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>
16 #include <linux/dma-map-ops.h>
18 #include <asm/fixmap.h>
19 #include <asm/tlbflush.h>
20 #include <asm/sections.h>
23 #include <asm/ptdump.h>
26 #include "../kernel/head.h"
28 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
30 EXPORT_SYMBOL(empty_zero_page);
33 #define DTB_EARLY_BASE_VA PGDIR_SIZE
34 void *dtb_early_va __initdata;
35 uintptr_t dtb_early_pa __initdata;
38 pte_t *(*get_pte_virt)(phys_addr_t pa);
39 phys_addr_t (*alloc_pte)(uintptr_t va);
40 #ifndef __PAGETABLE_PMD_FOLDED
41 pmd_t *(*get_pmd_virt)(phys_addr_t pa);
42 phys_addr_t (*alloc_pmd)(uintptr_t va);
46 static phys_addr_t dma32_phys_limit __ro_after_init;
48 static void __init zone_sizes_init(void)
50 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
52 #ifdef CONFIG_ZONE_DMA32
53 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
55 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
57 free_area_init(max_zone_pfns);
60 static void setup_zero_page(void)
62 memset((void *)empty_zero_page, 0, PAGE_SIZE);
65 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
66 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
68 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
72 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
74 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
78 static void print_vm_layout(void)
80 pr_notice("Virtual kernel memory layout:\n");
81 print_mlk("fixmap", (unsigned long)FIXADDR_START,
82 (unsigned long)FIXADDR_TOP);
83 print_mlm("pci io", (unsigned long)PCI_IO_START,
84 (unsigned long)PCI_IO_END);
85 print_mlm("vmemmap", (unsigned long)VMEMMAP_START,
86 (unsigned long)VMEMMAP_END);
87 print_mlm("vmalloc", (unsigned long)VMALLOC_START,
88 (unsigned long)VMALLOC_END);
89 print_mlm("lowmem", (unsigned long)PAGE_OFFSET,
90 (unsigned long)high_memory);
93 static void print_vm_layout(void) { }
94 #endif /* CONFIG_DEBUG_VM */
96 void __init mem_init(void)
100 #endif /* CONFIG_FLATMEM */
102 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
105 mem_init_print_info(NULL);
109 void __init setup_bootmem(void)
111 phys_addr_t vmlinux_end = __pa_symbol(&_end);
112 phys_addr_t vmlinux_start = __pa_symbol(&_start);
113 phys_addr_t dram_end = memblock_end_of_DRAM();
114 phys_addr_t max_mapped_addr = __pa(~(ulong)0);
116 /* The maximal physical memory size is -PAGE_OFFSET. */
117 memblock_enforce_memory_limit(-PAGE_OFFSET);
119 /* Reserve from the start of the kernel to the end of the kernel */
120 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
123 * memblock allocator is not aware of the fact that last 4K bytes of
124 * the addressable memory can not be mapped because of IS_ERR_VALUE
125 * macro. Make sure that last 4k bytes are not usable by memblock
126 * if end of dram is equal to maximum addressable memory.
128 if (max_mapped_addr == (dram_end - 1))
129 memblock_set_current_limit(max_mapped_addr - 4096);
131 max_pfn = PFN_DOWN(dram_end);
132 max_low_pfn = max_pfn;
133 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
134 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
136 reserve_initrd_mem();
138 * If DTB is built in, no need to reserve its memblock.
139 * Otherwise, do reserve it but avoid using
140 * early_init_fdt_reserve_self() since __pa() does
141 * not work for DTB pointers that are fixmap addresses
143 if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
144 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
146 early_init_fdt_scan_reserved_mem();
147 dma_contiguous_reserve(dma32_phys_limit);
148 memblock_allow_resize();
152 static struct pt_alloc_ops pt_ops;
154 unsigned long va_pa_offset;
155 EXPORT_SYMBOL(va_pa_offset);
156 unsigned long pfn_base;
157 EXPORT_SYMBOL(pfn_base);
159 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
160 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
161 pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
163 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
165 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
167 unsigned long addr = __fix_to_virt(idx);
170 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
172 ptep = &fixmap_pte[pte_index(addr)];
174 if (pgprot_val(prot))
175 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
177 pte_clear(&init_mm, addr, ptep);
178 local_flush_tlb_page(addr);
181 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
183 return (pte_t *)((uintptr_t)pa);
186 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
188 clear_fixmap(FIX_PTE);
189 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
192 static inline pte_t *get_pte_virt_late(phys_addr_t pa)
194 return (pte_t *) __va(pa);
197 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
200 * We only create PMD or PGD early mappings so we
201 * should never reach here with MMU disabled.
206 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
208 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
211 static phys_addr_t alloc_pte_late(uintptr_t va)
215 vaddr = __get_free_page(GFP_KERNEL);
216 if (!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)))
221 static void __init create_pte_mapping(pte_t *ptep,
222 uintptr_t va, phys_addr_t pa,
223 phys_addr_t sz, pgprot_t prot)
225 uintptr_t pte_idx = pte_index(va);
227 BUG_ON(sz != PAGE_SIZE);
229 if (pte_none(ptep[pte_idx]))
230 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
233 #ifndef __PAGETABLE_PMD_FOLDED
235 pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
236 pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
237 pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
238 pmd_t early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
240 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
242 /* Before MMU is enabled */
243 return (pmd_t *)((uintptr_t)pa);
246 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
248 clear_fixmap(FIX_PMD);
249 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
252 static pmd_t *get_pmd_virt_late(phys_addr_t pa)
254 return (pmd_t *) __va(pa);
257 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
259 BUG_ON((va - PAGE_OFFSET) >> PGDIR_SHIFT);
261 return (uintptr_t)early_pmd;
264 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
266 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
269 static phys_addr_t alloc_pmd_late(uintptr_t va)
273 vaddr = __get_free_page(GFP_KERNEL);
278 static void __init create_pmd_mapping(pmd_t *pmdp,
279 uintptr_t va, phys_addr_t pa,
280 phys_addr_t sz, pgprot_t prot)
283 phys_addr_t pte_phys;
284 uintptr_t pmd_idx = pmd_index(va);
286 if (sz == PMD_SIZE) {
287 if (pmd_none(pmdp[pmd_idx]))
288 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
292 if (pmd_none(pmdp[pmd_idx])) {
293 pte_phys = pt_ops.alloc_pte(va);
294 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
295 ptep = pt_ops.get_pte_virt(pte_phys);
296 memset(ptep, 0, PAGE_SIZE);
298 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
299 ptep = pt_ops.get_pte_virt(pte_phys);
302 create_pte_mapping(ptep, va, pa, sz, prot);
305 #define pgd_next_t pmd_t
306 #define alloc_pgd_next(__va) pt_ops.alloc_pmd(__va)
307 #define get_pgd_next_virt(__pa) pt_ops.get_pmd_virt(__pa)
308 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
309 create_pmd_mapping(__nextp, __va, __pa, __sz, __prot)
310 #define fixmap_pgd_next fixmap_pmd
312 #define pgd_next_t pte_t
313 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
314 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
315 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
316 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
317 #define fixmap_pgd_next fixmap_pte
320 void __init create_pgd_mapping(pgd_t *pgdp,
321 uintptr_t va, phys_addr_t pa,
322 phys_addr_t sz, pgprot_t prot)
325 phys_addr_t next_phys;
326 uintptr_t pgd_idx = pgd_index(va);
328 if (sz == PGDIR_SIZE) {
329 if (pgd_val(pgdp[pgd_idx]) == 0)
330 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
334 if (pgd_val(pgdp[pgd_idx]) == 0) {
335 next_phys = alloc_pgd_next(va);
336 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
337 nextp = get_pgd_next_virt(next_phys);
338 memset(nextp, 0, PAGE_SIZE);
340 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
341 nextp = get_pgd_next_virt(next_phys);
344 create_pgd_next_mapping(nextp, va, pa, sz, prot);
347 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
349 /* Upgrade to PMD_SIZE mappings whenever possible */
350 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
357 * setup_vm() is called from head.S with MMU-off.
359 * Following requirements should be honoured for setup_vm() to work
361 * 1) It should use PC-relative addressing for accessing kernel symbols.
362 * To achieve this we always use GCC cmodel=medany.
363 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
364 * so disable compiler instrumentation when FTRACE is enabled.
366 * Currently, the above requirements are honoured by using custom CFLAGS
367 * for init.o in mm/Makefile.
370 #ifndef __riscv_cmodel_medany
371 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
374 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
376 uintptr_t va, pa, end_va;
377 uintptr_t load_pa = (uintptr_t)(&_start);
378 uintptr_t load_sz = (uintptr_t)(&_end) - load_pa;
380 #ifndef __PAGETABLE_PMD_FOLDED
381 pmd_t fix_bmap_spmd, fix_bmap_epmd;
384 va_pa_offset = PAGE_OFFSET - load_pa;
385 pfn_base = PFN_DOWN(load_pa);
388 * Enforce boot alignment requirements of RV32 and
389 * RV64 by only allowing PMD or PGD mappings.
393 /* Sanity check alignment and size */
394 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
395 BUG_ON((load_pa % map_size) != 0);
397 pt_ops.alloc_pte = alloc_pte_early;
398 pt_ops.get_pte_virt = get_pte_virt_early;
399 #ifndef __PAGETABLE_PMD_FOLDED
400 pt_ops.alloc_pmd = alloc_pmd_early;
401 pt_ops.get_pmd_virt = get_pmd_virt_early;
403 /* Setup early PGD for fixmap */
404 create_pgd_mapping(early_pg_dir, FIXADDR_START,
405 (uintptr_t)fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
407 #ifndef __PAGETABLE_PMD_FOLDED
408 /* Setup fixmap PMD */
409 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
410 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
411 /* Setup trampoline PGD and PMD */
412 create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
413 (uintptr_t)trampoline_pmd, PGDIR_SIZE, PAGE_TABLE);
414 create_pmd_mapping(trampoline_pmd, PAGE_OFFSET,
415 load_pa, PMD_SIZE, PAGE_KERNEL_EXEC);
417 /* Setup trampoline PGD */
418 create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
419 load_pa, PGDIR_SIZE, PAGE_KERNEL_EXEC);
423 * Setup early PGD covering entire kernel which will allows
424 * us to reach paging_init(). We map all memory banks later
425 * in setup_vm_final() below.
427 end_va = PAGE_OFFSET + load_sz;
428 for (va = PAGE_OFFSET; va < end_va; va += map_size)
429 create_pgd_mapping(early_pg_dir, va,
430 load_pa + (va - PAGE_OFFSET),
431 map_size, PAGE_KERNEL_EXEC);
433 #ifndef __PAGETABLE_PMD_FOLDED
434 /* Setup early PMD for DTB */
435 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
436 (uintptr_t)early_dtb_pmd, PGDIR_SIZE, PAGE_TABLE);
437 #ifndef CONFIG_BUILTIN_DTB
438 /* Create two consecutive PMD mappings for FDT early scan */
439 pa = dtb_pa & ~(PMD_SIZE - 1);
440 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA,
441 pa, PMD_SIZE, PAGE_KERNEL);
442 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE,
443 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
444 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1));
445 #else /* CONFIG_BUILTIN_DTB */
446 dtb_early_va = __va(dtb_pa);
447 #endif /* CONFIG_BUILTIN_DTB */
449 #ifndef CONFIG_BUILTIN_DTB
450 /* Create two consecutive PGD mappings for FDT early scan */
451 pa = dtb_pa & ~(PGDIR_SIZE - 1);
452 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
453 pa, PGDIR_SIZE, PAGE_KERNEL);
454 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA + PGDIR_SIZE,
455 pa + PGDIR_SIZE, PGDIR_SIZE, PAGE_KERNEL);
456 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PGDIR_SIZE - 1));
457 #else /* CONFIG_BUILTIN_DTB */
458 dtb_early_va = __va(dtb_pa);
459 #endif /* CONFIG_BUILTIN_DTB */
461 dtb_early_pa = dtb_pa;
464 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
465 * range can not span multiple pmds.
467 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
468 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
470 #ifndef __PAGETABLE_PMD_FOLDED
472 * Early ioremap fixmap is already created as it lies within first 2MB
473 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
474 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
477 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
478 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
479 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
481 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
482 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
483 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
484 fix_to_virt(FIX_BTMAP_BEGIN));
485 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
486 fix_to_virt(FIX_BTMAP_END));
488 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
489 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
494 static void __init setup_vm_final(void)
496 uintptr_t va, map_size;
497 phys_addr_t pa, start, end;
501 * MMU is enabled at this point. But page table setup is not complete yet.
502 * fixmap page table alloc functions should be used at this point
504 pt_ops.alloc_pte = alloc_pte_fixmap;
505 pt_ops.get_pte_virt = get_pte_virt_fixmap;
506 #ifndef __PAGETABLE_PMD_FOLDED
507 pt_ops.alloc_pmd = alloc_pmd_fixmap;
508 pt_ops.get_pmd_virt = get_pmd_virt_fixmap;
510 /* Setup swapper PGD for fixmap */
511 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
512 __pa_symbol(fixmap_pgd_next),
513 PGDIR_SIZE, PAGE_TABLE);
515 /* Map all memory banks */
516 for_each_mem_range(i, &start, &end) {
519 if (start <= __pa(PAGE_OFFSET) &&
520 __pa(PAGE_OFFSET) < end)
521 start = __pa(PAGE_OFFSET);
523 map_size = best_map_size(start, end - start);
524 for (pa = start; pa < end; pa += map_size) {
525 va = (uintptr_t)__va(pa);
526 create_pgd_mapping(swapper_pg_dir, va, pa,
527 map_size, PAGE_KERNEL_EXEC);
531 /* Clear fixmap PTE and PMD mappings */
532 clear_fixmap(FIX_PTE);
533 clear_fixmap(FIX_PMD);
535 /* Move to swapper page table */
536 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE);
537 local_flush_tlb_all();
539 /* generic page allocation functions must be used to setup page table */
540 pt_ops.alloc_pte = alloc_pte_late;
541 pt_ops.get_pte_virt = get_pte_virt_late;
542 #ifndef __PAGETABLE_PMD_FOLDED
543 pt_ops.alloc_pmd = alloc_pmd_late;
544 pt_ops.get_pmd_virt = get_pmd_virt_late;
548 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
550 dtb_early_va = (void *)dtb_pa;
551 dtb_early_pa = dtb_pa;
554 static inline void setup_vm_final(void)
557 #endif /* CONFIG_MMU */
559 #ifdef CONFIG_STRICT_KERNEL_RWX
560 void protect_kernel_text_data(void)
562 unsigned long text_start = (unsigned long)_start;
563 unsigned long init_text_start = (unsigned long)__init_text_begin;
564 unsigned long init_data_start = (unsigned long)__init_data_begin;
565 unsigned long rodata_start = (unsigned long)__start_rodata;
566 unsigned long data_start = (unsigned long)_data;
567 unsigned long max_low = (unsigned long)(__va(PFN_PHYS(max_low_pfn)));
569 set_memory_ro(text_start, (init_text_start - text_start) >> PAGE_SHIFT);
570 set_memory_ro(init_text_start, (init_data_start - init_text_start) >> PAGE_SHIFT);
571 set_memory_nx(init_data_start, (rodata_start - init_data_start) >> PAGE_SHIFT);
572 /* rodata section is marked readonly in mark_rodata_ro */
573 set_memory_nx(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
574 set_memory_nx(data_start, (max_low - data_start) >> PAGE_SHIFT);
577 void mark_rodata_ro(void)
579 unsigned long rodata_start = (unsigned long)__start_rodata;
580 unsigned long data_start = (unsigned long)_data;
582 set_memory_ro(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
588 void __init paging_init(void)
594 void __init misc_mem_init(void)
602 #ifdef CONFIG_SPARSEMEM_VMEMMAP
603 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
604 struct vmem_altmap *altmap)
606 return vmemmap_populate_basepages(start, end, node, NULL);