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