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 __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39;
43 u64 satp_mode __ro_after_init = SATP_MODE_32;
45 EXPORT_SYMBOL(satp_mode);
47 bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
48 bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
49 EXPORT_SYMBOL(pgtable_l4_enabled);
50 EXPORT_SYMBOL(pgtable_l5_enabled);
52 phys_addr_t phys_ram_base __ro_after_init;
53 EXPORT_SYMBOL(phys_ram_base);
55 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
57 EXPORT_SYMBOL(empty_zero_page);
60 #define DTB_EARLY_BASE_VA PGDIR_SIZE
61 void *_dtb_early_va __initdata;
62 uintptr_t _dtb_early_pa __initdata;
64 static phys_addr_t dma32_phys_limit __initdata;
66 static void __init zone_sizes_init(void)
68 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
70 #ifdef CONFIG_ZONE_DMA32
71 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
73 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
75 free_area_init(max_zone_pfns);
78 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
80 #define LOG2_SZ_1K ilog2(SZ_1K)
81 #define LOG2_SZ_1M ilog2(SZ_1M)
82 #define LOG2_SZ_1G ilog2(SZ_1G)
83 #define LOG2_SZ_1T ilog2(SZ_1T)
85 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
87 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
88 (((t) - (b)) >> LOG2_SZ_1K));
91 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
93 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
94 (((t) - (b)) >> LOG2_SZ_1M));
97 static inline void print_mlg(char *name, unsigned long b, unsigned long t)
99 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t,
100 (((t) - (b)) >> LOG2_SZ_1G));
104 static inline void print_mlt(char *name, unsigned long b, unsigned long t)
106 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t,
107 (((t) - (b)) >> LOG2_SZ_1T));
110 #define print_mlt(n, b, t) do {} while (0)
113 static inline void print_ml(char *name, unsigned long b, unsigned long t)
115 unsigned long diff = t - b;
117 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10)
118 print_mlt(name, b, t);
119 else if ((diff >> LOG2_SZ_1G) >= 10)
120 print_mlg(name, b, t);
121 else if ((diff >> LOG2_SZ_1M) >= 10)
122 print_mlm(name, b, t);
124 print_mlk(name, b, t);
127 static void __init print_vm_layout(void)
129 pr_notice("Virtual kernel memory layout:\n");
130 print_ml("fixmap", (unsigned long)FIXADDR_START,
131 (unsigned long)FIXADDR_TOP);
132 print_ml("pci io", (unsigned long)PCI_IO_START,
133 (unsigned long)PCI_IO_END);
134 print_ml("vmemmap", (unsigned long)VMEMMAP_START,
135 (unsigned long)VMEMMAP_END);
136 print_ml("vmalloc", (unsigned long)VMALLOC_START,
137 (unsigned long)VMALLOC_END);
138 print_ml("lowmem", (unsigned long)PAGE_OFFSET,
139 (unsigned long)high_memory);
140 if (IS_ENABLED(CONFIG_64BIT)) {
142 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
145 print_ml("kernel", (unsigned long)KERNEL_LINK_ADDR,
146 (unsigned long)ADDRESS_SPACE_END);
150 static void print_vm_layout(void) { }
151 #endif /* CONFIG_DEBUG_VM */
153 void __init mem_init(void)
155 #ifdef CONFIG_FLATMEM
157 #endif /* CONFIG_FLATMEM */
159 swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE);
165 /* Limit the memory size via mem. */
166 static phys_addr_t memory_limit;
168 static int __init early_mem(char *p)
175 size = memparse(p, &p) & PAGE_MASK;
176 memory_limit = min_t(u64, size, memory_limit);
178 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
182 early_param("mem", early_mem);
184 static void __init setup_bootmem(void)
186 phys_addr_t vmlinux_end = __pa_symbol(&_end);
187 phys_addr_t max_mapped_addr;
188 phys_addr_t phys_ram_end, vmlinux_start;
190 if (IS_ENABLED(CONFIG_XIP_KERNEL))
191 vmlinux_start = __pa_symbol(&_sdata);
193 vmlinux_start = __pa_symbol(&_start);
195 memblock_enforce_memory_limit(memory_limit);
198 * Make sure we align the reservation on PMD_SIZE since we will
199 * map the kernel in the linear mapping as read-only: we do not want
200 * any allocation to happen between _end and the next pmd aligned page.
202 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
203 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
205 * Reserve from the start of the kernel to the end of the kernel
207 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
209 phys_ram_end = memblock_end_of_DRAM();
210 if (!IS_ENABLED(CONFIG_XIP_KERNEL))
211 phys_ram_base = memblock_start_of_DRAM();
213 * memblock allocator is not aware of the fact that last 4K bytes of
214 * the addressable memory can not be mapped because of IS_ERR_VALUE
215 * macro. Make sure that last 4k bytes are not usable by memblock
216 * if end of dram is equal to maximum addressable memory. For 64-bit
217 * kernel, this problem can't happen here as the end of the virtual
218 * address space is occupied by the kernel mapping then this check must
219 * be done as soon as the kernel mapping base address is determined.
221 if (!IS_ENABLED(CONFIG_64BIT)) {
222 max_mapped_addr = __pa(~(ulong)0);
223 if (max_mapped_addr == (phys_ram_end - 1))
224 memblock_set_current_limit(max_mapped_addr - 4096);
227 min_low_pfn = PFN_UP(phys_ram_base);
228 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
229 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
231 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
232 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
234 reserve_initrd_mem();
236 * If DTB is built in, no need to reserve its memblock.
237 * Otherwise, do reserve it but avoid using
238 * early_init_fdt_reserve_self() since __pa() does
239 * not work for DTB pointers that are fixmap addresses
241 if (!IS_ENABLED(CONFIG_BUILTIN_DTB)) {
243 * In case the DTB is not located in a memory region we won't
244 * be able to locate it later on via the linear mapping and
245 * get a segfault when accessing it via __va(dtb_early_pa).
246 * To avoid this situation copy DTB to a memory region.
247 * Note that memblock_phys_alloc will also reserve DTB region.
249 if (!memblock_is_memory(dtb_early_pa)) {
250 size_t fdt_size = fdt_totalsize(dtb_early_va);
251 phys_addr_t new_dtb_early_pa = memblock_phys_alloc(fdt_size, PAGE_SIZE);
252 void *new_dtb_early_va = early_memremap(new_dtb_early_pa, fdt_size);
254 memcpy(new_dtb_early_va, dtb_early_va, fdt_size);
255 early_memunmap(new_dtb_early_va, fdt_size);
256 _dtb_early_pa = new_dtb_early_pa;
258 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
261 early_init_fdt_scan_reserved_mem();
262 dma_contiguous_reserve(dma32_phys_limit);
263 if (IS_ENABLED(CONFIG_64BIT))
264 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
265 memblock_allow_resize();
269 struct pt_alloc_ops pt_ops __initdata;
271 unsigned long riscv_pfn_base __ro_after_init;
272 EXPORT_SYMBOL(riscv_pfn_base);
274 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
275 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
276 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
278 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
279 static p4d_t __maybe_unused early_dtb_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
280 static pud_t __maybe_unused early_dtb_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
281 static pmd_t __maybe_unused early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
283 #ifdef CONFIG_XIP_KERNEL
284 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
285 #define riscv_pfn_base (*(unsigned long *)XIP_FIXUP(&riscv_pfn_base))
286 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
287 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
288 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
289 #endif /* CONFIG_XIP_KERNEL */
291 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
293 unsigned long addr = __fix_to_virt(idx);
296 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
298 ptep = &fixmap_pte[pte_index(addr)];
300 if (pgprot_val(prot))
301 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
303 pte_clear(&init_mm, addr, ptep);
304 local_flush_tlb_page(addr);
307 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
309 return (pte_t *)((uintptr_t)pa);
312 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
314 clear_fixmap(FIX_PTE);
315 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
318 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
320 return (pte_t *) __va(pa);
323 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
326 * We only create PMD or PGD early mappings so we
327 * should never reach here with MMU disabled.
332 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
334 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
337 static phys_addr_t __init alloc_pte_late(uintptr_t va)
341 vaddr = __get_free_page(GFP_KERNEL);
342 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)));
347 static void __init create_pte_mapping(pte_t *ptep,
348 uintptr_t va, phys_addr_t pa,
349 phys_addr_t sz, pgprot_t prot)
351 uintptr_t pte_idx = pte_index(va);
353 BUG_ON(sz != PAGE_SIZE);
355 if (pte_none(ptep[pte_idx]))
356 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
359 #ifndef __PAGETABLE_PMD_FOLDED
361 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
362 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
363 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
365 #ifdef CONFIG_XIP_KERNEL
366 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
367 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
368 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
369 #endif /* CONFIG_XIP_KERNEL */
371 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
372 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
373 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
375 #ifdef CONFIG_XIP_KERNEL
376 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
377 #define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d))
378 #define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d))
379 #endif /* CONFIG_XIP_KERNEL */
381 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
382 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
383 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
385 #ifdef CONFIG_XIP_KERNEL
386 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
387 #define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
388 #define early_pud ((pud_t *)XIP_FIXUP(early_pud))
389 #endif /* CONFIG_XIP_KERNEL */
391 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
393 /* Before MMU is enabled */
394 return (pmd_t *)((uintptr_t)pa);
397 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
399 clear_fixmap(FIX_PMD);
400 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
403 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
405 return (pmd_t *) __va(pa);
408 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
410 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
412 return (uintptr_t)early_pmd;
415 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
417 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
420 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
424 vaddr = __get_free_page(GFP_KERNEL);
425 BUG_ON(!vaddr || !pgtable_pmd_page_ctor(virt_to_page(vaddr)));
430 static void __init create_pmd_mapping(pmd_t *pmdp,
431 uintptr_t va, phys_addr_t pa,
432 phys_addr_t sz, pgprot_t prot)
435 phys_addr_t pte_phys;
436 uintptr_t pmd_idx = pmd_index(va);
438 if (sz == PMD_SIZE) {
439 if (pmd_none(pmdp[pmd_idx]))
440 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
444 if (pmd_none(pmdp[pmd_idx])) {
445 pte_phys = pt_ops.alloc_pte(va);
446 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
447 ptep = pt_ops.get_pte_virt(pte_phys);
448 memset(ptep, 0, PAGE_SIZE);
450 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
451 ptep = pt_ops.get_pte_virt(pte_phys);
454 create_pte_mapping(ptep, va, pa, sz, prot);
457 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
459 return (pud_t *)((uintptr_t)pa);
462 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
464 clear_fixmap(FIX_PUD);
465 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
468 static pud_t *__init get_pud_virt_late(phys_addr_t pa)
470 return (pud_t *)__va(pa);
473 static phys_addr_t __init alloc_pud_early(uintptr_t va)
475 /* Only one PUD is available for early mapping */
476 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
478 return (uintptr_t)early_pud;
481 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
483 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
486 static phys_addr_t alloc_pud_late(uintptr_t va)
490 vaddr = __get_free_page(GFP_KERNEL);
495 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
497 return (p4d_t *)((uintptr_t)pa);
500 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
502 clear_fixmap(FIX_P4D);
503 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
506 static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
508 return (p4d_t *)__va(pa);
511 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
513 /* Only one P4D is available for early mapping */
514 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
516 return (uintptr_t)early_p4d;
519 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
521 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
524 static phys_addr_t alloc_p4d_late(uintptr_t va)
528 vaddr = __get_free_page(GFP_KERNEL);
533 static void __init create_pud_mapping(pud_t *pudp,
534 uintptr_t va, phys_addr_t pa,
535 phys_addr_t sz, pgprot_t prot)
538 phys_addr_t next_phys;
539 uintptr_t pud_index = pud_index(va);
541 if (sz == PUD_SIZE) {
542 if (pud_val(pudp[pud_index]) == 0)
543 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
547 if (pud_val(pudp[pud_index]) == 0) {
548 next_phys = pt_ops.alloc_pmd(va);
549 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
550 nextp = pt_ops.get_pmd_virt(next_phys);
551 memset(nextp, 0, PAGE_SIZE);
553 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
554 nextp = pt_ops.get_pmd_virt(next_phys);
557 create_pmd_mapping(nextp, va, pa, sz, prot);
560 static void __init create_p4d_mapping(p4d_t *p4dp,
561 uintptr_t va, phys_addr_t pa,
562 phys_addr_t sz, pgprot_t prot)
565 phys_addr_t next_phys;
566 uintptr_t p4d_index = p4d_index(va);
568 if (sz == P4D_SIZE) {
569 if (p4d_val(p4dp[p4d_index]) == 0)
570 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
574 if (p4d_val(p4dp[p4d_index]) == 0) {
575 next_phys = pt_ops.alloc_pud(va);
576 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
577 nextp = pt_ops.get_pud_virt(next_phys);
578 memset(nextp, 0, PAGE_SIZE);
580 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
581 nextp = pt_ops.get_pud_virt(next_phys);
584 create_pud_mapping(nextp, va, pa, sz, prot);
587 #define pgd_next_t p4d_t
588 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \
589 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \
590 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
591 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \
592 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \
593 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
594 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
595 (pgtable_l5_enabled ? \
596 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
597 (pgtable_l4_enabled ? \
598 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \
599 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
600 #define fixmap_pgd_next (pgtable_l5_enabled ? \
601 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \
602 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
603 #define trampoline_pgd_next (pgtable_l5_enabled ? \
604 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \
605 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
606 #define early_dtb_pgd_next (pgtable_l5_enabled ? \
607 (uintptr_t)early_dtb_p4d : (pgtable_l4_enabled ? \
608 (uintptr_t)early_dtb_pud : (uintptr_t)early_dtb_pmd))
610 #define pgd_next_t pte_t
611 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
612 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
613 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
614 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
615 #define fixmap_pgd_next ((uintptr_t)fixmap_pte)
616 #define early_dtb_pgd_next ((uintptr_t)early_dtb_pmd)
617 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
618 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
619 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
620 #endif /* __PAGETABLE_PMD_FOLDED */
622 void __init create_pgd_mapping(pgd_t *pgdp,
623 uintptr_t va, phys_addr_t pa,
624 phys_addr_t sz, pgprot_t prot)
627 phys_addr_t next_phys;
628 uintptr_t pgd_idx = pgd_index(va);
630 if (sz == PGDIR_SIZE) {
631 if (pgd_val(pgdp[pgd_idx]) == 0)
632 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
636 if (pgd_val(pgdp[pgd_idx]) == 0) {
637 next_phys = alloc_pgd_next(va);
638 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
639 nextp = get_pgd_next_virt(next_phys);
640 memset(nextp, 0, PAGE_SIZE);
642 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
643 nextp = get_pgd_next_virt(next_phys);
646 create_pgd_next_mapping(nextp, va, pa, sz, prot);
649 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
651 /* Upgrade to PMD_SIZE mappings whenever possible */
652 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
658 #ifdef CONFIG_XIP_KERNEL
659 #define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
660 extern char _xiprom[], _exiprom[], __data_loc;
662 /* called from head.S with MMU off */
663 asmlinkage void __init __copy_data(void)
665 void *from = (void *)(&__data_loc);
666 void *to = (void *)CONFIG_PHYS_RAM_BASE;
667 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
669 memcpy(to, from, sz);
673 #ifdef CONFIG_STRICT_KERNEL_RWX
674 static __init pgprot_t pgprot_from_va(uintptr_t va)
676 if (is_va_kernel_text(va))
677 return PAGE_KERNEL_READ_EXEC;
680 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
681 * we must protect its linear mapping alias from being executed and
683 * And rodata section is marked readonly in mark_rodata_ro.
685 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
686 return PAGE_KERNEL_READ;
691 void mark_rodata_ro(void)
693 set_kernel_memory(__start_rodata, _data, set_memory_ro);
694 if (IS_ENABLED(CONFIG_64BIT))
695 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
701 static __init pgprot_t pgprot_from_va(uintptr_t va)
703 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
706 return PAGE_KERNEL_EXEC;
708 #endif /* CONFIG_STRICT_KERNEL_RWX */
710 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
711 static void __init disable_pgtable_l5(void)
713 pgtable_l5_enabled = false;
714 kernel_map.page_offset = PAGE_OFFSET_L4;
715 satp_mode = SATP_MODE_48;
718 static void __init disable_pgtable_l4(void)
720 pgtable_l4_enabled = false;
721 kernel_map.page_offset = PAGE_OFFSET_L3;
722 satp_mode = SATP_MODE_39;
726 * There is a simple way to determine if 4-level is supported by the
727 * underlying hardware: establish 1:1 mapping in 4-level page table mode
728 * then read SATP to see if the configuration was taken into account
729 * meaning sv48 is supported.
731 static __init void set_satp_mode(void)
733 u64 identity_satp, hw_satp;
734 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
735 bool check_l4 = false;
737 create_p4d_mapping(early_p4d,
738 set_satp_mode_pmd, (uintptr_t)early_pud,
739 P4D_SIZE, PAGE_TABLE);
740 create_pud_mapping(early_pud,
741 set_satp_mode_pmd, (uintptr_t)early_pmd,
742 PUD_SIZE, PAGE_TABLE);
743 /* Handle the case where set_satp_mode straddles 2 PMDs */
744 create_pmd_mapping(early_pmd,
745 set_satp_mode_pmd, set_satp_mode_pmd,
746 PMD_SIZE, PAGE_KERNEL_EXEC);
747 create_pmd_mapping(early_pmd,
748 set_satp_mode_pmd + PMD_SIZE,
749 set_satp_mode_pmd + PMD_SIZE,
750 PMD_SIZE, PAGE_KERNEL_EXEC);
752 create_pgd_mapping(early_pg_dir,
754 check_l4 ? (uintptr_t)early_pud : (uintptr_t)early_p4d,
755 PGDIR_SIZE, PAGE_TABLE);
757 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
759 local_flush_tlb_all();
760 csr_write(CSR_SATP, identity_satp);
761 hw_satp = csr_swap(CSR_SATP, 0ULL);
762 local_flush_tlb_all();
764 if (hw_satp != identity_satp) {
766 disable_pgtable_l5();
768 memset(early_pg_dir, 0, PAGE_SIZE);
771 disable_pgtable_l4();
774 memset(early_pg_dir, 0, PAGE_SIZE);
775 memset(early_p4d, 0, PAGE_SIZE);
776 memset(early_pud, 0, PAGE_SIZE);
777 memset(early_pmd, 0, PAGE_SIZE);
782 * setup_vm() is called from head.S with MMU-off.
784 * Following requirements should be honoured for setup_vm() to work
786 * 1) It should use PC-relative addressing for accessing kernel symbols.
787 * To achieve this we always use GCC cmodel=medany.
788 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
789 * so disable compiler instrumentation when FTRACE is enabled.
791 * Currently, the above requirements are honoured by using custom CFLAGS
792 * for init.o in mm/Makefile.
795 #ifndef __riscv_cmodel_medany
796 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
799 #ifdef CONFIG_XIP_KERNEL
800 static void __init create_kernel_page_table(pgd_t *pgdir,
801 __always_unused bool early)
803 uintptr_t va, end_va;
805 /* Map the flash resident part */
806 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
807 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
808 create_pgd_mapping(pgdir, va,
809 kernel_map.xiprom + (va - kernel_map.virt_addr),
810 PMD_SIZE, PAGE_KERNEL_EXEC);
812 /* Map the data in RAM */
813 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
814 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
815 create_pgd_mapping(pgdir, va,
816 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
817 PMD_SIZE, PAGE_KERNEL);
820 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
822 uintptr_t va, end_va;
824 end_va = kernel_map.virt_addr + kernel_map.size;
825 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
826 create_pgd_mapping(pgdir, va,
827 kernel_map.phys_addr + (va - kernel_map.virt_addr),
830 PAGE_KERNEL_EXEC : pgprot_from_va(va));
835 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
836 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
839 static void __init create_fdt_early_page_table(pgd_t *pgdir, uintptr_t dtb_pa)
841 #ifndef CONFIG_BUILTIN_DTB
842 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
844 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
845 IS_ENABLED(CONFIG_64BIT) ? early_dtb_pgd_next : pa,
847 IS_ENABLED(CONFIG_64BIT) ? PAGE_TABLE : PAGE_KERNEL);
849 if (pgtable_l5_enabled)
850 create_p4d_mapping(early_dtb_p4d, DTB_EARLY_BASE_VA,
851 (uintptr_t)early_dtb_pud, P4D_SIZE, PAGE_TABLE);
853 if (pgtable_l4_enabled)
854 create_pud_mapping(early_dtb_pud, DTB_EARLY_BASE_VA,
855 (uintptr_t)early_dtb_pmd, PUD_SIZE, PAGE_TABLE);
857 if (IS_ENABLED(CONFIG_64BIT)) {
858 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA,
859 pa, PMD_SIZE, PAGE_KERNEL);
860 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE,
861 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
864 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1));
867 * For 64-bit kernel, __va can't be used since it would return a linear
868 * mapping address whereas dtb_early_va will be used before
869 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
870 * kernel is mapped in the linear mapping, that makes no difference.
872 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
875 dtb_early_pa = dtb_pa;
879 * MMU is not enabled, the page tables are allocated directly using
880 * early_pmd/pud/p4d and the address returned is the physical one.
882 static void __init pt_ops_set_early(void)
884 pt_ops.alloc_pte = alloc_pte_early;
885 pt_ops.get_pte_virt = get_pte_virt_early;
886 #ifndef __PAGETABLE_PMD_FOLDED
887 pt_ops.alloc_pmd = alloc_pmd_early;
888 pt_ops.get_pmd_virt = get_pmd_virt_early;
889 pt_ops.alloc_pud = alloc_pud_early;
890 pt_ops.get_pud_virt = get_pud_virt_early;
891 pt_ops.alloc_p4d = alloc_p4d_early;
892 pt_ops.get_p4d_virt = get_p4d_virt_early;
897 * MMU is enabled but page table setup is not complete yet.
898 * fixmap page table alloc functions must be used as a means to temporarily
899 * map the allocated physical pages since the linear mapping does not exist yet.
901 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
902 * but it will be used as described above.
904 static void __init pt_ops_set_fixmap(void)
906 pt_ops.alloc_pte = kernel_mapping_pa_to_va((uintptr_t)alloc_pte_fixmap);
907 pt_ops.get_pte_virt = kernel_mapping_pa_to_va((uintptr_t)get_pte_virt_fixmap);
908 #ifndef __PAGETABLE_PMD_FOLDED
909 pt_ops.alloc_pmd = kernel_mapping_pa_to_va((uintptr_t)alloc_pmd_fixmap);
910 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va((uintptr_t)get_pmd_virt_fixmap);
911 pt_ops.alloc_pud = kernel_mapping_pa_to_va((uintptr_t)alloc_pud_fixmap);
912 pt_ops.get_pud_virt = kernel_mapping_pa_to_va((uintptr_t)get_pud_virt_fixmap);
913 pt_ops.alloc_p4d = kernel_mapping_pa_to_va((uintptr_t)alloc_p4d_fixmap);
914 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va((uintptr_t)get_p4d_virt_fixmap);
919 * MMU is enabled and page table setup is complete, so from now, we can use
920 * generic page allocation functions to setup page table.
922 static void __init pt_ops_set_late(void)
924 pt_ops.alloc_pte = alloc_pte_late;
925 pt_ops.get_pte_virt = get_pte_virt_late;
926 #ifndef __PAGETABLE_PMD_FOLDED
927 pt_ops.alloc_pmd = alloc_pmd_late;
928 pt_ops.get_pmd_virt = get_pmd_virt_late;
929 pt_ops.alloc_pud = alloc_pud_late;
930 pt_ops.get_pud_virt = get_pud_virt_late;
931 pt_ops.alloc_p4d = alloc_p4d_late;
932 pt_ops.get_p4d_virt = get_p4d_virt_late;
936 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
938 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
940 kernel_map.virt_addr = KERNEL_LINK_ADDR;
941 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
943 #ifdef CONFIG_XIP_KERNEL
944 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
945 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
947 phys_ram_base = CONFIG_PHYS_RAM_BASE;
948 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
949 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
951 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
953 kernel_map.phys_addr = (uintptr_t)(&_start);
954 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
957 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
961 kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr;
962 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
964 riscv_pfn_base = PFN_DOWN(kernel_map.phys_addr);
967 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
968 * kernel, whereas for 64-bit kernel, the end of the virtual address
969 * space is occupied by the modules/BPF/kernel mappings which reduces
970 * the available size of the linear mapping.
972 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
974 /* Sanity check alignment and size */
975 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
976 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
980 * The last 4K bytes of the addressable memory can not be mapped because
981 * of IS_ERR_VALUE macro.
983 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
986 apply_early_boot_alternatives();
989 /* Setup early PGD for fixmap */
990 create_pgd_mapping(early_pg_dir, FIXADDR_START,
991 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
993 #ifndef __PAGETABLE_PMD_FOLDED
994 /* Setup fixmap P4D and PUD */
995 if (pgtable_l5_enabled)
996 create_p4d_mapping(fixmap_p4d, FIXADDR_START,
997 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
998 /* Setup fixmap PUD and PMD */
999 if (pgtable_l4_enabled)
1000 create_pud_mapping(fixmap_pud, FIXADDR_START,
1001 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1002 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1003 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1004 /* Setup trampoline PGD and PMD */
1005 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1006 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1007 if (pgtable_l5_enabled)
1008 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1009 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1010 if (pgtable_l4_enabled)
1011 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1012 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1013 #ifdef CONFIG_XIP_KERNEL
1014 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1015 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1017 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1018 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1021 /* Setup trampoline PGD */
1022 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1023 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1027 * Setup early PGD covering entire kernel which will allow
1028 * us to reach paging_init(). We map all memory banks later
1029 * in setup_vm_final() below.
1031 create_kernel_page_table(early_pg_dir, true);
1033 /* Setup early mapping for FDT early scan */
1034 create_fdt_early_page_table(early_pg_dir, dtb_pa);
1037 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1038 * range can not span multiple pmds.
1040 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1041 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1043 #ifndef __PAGETABLE_PMD_FOLDED
1045 * Early ioremap fixmap is already created as it lies within first 2MB
1046 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1047 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1050 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1051 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1052 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1054 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1055 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1056 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1057 fix_to_virt(FIX_BTMAP_BEGIN));
1058 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1059 fix_to_virt(FIX_BTMAP_END));
1061 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1062 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1066 pt_ops_set_fixmap();
1069 static void __init setup_vm_final(void)
1071 uintptr_t va, map_size;
1072 phys_addr_t pa, start, end;
1075 /* Setup swapper PGD for fixmap */
1076 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1077 __pa_symbol(fixmap_pgd_next),
1078 PGDIR_SIZE, PAGE_TABLE);
1080 /* Map all memory banks in the linear mapping */
1081 for_each_mem_range(i, &start, &end) {
1084 if (start <= __pa(PAGE_OFFSET) &&
1085 __pa(PAGE_OFFSET) < end)
1086 start = __pa(PAGE_OFFSET);
1087 if (end >= __pa(PAGE_OFFSET) + memory_limit)
1088 end = __pa(PAGE_OFFSET) + memory_limit;
1090 map_size = best_map_size(start, end - start);
1091 for (pa = start; pa < end; pa += map_size) {
1092 va = (uintptr_t)__va(pa);
1094 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1095 pgprot_from_va(va));
1099 /* Map the kernel */
1100 if (IS_ENABLED(CONFIG_64BIT))
1101 create_kernel_page_table(swapper_pg_dir, false);
1104 kasan_swapper_init();
1107 /* Clear fixmap PTE and PMD mappings */
1108 clear_fixmap(FIX_PTE);
1109 clear_fixmap(FIX_PMD);
1110 clear_fixmap(FIX_PUD);
1111 clear_fixmap(FIX_P4D);
1113 /* Move to swapper page table */
1114 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1115 local_flush_tlb_all();
1120 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1122 dtb_early_va = (void *)dtb_pa;
1123 dtb_early_pa = dtb_pa;
1126 static inline void setup_vm_final(void)
1129 #endif /* CONFIG_MMU */
1132 * reserve_crashkernel() - reserves memory for crash kernel
1134 * This function reserves memory area given in "crashkernel=" kernel command
1135 * line parameter. The memory reserved is used by dump capture kernel when
1136 * primary kernel is crashing.
1138 static void __init reserve_crashkernel(void)
1140 unsigned long long crash_base = 0;
1141 unsigned long long crash_size = 0;
1142 unsigned long search_start = memblock_start_of_DRAM();
1143 unsigned long search_end = memblock_end_of_DRAM();
1147 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1150 * Don't reserve a region for a crash kernel on a crash kernel
1151 * since it doesn't make much sense and we have limited memory
1154 if (is_kdump_kernel()) {
1155 pr_info("crashkernel: ignoring reservation request\n");
1159 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
1160 &crash_size, &crash_base);
1161 if (ret || !crash_size)
1164 crash_size = PAGE_ALIGN(crash_size);
1167 search_start = crash_base;
1168 search_end = crash_base + crash_size;
1172 * Current riscv boot protocol requires 2MB alignment for
1173 * RV64 and 4MB alignment for RV32 (hugepage size)
1175 * Try to alloc from 32bit addressible physical memory so that
1176 * swiotlb can work on the crash kernel.
1178 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1180 min(search_end, (unsigned long) SZ_4G));
1181 if (crash_base == 0) {
1182 /* Try again without restricting region to 32bit addressible memory */
1183 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
1184 search_start, search_end);
1185 if (crash_base == 0) {
1186 pr_warn("crashkernel: couldn't allocate %lldKB\n",
1192 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
1193 crash_base, crash_base + crash_size, crash_size >> 20);
1195 crashk_res.start = crash_base;
1196 crashk_res.end = crash_base + crash_size - 1;
1199 void __init paging_init(void)
1205 void __init misc_mem_init(void)
1207 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1211 reserve_crashkernel();
1212 memblock_dump_all();
1215 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1216 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1217 struct vmem_altmap *altmap)
1219 return vmemmap_populate_basepages(start, end, node, NULL);