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) ?
49 EXPORT_SYMBOL(pgtable_l4_enabled);
51 phys_addr_t phys_ram_base __ro_after_init;
52 EXPORT_SYMBOL(phys_ram_base);
54 #ifdef CONFIG_XIP_KERNEL
55 extern char _xiprom[], _exiprom[], __data_loc;
58 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
60 EXPORT_SYMBOL(empty_zero_page);
63 #define DTB_EARLY_BASE_VA PGDIR_SIZE
64 void *_dtb_early_va __initdata;
65 uintptr_t _dtb_early_pa __initdata;
67 static phys_addr_t dma32_phys_limit __initdata;
69 static void __init zone_sizes_init(void)
71 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
73 #ifdef CONFIG_ZONE_DMA32
74 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
76 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
78 free_area_init(max_zone_pfns);
81 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
82 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
84 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
88 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
90 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
94 static void __init print_vm_layout(void)
96 pr_notice("Virtual kernel memory layout:\n");
97 print_mlk("fixmap", (unsigned long)FIXADDR_START,
98 (unsigned long)FIXADDR_TOP);
99 print_mlm("pci io", (unsigned long)PCI_IO_START,
100 (unsigned long)PCI_IO_END);
101 print_mlm("vmemmap", (unsigned long)VMEMMAP_START,
102 (unsigned long)VMEMMAP_END);
103 print_mlm("vmalloc", (unsigned long)VMALLOC_START,
104 (unsigned long)VMALLOC_END);
105 print_mlm("lowmem", (unsigned long)PAGE_OFFSET,
106 (unsigned long)high_memory);
109 print_mlm("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
111 print_mlm("kernel", (unsigned long)KERNEL_LINK_ADDR,
112 (unsigned long)ADDRESS_SPACE_END);
116 static void print_vm_layout(void) { }
117 #endif /* CONFIG_DEBUG_VM */
119 void __init mem_init(void)
121 #ifdef CONFIG_FLATMEM
123 #endif /* CONFIG_FLATMEM */
125 #ifdef CONFIG_SWIOTLB
126 if (swiotlb_force == SWIOTLB_FORCE ||
127 max_pfn > PFN_DOWN(dma32_phys_limit))
130 swiotlb_force = SWIOTLB_NO_FORCE;
132 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
138 /* Limit the memory size via mem. */
139 static phys_addr_t memory_limit;
141 static int __init early_mem(char *p)
148 size = memparse(p, &p) & PAGE_MASK;
149 memory_limit = min_t(u64, size, memory_limit);
151 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
155 early_param("mem", early_mem);
157 static void __init setup_bootmem(void)
159 phys_addr_t vmlinux_end = __pa_symbol(&_end);
160 phys_addr_t vmlinux_start = __pa_symbol(&_start);
161 phys_addr_t __maybe_unused max_mapped_addr;
162 phys_addr_t phys_ram_end;
164 #ifdef CONFIG_XIP_KERNEL
165 vmlinux_start = __pa_symbol(&_sdata);
168 memblock_enforce_memory_limit(memory_limit);
171 * Reserve from the start of the kernel to the end of the kernel
173 #if defined(CONFIG_64BIT) && defined(CONFIG_STRICT_KERNEL_RWX)
175 * Make sure we align the reservation on PMD_SIZE since we will
176 * map the kernel in the linear mapping as read-only: we do not want
177 * any allocation to happen between _end and the next pmd aligned page.
179 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
181 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
184 phys_ram_end = memblock_end_of_DRAM();
186 #ifndef CONFIG_XIP_KERNEL
187 phys_ram_base = memblock_start_of_DRAM();
190 * memblock allocator is not aware of the fact that last 4K bytes of
191 * the addressable memory can not be mapped because of IS_ERR_VALUE
192 * macro. Make sure that last 4k bytes are not usable by memblock
193 * if end of dram is equal to maximum addressable memory. For 64-bit
194 * kernel, this problem can't happen here as the end of the virtual
195 * address space is occupied by the kernel mapping then this check must
196 * be done as soon as the kernel mapping base address is determined.
198 max_mapped_addr = __pa(~(ulong)0);
199 if (max_mapped_addr == (phys_ram_end - 1))
200 memblock_set_current_limit(max_mapped_addr - 4096);
203 min_low_pfn = PFN_UP(phys_ram_base);
204 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
206 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
207 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
209 reserve_initrd_mem();
211 * If DTB is built in, no need to reserve its memblock.
212 * Otherwise, do reserve it but avoid using
213 * early_init_fdt_reserve_self() since __pa() does
214 * not work for DTB pointers that are fixmap addresses
216 if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
217 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
219 early_init_fdt_scan_reserved_mem();
220 dma_contiguous_reserve(dma32_phys_limit);
221 if (IS_ENABLED(CONFIG_64BIT))
222 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
223 memblock_allow_resize();
227 struct pt_alloc_ops _pt_ops __initdata;
229 #ifdef CONFIG_XIP_KERNEL
230 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&_pt_ops))
232 #define pt_ops _pt_ops
235 unsigned long riscv_pfn_base __ro_after_init;
236 EXPORT_SYMBOL(riscv_pfn_base);
238 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
239 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
240 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
242 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
243 static pud_t __maybe_unused early_dtb_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
244 static pmd_t __maybe_unused early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
246 #ifdef CONFIG_XIP_KERNEL
247 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
248 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
249 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
250 #endif /* CONFIG_XIP_KERNEL */
252 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
254 unsigned long addr = __fix_to_virt(idx);
257 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
259 ptep = &fixmap_pte[pte_index(addr)];
261 if (pgprot_val(prot))
262 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
264 pte_clear(&init_mm, addr, ptep);
265 local_flush_tlb_page(addr);
268 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
270 return (pte_t *)((uintptr_t)pa);
273 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
275 clear_fixmap(FIX_PTE);
276 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
279 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
281 return (pte_t *) __va(pa);
284 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
287 * We only create PMD or PGD early mappings so we
288 * should never reach here with MMU disabled.
293 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
295 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
298 static phys_addr_t __init alloc_pte_late(uintptr_t va)
302 vaddr = __get_free_page(GFP_KERNEL);
303 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)));
308 static void __init create_pte_mapping(pte_t *ptep,
309 uintptr_t va, phys_addr_t pa,
310 phys_addr_t sz, pgprot_t prot)
312 uintptr_t pte_idx = pte_index(va);
314 BUG_ON(sz != PAGE_SIZE);
316 if (pte_none(ptep[pte_idx]))
317 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
320 #ifndef __PAGETABLE_PMD_FOLDED
322 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
323 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
324 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
326 #ifdef CONFIG_XIP_KERNEL
327 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
328 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
329 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
330 #endif /* CONFIG_XIP_KERNEL */
332 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
333 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
334 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
336 #ifdef CONFIG_XIP_KERNEL
337 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
338 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
339 #define early_pud ((pud_t *)XIP_FIXUP(early_pud))
340 #endif /* CONFIG_XIP_KERNEL */
342 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
344 /* Before MMU is enabled */
345 return (pmd_t *)((uintptr_t)pa);
348 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
350 clear_fixmap(FIX_PMD);
351 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
354 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
356 return (pmd_t *) __va(pa);
359 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
361 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
363 return (uintptr_t)early_pmd;
366 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
368 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
371 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
375 vaddr = __get_free_page(GFP_KERNEL);
380 static void __init create_pmd_mapping(pmd_t *pmdp,
381 uintptr_t va, phys_addr_t pa,
382 phys_addr_t sz, pgprot_t prot)
385 phys_addr_t pte_phys;
386 uintptr_t pmd_idx = pmd_index(va);
388 if (sz == PMD_SIZE) {
389 if (pmd_none(pmdp[pmd_idx]))
390 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
394 if (pmd_none(pmdp[pmd_idx])) {
395 pte_phys = pt_ops.alloc_pte(va);
396 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
397 ptep = pt_ops.get_pte_virt(pte_phys);
398 memset(ptep, 0, PAGE_SIZE);
400 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
401 ptep = pt_ops.get_pte_virt(pte_phys);
404 create_pte_mapping(ptep, va, pa, sz, prot);
407 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
409 return (pud_t *)((uintptr_t)pa);
412 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
414 clear_fixmap(FIX_PUD);
415 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
418 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
420 return (pud_t *)__va(pa);
423 static phys_addr_t __init alloc_pud_early(uintptr_t va)
425 /* Only one PUD is available for early mapping */
426 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
428 return (uintptr_t)early_pud;
431 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
433 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
436 static phys_addr_t alloc_pud_late(uintptr_t va)
440 vaddr = __get_free_page(GFP_KERNEL);
445 static void __init create_pud_mapping(pud_t *pudp,
446 uintptr_t va, phys_addr_t pa,
447 phys_addr_t sz, pgprot_t prot)
450 phys_addr_t next_phys;
451 uintptr_t pud_index = pud_index(va);
453 if (sz == PUD_SIZE) {
454 if (pud_val(pudp[pud_index]) == 0)
455 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
459 if (pud_val(pudp[pud_index]) == 0) {
460 next_phys = pt_ops.alloc_pmd(va);
461 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
462 nextp = pt_ops.get_pmd_virt(next_phys);
463 memset(nextp, 0, PAGE_SIZE);
465 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
466 nextp = pt_ops.get_pmd_virt(next_phys);
469 create_pmd_mapping(nextp, va, pa, sz, prot);
472 #define pgd_next_t pud_t
473 #define alloc_pgd_next(__va) (pgtable_l4_enabled ? \
474 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va))
475 #define get_pgd_next_virt(__pa) (pgtable_l4_enabled ? \
476 pt_ops.get_pud_virt(__pa) : (pgd_next_t *)pt_ops.get_pmd_virt(__pa))
477 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
478 (pgtable_l4_enabled ? \
479 create_pud_mapping(__nextp, __va, __pa, __sz, __prot) : \
480 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot))
481 #define fixmap_pgd_next (pgtable_l4_enabled ? \
482 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)
483 #define trampoline_pgd_next (pgtable_l4_enabled ? \
484 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)
485 #define early_dtb_pgd_next (pgtable_l4_enabled ? \
486 (uintptr_t)early_dtb_pud : (uintptr_t)early_dtb_pmd)
488 #define pgd_next_t pte_t
489 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
490 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
491 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
492 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
493 #define fixmap_pgd_next ((uintptr_t)fixmap_pte)
494 #define early_dtb_pgd_next ((uintptr_t)early_dtb_pmd)
495 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot)
496 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot)
497 #endif /* __PAGETABLE_PMD_FOLDED */
499 void __init create_pgd_mapping(pgd_t *pgdp,
500 uintptr_t va, phys_addr_t pa,
501 phys_addr_t sz, pgprot_t prot)
504 phys_addr_t next_phys;
505 uintptr_t pgd_idx = pgd_index(va);
507 if (sz == PGDIR_SIZE) {
508 if (pgd_val(pgdp[pgd_idx]) == 0)
509 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
513 if (pgd_val(pgdp[pgd_idx]) == 0) {
514 next_phys = alloc_pgd_next(va);
515 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
516 nextp = get_pgd_next_virt(next_phys);
517 memset(nextp, 0, PAGE_SIZE);
519 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
520 nextp = get_pgd_next_virt(next_phys);
523 create_pgd_next_mapping(nextp, va, pa, sz, prot);
526 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
528 /* Upgrade to PMD_SIZE mappings whenever possible */
529 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
535 #ifdef CONFIG_XIP_KERNEL
536 /* called from head.S with MMU off */
537 asmlinkage void __init __copy_data(void)
539 void *from = (void *)(&__data_loc);
540 void *to = (void *)CONFIG_PHYS_RAM_BASE;
541 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
543 memcpy(to, from, sz);
547 #ifdef CONFIG_STRICT_KERNEL_RWX
548 static __init pgprot_t pgprot_from_va(uintptr_t va)
550 if (is_va_kernel_text(va))
551 return PAGE_KERNEL_READ_EXEC;
554 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
555 * we must protect its linear mapping alias from being executed and
557 * And rodata section is marked readonly in mark_rodata_ro.
559 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
560 return PAGE_KERNEL_READ;
565 void mark_rodata_ro(void)
567 set_kernel_memory(__start_rodata, _data, set_memory_ro);
568 if (IS_ENABLED(CONFIG_64BIT))
569 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
575 static __init pgprot_t pgprot_from_va(uintptr_t va)
577 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
580 return PAGE_KERNEL_EXEC;
582 #endif /* CONFIG_STRICT_KERNEL_RWX */
585 static void __init disable_pgtable_l4(void)
587 pgtable_l4_enabled = false;
588 kernel_map.page_offset = PAGE_OFFSET_L3;
589 satp_mode = SATP_MODE_39;
593 * There is a simple way to determine if 4-level is supported by the
594 * underlying hardware: establish 1:1 mapping in 4-level page table mode
595 * then read SATP to see if the configuration was taken into account
596 * meaning sv48 is supported.
598 static __init void set_satp_mode(void)
600 u64 identity_satp, hw_satp;
601 uintptr_t set_satp_mode_pmd;
603 set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
604 create_pgd_mapping(early_pg_dir,
605 set_satp_mode_pmd, (uintptr_t)early_pud,
606 PGDIR_SIZE, PAGE_TABLE);
607 create_pud_mapping(early_pud,
608 set_satp_mode_pmd, (uintptr_t)early_pmd,
609 PUD_SIZE, PAGE_TABLE);
610 /* Handle the case where set_satp_mode straddles 2 PMDs */
611 create_pmd_mapping(early_pmd,
612 set_satp_mode_pmd, set_satp_mode_pmd,
613 PMD_SIZE, PAGE_KERNEL_EXEC);
614 create_pmd_mapping(early_pmd,
615 set_satp_mode_pmd + PMD_SIZE,
616 set_satp_mode_pmd + PMD_SIZE,
617 PMD_SIZE, PAGE_KERNEL_EXEC);
619 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
621 local_flush_tlb_all();
622 csr_write(CSR_SATP, identity_satp);
623 hw_satp = csr_swap(CSR_SATP, 0ULL);
624 local_flush_tlb_all();
626 if (hw_satp != identity_satp)
627 disable_pgtable_l4();
629 memset(early_pg_dir, 0, PAGE_SIZE);
630 memset(early_pud, 0, PAGE_SIZE);
631 memset(early_pmd, 0, PAGE_SIZE);
636 * setup_vm() is called from head.S with MMU-off.
638 * Following requirements should be honoured for setup_vm() to work
640 * 1) It should use PC-relative addressing for accessing kernel symbols.
641 * To achieve this we always use GCC cmodel=medany.
642 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
643 * so disable compiler instrumentation when FTRACE is enabled.
645 * Currently, the above requirements are honoured by using custom CFLAGS
646 * for init.o in mm/Makefile.
649 #ifndef __riscv_cmodel_medany
650 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
653 #ifdef CONFIG_XIP_KERNEL
654 static void __init create_kernel_page_table(pgd_t *pgdir,
655 __always_unused bool early)
657 uintptr_t va, end_va;
659 /* Map the flash resident part */
660 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
661 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
662 create_pgd_mapping(pgdir, va,
663 kernel_map.xiprom + (va - kernel_map.virt_addr),
664 PMD_SIZE, PAGE_KERNEL_EXEC);
666 /* Map the data in RAM */
667 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
668 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
669 create_pgd_mapping(pgdir, va,
670 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
671 PMD_SIZE, PAGE_KERNEL);
674 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
676 uintptr_t va, end_va;
678 end_va = kernel_map.virt_addr + kernel_map.size;
679 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
680 create_pgd_mapping(pgdir, va,
681 kernel_map.phys_addr + (va - kernel_map.virt_addr),
684 PAGE_KERNEL_EXEC : pgprot_from_va(va));
689 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
690 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
693 static void __init create_fdt_early_page_table(pgd_t *pgdir, uintptr_t dtb_pa)
695 #ifndef CONFIG_BUILTIN_DTB
696 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
698 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
699 IS_ENABLED(CONFIG_64BIT) ? early_dtb_pgd_next : pa,
701 IS_ENABLED(CONFIG_64BIT) ? PAGE_TABLE : PAGE_KERNEL);
703 if (pgtable_l4_enabled) {
704 create_pud_mapping(early_dtb_pud, DTB_EARLY_BASE_VA,
705 (uintptr_t)early_dtb_pmd, PUD_SIZE, PAGE_TABLE);
708 if (IS_ENABLED(CONFIG_64BIT)) {
709 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA,
710 pa, PMD_SIZE, PAGE_KERNEL);
711 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE,
712 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
715 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1));
718 * For 64-bit kernel, __va can't be used since it would return a linear
719 * mapping address whereas dtb_early_va will be used before
720 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
721 * kernel is mapped in the linear mapping, that makes no difference.
723 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
726 dtb_early_pa = dtb_pa;
730 * MMU is not enabled, the page tables are allocated directly using
731 * early_pmd/pud/p4d and the address returned is the physical one.
733 void pt_ops_set_early(void)
735 pt_ops.alloc_pte = alloc_pte_early;
736 pt_ops.get_pte_virt = get_pte_virt_early;
737 #ifndef __PAGETABLE_PMD_FOLDED
738 pt_ops.alloc_pmd = alloc_pmd_early;
739 pt_ops.get_pmd_virt = get_pmd_virt_early;
740 pt_ops.alloc_pud = alloc_pud_early;
741 pt_ops.get_pud_virt = get_pud_virt_early;
746 * MMU is enabled but page table setup is not complete yet.
747 * fixmap page table alloc functions must be used as a means to temporarily
748 * map the allocated physical pages since the linear mapping does not exist yet.
750 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
751 * but it will be used as described above.
753 void pt_ops_set_fixmap(void)
755 pt_ops.alloc_pte = kernel_mapping_pa_to_va((uintptr_t)alloc_pte_fixmap);
756 pt_ops.get_pte_virt = kernel_mapping_pa_to_va((uintptr_t)get_pte_virt_fixmap);
757 #ifndef __PAGETABLE_PMD_FOLDED
758 pt_ops.alloc_pmd = kernel_mapping_pa_to_va((uintptr_t)alloc_pmd_fixmap);
759 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va((uintptr_t)get_pmd_virt_fixmap);
760 pt_ops.alloc_pud = kernel_mapping_pa_to_va((uintptr_t)alloc_pud_fixmap);
761 pt_ops.get_pud_virt = kernel_mapping_pa_to_va((uintptr_t)get_pud_virt_fixmap);
766 * MMU is enabled and page table setup is complete, so from now, we can use
767 * generic page allocation functions to setup page table.
769 void pt_ops_set_late(void)
771 pt_ops.alloc_pte = alloc_pte_late;
772 pt_ops.get_pte_virt = get_pte_virt_late;
773 #ifndef __PAGETABLE_PMD_FOLDED
774 pt_ops.alloc_pmd = alloc_pmd_late;
775 pt_ops.get_pmd_virt = get_pmd_virt_late;
776 pt_ops.alloc_pud = alloc_pud_late;
777 pt_ops.get_pud_virt = get_pud_virt_late;
781 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
783 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
785 kernel_map.virt_addr = KERNEL_LINK_ADDR;
786 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
788 #ifdef CONFIG_XIP_KERNEL
789 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
790 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
792 phys_ram_base = CONFIG_PHYS_RAM_BASE;
793 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
794 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
796 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
798 kernel_map.phys_addr = (uintptr_t)(&_start);
799 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
802 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
806 kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr;
807 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
809 riscv_pfn_base = PFN_DOWN(kernel_map.phys_addr);
812 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
813 * kernel, whereas for 64-bit kernel, the end of the virtual address
814 * space is occupied by the modules/BPF/kernel mappings which reduces
815 * the available size of the linear mapping.
817 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
819 /* Sanity check alignment and size */
820 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
821 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
825 * The last 4K bytes of the addressable memory can not be mapped because
826 * of IS_ERR_VALUE macro.
828 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
833 /* Setup early PGD for fixmap */
834 create_pgd_mapping(early_pg_dir, FIXADDR_START,
835 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
837 #ifndef __PAGETABLE_PMD_FOLDED
838 /* Setup fixmap PUD and PMD */
839 if (pgtable_l4_enabled)
840 create_pud_mapping(fixmap_pud, FIXADDR_START,
841 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
842 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
843 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
844 /* Setup trampoline PGD and PMD */
845 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
846 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
847 if (pgtable_l4_enabled)
848 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
849 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
850 #ifdef CONFIG_XIP_KERNEL
851 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
852 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
854 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
855 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
858 /* Setup trampoline PGD */
859 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
860 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
864 * Setup early PGD covering entire kernel which will allow
865 * us to reach paging_init(). We map all memory banks later
866 * in setup_vm_final() below.
868 create_kernel_page_table(early_pg_dir, true);
870 /* Setup early mapping for FDT early scan */
871 create_fdt_early_page_table(early_pg_dir, dtb_pa);
874 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
875 * range can not span multiple pmds.
877 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
878 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
880 #ifndef __PAGETABLE_PMD_FOLDED
882 * Early ioremap fixmap is already created as it lies within first 2MB
883 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
884 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
887 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
888 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
889 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
891 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
892 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
893 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
894 fix_to_virt(FIX_BTMAP_BEGIN));
895 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
896 fix_to_virt(FIX_BTMAP_END));
898 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
899 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
906 static void __init setup_vm_final(void)
908 uintptr_t va, map_size;
909 phys_addr_t pa, start, end;
912 /* Setup swapper PGD for fixmap */
913 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
914 __pa_symbol(fixmap_pgd_next),
915 PGDIR_SIZE, PAGE_TABLE);
917 /* Map all memory banks in the linear mapping */
918 for_each_mem_range(i, &start, &end) {
921 if (start <= __pa(PAGE_OFFSET) &&
922 __pa(PAGE_OFFSET) < end)
923 start = __pa(PAGE_OFFSET);
924 if (end >= __pa(PAGE_OFFSET) + memory_limit)
925 end = __pa(PAGE_OFFSET) + memory_limit;
927 map_size = best_map_size(start, end - start);
928 for (pa = start; pa < end; pa += map_size) {
929 va = (uintptr_t)__va(pa);
931 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
938 create_kernel_page_table(swapper_pg_dir, false);
942 kasan_swapper_init();
945 /* Clear fixmap PTE and PMD mappings */
946 clear_fixmap(FIX_PTE);
947 clear_fixmap(FIX_PMD);
948 clear_fixmap(FIX_PUD);
950 /* Move to swapper page table */
951 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
952 local_flush_tlb_all();
957 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
959 dtb_early_va = (void *)dtb_pa;
960 dtb_early_pa = dtb_pa;
963 static inline void setup_vm_final(void)
966 #endif /* CONFIG_MMU */
968 #ifdef CONFIG_KEXEC_CORE
970 * reserve_crashkernel() - reserves memory for crash kernel
972 * This function reserves memory area given in "crashkernel=" kernel command
973 * line parameter. The memory reserved is used by dump capture kernel when
974 * primary kernel is crashing.
976 static void __init reserve_crashkernel(void)
978 unsigned long long crash_base = 0;
979 unsigned long long crash_size = 0;
980 unsigned long search_start = memblock_start_of_DRAM();
981 unsigned long search_end = memblock_end_of_DRAM();
986 * Don't reserve a region for a crash kernel on a crash kernel
987 * since it doesn't make much sense and we have limited memory
990 #ifdef CONFIG_CRASH_DUMP
991 if (is_kdump_kernel()) {
992 pr_info("crashkernel: ignoring reservation request\n");
997 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
998 &crash_size, &crash_base);
999 if (ret || !crash_size)
1002 crash_size = PAGE_ALIGN(crash_size);
1005 search_start = crash_base;
1006 search_end = crash_base + crash_size;
1010 * Current riscv boot protocol requires 2MB alignment for
1011 * RV64 and 4MB alignment for RV32 (hugepage size)
1013 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1014 search_start, search_end);
1015 if (crash_base == 0) {
1016 pr_warn("crashkernel: couldn't allocate %lldKB\n",
1021 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1022 crash_base, crash_base + crash_size, crash_size >> 20);
1024 crashk_res.start = crash_base;
1025 crashk_res.end = crash_base + crash_size - 1;
1027 #endif /* CONFIG_KEXEC_CORE */
1029 void __init paging_init(void)
1035 void __init misc_mem_init(void)
1037 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1041 #ifdef CONFIG_KEXEC_CORE
1042 reserve_crashkernel();
1044 memblock_dump_all();
1047 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1048 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1049 struct vmem_altmap *altmap)
1051 return vmemmap_populate_basepages(start, end, node, NULL);