1// SPDX-License-Identifier: GPL-2.0-only
2/*
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>
7 */
8
9#include <linux/init.h>
10#include <linux/mm.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>
23#ifdef CONFIG_RELOCATABLE
24#include <linux/elf.h>
25#endif
26#include <linux/kfence.h>
27
28#include <asm/fixmap.h>
29#include <asm/io.h>
30#include <asm/numa.h>
31#include <asm/pgtable.h>
32#include <asm/sections.h>
33#include <asm/soc.h>
34#include <asm/tlbflush.h>
35
36#include "../kernel/head.h"
37
38struct kernel_mapping kernel_map __ro_after_init;
39EXPORT_SYMBOL(kernel_map);
40#ifdef CONFIG_XIP_KERNEL
41#define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
42#endif
43
44#ifdef CONFIG_64BIT
45u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39;
46#else
47u64 satp_mode __ro_after_init = SATP_MODE_32;
48#endif
49EXPORT_SYMBOL(satp_mode);
50
51#ifdef CONFIG_64BIT
52bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
53bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
54EXPORT_SYMBOL(pgtable_l4_enabled);
55EXPORT_SYMBOL(pgtable_l5_enabled);
56#endif
57
58phys_addr_t phys_ram_base __ro_after_init;
59EXPORT_SYMBOL(phys_ram_base);
60
61unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
62 __page_aligned_bss;
63EXPORT_SYMBOL(empty_zero_page);
64
65extern char _start[];
66void *_dtb_early_va __initdata;
67uintptr_t _dtb_early_pa __initdata;
68
69phys_addr_t dma32_phys_limit __initdata;
70
71static void __init zone_sizes_init(void)
72{
73 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
74
75#ifdef CONFIG_ZONE_DMA32
76 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
77#endif
78 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
79
80 free_area_init(max_zone_pfn: max_zone_pfns);
81}
82
83#if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
84
85#define LOG2_SZ_1K ilog2(SZ_1K)
86#define LOG2_SZ_1M ilog2(SZ_1M)
87#define LOG2_SZ_1G ilog2(SZ_1G)
88#define LOG2_SZ_1T ilog2(SZ_1T)
89
90static inline void print_mlk(char *name, unsigned long b, unsigned long t)
91{
92 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
93 (((t) - (b)) >> LOG2_SZ_1K));
94}
95
96static inline void print_mlm(char *name, unsigned long b, unsigned long t)
97{
98 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
99 (((t) - (b)) >> LOG2_SZ_1M));
100}
101
102static inline void print_mlg(char *name, unsigned long b, unsigned long t)
103{
104 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t,
105 (((t) - (b)) >> LOG2_SZ_1G));
106}
107
108#ifdef CONFIG_64BIT
109static inline void print_mlt(char *name, unsigned long b, unsigned long t)
110{
111 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t,
112 (((t) - (b)) >> LOG2_SZ_1T));
113}
114#else
115#define print_mlt(n, b, t) do {} while (0)
116#endif
117
118static inline void print_ml(char *name, unsigned long b, unsigned long t)
119{
120 unsigned long diff = t - b;
121
122 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10)
123 print_mlt(name, b, t);
124 else if ((diff >> LOG2_SZ_1G) >= 10)
125 print_mlg(name, b, t);
126 else if ((diff >> LOG2_SZ_1M) >= 10)
127 print_mlm(name, b, t);
128 else
129 print_mlk(name, b, t);
130}
131
132static void __init print_vm_layout(void)
133{
134 pr_notice("Virtual kernel memory layout:\n");
135 print_ml(name: "fixmap", b: (unsigned long)FIXADDR_START,
136 t: (unsigned long)FIXADDR_TOP);
137 print_ml("pci io", (unsigned long)PCI_IO_START,
138 (unsigned long)PCI_IO_END);
139 print_ml("vmemmap", (unsigned long)VMEMMAP_START,
140 (unsigned long)VMEMMAP_END);
141 print_ml(name: "vmalloc", b: (unsigned long)VMALLOC_START,
142 t: (unsigned long)VMALLOC_END);
143#ifdef CONFIG_64BIT
144 print_ml(name: "modules", b: (unsigned long)MODULES_VADDR,
145 t: (unsigned long)MODULES_END);
146#endif
147 print_ml(name: "lowmem", b: (unsigned long)PAGE_OFFSET,
148 t: (unsigned long)high_memory);
149 if (IS_ENABLED(CONFIG_64BIT)) {
150#ifdef CONFIG_KASAN
151 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
152#endif
153
154 print_ml("kernel", (unsigned long)kernel_map.virt_addr,
155 (unsigned long)ADDRESS_SPACE_END);
156 }
157}
158#else
159static void print_vm_layout(void) { }
160#endif /* CONFIG_DEBUG_VM */
161
162void __init mem_init(void)
163{
164#ifdef CONFIG_FLATMEM
165 BUG_ON(!mem_map);
166#endif /* CONFIG_FLATMEM */
167
168 swiotlb_init(addressing_limited: max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE);
169 memblock_free_all();
170
171 print_vm_layout();
172}
173
174/* Limit the memory size via mem. */
175static phys_addr_t memory_limit;
176#ifdef CONFIG_XIP_KERNEL
177#define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit))
178#endif /* CONFIG_XIP_KERNEL */
179
180static int __init early_mem(char *p)
181{
182 u64 size;
183
184 if (!p)
185 return 1;
186
187 size = memparse(ptr: p, retptr: &p) & PAGE_MASK;
188 memory_limit = min_t(u64, size, memory_limit);
189
190 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
191
192 return 0;
193}
194early_param("mem", early_mem);
195
196static void __init setup_bootmem(void)
197{
198 phys_addr_t vmlinux_end = __pa_symbol(&_end);
199 phys_addr_t max_mapped_addr;
200 phys_addr_t phys_ram_end, vmlinux_start;
201
202 if (IS_ENABLED(CONFIG_XIP_KERNEL))
203 vmlinux_start = __pa_symbol(&_sdata);
204 else
205 vmlinux_start = __pa_symbol(&_start);
206
207 memblock_enforce_memory_limit(memory_limit);
208
209 /*
210 * Make sure we align the reservation on PMD_SIZE since we will
211 * map the kernel in the linear mapping as read-only: we do not want
212 * any allocation to happen between _end and the next pmd aligned page.
213 */
214 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
215 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
216 /*
217 * Reserve from the start of the kernel to the end of the kernel
218 */
219 memblock_reserve(base: vmlinux_start, size: vmlinux_end - vmlinux_start);
220
221 phys_ram_end = memblock_end_of_DRAM();
222
223 /*
224 * Make sure we align the start of the memory on a PMD boundary so that
225 * at worst, we map the linear mapping with PMD mappings.
226 */
227 if (!IS_ENABLED(CONFIG_XIP_KERNEL))
228 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK;
229
230 /*
231 * In 64-bit, any use of __va/__pa before this point is wrong as we
232 * did not know the start of DRAM before.
233 */
234 if (IS_ENABLED(CONFIG_64BIT))
235 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base;
236
237 /*
238 * memblock allocator is not aware of the fact that last 4K bytes of
239 * the addressable memory can not be mapped because of IS_ERR_VALUE
240 * macro. Make sure that last 4k bytes are not usable by memblock
241 * if end of dram is equal to maximum addressable memory. For 64-bit
242 * kernel, this problem can't happen here as the end of the virtual
243 * address space is occupied by the kernel mapping then this check must
244 * be done as soon as the kernel mapping base address is determined.
245 */
246 if (!IS_ENABLED(CONFIG_64BIT)) {
247 max_mapped_addr = __pa(~(ulong)0);
248 if (max_mapped_addr == (phys_ram_end - 1))
249 memblock_set_current_limit(limit: max_mapped_addr - 4096);
250 }
251
252 min_low_pfn = PFN_UP(phys_ram_base);
253 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
254 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
255
256 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
257 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
258
259 reserve_initrd_mem();
260
261 /*
262 * No allocation should be done before reserving the memory as defined
263 * in the device tree, otherwise the allocation could end up in a
264 * reserved region.
265 */
266 early_init_fdt_scan_reserved_mem();
267
268 /*
269 * If DTB is built in, no need to reserve its memblock.
270 * Otherwise, do reserve it but avoid using
271 * early_init_fdt_reserve_self() since __pa() does
272 * not work for DTB pointers that are fixmap addresses
273 */
274 if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
275 memblock_reserve(base: dtb_early_pa, fdt_totalsize(dtb_early_va));
276
277 dma_contiguous_reserve(addr_limit: dma32_phys_limit);
278 if (IS_ENABLED(CONFIG_64BIT))
279 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
280}
281
282#ifdef CONFIG_MMU
283struct pt_alloc_ops pt_ops __initdata;
284
285pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
286pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
287static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
288
289pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
290
291#ifdef CONFIG_XIP_KERNEL
292#define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
293#define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
294#define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
295#define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
296#endif /* CONFIG_XIP_KERNEL */
297
298static const pgprot_t protection_map[16] = {
299 [VM_NONE] = PAGE_NONE,
300 [VM_READ] = PAGE_READ,
301 [VM_WRITE] = PAGE_COPY,
302 [VM_WRITE | VM_READ] = PAGE_COPY,
303 [VM_EXEC] = PAGE_EXEC,
304 [VM_EXEC | VM_READ] = PAGE_READ_EXEC,
305 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC,
306 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC,
307 [VM_SHARED] = PAGE_NONE,
308 [VM_SHARED | VM_READ] = PAGE_READ,
309 [VM_SHARED | VM_WRITE] = PAGE_SHARED,
310 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED,
311 [VM_SHARED | VM_EXEC] = PAGE_EXEC,
312 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC,
313 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC,
314 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC
315};
316DECLARE_VM_GET_PAGE_PROT
317
318void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
319{
320 unsigned long addr = __fix_to_virt(idx);
321 pte_t *ptep;
322
323 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
324
325 ptep = &fixmap_pte[pte_index(address: addr)];
326
327 if (pgprot_val(prot))
328 set_pte(ptep, pte: pfn_pte(page_nr: phys >> PAGE_SHIFT, pgprot: prot));
329 else
330 pte_clear(mm: &init_mm, addr, ptep);
331 local_flush_tlb_page(addr);
332}
333
334static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
335{
336 return (pte_t *)((uintptr_t)pa);
337}
338
339static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
340{
341 clear_fixmap(FIX_PTE);
342 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
343}
344
345static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
346{
347 return (pte_t *) __va(pa);
348}
349
350static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
351{
352 /*
353 * We only create PMD or PGD early mappings so we
354 * should never reach here with MMU disabled.
355 */
356 BUG();
357}
358
359static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
360{
361 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
362}
363
364static phys_addr_t __init alloc_pte_late(uintptr_t va)
365{
366 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, order: 0);
367
368 BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc));
369 return __pa((pte_t *)ptdesc_address(ptdesc));
370}
371
372static void __init create_pte_mapping(pte_t *ptep,
373 uintptr_t va, phys_addr_t pa,
374 phys_addr_t sz, pgprot_t prot)
375{
376 uintptr_t pte_idx = pte_index(address: va);
377
378 BUG_ON(sz != PAGE_SIZE);
379
380 if (pte_none(pte: ptep[pte_idx]))
381 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), pgprot: prot);
382}
383
384#ifndef __PAGETABLE_PMD_FOLDED
385
386static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
387static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
388static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
389
390#ifdef CONFIG_XIP_KERNEL
391#define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
392#define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
393#define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
394#endif /* CONFIG_XIP_KERNEL */
395
396static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
397static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
398static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
399
400#ifdef CONFIG_XIP_KERNEL
401#define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
402#define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d))
403#define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d))
404#endif /* CONFIG_XIP_KERNEL */
405
406static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
407static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
408static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
409
410#ifdef CONFIG_XIP_KERNEL
411#define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
412#define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
413#define early_pud ((pud_t *)XIP_FIXUP(early_pud))
414#endif /* CONFIG_XIP_KERNEL */
415
416static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
417{
418 /* Before MMU is enabled */
419 return (pmd_t *)((uintptr_t)pa);
420}
421
422static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
423{
424 clear_fixmap(FIX_PMD);
425 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
426}
427
428static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
429{
430 return (pmd_t *) __va(pa);
431}
432
433static phys_addr_t __init alloc_pmd_early(uintptr_t va)
434{
435 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
436
437 return (uintptr_t)early_pmd;
438}
439
440static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
441{
442 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
443}
444
445static phys_addr_t __init alloc_pmd_late(uintptr_t va)
446{
447 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, order: 0);
448
449 BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc));
450 return __pa((pmd_t *)ptdesc_address(ptdesc));
451}
452
453static void __init create_pmd_mapping(pmd_t *pmdp,
454 uintptr_t va, phys_addr_t pa,
455 phys_addr_t sz, pgprot_t prot)
456{
457 pte_t *ptep;
458 phys_addr_t pte_phys;
459 uintptr_t pmd_idx = pmd_index(address: va);
460
461 if (sz == PMD_SIZE) {
462 if (pmd_none(pmd: pmdp[pmd_idx]))
463 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), pgprot: prot);
464 return;
465 }
466
467 if (pmd_none(pmd: pmdp[pmd_idx])) {
468 pte_phys = pt_ops.alloc_pte(va);
469 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), pgprot: PAGE_TABLE);
470 ptep = pt_ops.get_pte_virt(pte_phys);
471 memset(ptep, 0, PAGE_SIZE);
472 } else {
473 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
474 ptep = pt_ops.get_pte_virt(pte_phys);
475 }
476
477 create_pte_mapping(ptep, va, pa, sz, prot);
478}
479
480static pud_t *__init get_pud_virt_early(phys_addr_t pa)
481{
482 return (pud_t *)((uintptr_t)pa);
483}
484
485static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
486{
487 clear_fixmap(FIX_PUD);
488 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
489}
490
491static pud_t *__init get_pud_virt_late(phys_addr_t pa)
492{
493 return (pud_t *)__va(pa);
494}
495
496static phys_addr_t __init alloc_pud_early(uintptr_t va)
497{
498 /* Only one PUD is available for early mapping */
499 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
500
501 return (uintptr_t)early_pud;
502}
503
504static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
505{
506 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
507}
508
509static phys_addr_t alloc_pud_late(uintptr_t va)
510{
511 unsigned long vaddr;
512
513 vaddr = __get_free_page(GFP_KERNEL);
514 BUG_ON(!vaddr);
515 return __pa(vaddr);
516}
517
518static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
519{
520 return (p4d_t *)((uintptr_t)pa);
521}
522
523static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
524{
525 clear_fixmap(FIX_P4D);
526 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
527}
528
529static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
530{
531 return (p4d_t *)__va(pa);
532}
533
534static phys_addr_t __init alloc_p4d_early(uintptr_t va)
535{
536 /* Only one P4D is available for early mapping */
537 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
538
539 return (uintptr_t)early_p4d;
540}
541
542static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
543{
544 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
545}
546
547static phys_addr_t alloc_p4d_late(uintptr_t va)
548{
549 unsigned long vaddr;
550
551 vaddr = __get_free_page(GFP_KERNEL);
552 BUG_ON(!vaddr);
553 return __pa(vaddr);
554}
555
556static void __init create_pud_mapping(pud_t *pudp,
557 uintptr_t va, phys_addr_t pa,
558 phys_addr_t sz, pgprot_t prot)
559{
560 pmd_t *nextp;
561 phys_addr_t next_phys;
562 uintptr_t pud_index = pud_index(va);
563
564 if (sz == PUD_SIZE) {
565 if (pud_val(pud: pudp[pud_index]) == 0)
566 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), pgprot: prot);
567 return;
568 }
569
570 if (pud_val(pud: pudp[pud_index]) == 0) {
571 next_phys = pt_ops.alloc_pmd(va);
572 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), pgprot: PAGE_TABLE);
573 nextp = pt_ops.get_pmd_virt(next_phys);
574 memset(nextp, 0, PAGE_SIZE);
575 } else {
576 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
577 nextp = pt_ops.get_pmd_virt(next_phys);
578 }
579
580 create_pmd_mapping(pmdp: nextp, va, pa, sz, prot);
581}
582
583static void __init create_p4d_mapping(p4d_t *p4dp,
584 uintptr_t va, phys_addr_t pa,
585 phys_addr_t sz, pgprot_t prot)
586{
587 pud_t *nextp;
588 phys_addr_t next_phys;
589 uintptr_t p4d_index = p4d_index(va);
590
591 if (sz == P4D_SIZE) {
592 if (p4d_val(p4d: p4dp[p4d_index]) == 0)
593 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
594 return;
595 }
596
597 if (p4d_val(p4d: p4dp[p4d_index]) == 0) {
598 next_phys = pt_ops.alloc_pud(va);
599 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
600 nextp = pt_ops.get_pud_virt(next_phys);
601 memset(nextp, 0, PAGE_SIZE);
602 } else {
603 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
604 nextp = pt_ops.get_pud_virt(next_phys);
605 }
606
607 create_pud_mapping(pudp: nextp, va, pa, sz, prot);
608}
609
610#define pgd_next_t p4d_t
611#define alloc_pgd_next(__va) (pgtable_l5_enabled ? \
612 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \
613 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
614#define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \
615 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \
616 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
617#define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
618 (pgtable_l5_enabled ? \
619 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
620 (pgtable_l4_enabled ? \
621 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \
622 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
623#define fixmap_pgd_next (pgtable_l5_enabled ? \
624 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \
625 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
626#define trampoline_pgd_next (pgtable_l5_enabled ? \
627 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \
628 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
629#else
630#define pgd_next_t pte_t
631#define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
632#define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
633#define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
634 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
635#define fixmap_pgd_next ((uintptr_t)fixmap_pte)
636#define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
637#define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
638#define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
639#endif /* __PAGETABLE_PMD_FOLDED */
640
641void __init create_pgd_mapping(pgd_t *pgdp,
642 uintptr_t va, phys_addr_t pa,
643 phys_addr_t sz, pgprot_t prot)
644{
645 pgd_next_t *nextp;
646 phys_addr_t next_phys;
647 uintptr_t pgd_idx = pgd_index(va);
648
649 if (sz == PGDIR_SIZE) {
650 if (pgd_val(pgd: pgdp[pgd_idx]) == 0)
651 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
652 return;
653 }
654
655 if (pgd_val(pgd: pgdp[pgd_idx]) == 0) {
656 next_phys = alloc_pgd_next(va);
657 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
658 nextp = get_pgd_next_virt(next_phys);
659 memset(nextp, 0, PAGE_SIZE);
660 } else {
661 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
662 nextp = get_pgd_next_virt(next_phys);
663 }
664
665 create_pgd_next_mapping(nextp, va, pa, sz, prot);
666}
667
668static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va,
669 phys_addr_t size)
670{
671 if (pgtable_l5_enabled &&
672 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
673 return P4D_SIZE;
674
675 if (pgtable_l4_enabled &&
676 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
677 return PUD_SIZE;
678
679 if (IS_ENABLED(CONFIG_64BIT) &&
680 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
681 return PMD_SIZE;
682
683 return PAGE_SIZE;
684}
685
686#ifdef CONFIG_XIP_KERNEL
687#define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
688extern char _xiprom[], _exiprom[], __data_loc;
689
690/* called from head.S with MMU off */
691asmlinkage void __init __copy_data(void)
692{
693 void *from = (void *)(&__data_loc);
694 void *to = (void *)CONFIG_PHYS_RAM_BASE;
695 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
696
697 memcpy(to, from, sz);
698}
699#endif
700
701#ifdef CONFIG_STRICT_KERNEL_RWX
702static __init pgprot_t pgprot_from_va(uintptr_t va)
703{
704 if (is_va_kernel_text(va))
705 return PAGE_KERNEL_READ_EXEC;
706
707 /*
708 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
709 * we must protect its linear mapping alias from being executed and
710 * written.
711 * And rodata section is marked readonly in mark_rodata_ro.
712 */
713 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
714 return PAGE_KERNEL_READ;
715
716 return PAGE_KERNEL;
717}
718
719void mark_rodata_ro(void)
720{
721 set_kernel_memory(__start_rodata, _data, set_memory_ro);
722 if (IS_ENABLED(CONFIG_64BIT))
723 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
724 set_memory_ro);
725}
726#else
727static __init pgprot_t pgprot_from_va(uintptr_t va)
728{
729 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
730 return PAGE_KERNEL;
731
732 return PAGE_KERNEL_EXEC;
733}
734#endif /* CONFIG_STRICT_KERNEL_RWX */
735
736#if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
737u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
738
739static void __init disable_pgtable_l5(void)
740{
741 pgtable_l5_enabled = false;
742 kernel_map.page_offset = PAGE_OFFSET_L4;
743 satp_mode = SATP_MODE_48;
744}
745
746static void __init disable_pgtable_l4(void)
747{
748 pgtable_l4_enabled = false;
749 kernel_map.page_offset = PAGE_OFFSET_L3;
750 satp_mode = SATP_MODE_39;
751}
752
753static int __init print_no4lvl(char *p)
754{
755 pr_info("Disabled 4-level and 5-level paging");
756 return 0;
757}
758early_param("no4lvl", print_no4lvl);
759
760static int __init print_no5lvl(char *p)
761{
762 pr_info("Disabled 5-level paging");
763 return 0;
764}
765early_param("no5lvl", print_no5lvl);
766
767static void __init set_mmap_rnd_bits_max(void)
768{
769 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3;
770}
771
772/*
773 * There is a simple way to determine if 4-level is supported by the
774 * underlying hardware: establish 1:1 mapping in 4-level page table mode
775 * then read SATP to see if the configuration was taken into account
776 * meaning sv48 is supported.
777 */
778static __init void set_satp_mode(uintptr_t dtb_pa)
779{
780 u64 identity_satp, hw_satp;
781 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
782 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
783
784 if (satp_mode_cmdline == SATP_MODE_57) {
785 disable_pgtable_l5();
786 } else if (satp_mode_cmdline == SATP_MODE_48) {
787 disable_pgtable_l5();
788 disable_pgtable_l4();
789 return;
790 }
791
792 create_p4d_mapping(early_p4d,
793 set_satp_mode_pmd, (uintptr_t)early_pud,
794 P4D_SIZE, PAGE_TABLE);
795 create_pud_mapping(early_pud,
796 set_satp_mode_pmd, (uintptr_t)early_pmd,
797 PUD_SIZE, PAGE_TABLE);
798 /* Handle the case where set_satp_mode straddles 2 PMDs */
799 create_pmd_mapping(pmdp: early_pmd,
800 va: set_satp_mode_pmd, pa: set_satp_mode_pmd,
801 PMD_SIZE, PAGE_KERNEL_EXEC);
802 create_pmd_mapping(pmdp: early_pmd,
803 va: set_satp_mode_pmd + PMD_SIZE,
804 pa: set_satp_mode_pmd + PMD_SIZE,
805 PMD_SIZE, PAGE_KERNEL_EXEC);
806retry:
807 create_pgd_mapping(early_pg_dir,
808 set_satp_mode_pmd,
809 pgtable_l5_enabled ?
810 (uintptr_t)early_p4d : (uintptr_t)early_pud,
811 PGDIR_SIZE, PAGE_TABLE);
812
813 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
814
815 local_flush_tlb_all();
816 csr_write(CSR_SATP, identity_satp);
817 hw_satp = csr_swap(CSR_SATP, 0ULL);
818 local_flush_tlb_all();
819
820 if (hw_satp != identity_satp) {
821 if (pgtable_l5_enabled) {
822 disable_pgtable_l5();
823 memset(early_pg_dir, 0, PAGE_SIZE);
824 goto retry;
825 }
826 disable_pgtable_l4();
827 }
828
829 memset(early_pg_dir, 0, PAGE_SIZE);
830 memset(early_p4d, 0, PAGE_SIZE);
831 memset(early_pud, 0, PAGE_SIZE);
832 memset(early_pmd, 0, PAGE_SIZE);
833}
834#endif
835
836/*
837 * setup_vm() is called from head.S with MMU-off.
838 *
839 * Following requirements should be honoured for setup_vm() to work
840 * correctly:
841 * 1) It should use PC-relative addressing for accessing kernel symbols.
842 * To achieve this we always use GCC cmodel=medany.
843 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
844 * so disable compiler instrumentation when FTRACE is enabled.
845 *
846 * Currently, the above requirements are honoured by using custom CFLAGS
847 * for init.o in mm/Makefile.
848 */
849
850#ifndef __riscv_cmodel_medany
851#error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
852#endif
853
854#ifdef CONFIG_RELOCATABLE
855extern unsigned long __rela_dyn_start, __rela_dyn_end;
856
857static void __init relocate_kernel(void)
858{
859 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
860 /*
861 * This holds the offset between the linked virtual address and the
862 * relocated virtual address.
863 */
864 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
865 /*
866 * This holds the offset between kernel linked virtual address and
867 * physical address.
868 */
869 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
870
871 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
872 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
873 Elf64_Addr relocated_addr = rela->r_addend;
874
875 if (rela->r_info != R_RISCV_RELATIVE)
876 continue;
877
878 /*
879 * Make sure to not relocate vdso symbols like rt_sigreturn
880 * which are linked from the address 0 in vmlinux since
881 * vdso symbol addresses are actually used as an offset from
882 * mm->context.vdso in VDSO_OFFSET macro.
883 */
884 if (relocated_addr >= KERNEL_LINK_ADDR)
885 relocated_addr += reloc_offset;
886
887 *(Elf64_Addr *)addr = relocated_addr;
888 }
889}
890#endif /* CONFIG_RELOCATABLE */
891
892#ifdef CONFIG_XIP_KERNEL
893static void __init create_kernel_page_table(pgd_t *pgdir,
894 __always_unused bool early)
895{
896 uintptr_t va, end_va;
897
898 /* Map the flash resident part */
899 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
900 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
901 create_pgd_mapping(pgdir, va,
902 kernel_map.xiprom + (va - kernel_map.virt_addr),
903 PMD_SIZE, PAGE_KERNEL_EXEC);
904
905 /* Map the data in RAM */
906 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
907 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
908 create_pgd_mapping(pgdir, va,
909 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
910 PMD_SIZE, PAGE_KERNEL);
911}
912#else
913static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
914{
915 uintptr_t va, end_va;
916
917 end_va = kernel_map.virt_addr + kernel_map.size;
918 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
919 create_pgd_mapping(pgdp: pgdir, va,
920 pa: kernel_map.phys_addr + (va - kernel_map.virt_addr),
921 PMD_SIZE,
922 prot: early ?
923 PAGE_KERNEL_EXEC : pgprot_from_va(va));
924}
925#endif
926
927/*
928 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
929 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
930 * entry.
931 */
932static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
933 uintptr_t dtb_pa)
934{
935#ifndef CONFIG_BUILTIN_DTB
936 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
937
938 /* Make sure the fdt fixmap address is always aligned on PMD size */
939 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
940
941 /* In 32-bit only, the fdt lies in its own PGD */
942 if (!IS_ENABLED(CONFIG_64BIT)) {
943 create_pgd_mapping(early_pg_dir, fix_fdt_va,
944 pa, MAX_FDT_SIZE, PAGE_KERNEL);
945 } else {
946 create_pmd_mapping(pmdp: fixmap_pmd, va: fix_fdt_va,
947 pa, PMD_SIZE, PAGE_KERNEL);
948 create_pmd_mapping(pmdp: fixmap_pmd, va: fix_fdt_va + PMD_SIZE,
949 pa: pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
950 }
951
952 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
953#else
954 /*
955 * For 64-bit kernel, __va can't be used since it would return a linear
956 * mapping address whereas dtb_early_va will be used before
957 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
958 * kernel is mapped in the linear mapping, that makes no difference.
959 */
960 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa);
961#endif
962
963 dtb_early_pa = dtb_pa;
964}
965
966/*
967 * MMU is not enabled, the page tables are allocated directly using
968 * early_pmd/pud/p4d and the address returned is the physical one.
969 */
970static void __init pt_ops_set_early(void)
971{
972 pt_ops.alloc_pte = alloc_pte_early;
973 pt_ops.get_pte_virt = get_pte_virt_early;
974#ifndef __PAGETABLE_PMD_FOLDED
975 pt_ops.alloc_pmd = alloc_pmd_early;
976 pt_ops.get_pmd_virt = get_pmd_virt_early;
977 pt_ops.alloc_pud = alloc_pud_early;
978 pt_ops.get_pud_virt = get_pud_virt_early;
979 pt_ops.alloc_p4d = alloc_p4d_early;
980 pt_ops.get_p4d_virt = get_p4d_virt_early;
981#endif
982}
983
984/*
985 * MMU is enabled but page table setup is not complete yet.
986 * fixmap page table alloc functions must be used as a means to temporarily
987 * map the allocated physical pages since the linear mapping does not exist yet.
988 *
989 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
990 * but it will be used as described above.
991 */
992static void __init pt_ops_set_fixmap(void)
993{
994 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
995 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
996#ifndef __PAGETABLE_PMD_FOLDED
997 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
998 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
999 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
1000 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
1001 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
1002 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
1003#endif
1004}
1005
1006/*
1007 * MMU is enabled and page table setup is complete, so from now, we can use
1008 * generic page allocation functions to setup page table.
1009 */
1010static void __init pt_ops_set_late(void)
1011{
1012 pt_ops.alloc_pte = alloc_pte_late;
1013 pt_ops.get_pte_virt = get_pte_virt_late;
1014#ifndef __PAGETABLE_PMD_FOLDED
1015 pt_ops.alloc_pmd = alloc_pmd_late;
1016 pt_ops.get_pmd_virt = get_pmd_virt_late;
1017 pt_ops.alloc_pud = alloc_pud_late;
1018 pt_ops.get_pud_virt = get_pud_virt_late;
1019 pt_ops.alloc_p4d = alloc_p4d_late;
1020 pt_ops.get_p4d_virt = get_p4d_virt_late;
1021#endif
1022}
1023
1024#ifdef CONFIG_RANDOMIZE_BASE
1025extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa);
1026extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa);
1027
1028static int __init print_nokaslr(char *p)
1029{
1030 pr_info("Disabled KASLR");
1031 return 0;
1032}
1033early_param("nokaslr", print_nokaslr);
1034
1035unsigned long kaslr_offset(void)
1036{
1037 return kernel_map.virt_offset;
1038}
1039#endif
1040
1041asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1042{
1043 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1044
1045#ifdef CONFIG_RANDOMIZE_BASE
1046 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) {
1047 u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa);
1048 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
1049 u32 nr_pos;
1050
1051 /*
1052 * Compute the number of positions available: we are limited
1053 * by the early page table that only has one PUD and we must
1054 * be aligned on PMD_SIZE.
1055 */
1056 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE;
1057
1058 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE;
1059 }
1060#endif
1061
1062 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset;
1063
1064#ifdef CONFIG_XIP_KERNEL
1065#ifdef CONFIG_64BIT
1066 kernel_map.page_offset = PAGE_OFFSET_L3;
1067#else
1068 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1069#endif
1070 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1071 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1072
1073 phys_ram_base = CONFIG_PHYS_RAM_BASE;
1074 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1075 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
1076
1077 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1078#else
1079 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1080 kernel_map.phys_addr = (uintptr_t)(&_start);
1081 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1082#endif
1083
1084#if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1085 set_satp_mode(dtb_pa);
1086 set_mmap_rnd_bits_max();
1087#endif
1088
1089 /*
1090 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1091 * where we have the system memory layout: this allows us to align
1092 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1093 * for the linear mapping. This is only possible because the kernel
1094 * mapping lies outside the linear mapping.
1095 * In 32-bit however, as the kernel resides in the linear mapping,
1096 * setup_vm_final can not change the mapping established here,
1097 * otherwise the same kernel addresses would get mapped to different
1098 * physical addresses (if the start of dram is different from the
1099 * kernel physical address start).
1100 */
1101 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1102 0UL : PAGE_OFFSET - kernel_map.phys_addr;
1103 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1104
1105 /*
1106 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
1107 * kernel, whereas for 64-bit kernel, the end of the virtual address
1108 * space is occupied by the modules/BPF/kernel mappings which reduces
1109 * the available size of the linear mapping.
1110 */
1111 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
1112
1113 /* Sanity check alignment and size */
1114 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1115 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1116
1117#ifdef CONFIG_64BIT
1118 /*
1119 * The last 4K bytes of the addressable memory can not be mapped because
1120 * of IS_ERR_VALUE macro.
1121 */
1122 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1123#endif
1124
1125#ifdef CONFIG_RELOCATABLE
1126 /*
1127 * Early page table uses only one PUD, which makes it possible
1128 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1129 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1130 * since a part of the kernel would not get mapped.
1131 */
1132 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1133 relocate_kernel();
1134#endif
1135
1136 apply_early_boot_alternatives();
1137 pt_ops_set_early();
1138
1139 /* Setup early PGD for fixmap */
1140 create_pgd_mapping(early_pg_dir, FIXADDR_START,
1141 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1142
1143#ifndef __PAGETABLE_PMD_FOLDED
1144 /* Setup fixmap P4D and PUD */
1145 if (pgtable_l5_enabled)
1146 create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1147 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1148 /* Setup fixmap PUD and PMD */
1149 if (pgtable_l4_enabled)
1150 create_pud_mapping(fixmap_pud, FIXADDR_START,
1151 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1152 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1153 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1154 /* Setup trampoline PGD and PMD */
1155 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1156 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1157 if (pgtable_l5_enabled)
1158 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1159 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1160 if (pgtable_l4_enabled)
1161 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1162 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1163#ifdef CONFIG_XIP_KERNEL
1164 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1165 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1166#else
1167 create_pmd_mapping(pmdp: trampoline_pmd, va: kernel_map.virt_addr,
1168 pa: kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1169#endif
1170#else
1171 /* Setup trampoline PGD */
1172 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1173 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1174#endif
1175
1176 /*
1177 * Setup early PGD covering entire kernel which will allow
1178 * us to reach paging_init(). We map all memory banks later
1179 * in setup_vm_final() below.
1180 */
1181 create_kernel_page_table(pgdir: early_pg_dir, early: true);
1182
1183 /* Setup early mapping for FDT early scan */
1184 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1185
1186 /*
1187 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1188 * range can not span multiple pmds.
1189 */
1190 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1191 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1192
1193#ifndef __PAGETABLE_PMD_FOLDED
1194 /*
1195 * Early ioremap fixmap is already created as it lies within first 2MB
1196 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1197 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1198 * the user if not.
1199 */
1200 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1201 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1202 if (pmd_val(pmd: fix_bmap_spmd) != pmd_val(pmd: fix_bmap_epmd)) {
1203 WARN_ON(1);
1204 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1205 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1206 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1207 fix_to_virt(FIX_BTMAP_BEGIN));
1208 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1209 fix_to_virt(FIX_BTMAP_END));
1210
1211 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1212 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1213 }
1214#endif
1215
1216 pt_ops_set_fixmap();
1217}
1218
1219static void __init create_linear_mapping_range(phys_addr_t start,
1220 phys_addr_t end,
1221 uintptr_t fixed_map_size)
1222{
1223 phys_addr_t pa;
1224 uintptr_t va, map_size;
1225
1226 for (pa = start; pa < end; pa += map_size) {
1227 va = (uintptr_t)__va(pa);
1228 map_size = fixed_map_size ? fixed_map_size :
1229 best_map_size(pa, va, size: end - pa);
1230
1231 create_pgd_mapping(swapper_pg_dir, va, pa, sz: map_size,
1232 prot: pgprot_from_va(va));
1233 }
1234}
1235
1236static void __init create_linear_mapping_page_table(void)
1237{
1238 phys_addr_t start, end;
1239 phys_addr_t kfence_pool __maybe_unused;
1240 u64 i;
1241
1242#ifdef CONFIG_STRICT_KERNEL_RWX
1243 phys_addr_t ktext_start = __pa_symbol(_start);
1244 phys_addr_t ktext_size = __init_data_begin - _start;
1245 phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1246 phys_addr_t krodata_size = _data - __start_rodata;
1247
1248 /* Isolate kernel text and rodata so they don't get mapped with a PUD */
1249 memblock_mark_nomap(base: ktext_start, size: ktext_size);
1250 memblock_mark_nomap(base: krodata_start, size: krodata_size);
1251#endif
1252
1253#ifdef CONFIG_KFENCE
1254 /*
1255 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1256 * before we setup the linear mapping so that we avoid using hugepages
1257 * for this region.
1258 */
1259 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1260 BUG_ON(!kfence_pool);
1261
1262 memblock_mark_nomap(base: kfence_pool, KFENCE_POOL_SIZE);
1263 __kfence_pool = __va(kfence_pool);
1264#endif
1265
1266 /* Map all memory banks in the linear mapping */
1267 for_each_mem_range(i, &start, &end) {
1268 if (start >= end)
1269 break;
1270 if (start <= __pa(PAGE_OFFSET) &&
1271 __pa(PAGE_OFFSET) < end)
1272 start = __pa(PAGE_OFFSET);
1273 if (end >= __pa(PAGE_OFFSET) + memory_limit)
1274 end = __pa(PAGE_OFFSET) + memory_limit;
1275
1276 create_linear_mapping_range(start, end, fixed_map_size: 0);
1277 }
1278
1279#ifdef CONFIG_STRICT_KERNEL_RWX
1280 create_linear_mapping_range(start: ktext_start, end: ktext_start + ktext_size, fixed_map_size: 0);
1281 create_linear_mapping_range(start: krodata_start,
1282 end: krodata_start + krodata_size, fixed_map_size: 0);
1283
1284 memblock_clear_nomap(base: ktext_start, size: ktext_size);
1285 memblock_clear_nomap(base: krodata_start, size: krodata_size);
1286#endif
1287
1288#ifdef CONFIG_KFENCE
1289 create_linear_mapping_range(start: kfence_pool,
1290 end: kfence_pool + KFENCE_POOL_SIZE,
1291 PAGE_SIZE);
1292
1293 memblock_clear_nomap(base: kfence_pool, KFENCE_POOL_SIZE);
1294#endif
1295}
1296
1297static void __init setup_vm_final(void)
1298{
1299 /* Setup swapper PGD for fixmap */
1300#if !defined(CONFIG_64BIT)
1301 /*
1302 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1303 * directly in swapper_pg_dir in addition to the pgd entry that points
1304 * to fixmap_pte.
1305 */
1306 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1307
1308 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1309#endif
1310 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1311 __pa_symbol(fixmap_pgd_next),
1312 PGDIR_SIZE, PAGE_TABLE);
1313
1314 /* Map the linear mapping */
1315 create_linear_mapping_page_table();
1316
1317 /* Map the kernel */
1318 if (IS_ENABLED(CONFIG_64BIT))
1319 create_kernel_page_table(swapper_pg_dir, early: false);
1320
1321#ifdef CONFIG_KASAN
1322 kasan_swapper_init();
1323#endif
1324
1325 /* Clear fixmap PTE and PMD mappings */
1326 clear_fixmap(FIX_PTE);
1327 clear_fixmap(FIX_PMD);
1328 clear_fixmap(FIX_PUD);
1329 clear_fixmap(FIX_P4D);
1330
1331 /* Move to swapper page table */
1332 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1333 local_flush_tlb_all();
1334
1335 pt_ops_set_late();
1336}
1337#else
1338asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1339{
1340 dtb_early_va = (void *)dtb_pa;
1341 dtb_early_pa = dtb_pa;
1342}
1343
1344static inline void setup_vm_final(void)
1345{
1346}
1347#endif /* CONFIG_MMU */
1348
1349/*
1350 * reserve_crashkernel() - reserves memory for crash kernel
1351 *
1352 * This function reserves memory area given in "crashkernel=" kernel command
1353 * line parameter. The memory reserved is used by dump capture kernel when
1354 * primary kernel is crashing.
1355 */
1356static void __init arch_reserve_crashkernel(void)
1357{
1358 unsigned long long low_size = 0;
1359 unsigned long long crash_base, crash_size;
1360 char *cmdline = boot_command_line;
1361 bool high = false;
1362 int ret;
1363
1364 if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
1365 return;
1366
1367 ret = parse_crashkernel(cmdline, system_ram: memblock_phys_mem_size(),
1368 crash_size: &crash_size, crash_base: &crash_base,
1369 low_size: &low_size, high: &high);
1370 if (ret)
1371 return;
1372
1373 reserve_crashkernel_generic(cmdline, crash_size, crash_base,
1374 crash_low_size: low_size, high);
1375}
1376
1377void __init paging_init(void)
1378{
1379 setup_bootmem();
1380 setup_vm_final();
1381
1382 /* Depend on that Linear Mapping is ready */
1383 memblock_allow_resize();
1384}
1385
1386void __init misc_mem_init(void)
1387{
1388 early_memtest(start: min_low_pfn << PAGE_SHIFT, end: max_low_pfn << PAGE_SHIFT);
1389 arch_numa_init();
1390 sparse_init();
1391#ifdef CONFIG_SPARSEMEM_VMEMMAP
1392 /* The entire VMEMMAP region has been populated. Flush TLB for this region */
1393 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END);
1394#endif
1395 zone_sizes_init();
1396 arch_reserve_crashkernel();
1397 memblock_dump_all();
1398}
1399
1400#ifdef CONFIG_SPARSEMEM_VMEMMAP
1401void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node,
1402 unsigned long addr, unsigned long next)
1403{
1404 pmd_set_huge(pmd, virt_to_phys(address: p), PAGE_KERNEL);
1405}
1406
1407int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node,
1408 unsigned long addr, unsigned long next)
1409{
1410 vmemmap_verify((pte_t *)pmdp, node, addr, next);
1411 return 1;
1412}
1413
1414int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1415 struct vmem_altmap *altmap)
1416{
1417 /*
1418 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we
1419 * can't use hugepage mappings for 2-level page table because in case of
1420 * memory hotplug, we are not able to update all the page tables with
1421 * the new PMDs.
1422 */
1423 return vmemmap_populate_hugepages(start, end, node, NULL);
1424}
1425#endif
1426
1427#if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1428/*
1429 * Pre-allocates page-table pages for a specific area in the kernel
1430 * page-table. Only the level which needs to be synchronized between
1431 * all page-tables is allocated because the synchronization can be
1432 * expensive.
1433 */
1434static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1435 const char *area)
1436{
1437 unsigned long addr;
1438 const char *lvl;
1439
1440 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1441 pgd_t *pgd = pgd_offset_k(addr);
1442 p4d_t *p4d;
1443 pud_t *pud;
1444 pmd_t *pmd;
1445
1446 lvl = "p4d";
1447 p4d = p4d_alloc(mm: &init_mm, pgd, address: addr);
1448 if (!p4d)
1449 goto failed;
1450
1451 if (pgtable_l5_enabled)
1452 continue;
1453
1454 lvl = "pud";
1455 pud = pud_alloc(mm: &init_mm, p4d, address: addr);
1456 if (!pud)
1457 goto failed;
1458
1459 if (pgtable_l4_enabled)
1460 continue;
1461
1462 lvl = "pmd";
1463 pmd = pmd_alloc(mm: &init_mm, pud, address: addr);
1464 if (!pmd)
1465 goto failed;
1466 }
1467 return;
1468
1469failed:
1470 /*
1471 * The pages have to be there now or they will be missing in
1472 * process page-tables later.
1473 */
1474 panic(fmt: "Failed to pre-allocate %s pages for %s area\n", lvl, area);
1475}
1476
1477void __init pgtable_cache_init(void)
1478{
1479 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, area: "vmalloc");
1480 if (IS_ENABLED(CONFIG_MODULES))
1481 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, area: "bpf/modules");
1482}
1483#endif
1484

source code of linux/arch/riscv/mm/init.c