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.
7 #include <linux/init.h>
9 #include <linux/memblock.h>
10 #include <linux/initrd.h>
11 #include <linux/swap.h>
12 #include <linux/sizes.h>
13 #include <linux/of_fdt.h>
14 #include <linux/libfdt.h>
15 #include <linux/set_memory.h>
17 #include <asm/fixmap.h>
18 #include <asm/tlbflush.h>
19 #include <asm/sections.h>
22 #include <asm/ptdump.h>
24 #include "../kernel/head.h"
26 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
28 EXPORT_SYMBOL(empty_zero_page);
31 #define DTB_EARLY_BASE_VA PGDIR_SIZE
32 void *dtb_early_va __initdata;
33 uintptr_t dtb_early_pa __initdata;
36 pte_t *(*get_pte_virt)(phys_addr_t pa);
37 phys_addr_t (*alloc_pte)(uintptr_t va);
38 #ifndef __PAGETABLE_PMD_FOLDED
39 pmd_t *(*get_pmd_virt)(phys_addr_t pa);
40 phys_addr_t (*alloc_pmd)(uintptr_t va);
44 static void __init zone_sizes_init(void)
46 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
48 #ifdef CONFIG_ZONE_DMA32
49 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(min(4UL * SZ_1G,
50 (unsigned long) PFN_PHYS(max_low_pfn)));
52 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
54 free_area_init(max_zone_pfns);
57 static void setup_zero_page(void)
59 memset((void *)empty_zero_page, 0, PAGE_SIZE);
62 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
63 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
65 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
69 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
71 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
75 static void print_vm_layout(void)
77 pr_notice("Virtual kernel memory layout:\n");
78 print_mlk("fixmap", (unsigned long)FIXADDR_START,
79 (unsigned long)FIXADDR_TOP);
80 print_mlm("pci io", (unsigned long)PCI_IO_START,
81 (unsigned long)PCI_IO_END);
82 print_mlm("vmemmap", (unsigned long)VMEMMAP_START,
83 (unsigned long)VMEMMAP_END);
84 print_mlm("vmalloc", (unsigned long)VMALLOC_START,
85 (unsigned long)VMALLOC_END);
86 print_mlm("lowmem", (unsigned long)PAGE_OFFSET,
87 (unsigned long)high_memory);
90 static void print_vm_layout(void) { }
91 #endif /* CONFIG_DEBUG_VM */
93 void __init mem_init(void)
97 #endif /* CONFIG_FLATMEM */
99 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
102 mem_init_print_info(NULL);
106 #ifdef CONFIG_BLK_DEV_INITRD
107 static void __init setup_initrd(void)
112 /* Ignore the virtul address computed during device tree parsing */
113 initrd_start = initrd_end = 0;
115 if (!phys_initrd_size)
118 * Round the memory region to page boundaries as per free_initrd_mem()
119 * This allows us to detect whether the pages overlapping the initrd
120 * are in use, but more importantly, reserves the entire set of pages
121 * as we don't want these pages allocated for other purposes.
123 start = round_down(phys_initrd_start, PAGE_SIZE);
124 size = phys_initrd_size + (phys_initrd_start - start);
125 size = round_up(size, PAGE_SIZE);
127 if (!memblock_is_region_memory(start, size)) {
128 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
133 if (memblock_is_region_reserved(start, size)) {
134 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
139 memblock_reserve(start, size);
140 /* Now convert initrd to virtual addresses */
141 initrd_start = (unsigned long)__va(phys_initrd_start);
142 initrd_end = initrd_start + phys_initrd_size;
143 initrd_below_start_ok = 1;
145 pr_info("Initial ramdisk at: 0x%p (%lu bytes)\n",
146 (void *)(initrd_start), size);
149 pr_cont(" - disabling initrd\n");
153 #endif /* CONFIG_BLK_DEV_INITRD */
155 void __init setup_bootmem(void)
157 phys_addr_t mem_size = 0;
158 phys_addr_t total_mem = 0;
159 phys_addr_t mem_start, start, end = 0;
160 phys_addr_t vmlinux_end = __pa_symbol(&_end);
161 phys_addr_t vmlinux_start = __pa_symbol(&_start);
164 /* Find the memory region containing the kernel */
165 for_each_mem_range(i, &start, &end) {
166 phys_addr_t size = end - start;
169 if (start <= vmlinux_start && vmlinux_end <= end)
171 total_mem = total_mem + size;
175 * Remove memblock from the end of usable area to the
178 mem_size = min(total_mem, (phys_addr_t)-PAGE_OFFSET);
179 if (mem_start + mem_size < end)
180 memblock_remove(mem_start + mem_size,
181 end - mem_start - mem_size);
183 /* Reserve from the start of the kernel to the end of the kernel */
184 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
186 max_pfn = PFN_DOWN(memblock_end_of_DRAM());
187 max_low_pfn = max_pfn;
188 set_max_mapnr(max_low_pfn);
190 #ifdef CONFIG_BLK_DEV_INITRD
192 #endif /* CONFIG_BLK_DEV_INITRD */
195 * Avoid using early_init_fdt_reserve_self() since __pa() does
196 * not work for DTB pointers that are fixmap addresses
198 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
200 early_init_fdt_scan_reserved_mem();
201 memblock_allow_resize();
206 static struct pt_alloc_ops pt_ops;
208 unsigned long va_pa_offset;
209 EXPORT_SYMBOL(va_pa_offset);
210 unsigned long pfn_base;
211 EXPORT_SYMBOL(pfn_base);
213 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
214 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
215 pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
217 #define MAX_EARLY_MAPPING_SIZE SZ_128M
219 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
221 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
223 unsigned long addr = __fix_to_virt(idx);
226 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
228 ptep = &fixmap_pte[pte_index(addr)];
230 if (pgprot_val(prot))
231 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
233 pte_clear(&init_mm, addr, ptep);
234 local_flush_tlb_page(addr);
237 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
239 return (pte_t *)((uintptr_t)pa);
242 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
244 clear_fixmap(FIX_PTE);
245 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
248 static inline pte_t *get_pte_virt_late(phys_addr_t pa)
250 return (pte_t *) __va(pa);
253 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
256 * We only create PMD or PGD early mappings so we
257 * should never reach here with MMU disabled.
262 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
264 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
267 static phys_addr_t alloc_pte_late(uintptr_t va)
271 vaddr = __get_free_page(GFP_KERNEL);
272 if (!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)))
277 static void __init create_pte_mapping(pte_t *ptep,
278 uintptr_t va, phys_addr_t pa,
279 phys_addr_t sz, pgprot_t prot)
281 uintptr_t pte_idx = pte_index(va);
283 BUG_ON(sz != PAGE_SIZE);
285 if (pte_none(ptep[pte_idx]))
286 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
289 #ifndef __PAGETABLE_PMD_FOLDED
291 pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
292 pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
294 #if MAX_EARLY_MAPPING_SIZE < PGDIR_SIZE
295 #define NUM_EARLY_PMDS 1UL
297 #define NUM_EARLY_PMDS (1UL + MAX_EARLY_MAPPING_SIZE / PGDIR_SIZE)
299 pmd_t early_pmd[PTRS_PER_PMD * NUM_EARLY_PMDS] __initdata __aligned(PAGE_SIZE);
301 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
303 /* Before MMU is enabled */
304 return (pmd_t *)((uintptr_t)pa);
307 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
309 clear_fixmap(FIX_PMD);
310 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
313 static pmd_t *get_pmd_virt_late(phys_addr_t pa)
315 return (pmd_t *) __va(pa);
318 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
322 pmd_num = (va - PAGE_OFFSET) >> PGDIR_SHIFT;
323 BUG_ON(pmd_num >= NUM_EARLY_PMDS);
324 return (uintptr_t)&early_pmd[pmd_num * PTRS_PER_PMD];
327 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
329 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
332 static phys_addr_t alloc_pmd_late(uintptr_t va)
336 vaddr = __get_free_page(GFP_KERNEL);
341 static void __init create_pmd_mapping(pmd_t *pmdp,
342 uintptr_t va, phys_addr_t pa,
343 phys_addr_t sz, pgprot_t prot)
346 phys_addr_t pte_phys;
347 uintptr_t pmd_idx = pmd_index(va);
349 if (sz == PMD_SIZE) {
350 if (pmd_none(pmdp[pmd_idx]))
351 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
355 if (pmd_none(pmdp[pmd_idx])) {
356 pte_phys = pt_ops.alloc_pte(va);
357 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
358 ptep = pt_ops.get_pte_virt(pte_phys);
359 memset(ptep, 0, PAGE_SIZE);
361 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
362 ptep = pt_ops.get_pte_virt(pte_phys);
365 create_pte_mapping(ptep, va, pa, sz, prot);
368 #define pgd_next_t pmd_t
369 #define alloc_pgd_next(__va) pt_ops.alloc_pmd(__va)
370 #define get_pgd_next_virt(__pa) pt_ops.get_pmd_virt(__pa)
371 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
372 create_pmd_mapping(__nextp, __va, __pa, __sz, __prot)
373 #define fixmap_pgd_next fixmap_pmd
375 #define pgd_next_t pte_t
376 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
377 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
378 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
379 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
380 #define fixmap_pgd_next fixmap_pte
383 void __init create_pgd_mapping(pgd_t *pgdp,
384 uintptr_t va, phys_addr_t pa,
385 phys_addr_t sz, pgprot_t prot)
388 phys_addr_t next_phys;
389 uintptr_t pgd_idx = pgd_index(va);
391 if (sz == PGDIR_SIZE) {
392 if (pgd_val(pgdp[pgd_idx]) == 0)
393 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
397 if (pgd_val(pgdp[pgd_idx]) == 0) {
398 next_phys = alloc_pgd_next(va);
399 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
400 nextp = get_pgd_next_virt(next_phys);
401 memset(nextp, 0, PAGE_SIZE);
403 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
404 nextp = get_pgd_next_virt(next_phys);
407 create_pgd_next_mapping(nextp, va, pa, sz, prot);
410 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
412 /* Upgrade to PMD_SIZE mappings whenever possible */
413 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
420 * setup_vm() is called from head.S with MMU-off.
422 * Following requirements should be honoured for setup_vm() to work
424 * 1) It should use PC-relative addressing for accessing kernel symbols.
425 * To achieve this we always use GCC cmodel=medany.
426 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
427 * so disable compiler instrumentation when FTRACE is enabled.
429 * Currently, the above requirements are honoured by using custom CFLAGS
430 * for init.o in mm/Makefile.
433 #ifndef __riscv_cmodel_medany
434 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
437 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
439 uintptr_t va, pa, end_va;
440 uintptr_t load_pa = (uintptr_t)(&_start);
441 uintptr_t load_sz = (uintptr_t)(&_end) - load_pa;
442 uintptr_t map_size = best_map_size(load_pa, MAX_EARLY_MAPPING_SIZE);
443 #ifndef __PAGETABLE_PMD_FOLDED
444 pmd_t fix_bmap_spmd, fix_bmap_epmd;
447 va_pa_offset = PAGE_OFFSET - load_pa;
448 pfn_base = PFN_DOWN(load_pa);
451 * Enforce boot alignment requirements of RV32 and
452 * RV64 by only allowing PMD or PGD mappings.
454 BUG_ON(map_size == PAGE_SIZE);
456 /* Sanity check alignment and size */
457 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
458 BUG_ON((load_pa % map_size) != 0);
459 BUG_ON(load_sz > MAX_EARLY_MAPPING_SIZE);
461 pt_ops.alloc_pte = alloc_pte_early;
462 pt_ops.get_pte_virt = get_pte_virt_early;
463 #ifndef __PAGETABLE_PMD_FOLDED
464 pt_ops.alloc_pmd = alloc_pmd_early;
465 pt_ops.get_pmd_virt = get_pmd_virt_early;
467 /* Setup early PGD for fixmap */
468 create_pgd_mapping(early_pg_dir, FIXADDR_START,
469 (uintptr_t)fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
471 #ifndef __PAGETABLE_PMD_FOLDED
472 /* Setup fixmap PMD */
473 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
474 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
475 /* Setup trampoline PGD and PMD */
476 create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
477 (uintptr_t)trampoline_pmd, PGDIR_SIZE, PAGE_TABLE);
478 create_pmd_mapping(trampoline_pmd, PAGE_OFFSET,
479 load_pa, PMD_SIZE, PAGE_KERNEL_EXEC);
481 /* Setup trampoline PGD */
482 create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
483 load_pa, PGDIR_SIZE, PAGE_KERNEL_EXEC);
487 * Setup early PGD covering entire kernel which will allows
488 * us to reach paging_init(). We map all memory banks later
489 * in setup_vm_final() below.
491 end_va = PAGE_OFFSET + load_sz;
492 for (va = PAGE_OFFSET; va < end_va; va += map_size)
493 create_pgd_mapping(early_pg_dir, va,
494 load_pa + (va - PAGE_OFFSET),
495 map_size, PAGE_KERNEL_EXEC);
497 /* Create two consecutive PGD mappings for FDT early scan */
498 pa = dtb_pa & ~(PGDIR_SIZE - 1);
499 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
500 pa, PGDIR_SIZE, PAGE_KERNEL);
501 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA + PGDIR_SIZE,
502 pa + PGDIR_SIZE, PGDIR_SIZE, PAGE_KERNEL);
503 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PGDIR_SIZE - 1));
504 dtb_early_pa = dtb_pa;
507 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
508 * range can not span multiple pmds.
510 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
511 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
513 #ifndef __PAGETABLE_PMD_FOLDED
515 * Early ioremap fixmap is already created as it lies within first 2MB
516 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
517 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
520 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
521 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
522 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
524 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
525 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
526 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
527 fix_to_virt(FIX_BTMAP_BEGIN));
528 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
529 fix_to_virt(FIX_BTMAP_END));
531 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
532 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
537 static void __init setup_vm_final(void)
539 uintptr_t va, map_size;
540 phys_addr_t pa, start, end;
544 * MMU is enabled at this point. But page table setup is not complete yet.
545 * fixmap page table alloc functions should be used at this point
547 pt_ops.alloc_pte = alloc_pte_fixmap;
548 pt_ops.get_pte_virt = get_pte_virt_fixmap;
549 #ifndef __PAGETABLE_PMD_FOLDED
550 pt_ops.alloc_pmd = alloc_pmd_fixmap;
551 pt_ops.get_pmd_virt = get_pmd_virt_fixmap;
553 /* Setup swapper PGD for fixmap */
554 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
555 __pa_symbol(fixmap_pgd_next),
556 PGDIR_SIZE, PAGE_TABLE);
558 /* Map all memory banks */
559 for_each_mem_range(i, &start, &end) {
562 if (start <= __pa(PAGE_OFFSET) &&
563 __pa(PAGE_OFFSET) < end)
564 start = __pa(PAGE_OFFSET);
566 map_size = best_map_size(start, end - start);
567 for (pa = start; pa < end; pa += map_size) {
568 va = (uintptr_t)__va(pa);
569 create_pgd_mapping(swapper_pg_dir, va, pa,
570 map_size, PAGE_KERNEL_EXEC);
574 /* Clear fixmap PTE and PMD mappings */
575 clear_fixmap(FIX_PTE);
576 clear_fixmap(FIX_PMD);
578 /* Move to swapper page table */
579 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE);
580 local_flush_tlb_all();
582 /* generic page allocation functions must be used to setup page table */
583 pt_ops.alloc_pte = alloc_pte_late;
584 pt_ops.get_pte_virt = get_pte_virt_late;
585 #ifndef __PAGETABLE_PMD_FOLDED
586 pt_ops.alloc_pmd = alloc_pmd_late;
587 pt_ops.get_pmd_virt = get_pmd_virt_late;
591 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
593 #ifdef CONFIG_BUILTIN_DTB
594 dtb_early_va = soc_lookup_builtin_dtb();
596 /* Fallback to first available DTS */
597 dtb_early_va = (void *) __dtb_start;
600 dtb_early_va = (void *)dtb_pa;
602 dtb_early_pa = dtb_pa;
605 static inline void setup_vm_final(void)
608 #endif /* CONFIG_MMU */
610 #ifdef CONFIG_STRICT_KERNEL_RWX
611 void mark_rodata_ro(void)
613 unsigned long text_start = (unsigned long)_text;
614 unsigned long text_end = (unsigned long)_etext;
615 unsigned long rodata_start = (unsigned long)__start_rodata;
616 unsigned long data_start = (unsigned long)_data;
617 unsigned long max_low = (unsigned long)(__va(PFN_PHYS(max_low_pfn)));
619 set_memory_ro(text_start, (text_end - text_start) >> PAGE_SHIFT);
620 set_memory_ro(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
621 set_memory_nx(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
622 set_memory_nx(data_start, (max_low - data_start) >> PAGE_SHIFT);
628 static void __init resource_init(void)
630 struct memblock_region *region;
632 for_each_mem_region(region) {
633 struct resource *res;
635 res = memblock_alloc(sizeof(struct resource), SMP_CACHE_BYTES);
637 panic("%s: Failed to allocate %zu bytes\n", __func__,
638 sizeof(struct resource));
640 if (memblock_is_nomap(region)) {
641 res->name = "reserved";
642 res->flags = IORESOURCE_MEM;
644 res->name = "System RAM";
645 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
647 res->start = __pfn_to_phys(memblock_region_memory_base_pfn(region));
648 res->end = __pfn_to_phys(memblock_region_memory_end_pfn(region)) - 1;
650 request_resource(&iomem_resource, res);
654 void __init paging_init(void)
663 #ifdef CONFIG_SPARSEMEM_VMEMMAP
664 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
665 struct vmem_altmap *altmap)
667 return vmemmap_populate_basepages(start, end, node, NULL);