endif /* CONFIG_XIP_KERNEL */
331 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
333 /* Before MMU is enabled */
334 return (pmd_t *)((uintptr_t)pa);
337 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
339 clear_fixmap(FIX_PMD);
340 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
343 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
345 return (pmd_t *) __va(pa);
348 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
350 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
352 return (uintptr_t)early_pmd;
355 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
357 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
360 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
364 vaddr = __get_free_page(GFP_KERNEL);
365 BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page(vaddr)));
370 static void __init create_pmd_mapping(pmd_t *pmdp,
371 uintptr_t va, phys_addr_t pa,
372 phys_addr_t sz, pgprot_t prot)
375 phys_addr_t pte_phys;
376 uintptr_t pmd_idx = pmd_index(va);
378 if (sz == PMD_SIZE) {
379 if (pmd_none(pmdp[pmd_idx]))
380 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
384 if (pmd_none(pmdp[pmd_idx])) {
385 pte_phys = pt_ops.alloc_pte(va);
386 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
387 ptep = pt_ops.get_pte_virt(pte_phys);
388 memset(ptep, 0, PAGE_SIZE);
390 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
391 ptep = pt_ops.get_pte_virt(pte_phys);
394 create_pte_mapping(ptep, va, pa, sz, prot);
397 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
399 return (pud_t *)((uintptr_t)pa);
402 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
404 clear_fixmap(FIX_PUD);
405 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
408 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
410 return (pud_t *)__va(pa);
413 static phys_addr_t __init alloc_pud_early(uintptr_t va)
415 /* Only one PUD is available for early mapping */
416 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
418 return (uintptr_t)early_pud;
421 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
423 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
426 static phys_addr_t alloc_pud_late(uintptr_t va)
430 vaddr = __get_free_page(GFP_KERNEL);
435 static void __init create_pud_mapping(pud_t *pudp,
436 uintptr_t va, phys_addr_t pa,
437 phys_addr_t sz, pgprot_t prot)
440 phys_addr_t next_phys;
441 uintptr_t pud_index = pud_index(va);
443 if (sz == PUD_SIZE) {
444 if (pud_val(pudp[pud_index]) == 0)
445 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
449 if (pud_val(pudp[pud_index]) == 0) {
450 next_phys = pt_ops.alloc_pmd(va);
451 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
452 nextp = pt_ops.get_pmd_virt(next_phys);
453 memset(nextp, 0, PAGE_SIZE);
455 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
456 nextp = pt_ops.get_pmd_virt(next_phys);
459 create_pmd_mapping(nextp, va, pa, sz, prot);
462 #define pgd_next_t pud_t
463 #define alloc_pgd_next(__va) (pgtable_l4_enabled ? \
464 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))
465 #define get_pgd_next_virt(__pa) (pgtable_l4_enabled ? \
466 pt_ops.get_pud_virt(__pa) : (pgd_next_t *)pt_ops.get_pmd_virt(__pa))
467 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
468 (pgtable_l4_enabled ? \
469 create_pud_mapping(__nextp, __va, __pa, __sz, __prot) : \
470 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))
471 #define fixmap_pgd_next (pgtable_l4_enabled ? \
472 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)
473 #define trampoline_pgd_next (pgtable_l4_enabled ? \
474 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)
475 #define early_dtb_pgd_next (pgtable_l4_enabled ? \
476 (uintptr_t)early_dtb_pud : (uintptr_t)early_dtb_pmd)
478 #define pgd_next_t pte_t
479 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
480 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
481 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
482 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
483 #define fixmap_pgd_next ((uintptr_t)fixmap_pte)
484 #define early_dtb_pgd_next ((uintptr_t)early_dtb_pmd)
485 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot)
486 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot)
487 #endif /* __PAGETABLE_PMD_FOLDED */
489 void __init create_pgd_mapping(pgd_t *pgdp,
490 uintptr_t va, phys_addr_t pa,
491 phys_addr_t sz, pgprot_t prot)
494 phys_addr_t next_phys;
495 uintptr_t pgd_idx = pgd_index(va);
497 if (sz == PGDIR_SIZE) {
498 if (pgd_val(pgdp[pgd_idx]) == 0)
499 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
503 if (pgd_val(pgdp[pgd_idx]) == 0) {
504 next_phys = alloc_pgd_next(va);
505 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
506 nextp = get_pgd_next_virt(next_phys);
507 memset(nextp, 0, PAGE_SIZE);
509 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
510 nextp = get_pgd_next_virt(next_phys);
513 create_pgd_next_mapping(nextp, va, pa, sz, prot);
516 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
518 /* Upgrade to PMD_SIZE mappings whenever possible */
519 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
525 #ifdef CONFIG_XIP_KERNEL
526 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
527 extern char _xiprom[], _exiprom[], __data_loc;
529 /* called from head.S with MMU off */
530 asmlinkage void __init __copy_data(void)
532 void *from = (void *)(&__data_loc);
533 void *to = (void *)CONFIG_PHYS_RAM_BASE;
534 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
536 memcpy(to, from, sz);
540 #ifdef CONFIG_STRICT_KERNEL_RWX
541 static __init pgprot_t pgprot_from_va(uintptr_t va)
543 if (is_va_kernel_text(va))
544 return PAGE_KERNEL_READ_EXEC;
547 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
548 * we must protect its linear mapping alias from being executed and
550 * And rodata section is marked readonly in mark_rodata_ro.
552 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
553 return PAGE_KERNEL_READ;
558 void mark_rodata_ro(void)
560 set_kernel_memory(__start_rodata, _data, set_memory_ro);
561 if (IS_ENABLED(CONFIG_64BIT))
562 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
568 static __init pgprot_t pgprot_from_va(uintptr_t va)
570 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
573 return PAGE_KERNEL_EXEC;
575 #endif /* CONFIG_STRICT_KERNEL_RWX */
578 static void __init disable_pgtable_l4(void)
580 pgtable_l4_enabled = false;
581 kernel_map.page_offset = PAGE_OFFSET_L3;
582 satp_mode = SATP_MODE_39;
586 * There is a simple way to determine if 4-level is supported by the
587 * underlying hardware: establish 1:1 mapping in 4-level page table mode
588 * then read SATP to see if the configuration was taken into account
589 * meaning sv48 is supported.
591 static __init void set_satp_mode(void)
593 u64 identity_satp, hw_satp;
594 uintptr_t set_satp_mode_pmd;
596 set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
597 create_pgd_mapping(early_pg_dir,
598 set_satp_mode_pmd, (uintptr_t)early_pud,
599 PGDIR_SIZE, PAGE_TABLE);
600 create_pud_mapping(early_pud,
601 set_satp_mode_pmd, (uintptr_t)early_pmd,
602 PUD_SIZE, PAGE_TABLE);
603 /* Handle the case where set_satp_mode straddles 2 PMDs */
604 create_pmd_mapping(early_pmd,
605 set_satp_mode_pmd, set_satp_mode_pmd,
606 PMD_SIZE, PAGE_KERNEL_EXEC);
607 create_pmd_mapping(early_pmd,
608 set_satp_mode_pmd + PMD_SIZE,
609 set_satp_mode_pmd + PMD_SIZE,
610 PMD_SIZE, PAGE_KERNEL_EXEC);
612 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
614 local_flush_tlb_all();
615 csr_write(CSR_SATP, identity_satp);
616 hw_satp = csr_swap(CSR_SATP, 0ULL);
617 local_flush_tlb_all();
619 if (hw_satp != identity_satp)
620 disable_pgtable_l4();
622 memset(early_pg_dir, 0, PAGE_SIZE);
623 memset(early_pud, 0, PAGE_SIZE);
624 memset(early_pmd, 0, PAGE_SIZE);
629 * setup_vm() is called from head.S with MMU-off.
631 * Following requirements should be honoured for setup_vm() to work
633 * 1) It should use PC-relative addressing for accessing kernel symbols.
634 * To achieve this we always use GCC cmodel=medany.
635 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
636 * so disable compiler instrumentation when FTRACE is enabled.
638 * Currently, the above requirements are honoured by using custom CFLAGS
639 * for init.o in mm/Makefile.
642 #ifndef __riscv_cmodel_medany
643 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
646 #ifdef CONFIG_XIP_KERNEL
647 static void __init create_kernel_page_table(pgd_t *pgdir,
648 __always_unused bool early)
650 uintptr_t va, end_va;
652 /* Map the flash resident part */
653 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
654 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
655 create_pgd_mapping(pgdir, va,
656 kernel_map.xiprom + (va - kernel_map.virt_addr),
657 PMD_SIZE, PAGE_KERNEL_EXEC);
659 /* Map the data in RAM */
660 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
661 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
662 create_pgd_mapping(pgdir, va,
663 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
664 PMD_SIZE, PAGE_KERNEL);
667 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
669 uintptr_t va, end_va;
671 end_va = kernel_map.virt_addr + kernel_map.size;
672 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
673 create_pgd_mapping(pgdir, va,
674 kernel_map.phys_addr + (va - kernel_map.virt_addr),
677 PAGE_KERNEL_EXEC : pgprot_from_va(va));
682 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
683 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
686 static void __init create_fdt_early_page_table(pgd_t *pgdir, uintptr_t dtb_pa)
688 #ifndef CONFIG_BUILTIN_DTB
689 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
691 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
692 IS_ENABLED(CONFIG_64BIT) ? early_dtb_pgd_next : pa,
694 IS_ENABLED(CONFIG_64BIT) ? PAGE_TABLE : PAGE_KERNEL);
696 if (pgtable_l4_enabled) {
697 create_pud_mapping(early_dtb_pud, DTB_EARLY_BASE_VA,
698 (uintptr_t)early_dtb_pmd, PUD_SIZE, PAGE_TABLE);
701 if (IS_ENABLED(CONFIG_64BIT)) {
702 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA,
703 pa, PMD_SIZE, PAGE_KERNEL);
704 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE,
705 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
708 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1));
711 * For 64-bit kernel, __va can't be used since it would return a linear
712 * mapping address whereas dtb_early_va will be used before
713 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
714 * kernel is mapped in the linear mapping, that makes no difference.
716 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
719 dtb_early_pa = dtb_pa;
723 * MMU is not enabled, the page tables are allocated directly using
724 * early_pmd/pud/p4d and the address returned is the physical one.
726 void __init pt_ops_set_early(void)
728 pt_ops.alloc_pte = alloc_pte_early;
729 pt_ops.get_pte_virt = get_pte_virt_early;
730 #ifndef __PAGETABLE_PMD_FOLDED
731 pt_ops.alloc_pmd = alloc_pmd_early;
732 pt_ops.get_pmd_virt = get_pmd_virt_early;
733 pt_ops.alloc_pud = alloc_pud_early;
734 pt_ops.get_pud_virt = get_pud_virt_early;
739 * MMU is enabled but page table setup is not complete yet.
740 * fixmap page table alloc functions must be used as a means to temporarily
741 * map the allocated physical pages since the linear mapping does not exist yet.
743 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
744 * but it will be used as described above.
746 void __init pt_ops_set_fixmap(void)
748 pt_ops.alloc_pte = kernel_mapping_pa_to_va((uintptr_t)alloc_pte_fixmap);
749 pt_ops.get_pte_virt = kernel_mapping_pa_to_va((uintptr_t)get_pte_virt_fixmap);
750 #ifndef __PAGETABLE_PMD_FOLDED
751 pt_ops.alloc_pmd = kernel_mapping_pa_to_va((uintptr_t)alloc_pmd_fixmap);
752 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va((uintptr_t)get_pmd_virt_fixmap);
753 pt_ops.alloc_pud = kernel_mapping_pa_to_va((uintptr_t)alloc_pud_fixmap);
754 pt_ops.get_pud_virt = kernel_mapping_pa_to_va((uintptr_t)get_pud_virt_fixmap);
759 * MMU is enabled and page table setup is complete, so from now, we can use
760 * generic page allocation functions to setup page table.
762 void __init pt_ops_set_late(void)
764 pt_ops.alloc_pte = alloc_pte_late;
765 pt_ops.get_pte_virt = get_pte_virt_late;
766 #ifndef __PAGETABLE_PMD_FOLDED
767 pt_ops.alloc_pmd = alloc_pmd_late;
768 pt_ops.get_pmd_virt = get_pmd_virt_late;
769 pt_ops.alloc_pud = alloc_pud_late;
770 pt_ops.get_pud_virt = get_pud_virt_late;
774 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
776 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
778 kernel_map.virt_addr = KERNEL_LINK_ADDR;
779 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
781 #ifdef CONFIG_XIP_KERNEL
782 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
783 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
785 phys_ram_base = CONFIG_PHYS_RAM_BASE;
786 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
787 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
789 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
791 kernel_map.phys_addr = (uintptr_t)(&_start);
792 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
795 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
799 kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr;
800 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
802 riscv_pfn_base = PFN_DOWN(kernel_map.phys_addr);
805 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
806 * kernel, whereas for 64-bit kernel, the end of the virtual address
807 * space is occupied by the modules/BPF/kernel mappings which reduces
808 * the available size of the linear mapping.
810 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
812 /* Sanity check alignment and size */
813 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
814 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
818 * The last 4K bytes of the addressable memory can not be mapped because
819 * of IS_ERR_VALUE macro.
821 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
826 /* Setup early PGD for fixmap */
827 create_pgd_mapping(early_pg_dir, FIXADDR_START,
828 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
830 #ifndef __PAGETABLE_PMD_FOLDED
831 /* Setup fixmap PUD and PMD */
832 if (pgtable_l4_enabled)
833 create_pud_mapping(fixmap_pud, FIXADDR_START,
834 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
835 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
836 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
837 /* Setup trampoline PGD and PMD */
838 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
839 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
840 if (pgtable_l4_enabled)
841 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
842 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
843 #ifdef CONFIG_XIP_KERNEL
844 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
845 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
847 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
848 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
851 /* Setup trampoline PGD */
852 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
853 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
857 * Setup early PGD covering entire kernel which will allow
858 * us to reach paging_init(). We map all memory banks later
859 * in setup_vm_final() below.
861 create_kernel_page_table(early_pg_dir, true);
863 /* Setup early mapping for FDT early scan */
864 create_fdt_early_page_table(early_pg_dir, dtb_pa);
867 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
868 * range can not span multiple pmds.
870 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
871 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
873 #ifndef __PAGETABLE_PMD_FOLDED
875 * Early ioremap fixmap is already created as it lies within first 2MB
876 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
877 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
880 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
881 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
882 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
884 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
885 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
886 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
887 fix_to_virt(FIX_BTMAP_BEGIN));
888 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
889 fix_to_virt(FIX_BTMAP_END));
891 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
892 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
899 static void __init setup_vm_final(void)
901 uintptr_t va, map_size;
902 phys_addr_t pa, start, end;
905 /* Setup swapper PGD for fixmap */
906 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
907 __pa_symbol(fixmap_pgd_next),
908 PGDIR_SIZE, PAGE_TABLE);
910 /* Map all memory banks in the linear mapping */
911 for_each_mem_range(i, &start, &end) {
914 if (start <= __pa(PAGE_OFFSET) &&
915 __pa(PAGE_OFFSET) < end)
916 start = __pa(PAGE_OFFSET);
917 if (end >= __pa(PAGE_OFFSET) + memory_limit)
918 end = __pa(PAGE_OFFSET) + memory_limit;
920 map_size = best_map_size(start, end - start);
921 for (pa = start; pa < end; pa += map_size) {
922 va = (uintptr_t)__va(pa);
924 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
930 if (IS_ENABLED(CONFIG_64BIT))
931 create_kernel_page_table(swapper_pg_dir, false);
934 kasan_swapper_init();
937 /* Clear fixmap PTE and PMD mappings */
938 clear_fixmap(FIX_PTE);
939 clear_fixmap(FIX_PMD);
940 clear_fixmap(FIX_PUD);
942 /* Move to swapper page table */
943 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
944 local_flush_tlb_all();
949 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
951 dtb_early_va = (void *)dtb_pa;
952 dtb_early_pa = dtb_pa;
955 static inline void setup_vm_final(void)
958 #endif /* CONFIG_MMU */
960 #ifdef CONFIG_KEXEC_CORE
962 * reserve_crashkernel() - reserves memory for crash kernel
964 * This function reserves memory area given in "crashkernel=" kernel command
965 * line parameter. The memory reserved is used by dump capture kernel when
966 * primary kernel is crashing.
968 static void __init reserve_crashkernel(void)
970 unsigned long long crash_base = 0;
971 unsigned long long crash_size = 0;
972 unsigned long search_start = memblock_start_of_DRAM();
973 unsigned long search_end = memblock_end_of_DRAM();
978 * Don't reserve a region for a crash kernel on a crash kernel
979 * since it doesn't make much sense and we have limited memory
982 if (is_kdump_kernel()) {
983 pr_info("crashkernel: ignoring reservation request\n");
987 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
988 &crash_size, &crash_base);
989 if (ret || !crash_size)
992 crash_size = PAGE_ALIGN(crash_size);
995 search_start = crash_base;
996 search_end = crash_base + crash_size;
1000 * Current riscv boot protocol requires 2MB alignment for
1001 * RV64 and 4MB alignment for RV32 (hugepage size)
1003 * Try to alloc from 32bit addressible physical memory so that
1004 * swiotlb can work on the crash kernel.
1006 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1008 min(search_end, (unsigned long) SZ_4G));
1009 if (crash_base == 0) {
1010 /* Try again without restricting region to 32bit addressible memory */
1011 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1012 search_start, search_end);
1013 if (crash_base == 0) {
1014 pr_warn("crashkernel: couldn't allocate %lldKB\n",
1020 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1021 crash_base, crash_base + crash_size, crash_size >> 20);
1023 crashk_res.start = crash_base;
1024 crashk_res.end = crash_base + crash_size - 1;
1026 #endif /* CONFIG_KEXEC_CORE */
1028 void __init paging_init(void)
1034 void __init misc_mem_init(void)
1036 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1040 #ifdef CONFIG_KEXEC_CORE
1041 reserve_crashkernel();
1043 memblock_dump_all();
1046 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1047 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1048 struct vmem_altmap *altmap)
1050 return vmemmap_populate_basepages(start, end, node, NULL);