Loading...
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/ptdump.h>
33#include <asm/sections.h>
34#include <asm/soc.h>
35#include <asm/tlbflush.h>
36
37#include "../kernel/head.h"
38
39struct kernel_mapping kernel_map __ro_after_init;
40EXPORT_SYMBOL(kernel_map);
41#ifdef CONFIG_XIP_KERNEL
42#define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
43#endif
44
45#ifdef CONFIG_64BIT
46u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39;
47#else
48u64 satp_mode __ro_after_init = SATP_MODE_32;
49#endif
50EXPORT_SYMBOL(satp_mode);
51
52#ifdef CONFIG_64BIT
53bool pgtable_l4_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
54bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT) && !IS_ENABLED(CONFIG_XIP_KERNEL);
55EXPORT_SYMBOL(pgtable_l4_enabled);
56EXPORT_SYMBOL(pgtable_l5_enabled);
57#endif
58
59phys_addr_t phys_ram_base __ro_after_init;
60EXPORT_SYMBOL(phys_ram_base);
61
62unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
63 __page_aligned_bss;
64EXPORT_SYMBOL(empty_zero_page);
65
66extern char _start[];
67void *_dtb_early_va __initdata;
68uintptr_t _dtb_early_pa __initdata;
69
70phys_addr_t dma32_phys_limit __initdata;
71
72static void __init zone_sizes_init(void)
73{
74 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
75
76#ifdef CONFIG_ZONE_DMA32
77 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
78#endif
79 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
80
81 free_area_init(max_zone_pfns);
82}
83
84#if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
85
86#define LOG2_SZ_1K ilog2(SZ_1K)
87#define LOG2_SZ_1M ilog2(SZ_1M)
88#define LOG2_SZ_1G ilog2(SZ_1G)
89#define LOG2_SZ_1T ilog2(SZ_1T)
90
91static inline void print_mlk(char *name, unsigned long b, unsigned long t)
92{
93 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
94 (((t) - (b)) >> LOG2_SZ_1K));
95}
96
97static inline void print_mlm(char *name, unsigned long b, unsigned long t)
98{
99 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
100 (((t) - (b)) >> LOG2_SZ_1M));
101}
102
103static inline void print_mlg(char *name, unsigned long b, unsigned long t)
104{
105 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t,
106 (((t) - (b)) >> LOG2_SZ_1G));
107}
108
109#ifdef CONFIG_64BIT
110static inline void print_mlt(char *name, unsigned long b, unsigned long t)
111{
112 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t,
113 (((t) - (b)) >> LOG2_SZ_1T));
114}
115#else
116#define print_mlt(n, b, t) do {} while (0)
117#endif
118
119static inline void print_ml(char *name, unsigned long b, unsigned long t)
120{
121 unsigned long diff = t - b;
122
123 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10)
124 print_mlt(name, b, t);
125 else if ((diff >> LOG2_SZ_1G) >= 10)
126 print_mlg(name, b, t);
127 else if ((diff >> LOG2_SZ_1M) >= 10)
128 print_mlm(name, b, t);
129 else
130 print_mlk(name, b, t);
131}
132
133static void __init print_vm_layout(void)
134{
135 pr_notice("Virtual kernel memory layout:\n");
136 print_ml("fixmap", (unsigned long)FIXADDR_START,
137 (unsigned long)FIXADDR_TOP);
138 print_ml("pci io", (unsigned long)PCI_IO_START,
139 (unsigned long)PCI_IO_END);
140 print_ml("vmemmap", (unsigned long)VMEMMAP_START,
141 (unsigned long)VMEMMAP_END);
142 print_ml("vmalloc", (unsigned long)VMALLOC_START,
143 (unsigned long)VMALLOC_END);
144#ifdef CONFIG_64BIT
145 print_ml("modules", (unsigned long)MODULES_VADDR,
146 (unsigned long)MODULES_END);
147#endif
148 print_ml("lowmem", (unsigned long)PAGE_OFFSET,
149 (unsigned long)high_memory);
150 if (IS_ENABLED(CONFIG_64BIT)) {
151#ifdef CONFIG_KASAN
152 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
153#endif
154
155 print_ml("kernel", (unsigned long)kernel_map.virt_addr,
156 (unsigned long)ADDRESS_SPACE_END);
157 }
158}
159#else
160static void print_vm_layout(void) { }
161#endif /* CONFIG_DEBUG_VM */
162
163void __init mem_init(void)
164{
165#ifdef CONFIG_FLATMEM
166 BUG_ON(!mem_map);
167#endif /* CONFIG_FLATMEM */
168
169 swiotlb_init(max_pfn > PFN_DOWN(dma32_phys_limit), SWIOTLB_VERBOSE);
170 memblock_free_all();
171
172 print_vm_layout();
173}
174
175/* Limit the memory size via mem. */
176static phys_addr_t memory_limit;
177#ifdef CONFIG_XIP_KERNEL
178#define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit))
179#endif /* CONFIG_XIP_KERNEL */
180
181static int __init early_mem(char *p)
182{
183 u64 size;
184
185 if (!p)
186 return 1;
187
188 size = memparse(p, &p) & PAGE_MASK;
189 memory_limit = min_t(u64, size, memory_limit);
190
191 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
192
193 return 0;
194}
195early_param("mem", early_mem);
196
197static void __init setup_bootmem(void)
198{
199 phys_addr_t vmlinux_end = __pa_symbol(&_end);
200 phys_addr_t max_mapped_addr;
201 phys_addr_t phys_ram_end, vmlinux_start;
202
203 if (IS_ENABLED(CONFIG_XIP_KERNEL))
204 vmlinux_start = __pa_symbol(&_sdata);
205 else
206 vmlinux_start = __pa_symbol(&_start);
207
208 memblock_enforce_memory_limit(memory_limit);
209
210 /*
211 * Make sure we align the reservation on PMD_SIZE since we will
212 * map the kernel in the linear mapping as read-only: we do not want
213 * any allocation to happen between _end and the next pmd aligned page.
214 */
215 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
216 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
217 /*
218 * Reserve from the start of the kernel to the end of the kernel
219 */
220 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
221
222 phys_ram_end = memblock_end_of_DRAM();
223
224 /*
225 * Make sure we align the start of the memory on a PMD boundary so that
226 * at worst, we map the linear mapping with PMD mappings.
227 */
228 if (!IS_ENABLED(CONFIG_XIP_KERNEL))
229 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK;
230
231 /*
232 * In 64-bit, any use of __va/__pa before this point is wrong as we
233 * did not know the start of DRAM before.
234 */
235 if (IS_ENABLED(CONFIG_64BIT))
236 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base;
237
238 /*
239 * memblock allocator is not aware of the fact that last 4K bytes of
240 * the addressable memory can not be mapped because of IS_ERR_VALUE
241 * macro. Make sure that last 4k bytes are not usable by memblock
242 * if end of dram is equal to maximum addressable memory. For 64-bit
243 * kernel, this problem can't happen here as the end of the virtual
244 * address space is occupied by the kernel mapping then this check must
245 * be done as soon as the kernel mapping base address is determined.
246 */
247 if (!IS_ENABLED(CONFIG_64BIT)) {
248 max_mapped_addr = __pa(~(ulong)0);
249 if (max_mapped_addr == (phys_ram_end - 1))
250 memblock_set_current_limit(max_mapped_addr - 4096);
251 }
252
253 min_low_pfn = PFN_UP(phys_ram_base);
254 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
255 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
256
257 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
258 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
259
260 reserve_initrd_mem();
261
262 /*
263 * No allocation should be done before reserving the memory as defined
264 * in the device tree, otherwise the allocation could end up in a
265 * reserved region.
266 */
267 early_init_fdt_scan_reserved_mem();
268
269 /*
270 * If DTB is built in, no need to reserve its memblock.
271 * Otherwise, do reserve it but avoid using
272 * early_init_fdt_reserve_self() since __pa() does
273 * not work for DTB pointers that are fixmap addresses
274 */
275 if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
276 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
277
278 dma_contiguous_reserve(dma32_phys_limit);
279 if (IS_ENABLED(CONFIG_64BIT))
280 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
281}
282
283#ifdef CONFIG_MMU
284struct pt_alloc_ops pt_ops __initdata;
285
286pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
287pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
288static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
289
290pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
291
292#ifdef CONFIG_XIP_KERNEL
293#define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
294#define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
295#define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
296#define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
297#endif /* CONFIG_XIP_KERNEL */
298
299static const pgprot_t protection_map[16] = {
300 [VM_NONE] = PAGE_NONE,
301 [VM_READ] = PAGE_READ,
302 [VM_WRITE] = PAGE_COPY,
303 [VM_WRITE | VM_READ] = PAGE_COPY,
304 [VM_EXEC] = PAGE_EXEC,
305 [VM_EXEC | VM_READ] = PAGE_READ_EXEC,
306 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC,
307 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC,
308 [VM_SHARED] = PAGE_NONE,
309 [VM_SHARED | VM_READ] = PAGE_READ,
310 [VM_SHARED | VM_WRITE] = PAGE_SHARED,
311 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED,
312 [VM_SHARED | VM_EXEC] = PAGE_EXEC,
313 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC,
314 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC,
315 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC
316};
317DECLARE_VM_GET_PAGE_PROT
318
319void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
320{
321 unsigned long addr = __fix_to_virt(idx);
322 pte_t *ptep;
323
324 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
325
326 ptep = &fixmap_pte[pte_index(addr)];
327
328 if (pgprot_val(prot))
329 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
330 else
331 pte_clear(&init_mm, addr, ptep);
332 local_flush_tlb_page(addr);
333}
334
335static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
336{
337 return (pte_t *)((uintptr_t)pa);
338}
339
340static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
341{
342 clear_fixmap(FIX_PTE);
343 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
344}
345
346static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
347{
348 return (pte_t *) __va(pa);
349}
350
351static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
352{
353 /*
354 * We only create PMD or PGD early mappings so we
355 * should never reach here with MMU disabled.
356 */
357 BUG();
358}
359
360static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
361{
362 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
363}
364
365static phys_addr_t __init alloc_pte_late(uintptr_t va)
366{
367 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
368
369 BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc));
370 return __pa((pte_t *)ptdesc_address(ptdesc));
371}
372
373static void __init create_pte_mapping(pte_t *ptep,
374 uintptr_t va, phys_addr_t pa,
375 phys_addr_t sz, pgprot_t prot)
376{
377 uintptr_t pte_idx = pte_index(va);
378
379 BUG_ON(sz != PAGE_SIZE);
380
381 if (pte_none(ptep[pte_idx]))
382 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
383}
384
385#ifndef __PAGETABLE_PMD_FOLDED
386
387static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
388static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
389static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
390
391#ifdef CONFIG_XIP_KERNEL
392#define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
393#define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
394#define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
395#endif /* CONFIG_XIP_KERNEL */
396
397static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
398static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
399static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
400
401#ifdef CONFIG_XIP_KERNEL
402#define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
403#define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d))
404#define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d))
405#endif /* CONFIG_XIP_KERNEL */
406
407static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
408static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
409static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
410
411#ifdef CONFIG_XIP_KERNEL
412#define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
413#define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
414#define early_pud ((pud_t *)XIP_FIXUP(early_pud))
415#endif /* CONFIG_XIP_KERNEL */
416
417static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
418{
419 /* Before MMU is enabled */
420 return (pmd_t *)((uintptr_t)pa);
421}
422
423static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
424{
425 clear_fixmap(FIX_PMD);
426 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
427}
428
429static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
430{
431 return (pmd_t *) __va(pa);
432}
433
434static phys_addr_t __init alloc_pmd_early(uintptr_t va)
435{
436 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
437
438 return (uintptr_t)early_pmd;
439}
440
441static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
442{
443 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
444}
445
446static phys_addr_t __init alloc_pmd_late(uintptr_t va)
447{
448 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
449
450 BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc));
451 return __pa((pmd_t *)ptdesc_address(ptdesc));
452}
453
454static void __init create_pmd_mapping(pmd_t *pmdp,
455 uintptr_t va, phys_addr_t pa,
456 phys_addr_t sz, pgprot_t prot)
457{
458 pte_t *ptep;
459 phys_addr_t pte_phys;
460 uintptr_t pmd_idx = pmd_index(va);
461
462 if (sz == PMD_SIZE) {
463 if (pmd_none(pmdp[pmd_idx]))
464 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
465 return;
466 }
467
468 if (pmd_none(pmdp[pmd_idx])) {
469 pte_phys = pt_ops.alloc_pte(va);
470 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
471 ptep = pt_ops.get_pte_virt(pte_phys);
472 memset(ptep, 0, PAGE_SIZE);
473 } else {
474 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
475 ptep = pt_ops.get_pte_virt(pte_phys);
476 }
477
478 create_pte_mapping(ptep, va, pa, sz, prot);
479}
480
481static pud_t *__init get_pud_virt_early(phys_addr_t pa)
482{
483 return (pud_t *)((uintptr_t)pa);
484}
485
486static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
487{
488 clear_fixmap(FIX_PUD);
489 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
490}
491
492static pud_t *__init get_pud_virt_late(phys_addr_t pa)
493{
494 return (pud_t *)__va(pa);
495}
496
497static phys_addr_t __init alloc_pud_early(uintptr_t va)
498{
499 /* Only one PUD is available for early mapping */
500 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
501
502 return (uintptr_t)early_pud;
503}
504
505static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
506{
507 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
508}
509
510static phys_addr_t alloc_pud_late(uintptr_t va)
511{
512 unsigned long vaddr;
513
514 vaddr = __get_free_page(GFP_KERNEL);
515 BUG_ON(!vaddr);
516 return __pa(vaddr);
517}
518
519static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
520{
521 return (p4d_t *)((uintptr_t)pa);
522}
523
524static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
525{
526 clear_fixmap(FIX_P4D);
527 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
528}
529
530static p4d_t *__init get_p4d_virt_late(phys_addr_t pa)
531{
532 return (p4d_t *)__va(pa);
533}
534
535static phys_addr_t __init alloc_p4d_early(uintptr_t va)
536{
537 /* Only one P4D is available for early mapping */
538 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
539
540 return (uintptr_t)early_p4d;
541}
542
543static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
544{
545 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
546}
547
548static phys_addr_t alloc_p4d_late(uintptr_t va)
549{
550 unsigned long vaddr;
551
552 vaddr = __get_free_page(GFP_KERNEL);
553 BUG_ON(!vaddr);
554 return __pa(vaddr);
555}
556
557static void __init create_pud_mapping(pud_t *pudp,
558 uintptr_t va, phys_addr_t pa,
559 phys_addr_t sz, pgprot_t prot)
560{
561 pmd_t *nextp;
562 phys_addr_t next_phys;
563 uintptr_t pud_index = pud_index(va);
564
565 if (sz == PUD_SIZE) {
566 if (pud_val(pudp[pud_index]) == 0)
567 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
568 return;
569 }
570
571 if (pud_val(pudp[pud_index]) == 0) {
572 next_phys = pt_ops.alloc_pmd(va);
573 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
574 nextp = pt_ops.get_pmd_virt(next_phys);
575 memset(nextp, 0, PAGE_SIZE);
576 } else {
577 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
578 nextp = pt_ops.get_pmd_virt(next_phys);
579 }
580
581 create_pmd_mapping(nextp, va, pa, sz, prot);
582}
583
584static void __init create_p4d_mapping(p4d_t *p4dp,
585 uintptr_t va, phys_addr_t pa,
586 phys_addr_t sz, pgprot_t prot)
587{
588 pud_t *nextp;
589 phys_addr_t next_phys;
590 uintptr_t p4d_index = p4d_index(va);
591
592 if (sz == P4D_SIZE) {
593 if (p4d_val(p4dp[p4d_index]) == 0)
594 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
595 return;
596 }
597
598 if (p4d_val(p4dp[p4d_index]) == 0) {
599 next_phys = pt_ops.alloc_pud(va);
600 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
601 nextp = pt_ops.get_pud_virt(next_phys);
602 memset(nextp, 0, PAGE_SIZE);
603 } else {
604 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
605 nextp = pt_ops.get_pud_virt(next_phys);
606 }
607
608 create_pud_mapping(nextp, va, pa, sz, prot);
609}
610
611#define pgd_next_t p4d_t
612#define alloc_pgd_next(__va) (pgtable_l5_enabled ? \
613 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \
614 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
615#define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \
616 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \
617 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
618#define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
619 (pgtable_l5_enabled ? \
620 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
621 (pgtable_l4_enabled ? \
622 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \
623 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
624#define fixmap_pgd_next (pgtable_l5_enabled ? \
625 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \
626 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
627#define trampoline_pgd_next (pgtable_l5_enabled ? \
628 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \
629 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
630#else
631#define pgd_next_t pte_t
632#define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
633#define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
634#define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
635 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
636#define fixmap_pgd_next ((uintptr_t)fixmap_pte)
637#define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
638#define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
639#define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
640#endif /* __PAGETABLE_PMD_FOLDED */
641
642void __init create_pgd_mapping(pgd_t *pgdp,
643 uintptr_t va, phys_addr_t pa,
644 phys_addr_t sz, pgprot_t prot)
645{
646 pgd_next_t *nextp;
647 phys_addr_t next_phys;
648 uintptr_t pgd_idx = pgd_index(va);
649
650 if (sz == PGDIR_SIZE) {
651 if (pgd_val(pgdp[pgd_idx]) == 0)
652 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
653 return;
654 }
655
656 if (pgd_val(pgdp[pgd_idx]) == 0) {
657 next_phys = alloc_pgd_next(va);
658 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
659 nextp = get_pgd_next_virt(next_phys);
660 memset(nextp, 0, PAGE_SIZE);
661 } else {
662 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
663 nextp = get_pgd_next_virt(next_phys);
664 }
665
666 create_pgd_next_mapping(nextp, va, pa, sz, prot);
667}
668
669static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va,
670 phys_addr_t size)
671{
672 if (pgtable_l5_enabled &&
673 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
674 return P4D_SIZE;
675
676 if (pgtable_l4_enabled &&
677 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
678 return PUD_SIZE;
679
680 if (IS_ENABLED(CONFIG_64BIT) &&
681 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
682 return PMD_SIZE;
683
684 return PAGE_SIZE;
685}
686
687#ifdef CONFIG_XIP_KERNEL
688#define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
689extern char _xiprom[], _exiprom[], __data_loc;
690
691/* called from head.S with MMU off */
692asmlinkage void __init __copy_data(void)
693{
694 void *from = (void *)(&__data_loc);
695 void *to = (void *)CONFIG_PHYS_RAM_BASE;
696 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
697
698 memcpy(to, from, sz);
699}
700#endif
701
702#ifdef CONFIG_STRICT_KERNEL_RWX
703static __init pgprot_t pgprot_from_va(uintptr_t va)
704{
705 if (is_va_kernel_text(va))
706 return PAGE_KERNEL_READ_EXEC;
707
708 /*
709 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
710 * we must protect its linear mapping alias from being executed and
711 * written.
712 * And rodata section is marked readonly in mark_rodata_ro.
713 */
714 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
715 return PAGE_KERNEL_READ;
716
717 return PAGE_KERNEL;
718}
719
720void mark_rodata_ro(void)
721{
722 set_kernel_memory(__start_rodata, _data, set_memory_ro);
723 if (IS_ENABLED(CONFIG_64BIT))
724 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
725 set_memory_ro);
726
727 debug_checkwx();
728}
729#else
730static __init pgprot_t pgprot_from_va(uintptr_t va)
731{
732 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
733 return PAGE_KERNEL;
734
735 return PAGE_KERNEL_EXEC;
736}
737#endif /* CONFIG_STRICT_KERNEL_RWX */
738
739#if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
740u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
741
742static void __init disable_pgtable_l5(void)
743{
744 pgtable_l5_enabled = false;
745 kernel_map.page_offset = PAGE_OFFSET_L4;
746 satp_mode = SATP_MODE_48;
747}
748
749static void __init disable_pgtable_l4(void)
750{
751 pgtable_l4_enabled = false;
752 kernel_map.page_offset = PAGE_OFFSET_L3;
753 satp_mode = SATP_MODE_39;
754}
755
756static int __init print_no4lvl(char *p)
757{
758 pr_info("Disabled 4-level and 5-level paging");
759 return 0;
760}
761early_param("no4lvl", print_no4lvl);
762
763static int __init print_no5lvl(char *p)
764{
765 pr_info("Disabled 5-level paging");
766 return 0;
767}
768early_param("no5lvl", print_no5lvl);
769
770/*
771 * There is a simple way to determine if 4-level is supported by the
772 * underlying hardware: establish 1:1 mapping in 4-level page table mode
773 * then read SATP to see if the configuration was taken into account
774 * meaning sv48 is supported.
775 */
776static __init void set_satp_mode(uintptr_t dtb_pa)
777{
778 u64 identity_satp, hw_satp;
779 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
780 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
781
782 if (satp_mode_cmdline == SATP_MODE_57) {
783 disable_pgtable_l5();
784 } else if (satp_mode_cmdline == SATP_MODE_48) {
785 disable_pgtable_l5();
786 disable_pgtable_l4();
787 return;
788 }
789
790 create_p4d_mapping(early_p4d,
791 set_satp_mode_pmd, (uintptr_t)early_pud,
792 P4D_SIZE, PAGE_TABLE);
793 create_pud_mapping(early_pud,
794 set_satp_mode_pmd, (uintptr_t)early_pmd,
795 PUD_SIZE, PAGE_TABLE);
796 /* Handle the case where set_satp_mode straddles 2 PMDs */
797 create_pmd_mapping(early_pmd,
798 set_satp_mode_pmd, set_satp_mode_pmd,
799 PMD_SIZE, PAGE_KERNEL_EXEC);
800 create_pmd_mapping(early_pmd,
801 set_satp_mode_pmd + PMD_SIZE,
802 set_satp_mode_pmd + PMD_SIZE,
803 PMD_SIZE, PAGE_KERNEL_EXEC);
804retry:
805 create_pgd_mapping(early_pg_dir,
806 set_satp_mode_pmd,
807 pgtable_l5_enabled ?
808 (uintptr_t)early_p4d : (uintptr_t)early_pud,
809 PGDIR_SIZE, PAGE_TABLE);
810
811 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
812
813 local_flush_tlb_all();
814 csr_write(CSR_SATP, identity_satp);
815 hw_satp = csr_swap(CSR_SATP, 0ULL);
816 local_flush_tlb_all();
817
818 if (hw_satp != identity_satp) {
819 if (pgtable_l5_enabled) {
820 disable_pgtable_l5();
821 memset(early_pg_dir, 0, PAGE_SIZE);
822 goto retry;
823 }
824 disable_pgtable_l4();
825 }
826
827 memset(early_pg_dir, 0, PAGE_SIZE);
828 memset(early_p4d, 0, PAGE_SIZE);
829 memset(early_pud, 0, PAGE_SIZE);
830 memset(early_pmd, 0, PAGE_SIZE);
831}
832#endif
833
834/*
835 * setup_vm() is called from head.S with MMU-off.
836 *
837 * Following requirements should be honoured for setup_vm() to work
838 * correctly:
839 * 1) It should use PC-relative addressing for accessing kernel symbols.
840 * To achieve this we always use GCC cmodel=medany.
841 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
842 * so disable compiler instrumentation when FTRACE is enabled.
843 *
844 * Currently, the above requirements are honoured by using custom CFLAGS
845 * for init.o in mm/Makefile.
846 */
847
848#ifndef __riscv_cmodel_medany
849#error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
850#endif
851
852#ifdef CONFIG_RELOCATABLE
853extern unsigned long __rela_dyn_start, __rela_dyn_end;
854
855static void __init relocate_kernel(void)
856{
857 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
858 /*
859 * This holds the offset between the linked virtual address and the
860 * relocated virtual address.
861 */
862 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
863 /*
864 * This holds the offset between kernel linked virtual address and
865 * physical address.
866 */
867 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
868
869 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
870 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
871 Elf64_Addr relocated_addr = rela->r_addend;
872
873 if (rela->r_info != R_RISCV_RELATIVE)
874 continue;
875
876 /*
877 * Make sure to not relocate vdso symbols like rt_sigreturn
878 * which are linked from the address 0 in vmlinux since
879 * vdso symbol addresses are actually used as an offset from
880 * mm->context.vdso in VDSO_OFFSET macro.
881 */
882 if (relocated_addr >= KERNEL_LINK_ADDR)
883 relocated_addr += reloc_offset;
884
885 *(Elf64_Addr *)addr = relocated_addr;
886 }
887}
888#endif /* CONFIG_RELOCATABLE */
889
890#ifdef CONFIG_XIP_KERNEL
891static void __init create_kernel_page_table(pgd_t *pgdir,
892 __always_unused bool early)
893{
894 uintptr_t va, end_va;
895
896 /* Map the flash resident part */
897 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
898 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
899 create_pgd_mapping(pgdir, va,
900 kernel_map.xiprom + (va - kernel_map.virt_addr),
901 PMD_SIZE, PAGE_KERNEL_EXEC);
902
903 /* Map the data in RAM */
904 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
905 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
906 create_pgd_mapping(pgdir, va,
907 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
908 PMD_SIZE, PAGE_KERNEL);
909}
910#else
911static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
912{
913 uintptr_t va, end_va;
914
915 end_va = kernel_map.virt_addr + kernel_map.size;
916 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
917 create_pgd_mapping(pgdir, va,
918 kernel_map.phys_addr + (va - kernel_map.virt_addr),
919 PMD_SIZE,
920 early ?
921 PAGE_KERNEL_EXEC : pgprot_from_va(va));
922}
923#endif
924
925/*
926 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
927 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
928 * entry.
929 */
930static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
931 uintptr_t dtb_pa)
932{
933#ifndef CONFIG_BUILTIN_DTB
934 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
935
936 /* Make sure the fdt fixmap address is always aligned on PMD size */
937 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
938
939 /* In 32-bit only, the fdt lies in its own PGD */
940 if (!IS_ENABLED(CONFIG_64BIT)) {
941 create_pgd_mapping(early_pg_dir, fix_fdt_va,
942 pa, MAX_FDT_SIZE, PAGE_KERNEL);
943 } else {
944 create_pmd_mapping(fixmap_pmd, fix_fdt_va,
945 pa, PMD_SIZE, PAGE_KERNEL);
946 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
947 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
948 }
949
950 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
951#else
952 /*
953 * For 64-bit kernel, __va can't be used since it would return a linear
954 * mapping address whereas dtb_early_va will be used before
955 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
956 * kernel is mapped in the linear mapping, that makes no difference.
957 */
958 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa);
959#endif
960
961 dtb_early_pa = dtb_pa;
962}
963
964/*
965 * MMU is not enabled, the page tables are allocated directly using
966 * early_pmd/pud/p4d and the address returned is the physical one.
967 */
968static void __init pt_ops_set_early(void)
969{
970 pt_ops.alloc_pte = alloc_pte_early;
971 pt_ops.get_pte_virt = get_pte_virt_early;
972#ifndef __PAGETABLE_PMD_FOLDED
973 pt_ops.alloc_pmd = alloc_pmd_early;
974 pt_ops.get_pmd_virt = get_pmd_virt_early;
975 pt_ops.alloc_pud = alloc_pud_early;
976 pt_ops.get_pud_virt = get_pud_virt_early;
977 pt_ops.alloc_p4d = alloc_p4d_early;
978 pt_ops.get_p4d_virt = get_p4d_virt_early;
979#endif
980}
981
982/*
983 * MMU is enabled but page table setup is not complete yet.
984 * fixmap page table alloc functions must be used as a means to temporarily
985 * map the allocated physical pages since the linear mapping does not exist yet.
986 *
987 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
988 * but it will be used as described above.
989 */
990static void __init pt_ops_set_fixmap(void)
991{
992 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
993 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
994#ifndef __PAGETABLE_PMD_FOLDED
995 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
996 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
997 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
998 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
999 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
1000 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
1001#endif
1002}
1003
1004/*
1005 * MMU is enabled and page table setup is complete, so from now, we can use
1006 * generic page allocation functions to setup page table.
1007 */
1008static void __init pt_ops_set_late(void)
1009{
1010 pt_ops.alloc_pte = alloc_pte_late;
1011 pt_ops.get_pte_virt = get_pte_virt_late;
1012#ifndef __PAGETABLE_PMD_FOLDED
1013 pt_ops.alloc_pmd = alloc_pmd_late;
1014 pt_ops.get_pmd_virt = get_pmd_virt_late;
1015 pt_ops.alloc_pud = alloc_pud_late;
1016 pt_ops.get_pud_virt = get_pud_virt_late;
1017 pt_ops.alloc_p4d = alloc_p4d_late;
1018 pt_ops.get_p4d_virt = get_p4d_virt_late;
1019#endif
1020}
1021
1022#ifdef CONFIG_RANDOMIZE_BASE
1023extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa);
1024extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa);
1025
1026static int __init print_nokaslr(char *p)
1027{
1028 pr_info("Disabled KASLR");
1029 return 0;
1030}
1031early_param("nokaslr", print_nokaslr);
1032
1033unsigned long kaslr_offset(void)
1034{
1035 return kernel_map.virt_offset;
1036}
1037#endif
1038
1039asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1040{
1041 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1042
1043#ifdef CONFIG_RANDOMIZE_BASE
1044 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) {
1045 u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa);
1046 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
1047 u32 nr_pos;
1048
1049 /*
1050 * Compute the number of positions available: we are limited
1051 * by the early page table that only has one PUD and we must
1052 * be aligned on PMD_SIZE.
1053 */
1054 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE;
1055
1056 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE;
1057 }
1058#endif
1059
1060 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset;
1061
1062#ifdef CONFIG_XIP_KERNEL
1063#ifdef CONFIG_64BIT
1064 kernel_map.page_offset = PAGE_OFFSET_L3;
1065#else
1066 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1067#endif
1068 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1069 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1070
1071 phys_ram_base = CONFIG_PHYS_RAM_BASE;
1072 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1073 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
1074
1075 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1076#else
1077 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1078 kernel_map.phys_addr = (uintptr_t)(&_start);
1079 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1080#endif
1081
1082#if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1083 set_satp_mode(dtb_pa);
1084#endif
1085
1086 /*
1087 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1088 * where we have the system memory layout: this allows us to align
1089 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1090 * for the linear mapping. This is only possible because the kernel
1091 * mapping lies outside the linear mapping.
1092 * In 32-bit however, as the kernel resides in the linear mapping,
1093 * setup_vm_final can not change the mapping established here,
1094 * otherwise the same kernel addresses would get mapped to different
1095 * physical addresses (if the start of dram is different from the
1096 * kernel physical address start).
1097 */
1098 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1099 0UL : PAGE_OFFSET - kernel_map.phys_addr;
1100 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1101
1102 /*
1103 * The default maximal physical memory size is KERN_VIRT_SIZE for 32-bit
1104 * kernel, whereas for 64-bit kernel, the end of the virtual address
1105 * space is occupied by the modules/BPF/kernel mappings which reduces
1106 * the available size of the linear mapping.
1107 */
1108 memory_limit = KERN_VIRT_SIZE - (IS_ENABLED(CONFIG_64BIT) ? SZ_4G : 0);
1109
1110 /* Sanity check alignment and size */
1111 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1112 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1113
1114#ifdef CONFIG_64BIT
1115 /*
1116 * The last 4K bytes of the addressable memory can not be mapped because
1117 * of IS_ERR_VALUE macro.
1118 */
1119 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1120#endif
1121
1122#ifdef CONFIG_RELOCATABLE
1123 /*
1124 * Early page table uses only one PUD, which makes it possible
1125 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1126 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1127 * since a part of the kernel would not get mapped.
1128 */
1129 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1130 relocate_kernel();
1131#endif
1132
1133 apply_early_boot_alternatives();
1134 pt_ops_set_early();
1135
1136 /* Setup early PGD for fixmap */
1137 create_pgd_mapping(early_pg_dir, FIXADDR_START,
1138 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1139
1140#ifndef __PAGETABLE_PMD_FOLDED
1141 /* Setup fixmap P4D and PUD */
1142 if (pgtable_l5_enabled)
1143 create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1144 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1145 /* Setup fixmap PUD and PMD */
1146 if (pgtable_l4_enabled)
1147 create_pud_mapping(fixmap_pud, FIXADDR_START,
1148 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1149 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1150 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1151 /* Setup trampoline PGD and PMD */
1152 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1153 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1154 if (pgtable_l5_enabled)
1155 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1156 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1157 if (pgtable_l4_enabled)
1158 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1159 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1160#ifdef CONFIG_XIP_KERNEL
1161 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1162 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1163#else
1164 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1165 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1166#endif
1167#else
1168 /* Setup trampoline PGD */
1169 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1170 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1171#endif
1172
1173 /*
1174 * Setup early PGD covering entire kernel which will allow
1175 * us to reach paging_init(). We map all memory banks later
1176 * in setup_vm_final() below.
1177 */
1178 create_kernel_page_table(early_pg_dir, true);
1179
1180 /* Setup early mapping for FDT early scan */
1181 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1182
1183 /*
1184 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1185 * range can not span multiple pmds.
1186 */
1187 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1188 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1189
1190#ifndef __PAGETABLE_PMD_FOLDED
1191 /*
1192 * Early ioremap fixmap is already created as it lies within first 2MB
1193 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1194 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1195 * the user if not.
1196 */
1197 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1198 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1199 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1200 WARN_ON(1);
1201 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1202 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1203 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1204 fix_to_virt(FIX_BTMAP_BEGIN));
1205 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1206 fix_to_virt(FIX_BTMAP_END));
1207
1208 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1209 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1210 }
1211#endif
1212
1213 pt_ops_set_fixmap();
1214}
1215
1216static void __init create_linear_mapping_range(phys_addr_t start,
1217 phys_addr_t end,
1218 uintptr_t fixed_map_size)
1219{
1220 phys_addr_t pa;
1221 uintptr_t va, map_size;
1222
1223 for (pa = start; pa < end; pa += map_size) {
1224 va = (uintptr_t)__va(pa);
1225 map_size = fixed_map_size ? fixed_map_size :
1226 best_map_size(pa, va, end - pa);
1227
1228 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1229 pgprot_from_va(va));
1230 }
1231}
1232
1233static void __init create_linear_mapping_page_table(void)
1234{
1235 phys_addr_t start, end;
1236 phys_addr_t kfence_pool __maybe_unused;
1237 u64 i;
1238
1239#ifdef CONFIG_STRICT_KERNEL_RWX
1240 phys_addr_t ktext_start = __pa_symbol(_start);
1241 phys_addr_t ktext_size = __init_data_begin - _start;
1242 phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1243 phys_addr_t krodata_size = _data - __start_rodata;
1244
1245 /* Isolate kernel text and rodata so they don't get mapped with a PUD */
1246 memblock_mark_nomap(ktext_start, ktext_size);
1247 memblock_mark_nomap(krodata_start, krodata_size);
1248#endif
1249
1250#ifdef CONFIG_KFENCE
1251 /*
1252 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1253 * before we setup the linear mapping so that we avoid using hugepages
1254 * for this region.
1255 */
1256 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1257 BUG_ON(!kfence_pool);
1258
1259 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1260 __kfence_pool = __va(kfence_pool);
1261#endif
1262
1263 /* Map all memory banks in the linear mapping */
1264 for_each_mem_range(i, &start, &end) {
1265 if (start >= end)
1266 break;
1267 if (start <= __pa(PAGE_OFFSET) &&
1268 __pa(PAGE_OFFSET) < end)
1269 start = __pa(PAGE_OFFSET);
1270 if (end >= __pa(PAGE_OFFSET) + memory_limit)
1271 end = __pa(PAGE_OFFSET) + memory_limit;
1272
1273 create_linear_mapping_range(start, end, 0);
1274 }
1275
1276#ifdef CONFIG_STRICT_KERNEL_RWX
1277 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0);
1278 create_linear_mapping_range(krodata_start,
1279 krodata_start + krodata_size, 0);
1280
1281 memblock_clear_nomap(ktext_start, ktext_size);
1282 memblock_clear_nomap(krodata_start, krodata_size);
1283#endif
1284
1285#ifdef CONFIG_KFENCE
1286 create_linear_mapping_range(kfence_pool,
1287 kfence_pool + KFENCE_POOL_SIZE,
1288 PAGE_SIZE);
1289
1290 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1291#endif
1292}
1293
1294static void __init setup_vm_final(void)
1295{
1296 /* Setup swapper PGD for fixmap */
1297#if !defined(CONFIG_64BIT)
1298 /*
1299 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1300 * directly in swapper_pg_dir in addition to the pgd entry that points
1301 * to fixmap_pte.
1302 */
1303 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1304
1305 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1306#endif
1307 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1308 __pa_symbol(fixmap_pgd_next),
1309 PGDIR_SIZE, PAGE_TABLE);
1310
1311 /* Map the linear mapping */
1312 create_linear_mapping_page_table();
1313
1314 /* Map the kernel */
1315 if (IS_ENABLED(CONFIG_64BIT))
1316 create_kernel_page_table(swapper_pg_dir, false);
1317
1318#ifdef CONFIG_KASAN
1319 kasan_swapper_init();
1320#endif
1321
1322 /* Clear fixmap PTE and PMD mappings */
1323 clear_fixmap(FIX_PTE);
1324 clear_fixmap(FIX_PMD);
1325 clear_fixmap(FIX_PUD);
1326 clear_fixmap(FIX_P4D);
1327
1328 /* Move to swapper page table */
1329 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1330 local_flush_tlb_all();
1331
1332 pt_ops_set_late();
1333}
1334#else
1335asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1336{
1337 dtb_early_va = (void *)dtb_pa;
1338 dtb_early_pa = dtb_pa;
1339}
1340
1341static inline void setup_vm_final(void)
1342{
1343}
1344#endif /* CONFIG_MMU */
1345
1346/*
1347 * reserve_crashkernel() - reserves memory for crash kernel
1348 *
1349 * This function reserves memory area given in "crashkernel=" kernel command
1350 * line parameter. The memory reserved is used by dump capture kernel when
1351 * primary kernel is crashing.
1352 */
1353static void __init arch_reserve_crashkernel(void)
1354{
1355 unsigned long long low_size = 0;
1356 unsigned long long crash_base, crash_size;
1357 char *cmdline = boot_command_line;
1358 bool high = false;
1359 int ret;
1360
1361 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
1362 return;
1363
1364 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(),
1365 &crash_size, &crash_base,
1366 &low_size, &high);
1367 if (ret)
1368 return;
1369
1370 reserve_crashkernel_generic(cmdline, crash_size, crash_base,
1371 low_size, high);
1372}
1373
1374void __init paging_init(void)
1375{
1376 setup_bootmem();
1377 setup_vm_final();
1378
1379 /* Depend on that Linear Mapping is ready */
1380 memblock_allow_resize();
1381}
1382
1383void __init misc_mem_init(void)
1384{
1385 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1386 arch_numa_init();
1387 sparse_init();
1388#ifdef CONFIG_SPARSEMEM_VMEMMAP
1389 /* The entire VMEMMAP region has been populated. Flush TLB for this region */
1390 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END);
1391#endif
1392 zone_sizes_init();
1393 arch_reserve_crashkernel();
1394 memblock_dump_all();
1395}
1396
1397#ifdef CONFIG_SPARSEMEM_VMEMMAP
1398void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node,
1399 unsigned long addr, unsigned long next)
1400{
1401 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL);
1402}
1403
1404int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node,
1405 unsigned long addr, unsigned long next)
1406{
1407 vmemmap_verify((pte_t *)pmdp, node, addr, next);
1408 return 1;
1409}
1410
1411int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1412 struct vmem_altmap *altmap)
1413{
1414 /*
1415 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we
1416 * can't use hugepage mappings for 2-level page table because in case of
1417 * memory hotplug, we are not able to update all the page tables with
1418 * the new PMDs.
1419 */
1420 return vmemmap_populate_hugepages(start, end, node, NULL);
1421}
1422#endif
1423
1424#if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1425/*
1426 * Pre-allocates page-table pages for a specific area in the kernel
1427 * page-table. Only the level which needs to be synchronized between
1428 * all page-tables is allocated because the synchronization can be
1429 * expensive.
1430 */
1431static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1432 const char *area)
1433{
1434 unsigned long addr;
1435 const char *lvl;
1436
1437 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1438 pgd_t *pgd = pgd_offset_k(addr);
1439 p4d_t *p4d;
1440 pud_t *pud;
1441 pmd_t *pmd;
1442
1443 lvl = "p4d";
1444 p4d = p4d_alloc(&init_mm, pgd, addr);
1445 if (!p4d)
1446 goto failed;
1447
1448 if (pgtable_l5_enabled)
1449 continue;
1450
1451 lvl = "pud";
1452 pud = pud_alloc(&init_mm, p4d, addr);
1453 if (!pud)
1454 goto failed;
1455
1456 if (pgtable_l4_enabled)
1457 continue;
1458
1459 lvl = "pmd";
1460 pmd = pmd_alloc(&init_mm, pud, addr);
1461 if (!pmd)
1462 goto failed;
1463 }
1464 return;
1465
1466failed:
1467 /*
1468 * The pages have to be there now or they will be missing in
1469 * process page-tables later.
1470 */
1471 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1472}
1473
1474void __init pgtable_cache_init(void)
1475{
1476 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1477 if (IS_ENABLED(CONFIG_MODULES))
1478 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
1479}
1480#endif
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#include <linux/execmem.h>
28
29#include <asm/fixmap.h>
30#include <asm/io.h>
31#include <asm/kasan.h>
32#include <asm/numa.h>
33#include <asm/pgtable.h>
34#include <asm/sections.h>
35#include <asm/soc.h>
36#include <asm/sparsemem.h>
37#include <asm/tlbflush.h>
38
39#include "../kernel/head.h"
40
41u64 new_vmalloc[NR_CPUS / sizeof(u64) + 1];
42
43struct kernel_mapping kernel_map __ro_after_init;
44EXPORT_SYMBOL(kernel_map);
45#ifdef CONFIG_XIP_KERNEL
46#define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
47#endif
48
49#ifdef CONFIG_64BIT
50u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39;
51#else
52u64 satp_mode __ro_after_init = SATP_MODE_32;
53#endif
54EXPORT_SYMBOL(satp_mode);
55
56#ifdef CONFIG_64BIT
57bool pgtable_l4_enabled __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL);
58bool pgtable_l5_enabled __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL);
59EXPORT_SYMBOL(pgtable_l4_enabled);
60EXPORT_SYMBOL(pgtable_l5_enabled);
61#endif
62
63phys_addr_t phys_ram_base __ro_after_init;
64EXPORT_SYMBOL(phys_ram_base);
65
66#ifdef CONFIG_SPARSEMEM_VMEMMAP
67#define VMEMMAP_ADDR_ALIGN (1ULL << SECTION_SIZE_BITS)
68
69unsigned long vmemmap_start_pfn __ro_after_init;
70EXPORT_SYMBOL(vmemmap_start_pfn);
71#endif
72
73unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
74 __page_aligned_bss;
75EXPORT_SYMBOL(empty_zero_page);
76
77extern char _start[];
78void *_dtb_early_va __initdata;
79uintptr_t _dtb_early_pa __initdata;
80
81phys_addr_t dma32_phys_limit __initdata;
82
83static void __init zone_sizes_init(void)
84{
85 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
86
87#ifdef CONFIG_ZONE_DMA32
88 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
89#endif
90 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
91
92 free_area_init(max_zone_pfns);
93}
94
95#if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
96
97#define LOG2_SZ_1K ilog2(SZ_1K)
98#define LOG2_SZ_1M ilog2(SZ_1M)
99#define LOG2_SZ_1G ilog2(SZ_1G)
100#define LOG2_SZ_1T ilog2(SZ_1T)
101
102static inline void print_mlk(char *name, unsigned long b, unsigned long t)
103{
104 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
105 (((t) - (b)) >> LOG2_SZ_1K));
106}
107
108static inline void print_mlm(char *name, unsigned long b, unsigned long t)
109{
110 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
111 (((t) - (b)) >> LOG2_SZ_1M));
112}
113
114static inline void print_mlg(char *name, unsigned long b, unsigned long t)
115{
116 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld GB)\n", name, b, t,
117 (((t) - (b)) >> LOG2_SZ_1G));
118}
119
120#ifdef CONFIG_64BIT
121static inline void print_mlt(char *name, unsigned long b, unsigned long t)
122{
123 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld TB)\n", name, b, t,
124 (((t) - (b)) >> LOG2_SZ_1T));
125}
126#else
127#define print_mlt(n, b, t) do {} while (0)
128#endif
129
130static inline void print_ml(char *name, unsigned long b, unsigned long t)
131{
132 unsigned long diff = t - b;
133
134 if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10)
135 print_mlt(name, b, t);
136 else if ((diff >> LOG2_SZ_1G) >= 10)
137 print_mlg(name, b, t);
138 else if ((diff >> LOG2_SZ_1M) >= 10)
139 print_mlm(name, b, t);
140 else
141 print_mlk(name, b, t);
142}
143
144static void __init print_vm_layout(void)
145{
146 pr_notice("Virtual kernel memory layout:\n");
147 print_ml("fixmap", (unsigned long)FIXADDR_START,
148 (unsigned long)FIXADDR_TOP);
149 print_ml("pci io", (unsigned long)PCI_IO_START,
150 (unsigned long)PCI_IO_END);
151 print_ml("vmemmap", (unsigned long)VMEMMAP_START,
152 (unsigned long)VMEMMAP_END);
153 print_ml("vmalloc", (unsigned long)VMALLOC_START,
154 (unsigned long)VMALLOC_END);
155#ifdef CONFIG_64BIT
156 print_ml("modules", (unsigned long)MODULES_VADDR,
157 (unsigned long)MODULES_END);
158#endif
159 print_ml("lowmem", (unsigned long)PAGE_OFFSET,
160 (unsigned long)high_memory);
161 if (IS_ENABLED(CONFIG_64BIT)) {
162#ifdef CONFIG_KASAN
163 print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
164#endif
165
166 print_ml("kernel", (unsigned long)kernel_map.virt_addr,
167 (unsigned long)ADDRESS_SPACE_END);
168 }
169}
170#else
171static void print_vm_layout(void) { }
172#endif /* CONFIG_DEBUG_VM */
173
174void __init mem_init(void)
175{
176 bool swiotlb = max_pfn > PFN_DOWN(dma32_phys_limit);
177#ifdef CONFIG_FLATMEM
178 BUG_ON(!mem_map);
179#endif /* CONFIG_FLATMEM */
180
181 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && !swiotlb &&
182 dma_cache_alignment != 1) {
183 /*
184 * If no bouncing needed for ZONE_DMA, allocate 1MB swiotlb
185 * buffer per 1GB of RAM for kmalloc() bouncing on
186 * non-coherent platforms.
187 */
188 unsigned long size =
189 DIV_ROUND_UP(memblock_phys_mem_size(), 1024);
190 swiotlb_adjust_size(min(swiotlb_size_or_default(), size));
191 swiotlb = true;
192 }
193
194 swiotlb_init(swiotlb, SWIOTLB_VERBOSE);
195 memblock_free_all();
196
197 print_vm_layout();
198}
199
200/* Limit the memory size via mem. */
201static phys_addr_t memory_limit;
202#ifdef CONFIG_XIP_KERNEL
203#define memory_limit (*(phys_addr_t *)XIP_FIXUP(&memory_limit))
204#endif /* CONFIG_XIP_KERNEL */
205
206static int __init early_mem(char *p)
207{
208 u64 size;
209
210 if (!p)
211 return 1;
212
213 size = memparse(p, &p) & PAGE_MASK;
214 memory_limit = min_t(u64, size, memory_limit);
215
216 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
217
218 return 0;
219}
220early_param("mem", early_mem);
221
222static void __init setup_bootmem(void)
223{
224 phys_addr_t vmlinux_end = __pa_symbol(&_end);
225 phys_addr_t max_mapped_addr;
226 phys_addr_t phys_ram_end, vmlinux_start;
227
228 if (IS_ENABLED(CONFIG_XIP_KERNEL))
229 vmlinux_start = __pa_symbol(&_sdata);
230 else
231 vmlinux_start = __pa_symbol(&_start);
232
233 memblock_enforce_memory_limit(memory_limit);
234
235 /*
236 * Make sure we align the reservation on PMD_SIZE since we will
237 * map the kernel in the linear mapping as read-only: we do not want
238 * any allocation to happen between _end and the next pmd aligned page.
239 */
240 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
241 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
242 /*
243 * Reserve from the start of the kernel to the end of the kernel
244 */
245 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
246
247 /*
248 * Make sure we align the start of the memory on a PMD boundary so that
249 * at worst, we map the linear mapping with PMD mappings.
250 */
251 if (!IS_ENABLED(CONFIG_XIP_KERNEL)) {
252 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK;
253#ifdef CONFIG_SPARSEMEM_VMEMMAP
254 vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT;
255#endif
256 }
257
258 /*
259 * In 64-bit, any use of __va/__pa before this point is wrong as we
260 * did not know the start of DRAM before.
261 */
262 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU))
263 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base;
264
265 /*
266 * The size of the linear page mapping may restrict the amount of
267 * usable RAM.
268 */
269 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) {
270 max_mapped_addr = __pa(PAGE_OFFSET) + KERN_VIRT_SIZE;
271 memblock_cap_memory_range(phys_ram_base,
272 max_mapped_addr - phys_ram_base);
273 }
274
275 /*
276 * Reserve physical address space that would be mapped to virtual
277 * addresses greater than (void *)(-PAGE_SIZE) because:
278 * - This memory would overlap with ERR_PTR
279 * - This memory belongs to high memory, which is not supported
280 *
281 * This is not applicable to 64-bit kernel, because virtual addresses
282 * after (void *)(-PAGE_SIZE) are not linearly mapped: they are
283 * occupied by kernel mapping. Also it is unrealistic for high memory
284 * to exist on 64-bit platforms.
285 */
286 if (!IS_ENABLED(CONFIG_64BIT)) {
287 max_mapped_addr = __va_to_pa_nodebug(-PAGE_SIZE);
288 memblock_reserve(max_mapped_addr, (phys_addr_t)-max_mapped_addr);
289 }
290
291 phys_ram_end = memblock_end_of_DRAM();
292 min_low_pfn = PFN_UP(phys_ram_base);
293 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
294 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
295
296 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
297 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
298
299 reserve_initrd_mem();
300
301 /*
302 * No allocation should be done before reserving the memory as defined
303 * in the device tree, otherwise the allocation could end up in a
304 * reserved region.
305 */
306 early_init_fdt_scan_reserved_mem();
307
308 /*
309 * If DTB is built in, no need to reserve its memblock.
310 * Otherwise, do reserve it but avoid using
311 * early_init_fdt_reserve_self() since __pa() does
312 * not work for DTB pointers that are fixmap addresses
313 */
314 if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
315 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
316
317 dma_contiguous_reserve(dma32_phys_limit);
318 if (IS_ENABLED(CONFIG_64BIT))
319 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
320}
321
322#ifdef CONFIG_MMU
323struct pt_alloc_ops pt_ops __meminitdata;
324
325pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
326pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
327static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
328
329pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
330
331#ifdef CONFIG_XIP_KERNEL
332#define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
333#define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
334#define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
335#define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
336#endif /* CONFIG_XIP_KERNEL */
337
338static const pgprot_t protection_map[16] = {
339 [VM_NONE] = PAGE_NONE,
340 [VM_READ] = PAGE_READ,
341 [VM_WRITE] = PAGE_COPY,
342 [VM_WRITE | VM_READ] = PAGE_COPY,
343 [VM_EXEC] = PAGE_EXEC,
344 [VM_EXEC | VM_READ] = PAGE_READ_EXEC,
345 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC,
346 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC,
347 [VM_SHARED] = PAGE_NONE,
348 [VM_SHARED | VM_READ] = PAGE_READ,
349 [VM_SHARED | VM_WRITE] = PAGE_SHARED,
350 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED,
351 [VM_SHARED | VM_EXEC] = PAGE_EXEC,
352 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC,
353 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC,
354 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC
355};
356DECLARE_VM_GET_PAGE_PROT
357
358void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
359{
360 unsigned long addr = __fix_to_virt(idx);
361 pte_t *ptep;
362
363 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
364
365 ptep = &fixmap_pte[pte_index(addr)];
366
367 if (pgprot_val(prot))
368 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
369 else
370 pte_clear(&init_mm, addr, ptep);
371 local_flush_tlb_page(addr);
372}
373
374static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
375{
376 return (pte_t *)((uintptr_t)pa);
377}
378
379static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
380{
381 clear_fixmap(FIX_PTE);
382 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
383}
384
385static inline pte_t *__meminit get_pte_virt_late(phys_addr_t pa)
386{
387 return (pte_t *) __va(pa);
388}
389
390static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
391{
392 /*
393 * We only create PMD or PGD early mappings so we
394 * should never reach here with MMU disabled.
395 */
396 BUG();
397}
398
399static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
400{
401 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
402}
403
404static phys_addr_t __meminit alloc_pte_late(uintptr_t va)
405{
406 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
407
408 BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc));
409 return __pa((pte_t *)ptdesc_address(ptdesc));
410}
411
412static void __meminit create_pte_mapping(pte_t *ptep, uintptr_t va, phys_addr_t pa, phys_addr_t sz,
413 pgprot_t prot)
414{
415 uintptr_t pte_idx = pte_index(va);
416
417 BUG_ON(sz != PAGE_SIZE);
418
419 if (pte_none(ptep[pte_idx]))
420 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
421}
422
423#ifndef __PAGETABLE_PMD_FOLDED
424
425static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
426static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
427static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
428
429#ifdef CONFIG_XIP_KERNEL
430#define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
431#define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
432#define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
433#endif /* CONFIG_XIP_KERNEL */
434
435static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
436static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
437static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
438
439#ifdef CONFIG_XIP_KERNEL
440#define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
441#define fixmap_p4d ((p4d_t *)XIP_FIXUP(fixmap_p4d))
442#define early_p4d ((p4d_t *)XIP_FIXUP(early_p4d))
443#endif /* CONFIG_XIP_KERNEL */
444
445static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
446static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
447static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
448
449#ifdef CONFIG_XIP_KERNEL
450#define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
451#define fixmap_pud ((pud_t *)XIP_FIXUP(fixmap_pud))
452#define early_pud ((pud_t *)XIP_FIXUP(early_pud))
453#endif /* CONFIG_XIP_KERNEL */
454
455static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
456{
457 /* Before MMU is enabled */
458 return (pmd_t *)((uintptr_t)pa);
459}
460
461static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
462{
463 clear_fixmap(FIX_PMD);
464 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
465}
466
467static pmd_t *__meminit get_pmd_virt_late(phys_addr_t pa)
468{
469 return (pmd_t *) __va(pa);
470}
471
472static phys_addr_t __init alloc_pmd_early(uintptr_t va)
473{
474 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
475
476 return (uintptr_t)early_pmd;
477}
478
479static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
480{
481 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
482}
483
484static phys_addr_t __meminit alloc_pmd_late(uintptr_t va)
485{
486 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
487
488 BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc));
489 return __pa((pmd_t *)ptdesc_address(ptdesc));
490}
491
492static void __meminit create_pmd_mapping(pmd_t *pmdp,
493 uintptr_t va, phys_addr_t pa,
494 phys_addr_t sz, pgprot_t prot)
495{
496 pte_t *ptep;
497 phys_addr_t pte_phys;
498 uintptr_t pmd_idx = pmd_index(va);
499
500 if (sz == PMD_SIZE) {
501 if (pmd_none(pmdp[pmd_idx]))
502 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
503 return;
504 }
505
506 if (pmd_none(pmdp[pmd_idx])) {
507 pte_phys = pt_ops.alloc_pte(va);
508 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
509 ptep = pt_ops.get_pte_virt(pte_phys);
510 memset(ptep, 0, PAGE_SIZE);
511 } else {
512 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
513 ptep = pt_ops.get_pte_virt(pte_phys);
514 }
515
516 create_pte_mapping(ptep, va, pa, sz, prot);
517}
518
519static pud_t *__init get_pud_virt_early(phys_addr_t pa)
520{
521 return (pud_t *)((uintptr_t)pa);
522}
523
524static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
525{
526 clear_fixmap(FIX_PUD);
527 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
528}
529
530static pud_t *__meminit get_pud_virt_late(phys_addr_t pa)
531{
532 return (pud_t *)__va(pa);
533}
534
535static phys_addr_t __init alloc_pud_early(uintptr_t va)
536{
537 /* Only one PUD is available for early mapping */
538 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
539
540 return (uintptr_t)early_pud;
541}
542
543static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
544{
545 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
546}
547
548static phys_addr_t __meminit alloc_pud_late(uintptr_t va)
549{
550 unsigned long vaddr;
551
552 vaddr = __get_free_page(GFP_KERNEL);
553 BUG_ON(!vaddr);
554 return __pa(vaddr);
555}
556
557static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
558{
559 return (p4d_t *)((uintptr_t)pa);
560}
561
562static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
563{
564 clear_fixmap(FIX_P4D);
565 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
566}
567
568static p4d_t *__meminit get_p4d_virt_late(phys_addr_t pa)
569{
570 return (p4d_t *)__va(pa);
571}
572
573static phys_addr_t __init alloc_p4d_early(uintptr_t va)
574{
575 /* Only one P4D is available for early mapping */
576 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
577
578 return (uintptr_t)early_p4d;
579}
580
581static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
582{
583 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
584}
585
586static phys_addr_t __meminit alloc_p4d_late(uintptr_t va)
587{
588 unsigned long vaddr;
589
590 vaddr = __get_free_page(GFP_KERNEL);
591 BUG_ON(!vaddr);
592 return __pa(vaddr);
593}
594
595static void __meminit create_pud_mapping(pud_t *pudp, uintptr_t va, phys_addr_t pa, phys_addr_t sz,
596 pgprot_t prot)
597{
598 pmd_t *nextp;
599 phys_addr_t next_phys;
600 uintptr_t pud_index = pud_index(va);
601
602 if (sz == PUD_SIZE) {
603 if (pud_val(pudp[pud_index]) == 0)
604 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
605 return;
606 }
607
608 if (pud_val(pudp[pud_index]) == 0) {
609 next_phys = pt_ops.alloc_pmd(va);
610 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
611 nextp = pt_ops.get_pmd_virt(next_phys);
612 memset(nextp, 0, PAGE_SIZE);
613 } else {
614 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
615 nextp = pt_ops.get_pmd_virt(next_phys);
616 }
617
618 create_pmd_mapping(nextp, va, pa, sz, prot);
619}
620
621static void __meminit create_p4d_mapping(p4d_t *p4dp, uintptr_t va, phys_addr_t pa, phys_addr_t sz,
622 pgprot_t prot)
623{
624 pud_t *nextp;
625 phys_addr_t next_phys;
626 uintptr_t p4d_index = p4d_index(va);
627
628 if (sz == P4D_SIZE) {
629 if (p4d_val(p4dp[p4d_index]) == 0)
630 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
631 return;
632 }
633
634 if (p4d_val(p4dp[p4d_index]) == 0) {
635 next_phys = pt_ops.alloc_pud(va);
636 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
637 nextp = pt_ops.get_pud_virt(next_phys);
638 memset(nextp, 0, PAGE_SIZE);
639 } else {
640 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
641 nextp = pt_ops.get_pud_virt(next_phys);
642 }
643
644 create_pud_mapping(nextp, va, pa, sz, prot);
645}
646
647#define pgd_next_t p4d_t
648#define alloc_pgd_next(__va) (pgtable_l5_enabled ? \
649 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \
650 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
651#define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \
652 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \
653 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
654#define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
655 (pgtable_l5_enabled ? \
656 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
657 (pgtable_l4_enabled ? \
658 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \
659 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
660#define fixmap_pgd_next (pgtable_l5_enabled ? \
661 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \
662 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
663#define trampoline_pgd_next (pgtable_l5_enabled ? \
664 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \
665 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
666#else
667#define pgd_next_t pte_t
668#define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
669#define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
670#define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
671 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
672#define fixmap_pgd_next ((uintptr_t)fixmap_pte)
673#define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
674#define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
675#define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
676#endif /* __PAGETABLE_PMD_FOLDED */
677
678void __meminit create_pgd_mapping(pgd_t *pgdp, uintptr_t va, phys_addr_t pa, phys_addr_t sz,
679 pgprot_t prot)
680{
681 pgd_next_t *nextp;
682 phys_addr_t next_phys;
683 uintptr_t pgd_idx = pgd_index(va);
684
685 if (sz == PGDIR_SIZE) {
686 if (pgd_val(pgdp[pgd_idx]) == 0)
687 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
688 return;
689 }
690
691 if (pgd_val(pgdp[pgd_idx]) == 0) {
692 next_phys = alloc_pgd_next(va);
693 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
694 nextp = get_pgd_next_virt(next_phys);
695 memset(nextp, 0, PAGE_SIZE);
696 } else {
697 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
698 nextp = get_pgd_next_virt(next_phys);
699 }
700
701 create_pgd_next_mapping(nextp, va, pa, sz, prot);
702}
703
704static uintptr_t __meminit best_map_size(phys_addr_t pa, uintptr_t va, phys_addr_t size)
705{
706 if (debug_pagealloc_enabled())
707 return PAGE_SIZE;
708
709 if (pgtable_l5_enabled &&
710 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
711 return P4D_SIZE;
712
713 if (pgtable_l4_enabled &&
714 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
715 return PUD_SIZE;
716
717 if (IS_ENABLED(CONFIG_64BIT) &&
718 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
719 return PMD_SIZE;
720
721 return PAGE_SIZE;
722}
723
724#ifdef CONFIG_XIP_KERNEL
725#define phys_ram_base (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
726extern char _xiprom[], _exiprom[], __data_loc;
727
728/* called from head.S with MMU off */
729asmlinkage void __init __copy_data(void)
730{
731 void *from = (void *)(&__data_loc);
732 void *to = (void *)CONFIG_PHYS_RAM_BASE;
733 size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
734
735 memcpy(to, from, sz);
736}
737#endif
738
739#ifdef CONFIG_STRICT_KERNEL_RWX
740static __meminit pgprot_t pgprot_from_va(uintptr_t va)
741{
742 if (is_va_kernel_text(va))
743 return PAGE_KERNEL_READ_EXEC;
744
745 /*
746 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
747 * we must protect its linear mapping alias from being executed and
748 * written.
749 * And rodata section is marked readonly in mark_rodata_ro.
750 */
751 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
752 return PAGE_KERNEL_READ;
753
754 return PAGE_KERNEL;
755}
756
757void mark_rodata_ro(void)
758{
759 set_kernel_memory(__start_rodata, _data, set_memory_ro);
760 if (IS_ENABLED(CONFIG_64BIT))
761 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
762 set_memory_ro);
763}
764#else
765static __meminit pgprot_t pgprot_from_va(uintptr_t va)
766{
767 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
768 return PAGE_KERNEL;
769
770 return PAGE_KERNEL_EXEC;
771}
772#endif /* CONFIG_STRICT_KERNEL_RWX */
773
774#if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
775u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
776
777static void __init disable_pgtable_l5(void)
778{
779 pgtable_l5_enabled = false;
780 kernel_map.page_offset = PAGE_OFFSET_L4;
781 satp_mode = SATP_MODE_48;
782}
783
784static void __init disable_pgtable_l4(void)
785{
786 pgtable_l4_enabled = false;
787 kernel_map.page_offset = PAGE_OFFSET_L3;
788 satp_mode = SATP_MODE_39;
789}
790
791static int __init print_no4lvl(char *p)
792{
793 pr_info("Disabled 4-level and 5-level paging");
794 return 0;
795}
796early_param("no4lvl", print_no4lvl);
797
798static int __init print_no5lvl(char *p)
799{
800 pr_info("Disabled 5-level paging");
801 return 0;
802}
803early_param("no5lvl", print_no5lvl);
804
805static void __init set_mmap_rnd_bits_max(void)
806{
807 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3;
808}
809
810/*
811 * There is a simple way to determine if 4-level is supported by the
812 * underlying hardware: establish 1:1 mapping in 4-level page table mode
813 * then read SATP to see if the configuration was taken into account
814 * meaning sv48 is supported.
815 */
816static __init void set_satp_mode(uintptr_t dtb_pa)
817{
818 u64 identity_satp, hw_satp;
819 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
820 u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
821
822 if (satp_mode_cmdline == SATP_MODE_57) {
823 disable_pgtable_l5();
824 } else if (satp_mode_cmdline == SATP_MODE_48) {
825 disable_pgtable_l5();
826 disable_pgtable_l4();
827 return;
828 }
829
830 create_p4d_mapping(early_p4d,
831 set_satp_mode_pmd, (uintptr_t)early_pud,
832 P4D_SIZE, PAGE_TABLE);
833 create_pud_mapping(early_pud,
834 set_satp_mode_pmd, (uintptr_t)early_pmd,
835 PUD_SIZE, PAGE_TABLE);
836 /* Handle the case where set_satp_mode straddles 2 PMDs */
837 create_pmd_mapping(early_pmd,
838 set_satp_mode_pmd, set_satp_mode_pmd,
839 PMD_SIZE, PAGE_KERNEL_EXEC);
840 create_pmd_mapping(early_pmd,
841 set_satp_mode_pmd + PMD_SIZE,
842 set_satp_mode_pmd + PMD_SIZE,
843 PMD_SIZE, PAGE_KERNEL_EXEC);
844retry:
845 create_pgd_mapping(early_pg_dir,
846 set_satp_mode_pmd,
847 pgtable_l5_enabled ?
848 (uintptr_t)early_p4d : (uintptr_t)early_pud,
849 PGDIR_SIZE, PAGE_TABLE);
850
851 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
852
853 local_flush_tlb_all();
854 csr_write(CSR_SATP, identity_satp);
855 hw_satp = csr_swap(CSR_SATP, 0ULL);
856 local_flush_tlb_all();
857
858 if (hw_satp != identity_satp) {
859 if (pgtable_l5_enabled) {
860 disable_pgtable_l5();
861 memset(early_pg_dir, 0, PAGE_SIZE);
862 goto retry;
863 }
864 disable_pgtable_l4();
865 }
866
867 memset(early_pg_dir, 0, PAGE_SIZE);
868 memset(early_p4d, 0, PAGE_SIZE);
869 memset(early_pud, 0, PAGE_SIZE);
870 memset(early_pmd, 0, PAGE_SIZE);
871}
872#endif
873
874/*
875 * setup_vm() is called from head.S with MMU-off.
876 *
877 * Following requirements should be honoured for setup_vm() to work
878 * correctly:
879 * 1) It should use PC-relative addressing for accessing kernel symbols.
880 * To achieve this we always use GCC cmodel=medany.
881 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
882 * so disable compiler instrumentation when FTRACE is enabled.
883 *
884 * Currently, the above requirements are honoured by using custom CFLAGS
885 * for init.o in mm/Makefile.
886 */
887
888#ifndef __riscv_cmodel_medany
889#error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
890#endif
891
892#ifdef CONFIG_RELOCATABLE
893extern unsigned long __rela_dyn_start, __rela_dyn_end;
894
895static void __init relocate_kernel(void)
896{
897 Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
898 /*
899 * This holds the offset between the linked virtual address and the
900 * relocated virtual address.
901 */
902 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
903 /*
904 * This holds the offset between kernel linked virtual address and
905 * physical address.
906 */
907 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
908
909 for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
910 Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
911 Elf64_Addr relocated_addr = rela->r_addend;
912
913 if (rela->r_info != R_RISCV_RELATIVE)
914 continue;
915
916 /*
917 * Make sure to not relocate vdso symbols like rt_sigreturn
918 * which are linked from the address 0 in vmlinux since
919 * vdso symbol addresses are actually used as an offset from
920 * mm->context.vdso in VDSO_OFFSET macro.
921 */
922 if (relocated_addr >= KERNEL_LINK_ADDR)
923 relocated_addr += reloc_offset;
924
925 *(Elf64_Addr *)addr = relocated_addr;
926 }
927}
928#endif /* CONFIG_RELOCATABLE */
929
930#ifdef CONFIG_XIP_KERNEL
931static void __init create_kernel_page_table(pgd_t *pgdir,
932 __always_unused bool early)
933{
934 uintptr_t va, start_va, end_va;
935
936 /* Map the flash resident part */
937 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
938 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
939 create_pgd_mapping(pgdir, va,
940 kernel_map.xiprom + (va - kernel_map.virt_addr),
941 PMD_SIZE, PAGE_KERNEL_EXEC);
942
943 /* Map the data in RAM */
944 start_va = kernel_map.virt_addr + (uintptr_t)&_sdata - (uintptr_t)&_start;
945 end_va = kernel_map.virt_addr + kernel_map.size;
946 for (va = start_va; va < end_va; va += PMD_SIZE)
947 create_pgd_mapping(pgdir, va,
948 kernel_map.phys_addr + (va - start_va),
949 PMD_SIZE, PAGE_KERNEL);
950}
951#else
952static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
953{
954 uintptr_t va, end_va;
955
956 end_va = kernel_map.virt_addr + kernel_map.size;
957 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
958 create_pgd_mapping(pgdir, va,
959 kernel_map.phys_addr + (va - kernel_map.virt_addr),
960 PMD_SIZE,
961 early ?
962 PAGE_KERNEL_EXEC : pgprot_from_va(va));
963}
964#endif
965
966/*
967 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
968 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
969 * entry.
970 */
971static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
972 uintptr_t dtb_pa)
973{
974#ifndef CONFIG_BUILTIN_DTB
975 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
976
977 /* Make sure the fdt fixmap address is always aligned on PMD size */
978 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
979
980 /* In 32-bit only, the fdt lies in its own PGD */
981 if (!IS_ENABLED(CONFIG_64BIT)) {
982 create_pgd_mapping(early_pg_dir, fix_fdt_va,
983 pa, MAX_FDT_SIZE, PAGE_KERNEL);
984 } else {
985 create_pmd_mapping(fixmap_pmd, fix_fdt_va,
986 pa, PMD_SIZE, PAGE_KERNEL);
987 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
988 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
989 }
990
991 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
992#else
993 /*
994 * For 64-bit kernel, __va can't be used since it would return a linear
995 * mapping address whereas dtb_early_va will be used before
996 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
997 * kernel is mapped in the linear mapping, that makes no difference.
998 */
999 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa);
1000#endif
1001
1002 dtb_early_pa = dtb_pa;
1003}
1004
1005/*
1006 * MMU is not enabled, the page tables are allocated directly using
1007 * early_pmd/pud/p4d and the address returned is the physical one.
1008 */
1009static void __init pt_ops_set_early(void)
1010{
1011 pt_ops.alloc_pte = alloc_pte_early;
1012 pt_ops.get_pte_virt = get_pte_virt_early;
1013#ifndef __PAGETABLE_PMD_FOLDED
1014 pt_ops.alloc_pmd = alloc_pmd_early;
1015 pt_ops.get_pmd_virt = get_pmd_virt_early;
1016 pt_ops.alloc_pud = alloc_pud_early;
1017 pt_ops.get_pud_virt = get_pud_virt_early;
1018 pt_ops.alloc_p4d = alloc_p4d_early;
1019 pt_ops.get_p4d_virt = get_p4d_virt_early;
1020#endif
1021}
1022
1023/*
1024 * MMU is enabled but page table setup is not complete yet.
1025 * fixmap page table alloc functions must be used as a means to temporarily
1026 * map the allocated physical pages since the linear mapping does not exist yet.
1027 *
1028 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
1029 * but it will be used as described above.
1030 */
1031static void __init pt_ops_set_fixmap(void)
1032{
1033 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
1034 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
1035#ifndef __PAGETABLE_PMD_FOLDED
1036 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
1037 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
1038 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
1039 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
1040 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
1041 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
1042#endif
1043}
1044
1045/*
1046 * MMU is enabled and page table setup is complete, so from now, we can use
1047 * generic page allocation functions to setup page table.
1048 */
1049static void __init pt_ops_set_late(void)
1050{
1051 pt_ops.alloc_pte = alloc_pte_late;
1052 pt_ops.get_pte_virt = get_pte_virt_late;
1053#ifndef __PAGETABLE_PMD_FOLDED
1054 pt_ops.alloc_pmd = alloc_pmd_late;
1055 pt_ops.get_pmd_virt = get_pmd_virt_late;
1056 pt_ops.alloc_pud = alloc_pud_late;
1057 pt_ops.get_pud_virt = get_pud_virt_late;
1058 pt_ops.alloc_p4d = alloc_p4d_late;
1059 pt_ops.get_p4d_virt = get_p4d_virt_late;
1060#endif
1061}
1062
1063#ifdef CONFIG_RANDOMIZE_BASE
1064extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa);
1065extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa);
1066extern u64 __init __pi_get_kaslr_seed_zkr(const uintptr_t dtb_pa);
1067
1068static int __init print_nokaslr(char *p)
1069{
1070 pr_info("Disabled KASLR");
1071 return 0;
1072}
1073early_param("nokaslr", print_nokaslr);
1074
1075unsigned long kaslr_offset(void)
1076{
1077 return kernel_map.virt_offset;
1078}
1079#endif
1080
1081asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1082{
1083 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1084
1085#ifdef CONFIG_RANDOMIZE_BASE
1086 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) {
1087 u64 kaslr_seed = __pi_get_kaslr_seed_zkr(dtb_pa);
1088 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
1089 u32 nr_pos;
1090
1091 if (kaslr_seed == 0)
1092 kaslr_seed = __pi_get_kaslr_seed(dtb_pa);
1093 /*
1094 * Compute the number of positions available: we are limited
1095 * by the early page table that only has one PUD and we must
1096 * be aligned on PMD_SIZE.
1097 */
1098 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE;
1099
1100 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE;
1101 }
1102#endif
1103
1104 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset;
1105
1106#ifdef CONFIG_XIP_KERNEL
1107#ifdef CONFIG_64BIT
1108 kernel_map.page_offset = PAGE_OFFSET_L3;
1109#else
1110 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1111#endif
1112 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1113 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1114
1115 phys_ram_base = CONFIG_PHYS_RAM_BASE;
1116#ifdef CONFIG_SPARSEMEM_VMEMMAP
1117 vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT;
1118#endif
1119 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1120 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
1121
1122 kernel_map.va_kernel_xip_text_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1123 kernel_map.va_kernel_xip_data_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr
1124 + (uintptr_t)&_sdata - (uintptr_t)&_start;
1125#else
1126 kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1127 kernel_map.phys_addr = (uintptr_t)(&_start);
1128 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1129 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1130#endif
1131
1132#if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1133 set_satp_mode(dtb_pa);
1134 set_mmap_rnd_bits_max();
1135#endif
1136
1137 /*
1138 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1139 * where we have the system memory layout: this allows us to align
1140 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1141 * for the linear mapping. This is only possible because the kernel
1142 * mapping lies outside the linear mapping.
1143 * In 32-bit however, as the kernel resides in the linear mapping,
1144 * setup_vm_final can not change the mapping established here,
1145 * otherwise the same kernel addresses would get mapped to different
1146 * physical addresses (if the start of dram is different from the
1147 * kernel physical address start).
1148 */
1149 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1150 0UL : PAGE_OFFSET - kernel_map.phys_addr;
1151
1152 memory_limit = KERN_VIRT_SIZE;
1153
1154 /* Sanity check alignment and size */
1155 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1156 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1157
1158#ifdef CONFIG_64BIT
1159 /*
1160 * The last 4K bytes of the addressable memory can not be mapped because
1161 * of IS_ERR_VALUE macro.
1162 */
1163 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1164#endif
1165
1166#ifdef CONFIG_RELOCATABLE
1167 /*
1168 * Early page table uses only one PUD, which makes it possible
1169 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1170 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1171 * since a part of the kernel would not get mapped.
1172 */
1173 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1174 relocate_kernel();
1175#endif
1176
1177 apply_early_boot_alternatives();
1178 pt_ops_set_early();
1179
1180 /* Setup early PGD for fixmap */
1181 create_pgd_mapping(early_pg_dir, FIXADDR_START,
1182 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1183
1184#ifndef __PAGETABLE_PMD_FOLDED
1185 /* Setup fixmap P4D and PUD */
1186 if (pgtable_l5_enabled)
1187 create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1188 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1189 /* Setup fixmap PUD and PMD */
1190 if (pgtable_l4_enabled)
1191 create_pud_mapping(fixmap_pud, FIXADDR_START,
1192 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1193 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1194 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1195 /* Setup trampoline PGD and PMD */
1196 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1197 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1198 if (pgtable_l5_enabled)
1199 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1200 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1201 if (pgtable_l4_enabled)
1202 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1203 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1204#ifdef CONFIG_XIP_KERNEL
1205 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1206 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1207#else
1208 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1209 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1210#endif
1211#else
1212 /* Setup trampoline PGD */
1213 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1214 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1215#endif
1216
1217 /*
1218 * Setup early PGD covering entire kernel which will allow
1219 * us to reach paging_init(). We map all memory banks later
1220 * in setup_vm_final() below.
1221 */
1222 create_kernel_page_table(early_pg_dir, true);
1223
1224 /* Setup early mapping for FDT early scan */
1225 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1226
1227 /*
1228 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1229 * range can not span multiple pmds.
1230 */
1231 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1232 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1233
1234#ifndef __PAGETABLE_PMD_FOLDED
1235 /*
1236 * Early ioremap fixmap is already created as it lies within first 2MB
1237 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1238 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1239 * the user if not.
1240 */
1241 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1242 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1243 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1244 WARN_ON(1);
1245 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1246 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1247 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1248 fix_to_virt(FIX_BTMAP_BEGIN));
1249 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1250 fix_to_virt(FIX_BTMAP_END));
1251
1252 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1253 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1254 }
1255#endif
1256
1257 pt_ops_set_fixmap();
1258}
1259
1260static void __meminit create_linear_mapping_range(phys_addr_t start, phys_addr_t end,
1261 uintptr_t fixed_map_size, const pgprot_t *pgprot)
1262{
1263 phys_addr_t pa;
1264 uintptr_t va, map_size;
1265
1266 for (pa = start; pa < end; pa += map_size) {
1267 va = (uintptr_t)__va(pa);
1268 map_size = fixed_map_size ? fixed_map_size :
1269 best_map_size(pa, va, end - pa);
1270
1271 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1272 pgprot ? *pgprot : pgprot_from_va(va));
1273 }
1274}
1275
1276static void __init create_linear_mapping_page_table(void)
1277{
1278 phys_addr_t start, end;
1279 phys_addr_t kfence_pool __maybe_unused;
1280 u64 i;
1281
1282#ifdef CONFIG_STRICT_KERNEL_RWX
1283 phys_addr_t ktext_start = __pa_symbol(_start);
1284 phys_addr_t ktext_size = __init_data_begin - _start;
1285 phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1286 phys_addr_t krodata_size = _data - __start_rodata;
1287
1288 /* Isolate kernel text and rodata so they don't get mapped with a PUD */
1289 memblock_mark_nomap(ktext_start, ktext_size);
1290 memblock_mark_nomap(krodata_start, krodata_size);
1291#endif
1292
1293#ifdef CONFIG_KFENCE
1294 /*
1295 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1296 * before we setup the linear mapping so that we avoid using hugepages
1297 * for this region.
1298 */
1299 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1300 BUG_ON(!kfence_pool);
1301
1302 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1303 __kfence_pool = __va(kfence_pool);
1304#endif
1305
1306 /* Map all memory banks in the linear mapping */
1307 for_each_mem_range(i, &start, &end) {
1308 if (start >= end)
1309 break;
1310 if (start <= __pa(PAGE_OFFSET) &&
1311 __pa(PAGE_OFFSET) < end)
1312 start = __pa(PAGE_OFFSET);
1313
1314 create_linear_mapping_range(start, end, 0, NULL);
1315 }
1316
1317#ifdef CONFIG_STRICT_KERNEL_RWX
1318 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0, NULL);
1319 create_linear_mapping_range(krodata_start, krodata_start + krodata_size, 0, NULL);
1320
1321 memblock_clear_nomap(ktext_start, ktext_size);
1322 memblock_clear_nomap(krodata_start, krodata_size);
1323#endif
1324
1325#ifdef CONFIG_KFENCE
1326 create_linear_mapping_range(kfence_pool, kfence_pool + KFENCE_POOL_SIZE, PAGE_SIZE, NULL);
1327
1328 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1329#endif
1330}
1331
1332static void __init setup_vm_final(void)
1333{
1334 /* Setup swapper PGD for fixmap */
1335#if !defined(CONFIG_64BIT)
1336 /*
1337 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1338 * directly in swapper_pg_dir in addition to the pgd entry that points
1339 * to fixmap_pte.
1340 */
1341 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1342
1343 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1344#endif
1345 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1346 __pa_symbol(fixmap_pgd_next),
1347 PGDIR_SIZE, PAGE_TABLE);
1348
1349 /* Map the linear mapping */
1350 create_linear_mapping_page_table();
1351
1352 /* Map the kernel */
1353 if (IS_ENABLED(CONFIG_64BIT))
1354 create_kernel_page_table(swapper_pg_dir, false);
1355
1356#ifdef CONFIG_KASAN
1357 kasan_swapper_init();
1358#endif
1359
1360 /* Clear fixmap PTE and PMD mappings */
1361 clear_fixmap(FIX_PTE);
1362 clear_fixmap(FIX_PMD);
1363 clear_fixmap(FIX_PUD);
1364 clear_fixmap(FIX_P4D);
1365
1366 /* Move to swapper page table */
1367 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1368 local_flush_tlb_all();
1369
1370 pt_ops_set_late();
1371}
1372#else
1373asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1374{
1375 dtb_early_va = (void *)dtb_pa;
1376 dtb_early_pa = dtb_pa;
1377}
1378
1379static inline void setup_vm_final(void)
1380{
1381}
1382#endif /* CONFIG_MMU */
1383
1384/*
1385 * reserve_crashkernel() - reserves memory for crash kernel
1386 *
1387 * This function reserves memory area given in "crashkernel=" kernel command
1388 * line parameter. The memory reserved is used by dump capture kernel when
1389 * primary kernel is crashing.
1390 */
1391static void __init arch_reserve_crashkernel(void)
1392{
1393 unsigned long long low_size = 0;
1394 unsigned long long crash_base, crash_size;
1395 char *cmdline = boot_command_line;
1396 bool high = false;
1397 int ret;
1398
1399 if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
1400 return;
1401
1402 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(),
1403 &crash_size, &crash_base,
1404 &low_size, &high);
1405 if (ret)
1406 return;
1407
1408 reserve_crashkernel_generic(cmdline, crash_size, crash_base,
1409 low_size, high);
1410}
1411
1412void __init paging_init(void)
1413{
1414 setup_bootmem();
1415 setup_vm_final();
1416
1417 /* Depend on that Linear Mapping is ready */
1418 memblock_allow_resize();
1419}
1420
1421void __init misc_mem_init(void)
1422{
1423 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1424 arch_numa_init();
1425 sparse_init();
1426#ifdef CONFIG_SPARSEMEM_VMEMMAP
1427 /* The entire VMEMMAP region has been populated. Flush TLB for this region */
1428 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END);
1429#endif
1430 zone_sizes_init();
1431 arch_reserve_crashkernel();
1432 memblock_dump_all();
1433}
1434
1435#ifdef CONFIG_SPARSEMEM_VMEMMAP
1436void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node,
1437 unsigned long addr, unsigned long next)
1438{
1439 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL);
1440}
1441
1442int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node,
1443 unsigned long addr, unsigned long next)
1444{
1445 vmemmap_verify((pte_t *)pmdp, node, addr, next);
1446 return 1;
1447}
1448
1449int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1450 struct vmem_altmap *altmap)
1451{
1452 /*
1453 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we
1454 * can't use hugepage mappings for 2-level page table because in case of
1455 * memory hotplug, we are not able to update all the page tables with
1456 * the new PMDs.
1457 */
1458 return vmemmap_populate_hugepages(start, end, node, altmap);
1459}
1460#endif
1461
1462#if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1463/*
1464 * Pre-allocates page-table pages for a specific area in the kernel
1465 * page-table. Only the level which needs to be synchronized between
1466 * all page-tables is allocated because the synchronization can be
1467 * expensive.
1468 */
1469static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1470 const char *area)
1471{
1472 unsigned long addr;
1473 const char *lvl;
1474
1475 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1476 pgd_t *pgd = pgd_offset_k(addr);
1477 p4d_t *p4d;
1478 pud_t *pud;
1479 pmd_t *pmd;
1480
1481 lvl = "p4d";
1482 p4d = p4d_alloc(&init_mm, pgd, addr);
1483 if (!p4d)
1484 goto failed;
1485
1486 if (pgtable_l5_enabled)
1487 continue;
1488
1489 lvl = "pud";
1490 pud = pud_alloc(&init_mm, p4d, addr);
1491 if (!pud)
1492 goto failed;
1493
1494 if (pgtable_l4_enabled)
1495 continue;
1496
1497 lvl = "pmd";
1498 pmd = pmd_alloc(&init_mm, pud, addr);
1499 if (!pmd)
1500 goto failed;
1501 }
1502 return;
1503
1504failed:
1505 /*
1506 * The pages have to be there now or they will be missing in
1507 * process page-tables later.
1508 */
1509 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1510}
1511
1512#define PAGE_END KASAN_SHADOW_START
1513
1514void __init pgtable_cache_init(void)
1515{
1516 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1517 if (IS_ENABLED(CONFIG_MODULES))
1518 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
1519 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) {
1520 preallocate_pgd_pages_range(VMEMMAP_START, VMEMMAP_END, "vmemmap");
1521 preallocate_pgd_pages_range(PAGE_OFFSET, PAGE_END, "direct map");
1522 if (IS_ENABLED(CONFIG_KASAN))
1523 preallocate_pgd_pages_range(KASAN_SHADOW_START, KASAN_SHADOW_END, "kasan");
1524 }
1525}
1526#endif
1527
1528#ifdef CONFIG_EXECMEM
1529#ifdef CONFIG_MMU
1530static struct execmem_info execmem_info __ro_after_init;
1531
1532struct execmem_info __init *execmem_arch_setup(void)
1533{
1534 execmem_info = (struct execmem_info){
1535 .ranges = {
1536 [EXECMEM_DEFAULT] = {
1537 .start = MODULES_VADDR,
1538 .end = MODULES_END,
1539 .pgprot = PAGE_KERNEL,
1540 .alignment = 1,
1541 },
1542 [EXECMEM_KPROBES] = {
1543 .start = VMALLOC_START,
1544 .end = VMALLOC_END,
1545 .pgprot = PAGE_KERNEL_READ_EXEC,
1546 .alignment = 1,
1547 },
1548 [EXECMEM_BPF] = {
1549 .start = BPF_JIT_REGION_START,
1550 .end = BPF_JIT_REGION_END,
1551 .pgprot = PAGE_KERNEL,
1552 .alignment = PAGE_SIZE,
1553 },
1554 },
1555 };
1556
1557 return &execmem_info;
1558}
1559#endif /* CONFIG_MMU */
1560#endif /* CONFIG_EXECMEM */
1561
1562#ifdef CONFIG_MEMORY_HOTPLUG
1563static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
1564{
1565 struct page *page = pmd_page(*pmd);
1566 struct ptdesc *ptdesc = page_ptdesc(page);
1567 pte_t *pte;
1568 int i;
1569
1570 for (i = 0; i < PTRS_PER_PTE; i++) {
1571 pte = pte_start + i;
1572 if (!pte_none(*pte))
1573 return;
1574 }
1575
1576 pagetable_pte_dtor(ptdesc);
1577 if (PageReserved(page))
1578 free_reserved_page(page);
1579 else
1580 pagetable_free(ptdesc);
1581 pmd_clear(pmd);
1582}
1583
1584static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud, bool is_vmemmap)
1585{
1586 struct page *page = pud_page(*pud);
1587 struct ptdesc *ptdesc = page_ptdesc(page);
1588 pmd_t *pmd;
1589 int i;
1590
1591 for (i = 0; i < PTRS_PER_PMD; i++) {
1592 pmd = pmd_start + i;
1593 if (!pmd_none(*pmd))
1594 return;
1595 }
1596
1597 if (!is_vmemmap)
1598 pagetable_pmd_dtor(ptdesc);
1599 if (PageReserved(page))
1600 free_reserved_page(page);
1601 else
1602 pagetable_free(ptdesc);
1603 pud_clear(pud);
1604}
1605
1606static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d)
1607{
1608 struct page *page = p4d_page(*p4d);
1609 pud_t *pud;
1610 int i;
1611
1612 for (i = 0; i < PTRS_PER_PUD; i++) {
1613 pud = pud_start + i;
1614 if (!pud_none(*pud))
1615 return;
1616 }
1617
1618 if (PageReserved(page))
1619 free_reserved_page(page);
1620 else
1621 free_pages((unsigned long)page_address(page), 0);
1622 p4d_clear(p4d);
1623}
1624
1625static void __meminit free_vmemmap_storage(struct page *page, size_t size,
1626 struct vmem_altmap *altmap)
1627{
1628 int order = get_order(size);
1629
1630 if (altmap) {
1631 vmem_altmap_free(altmap, size >> PAGE_SHIFT);
1632 return;
1633 }
1634
1635 if (PageReserved(page)) {
1636 unsigned int nr_pages = 1 << order;
1637
1638 while (nr_pages--)
1639 free_reserved_page(page++);
1640 return;
1641 }
1642
1643 free_pages((unsigned long)page_address(page), order);
1644}
1645
1646static void __meminit remove_pte_mapping(pte_t *pte_base, unsigned long addr, unsigned long end,
1647 bool is_vmemmap, struct vmem_altmap *altmap)
1648{
1649 unsigned long next;
1650 pte_t *ptep, pte;
1651
1652 for (; addr < end; addr = next) {
1653 next = (addr + PAGE_SIZE) & PAGE_MASK;
1654 if (next > end)
1655 next = end;
1656
1657 ptep = pte_base + pte_index(addr);
1658 pte = ptep_get(ptep);
1659 if (!pte_present(*ptep))
1660 continue;
1661
1662 pte_clear(&init_mm, addr, ptep);
1663 if (is_vmemmap)
1664 free_vmemmap_storage(pte_page(pte), PAGE_SIZE, altmap);
1665 }
1666}
1667
1668static void __meminit remove_pmd_mapping(pmd_t *pmd_base, unsigned long addr, unsigned long end,
1669 bool is_vmemmap, struct vmem_altmap *altmap)
1670{
1671 unsigned long next;
1672 pte_t *pte_base;
1673 pmd_t *pmdp, pmd;
1674
1675 for (; addr < end; addr = next) {
1676 next = pmd_addr_end(addr, end);
1677 pmdp = pmd_base + pmd_index(addr);
1678 pmd = pmdp_get(pmdp);
1679 if (!pmd_present(pmd))
1680 continue;
1681
1682 if (pmd_leaf(pmd)) {
1683 pmd_clear(pmdp);
1684 if (is_vmemmap)
1685 free_vmemmap_storage(pmd_page(pmd), PMD_SIZE, altmap);
1686 continue;
1687 }
1688
1689 pte_base = (pte_t *)pmd_page_vaddr(*pmdp);
1690 remove_pte_mapping(pte_base, addr, next, is_vmemmap, altmap);
1691 free_pte_table(pte_base, pmdp);
1692 }
1693}
1694
1695static void __meminit remove_pud_mapping(pud_t *pud_base, unsigned long addr, unsigned long end,
1696 bool is_vmemmap, struct vmem_altmap *altmap)
1697{
1698 unsigned long next;
1699 pud_t *pudp, pud;
1700 pmd_t *pmd_base;
1701
1702 for (; addr < end; addr = next) {
1703 next = pud_addr_end(addr, end);
1704 pudp = pud_base + pud_index(addr);
1705 pud = pudp_get(pudp);
1706 if (!pud_present(pud))
1707 continue;
1708
1709 if (pud_leaf(pud)) {
1710 if (pgtable_l4_enabled) {
1711 pud_clear(pudp);
1712 if (is_vmemmap)
1713 free_vmemmap_storage(pud_page(pud), PUD_SIZE, altmap);
1714 }
1715 continue;
1716 }
1717
1718 pmd_base = pmd_offset(pudp, 0);
1719 remove_pmd_mapping(pmd_base, addr, next, is_vmemmap, altmap);
1720
1721 if (pgtable_l4_enabled)
1722 free_pmd_table(pmd_base, pudp, is_vmemmap);
1723 }
1724}
1725
1726static void __meminit remove_p4d_mapping(p4d_t *p4d_base, unsigned long addr, unsigned long end,
1727 bool is_vmemmap, struct vmem_altmap *altmap)
1728{
1729 unsigned long next;
1730 p4d_t *p4dp, p4d;
1731 pud_t *pud_base;
1732
1733 for (; addr < end; addr = next) {
1734 next = p4d_addr_end(addr, end);
1735 p4dp = p4d_base + p4d_index(addr);
1736 p4d = p4dp_get(p4dp);
1737 if (!p4d_present(p4d))
1738 continue;
1739
1740 if (p4d_leaf(p4d)) {
1741 if (pgtable_l5_enabled) {
1742 p4d_clear(p4dp);
1743 if (is_vmemmap)
1744 free_vmemmap_storage(p4d_page(p4d), P4D_SIZE, altmap);
1745 }
1746 continue;
1747 }
1748
1749 pud_base = pud_offset(p4dp, 0);
1750 remove_pud_mapping(pud_base, addr, next, is_vmemmap, altmap);
1751
1752 if (pgtable_l5_enabled)
1753 free_pud_table(pud_base, p4dp);
1754 }
1755}
1756
1757static void __meminit remove_pgd_mapping(unsigned long va, unsigned long end, bool is_vmemmap,
1758 struct vmem_altmap *altmap)
1759{
1760 unsigned long addr, next;
1761 p4d_t *p4d_base;
1762 pgd_t *pgd;
1763
1764 for (addr = va; addr < end; addr = next) {
1765 next = pgd_addr_end(addr, end);
1766 pgd = pgd_offset_k(addr);
1767
1768 if (!pgd_present(*pgd))
1769 continue;
1770
1771 if (pgd_leaf(*pgd))
1772 continue;
1773
1774 p4d_base = p4d_offset(pgd, 0);
1775 remove_p4d_mapping(p4d_base, addr, next, is_vmemmap, altmap);
1776 }
1777
1778 flush_tlb_all();
1779}
1780
1781static void __meminit remove_linear_mapping(phys_addr_t start, u64 size)
1782{
1783 unsigned long va = (unsigned long)__va(start);
1784 unsigned long end = (unsigned long)__va(start + size);
1785
1786 remove_pgd_mapping(va, end, false, NULL);
1787}
1788
1789struct range arch_get_mappable_range(void)
1790{
1791 struct range mhp_range;
1792
1793 mhp_range.start = __pa(PAGE_OFFSET);
1794 mhp_range.end = __pa(PAGE_END - 1);
1795 return mhp_range;
1796}
1797
1798int __ref arch_add_memory(int nid, u64 start, u64 size, struct mhp_params *params)
1799{
1800 int ret = 0;
1801
1802 create_linear_mapping_range(start, start + size, 0, ¶ms->pgprot);
1803 ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT, params);
1804 if (ret) {
1805 remove_linear_mapping(start, size);
1806 goto out;
1807 }
1808
1809 max_pfn = PFN_UP(start + size);
1810 max_low_pfn = max_pfn;
1811
1812 out:
1813 flush_tlb_all();
1814 return ret;
1815}
1816
1817void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
1818{
1819 __remove_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT, altmap);
1820 remove_linear_mapping(start, size);
1821 flush_tlb_all();
1822}
1823
1824void __ref vmemmap_free(unsigned long start, unsigned long end, struct vmem_altmap *altmap)
1825{
1826 remove_pgd_mapping(start, end, true, altmap);
1827}
1828#endif /* CONFIG_MEMORY_HOTPLUG */