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 #include <linux/kfence.h>
24 #include <linux/execmem.h>
25
26 #include <asm/alternative.h>
27 #include <asm/fixmap.h>
28 #include <asm/io.h>
29 #include <asm/kasan.h>
30 #include <asm/module.h>
31 #include <asm/numa.h>
32 #include <asm/pgtable.h>
33 #include <asm/sections.h>
34 #include <asm/soc.h>
35 #include <asm/sparsemem.h>
36 #include <asm/tlbflush.h>
37
38 #include "../kernel/head.h"
39
40 u64 new_vmalloc[NR_CPUS / sizeof(u64) + 1];
41
42 struct kernel_mapping kernel_map __ro_after_init;
43 EXPORT_SYMBOL(kernel_map);
44
45 #ifdef CONFIG_64BIT
46 u64 satp_mode __ro_after_init = SATP_MODE_57;
47 #else
48 u64 satp_mode __ro_after_init = SATP_MODE_32;
49 #endif
50 EXPORT_SYMBOL(satp_mode);
51
52 #ifdef CONFIG_64BIT
53 bool pgtable_l4_enabled __ro_after_init = true;
54 bool pgtable_l5_enabled __ro_after_init = true;
55 EXPORT_SYMBOL(pgtable_l4_enabled);
56 EXPORT_SYMBOL(pgtable_l5_enabled);
57 #endif
58
59 phys_addr_t phys_ram_base __ro_after_init;
60 EXPORT_SYMBOL(phys_ram_base);
61
62 #ifdef CONFIG_SPARSEMEM_VMEMMAP
63 #define VMEMMAP_ADDR_ALIGN max(1ULL << SECTION_SIZE_BITS, \
64 MAX_FOLIO_VMEMMAP_ALIGN)
65
66 unsigned long vmemmap_start_pfn __ro_after_init;
67 EXPORT_SYMBOL(vmemmap_start_pfn);
68 #endif
69
70 extern char _start[];
71 void *_dtb_early_va __initdata;
72 uintptr_t _dtb_early_pa __initdata;
73
74 phys_addr_t dma32_phys_limit __initdata;
75
arch_zone_limits_init(unsigned long * max_zone_pfns)76 void __init arch_zone_limits_init(unsigned long *max_zone_pfns)
77 {
78 #ifdef CONFIG_ZONE_DMA32
79 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
80 #endif
81 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
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
print_mlk(char * name,unsigned long b,unsigned long t)91 static 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
print_mlm(char * name,unsigned long b,unsigned long t)97 static 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
print_mlg(char * name,unsigned long b,unsigned long t)103 static 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
print_mlt(char * name,unsigned long b,unsigned long t)110 static 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
print_ml(char * name,unsigned long b,unsigned long t)119 static 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
print_vm_layout(void)133 static 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
print_vm_layout(void)160 static void print_vm_layout(void) { }
161 #endif /* CONFIG_DEBUG_VM */
162
arch_mm_preinit(void)163 void __init arch_mm_preinit(void)
164 {
165 bool swiotlb = max_pfn > PFN_DOWN(dma32_phys_limit);
166 #ifdef CONFIG_FLATMEM
167 BUG_ON(!mem_map);
168 #endif /* CONFIG_FLATMEM */
169
170 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && !swiotlb &&
171 dma_cache_alignment != 1) {
172 /*
173 * If no bouncing needed for ZONE_DMA, allocate 1MB swiotlb
174 * buffer per 1GB of RAM for kmalloc() bouncing on
175 * non-coherent platforms.
176 */
177 unsigned long size =
178 DIV_ROUND_UP(memblock_phys_mem_size(), 1024);
179 swiotlb_adjust_size(min(swiotlb_size_or_default(), size));
180 swiotlb = true;
181 }
182
183 swiotlb_init(swiotlb, SWIOTLB_VERBOSE);
184
185 print_vm_layout();
186 }
187
188 /* Limit the memory size via mem. */
189 static phys_addr_t memory_limit;
190
early_mem(char * p)191 static int __init early_mem(char *p)
192 {
193 u64 size;
194
195 if (!p)
196 return 1;
197
198 size = memparse(p, &p) & PAGE_MASK;
199 memory_limit = min_t(u64, size, memory_limit);
200
201 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
202
203 return 0;
204 }
205 early_param("mem", early_mem);
206
setup_bootmem(void)207 static void __init setup_bootmem(void)
208 {
209 phys_addr_t vmlinux_end = __pa_symbol(&_end);
210 phys_addr_t max_mapped_addr;
211 phys_addr_t phys_ram_end, vmlinux_start;
212
213 vmlinux_start = __pa_symbol(&_start);
214
215 memblock_enforce_memory_limit(memory_limit);
216
217 /*
218 * Make sure we align the reservation on PMD_SIZE since we will
219 * map the kernel in the linear mapping as read-only: we do not want
220 * any allocation to happen between _end and the next pmd aligned page.
221 */
222 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
223 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
224 /*
225 * Reserve from the start of the kernel to the end of the kernel
226 */
227 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
228
229 /*
230 * Make sure we align the start of the memory on a PMD boundary so that
231 * at worst, we map the linear mapping with PMD mappings.
232 */
233 phys_ram_base = memblock_start_of_DRAM() & PMD_MASK;
234 #ifdef CONFIG_SPARSEMEM_VMEMMAP
235 vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT;
236 #endif
237
238 /*
239 * In 64-bit, any use of __va/__pa before this point is wrong as we
240 * did not know the start of DRAM before.
241 */
242 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU))
243 kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base;
244
245 /*
246 * The size of the linear page mapping may restrict the amount of
247 * usable RAM.
248 */
249 if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) {
250 max_mapped_addr = __pa(PAGE_OFFSET) + KERN_VIRT_SIZE;
251 if (memblock_end_of_DRAM() > max_mapped_addr) {
252 memblock_cap_memory_range(phys_ram_base,
253 max_mapped_addr - phys_ram_base);
254 pr_warn("Physical memory overflows the linear mapping size: region above %pa removed",
255 &max_mapped_addr);
256 }
257 }
258
259 /*
260 * Reserve physical address space that would be mapped to virtual
261 * addresses greater than (void *)(-PAGE_SIZE) because:
262 * - This memory would overlap with ERR_PTR
263 * - This memory belongs to high memory, which is not supported
264 *
265 * This is not applicable to 64-bit kernel, because virtual addresses
266 * after (void *)(-PAGE_SIZE) are not linearly mapped: they are
267 * occupied by kernel mapping. Also it is unrealistic for high memory
268 * to exist on 64-bit platforms.
269 */
270 if (!IS_ENABLED(CONFIG_64BIT)) {
271 max_mapped_addr = __va_to_pa_nodebug(-PAGE_SIZE);
272 memblock_reserve(max_mapped_addr, (phys_addr_t)-max_mapped_addr);
273 }
274
275 phys_ram_end = memblock_end_of_DRAM();
276 min_low_pfn = PFN_UP(phys_ram_base);
277 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
278
279 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
280
281 reserve_initrd_mem();
282
283 /*
284 * No allocation should be done before reserving the memory as defined
285 * in the device tree, otherwise the allocation could end up in a
286 * reserved region.
287 */
288 early_init_fdt_scan_reserved_mem();
289
290 /*
291 * If DTB is built in, no need to reserve its memblock.
292 * Otherwise, do reserve it but avoid using
293 * early_init_fdt_reserve_self() since __pa() does
294 * not work for DTB pointers that are fixmap addresses
295 */
296 if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
297 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
298
299 dma_contiguous_reserve(dma32_phys_limit);
300 }
301
302 #ifdef CONFIG_RELOCATABLE
303 extern unsigned long __rela_dyn_start, __rela_dyn_end;
304
relocate_kernel(void)305 static void __init relocate_kernel(void)
306 {
307 Elf_Rela *rela = (Elf_Rela *)&__rela_dyn_start;
308 /*
309 * This holds the offset between the linked virtual address and the
310 * relocated virtual address.
311 */
312 uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
313 /*
314 * This holds the offset between kernel linked virtual address and
315 * physical address.
316 */
317 uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
318
319 for ( ; rela < (Elf_Rela *)&__rela_dyn_end; rela++) {
320 Elf_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
321 Elf_Addr relocated_addr = rela->r_addend;
322
323 if (rela->r_info != R_RISCV_RELATIVE)
324 continue;
325
326 /*
327 * Make sure to not relocate vdso symbols like rt_sigreturn
328 * which are linked from the address 0 in vmlinux since
329 * vdso symbol addresses are actually used as an offset from
330 * mm->context.vdso in VDSO_OFFSET macro.
331 */
332 if (relocated_addr >= KERNEL_LINK_ADDR)
333 relocated_addr += reloc_offset;
334
335 *(Elf_Addr *)addr = relocated_addr;
336 }
337 }
338 #endif /* CONFIG_RELOCATABLE */
339
340 #ifdef CONFIG_MMU
341 struct pt_alloc_ops pt_ops __meminitdata;
342
343 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
344 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
345 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
346
347 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
348
349 static const pgprot_t protection_map[16] = {
350 [VM_NONE] = PAGE_NONE,
351 [VM_READ] = PAGE_READ,
352 [VM_WRITE] = PAGE_SHADOWSTACK,
353 [VM_WRITE | VM_READ] = PAGE_COPY,
354 [VM_EXEC] = PAGE_EXEC,
355 [VM_EXEC | VM_READ] = PAGE_READ_EXEC,
356 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC,
357 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC,
358 [VM_SHARED] = PAGE_NONE,
359 [VM_SHARED | VM_READ] = PAGE_READ,
360 [VM_SHARED | VM_WRITE] = PAGE_SHARED,
361 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED,
362 [VM_SHARED | VM_EXEC] = PAGE_EXEC,
363 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_READ_EXEC,
364 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_SHARED_EXEC,
365 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_SHARED_EXEC
366 };
367 DECLARE_VM_GET_PAGE_PROT
368
__set_fixmap(enum fixed_addresses idx,phys_addr_t phys,pgprot_t prot)369 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
370 {
371 unsigned long addr = __fix_to_virt(idx);
372 pte_t *ptep;
373
374 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
375
376 ptep = &fixmap_pte[pte_index(addr)];
377
378 if (pgprot_val(prot))
379 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
380 else
381 pte_clear(&init_mm, addr, ptep);
382 local_flush_tlb_page(addr);
383 }
384
get_pte_virt_early(phys_addr_t pa)385 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
386 {
387 return (pte_t *)((uintptr_t)pa);
388 }
389
get_pte_virt_fixmap(phys_addr_t pa)390 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
391 {
392 clear_fixmap(FIX_PTE);
393 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
394 }
395
get_pte_virt_late(phys_addr_t pa)396 static inline pte_t *__meminit get_pte_virt_late(phys_addr_t pa)
397 {
398 return (pte_t *) __va(pa);
399 }
400
alloc_pte_early(uintptr_t va)401 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
402 {
403 /*
404 * We only create PMD or PGD early mappings so we
405 * should never reach here with MMU disabled.
406 */
407 BUG();
408 }
409
alloc_pte_fixmap(uintptr_t va)410 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
411 {
412 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
413 }
414
alloc_pte_late(uintptr_t va)415 static phys_addr_t __meminit alloc_pte_late(uintptr_t va)
416 {
417 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0);
418
419 /*
420 * We do not know which mm the PTE page is associated to at this point.
421 * Passing NULL to the ctor is the safe option, though it may result
422 * in unnecessary work (e.g. initialising the ptlock for init_mm).
423 */
424 BUG_ON(!ptdesc || !pagetable_pte_ctor(NULL, ptdesc));
425 return __pa((pte_t *)ptdesc_address(ptdesc));
426 }
427
create_pte_mapping(pte_t * ptep,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)428 static void __meminit create_pte_mapping(pte_t *ptep, uintptr_t va, phys_addr_t pa, phys_addr_t sz,
429 pgprot_t prot)
430 {
431 uintptr_t pte_idx = pte_index(va);
432
433 BUG_ON(sz != PAGE_SIZE);
434
435 if (pte_none(ptep[pte_idx]))
436 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
437 }
438
439 #ifndef __PAGETABLE_PMD_FOLDED
440
441 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
442 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
443 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
444
445 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
446 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
447 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
448
449 static pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
450 static pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
451 static pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
452
get_pmd_virt_early(phys_addr_t pa)453 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
454 {
455 /* Before MMU is enabled */
456 return (pmd_t *)((uintptr_t)pa);
457 }
458
get_pmd_virt_fixmap(phys_addr_t pa)459 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
460 {
461 clear_fixmap(FIX_PMD);
462 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
463 }
464
get_pmd_virt_late(phys_addr_t pa)465 static pmd_t *__meminit get_pmd_virt_late(phys_addr_t pa)
466 {
467 return (pmd_t *) __va(pa);
468 }
469
alloc_pmd_early(uintptr_t va)470 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
471 {
472 BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
473
474 return (uintptr_t)early_pmd;
475 }
476
alloc_pmd_fixmap(uintptr_t va)477 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
478 {
479 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
480 }
481
alloc_pmd_late(uintptr_t va)482 static phys_addr_t __meminit alloc_pmd_late(uintptr_t va)
483 {
484 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0);
485
486 /* See comment in alloc_pte_late() regarding NULL passed the ctor */
487 BUG_ON(!ptdesc || !pagetable_pmd_ctor(NULL, ptdesc));
488 return __pa((pmd_t *)ptdesc_address(ptdesc));
489 }
490
create_pmd_mapping(pmd_t * pmdp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)491 static void __meminit create_pmd_mapping(pmd_t *pmdp,
492 uintptr_t va, phys_addr_t pa,
493 phys_addr_t sz, pgprot_t prot)
494 {
495 pte_t *ptep;
496 phys_addr_t pte_phys;
497 uintptr_t pmd_idx = pmd_index(va);
498
499 if (sz == PMD_SIZE) {
500 if (pmd_none(pmdp[pmd_idx]))
501 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
502 return;
503 }
504
505 if (pmd_none(pmdp[pmd_idx])) {
506 pte_phys = pt_ops.alloc_pte(va);
507 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
508 ptep = pt_ops.get_pte_virt(pte_phys);
509 memset(ptep, 0, PAGE_SIZE);
510 } else {
511 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
512 ptep = pt_ops.get_pte_virt(pte_phys);
513 }
514
515 create_pte_mapping(ptep, va, pa, sz, prot);
516 }
517
get_pud_virt_early(phys_addr_t pa)518 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
519 {
520 return (pud_t *)((uintptr_t)pa);
521 }
522
get_pud_virt_fixmap(phys_addr_t pa)523 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
524 {
525 clear_fixmap(FIX_PUD);
526 return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
527 }
528
get_pud_virt_late(phys_addr_t pa)529 static pud_t *__meminit get_pud_virt_late(phys_addr_t pa)
530 {
531 return (pud_t *)__va(pa);
532 }
533
alloc_pud_early(uintptr_t va)534 static phys_addr_t __init alloc_pud_early(uintptr_t va)
535 {
536 /* Only one PUD is available for early mapping */
537 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
538
539 return (uintptr_t)early_pud;
540 }
541
alloc_pud_fixmap(uintptr_t va)542 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
543 {
544 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
545 }
546
alloc_pud_late(uintptr_t va)547 static phys_addr_t __meminit alloc_pud_late(uintptr_t va)
548 {
549 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0);
550
551 BUG_ON(!ptdesc);
552 pagetable_pud_ctor(ptdesc);
553 return __pa((pud_t *)ptdesc_address(ptdesc));
554 }
555
get_p4d_virt_early(phys_addr_t pa)556 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
557 {
558 return (p4d_t *)((uintptr_t)pa);
559 }
560
get_p4d_virt_fixmap(phys_addr_t pa)561 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
562 {
563 clear_fixmap(FIX_P4D);
564 return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
565 }
566
get_p4d_virt_late(phys_addr_t pa)567 static p4d_t *__meminit get_p4d_virt_late(phys_addr_t pa)
568 {
569 return (p4d_t *)__va(pa);
570 }
571
alloc_p4d_early(uintptr_t va)572 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
573 {
574 /* Only one P4D is available for early mapping */
575 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
576
577 return (uintptr_t)early_p4d;
578 }
579
alloc_p4d_fixmap(uintptr_t va)580 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
581 {
582 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
583 }
584
alloc_p4d_late(uintptr_t va)585 static phys_addr_t __meminit alloc_p4d_late(uintptr_t va)
586 {
587 struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL, 0);
588
589 BUG_ON(!ptdesc);
590 pagetable_p4d_ctor(ptdesc);
591 return __pa((p4d_t *)ptdesc_address(ptdesc));
592 }
593
create_pud_mapping(pud_t * pudp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)594 static void __meminit create_pud_mapping(pud_t *pudp, uintptr_t va, phys_addr_t pa, phys_addr_t sz,
595 pgprot_t prot)
596 {
597 pmd_t *nextp;
598 phys_addr_t next_phys;
599 uintptr_t pud_index = pud_index(va);
600
601 if (sz == PUD_SIZE) {
602 if (pud_val(pudp[pud_index]) == 0)
603 pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
604 return;
605 }
606
607 if (pud_val(pudp[pud_index]) == 0) {
608 next_phys = pt_ops.alloc_pmd(va);
609 pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
610 nextp = pt_ops.get_pmd_virt(next_phys);
611 memset(nextp, 0, PAGE_SIZE);
612 } else {
613 next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
614 nextp = pt_ops.get_pmd_virt(next_phys);
615 }
616
617 create_pmd_mapping(nextp, va, pa, sz, prot);
618 }
619
create_p4d_mapping(p4d_t * p4dp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)620 static void __meminit create_p4d_mapping(p4d_t *p4dp, uintptr_t va, phys_addr_t pa, phys_addr_t sz,
621 pgprot_t prot)
622 {
623 pud_t *nextp;
624 phys_addr_t next_phys;
625 uintptr_t p4d_index = p4d_index(va);
626
627 if (sz == P4D_SIZE) {
628 if (p4d_val(p4dp[p4d_index]) == 0)
629 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
630 return;
631 }
632
633 if (p4d_val(p4dp[p4d_index]) == 0) {
634 next_phys = pt_ops.alloc_pud(va);
635 p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
636 nextp = pt_ops.get_pud_virt(next_phys);
637 memset(nextp, 0, PAGE_SIZE);
638 } else {
639 next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
640 nextp = pt_ops.get_pud_virt(next_phys);
641 }
642
643 create_pud_mapping(nextp, va, pa, sz, prot);
644 }
645
646 #define pgd_next_t p4d_t
647 #define alloc_pgd_next(__va) (pgtable_l5_enabled ? \
648 pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ? \
649 pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
650 #define get_pgd_next_virt(__pa) (pgtable_l5_enabled ? \
651 pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ? \
652 pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
653 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
654 (pgtable_l5_enabled ? \
655 create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
656 (pgtable_l4_enabled ? \
657 create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) : \
658 create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
659 #define fixmap_pgd_next (pgtable_l5_enabled ? \
660 (uintptr_t)fixmap_p4d : (pgtable_l4_enabled ? \
661 (uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
662 #define trampoline_pgd_next (pgtable_l5_enabled ? \
663 (uintptr_t)trampoline_p4d : (pgtable_l4_enabled ? \
664 (uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
665 #else
666 #define pgd_next_t pte_t
667 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
668 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
669 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
670 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
671 #define fixmap_pgd_next ((uintptr_t)fixmap_pte)
672 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
673 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
674 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
675 #endif /* __PAGETABLE_PMD_FOLDED */
676
create_pgd_mapping(pgd_t * pgdp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)677 void __meminit create_pgd_mapping(pgd_t *pgdp, uintptr_t va, phys_addr_t pa, phys_addr_t sz,
678 pgprot_t prot)
679 {
680 pgd_next_t *nextp;
681 phys_addr_t next_phys;
682 uintptr_t pgd_idx = pgd_index(va);
683
684 if (sz == PGDIR_SIZE) {
685 if (pgd_val(pgdp[pgd_idx]) == 0)
686 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
687 return;
688 }
689
690 if (pgd_val(pgdp[pgd_idx]) == 0) {
691 next_phys = alloc_pgd_next(va);
692 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
693 nextp = get_pgd_next_virt(next_phys);
694 memset(nextp, 0, PAGE_SIZE);
695 } else {
696 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
697 nextp = get_pgd_next_virt(next_phys);
698 }
699
700 create_pgd_next_mapping(nextp, va, pa, sz, prot);
701 }
702
best_map_size(phys_addr_t pa,uintptr_t va,phys_addr_t size)703 static uintptr_t __meminit best_map_size(phys_addr_t pa, uintptr_t va, phys_addr_t size)
704 {
705 if (debug_pagealloc_enabled())
706 return PAGE_SIZE;
707
708 if (pgtable_l5_enabled &&
709 !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
710 return P4D_SIZE;
711
712 if (pgtable_l4_enabled &&
713 !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
714 return PUD_SIZE;
715
716 if (IS_ENABLED(CONFIG_64BIT) &&
717 !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
718 return PMD_SIZE;
719
720 return PAGE_SIZE;
721 }
722
723 #ifdef CONFIG_STRICT_KERNEL_RWX
pgprot_from_va(uintptr_t va)724 static __meminit pgprot_t pgprot_from_va(uintptr_t va)
725 {
726 if (is_va_kernel_text(va))
727 return PAGE_KERNEL_READ_EXEC;
728
729 /*
730 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
731 * we must protect its linear mapping alias from being executed and
732 * written.
733 * And rodata section is marked readonly in mark_rodata_ro.
734 */
735 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
736 return PAGE_KERNEL_READ;
737
738 return PAGE_KERNEL;
739 }
740
mark_rodata_ro(void)741 void mark_rodata_ro(void)
742 {
743 set_kernel_memory(__start_rodata, _data, set_memory_ro);
744 if (IS_ENABLED(CONFIG_64BIT))
745 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
746 set_memory_ro);
747 }
748 #else
pgprot_from_va(uintptr_t va)749 static __meminit pgprot_t pgprot_from_va(uintptr_t va)
750 {
751 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
752 return PAGE_KERNEL;
753
754 return PAGE_KERNEL_EXEC;
755 }
756 #endif /* CONFIG_STRICT_KERNEL_RWX */
757
758 #if defined(CONFIG_64BIT)
759 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
760 u64 __pi_set_satp_mode_from_fdt(uintptr_t dtb_pa);
761
disable_pgtable_l5(void)762 static void __init disable_pgtable_l5(void)
763 {
764 pgtable_l5_enabled = false;
765 kernel_map.page_offset = PAGE_OFFSET_L4;
766 satp_mode = SATP_MODE_48;
767 }
768
disable_pgtable_l4(void)769 static void __init disable_pgtable_l4(void)
770 {
771 pgtable_l4_enabled = false;
772 kernel_map.page_offset = PAGE_OFFSET_L3;
773 satp_mode = SATP_MODE_39;
774 }
775
print_no4lvl(char * p)776 static int __init print_no4lvl(char *p)
777 {
778 pr_info("Disabled 4-level and 5-level paging");
779 return 0;
780 }
781 early_param("no4lvl", print_no4lvl);
782
print_no5lvl(char * p)783 static int __init print_no5lvl(char *p)
784 {
785 pr_info("Disabled 5-level paging");
786 return 0;
787 }
788 early_param("no5lvl", print_no5lvl);
789
set_mmap_rnd_bits_max(void)790 static void __init set_mmap_rnd_bits_max(void)
791 {
792 mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3;
793 }
794
is_vaddr_valid(unsigned long va)795 static bool __init is_vaddr_valid(unsigned long va)
796 {
797 unsigned long up = 0;
798
799 switch (satp_mode) {
800 case SATP_MODE_39:
801 up = 1UL << 38;
802 break;
803 case SATP_MODE_48:
804 up = 1UL << 47;
805 break;
806 case SATP_MODE_57:
807 up = 1UL << 56;
808 break;
809 default:
810 return false;
811 }
812
813 return (va < up) || (va >= (ULONG_MAX - up + 1));
814 }
815
816 /*
817 * There is a simple way to determine if 4-level is supported by the
818 * underlying hardware: establish 1:1 mapping in 4-level page table mode
819 * then read SATP to see if the configuration was taken into account
820 * meaning sv48 is supported.
821 * The maximum SATP mode is limited by both the command line and the "mmu-type"
822 * property in the device tree, since some platforms may hang if an unsupported
823 * SATP mode is attempted.
824 */
set_satp_mode(uintptr_t dtb_pa)825 static __init void set_satp_mode(uintptr_t dtb_pa)
826 {
827 u64 identity_satp, hw_satp;
828 uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
829 u64 satp_mode_limit = min_not_zero(__pi_set_satp_mode_from_cmdline(dtb_pa),
830 __pi_set_satp_mode_from_fdt(dtb_pa));
831
832 kernel_map.page_offset = PAGE_OFFSET_L5;
833
834 if (satp_mode_limit == SATP_MODE_48) {
835 disable_pgtable_l5();
836 } else if (satp_mode_limit == SATP_MODE_39) {
837 disable_pgtable_l5();
838 disable_pgtable_l4();
839 return;
840 }
841
842 create_p4d_mapping(early_p4d,
843 set_satp_mode_pmd, (uintptr_t)early_pud,
844 P4D_SIZE, PAGE_TABLE);
845 create_pud_mapping(early_pud,
846 set_satp_mode_pmd, (uintptr_t)early_pmd,
847 PUD_SIZE, PAGE_TABLE);
848 /* Handle the case where set_satp_mode straddles 2 PMDs */
849 create_pmd_mapping(early_pmd,
850 set_satp_mode_pmd, set_satp_mode_pmd,
851 PMD_SIZE, PAGE_KERNEL_EXEC);
852 create_pmd_mapping(early_pmd,
853 set_satp_mode_pmd + PMD_SIZE,
854 set_satp_mode_pmd + PMD_SIZE,
855 PMD_SIZE, PAGE_KERNEL_EXEC);
856 retry:
857 if (!is_vaddr_valid(set_satp_mode_pmd))
858 goto out;
859
860 create_pgd_mapping(early_pg_dir,
861 set_satp_mode_pmd,
862 pgtable_l5_enabled ?
863 (uintptr_t)early_p4d : (uintptr_t)early_pud,
864 PGDIR_SIZE, PAGE_TABLE);
865
866 identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
867
868 local_flush_tlb_all();
869 csr_write(CSR_SATP, identity_satp);
870 hw_satp = csr_swap(CSR_SATP, 0ULL);
871 local_flush_tlb_all();
872
873 if (hw_satp != identity_satp) {
874 if (pgtable_l5_enabled) {
875 disable_pgtable_l5();
876 memset(early_pg_dir, 0, PAGE_SIZE);
877 goto retry;
878 }
879 disable_pgtable_l4();
880 }
881
882 out:
883 memset(early_pg_dir, 0, PAGE_SIZE);
884 memset(early_p4d, 0, PAGE_SIZE);
885 memset(early_pud, 0, PAGE_SIZE);
886 memset(early_pmd, 0, PAGE_SIZE);
887 }
888 #endif
889
890 /*
891 * setup_vm() is called from head.S with MMU-off.
892 *
893 * Following requirements should be honoured for setup_vm() to work
894 * correctly:
895 * 1) It should use PC-relative addressing for accessing kernel symbols.
896 * To achieve this we always use GCC cmodel=medany.
897 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
898 * so disable compiler instrumentation when FTRACE is enabled.
899 *
900 * Currently, the above requirements are honoured by using custom CFLAGS
901 * for init.o in mm/Makefile.
902 */
903
904 #ifndef __riscv_cmodel_medany
905 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
906 #endif
907
create_kernel_page_table(pgd_t * pgdir,bool early)908 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
909 {
910 uintptr_t va, end_va;
911
912 end_va = kernel_map.virt_addr + kernel_map.size;
913 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
914 create_pgd_mapping(pgdir, va,
915 kernel_map.phys_addr + (va - kernel_map.virt_addr),
916 PMD_SIZE,
917 early ?
918 PAGE_KERNEL_EXEC : pgprot_from_va(va));
919 }
920
921 /*
922 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
923 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
924 * entry.
925 */
create_fdt_early_page_table(uintptr_t fix_fdt_va,uintptr_t dtb_pa)926 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
927 uintptr_t dtb_pa)
928 {
929 #ifndef CONFIG_BUILTIN_DTB
930 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
931
932 /* Make sure the fdt fixmap address is always aligned on PMD size */
933 BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
934
935 /* In 32-bit only, the fdt lies in its own PGD */
936 if (!IS_ENABLED(CONFIG_64BIT)) {
937 create_pgd_mapping(early_pg_dir, fix_fdt_va,
938 pa, MAX_FDT_SIZE, PAGE_KERNEL);
939 } else {
940 create_pmd_mapping(fixmap_pmd, fix_fdt_va,
941 pa, PMD_SIZE, PAGE_KERNEL);
942 create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
943 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
944 }
945
946 dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
947 #else
948 /*
949 * For 64-bit kernel, __va can't be used since it would return a linear
950 * mapping address whereas dtb_early_va will be used before
951 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
952 * kernel is mapped in the linear mapping, that makes no difference.
953 */
954 dtb_early_va = kernel_mapping_pa_to_va(dtb_pa);
955 #endif
956
957 dtb_early_pa = dtb_pa;
958 }
959
960 /*
961 * MMU is not enabled, the page tables are allocated directly using
962 * early_pmd/pud/p4d and the address returned is the physical one.
963 */
pt_ops_set_early(void)964 static void __init pt_ops_set_early(void)
965 {
966 pt_ops.alloc_pte = alloc_pte_early;
967 pt_ops.get_pte_virt = get_pte_virt_early;
968 #ifndef __PAGETABLE_PMD_FOLDED
969 pt_ops.alloc_pmd = alloc_pmd_early;
970 pt_ops.get_pmd_virt = get_pmd_virt_early;
971 pt_ops.alloc_pud = alloc_pud_early;
972 pt_ops.get_pud_virt = get_pud_virt_early;
973 pt_ops.alloc_p4d = alloc_p4d_early;
974 pt_ops.get_p4d_virt = get_p4d_virt_early;
975 #endif
976 }
977
978 /*
979 * MMU is enabled but page table setup is not complete yet.
980 * fixmap page table alloc functions must be used as a means to temporarily
981 * map the allocated physical pages since the linear mapping does not exist yet.
982 *
983 * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
984 * but it will be used as described above.
985 */
pt_ops_set_fixmap(void)986 static void __init pt_ops_set_fixmap(void)
987 {
988 pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
989 pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
990 #ifndef __PAGETABLE_PMD_FOLDED
991 pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
992 pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
993 pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
994 pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
995 pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
996 pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
997 #endif
998 }
999
1000 /*
1001 * MMU is enabled and page table setup is complete, so from now, we can use
1002 * generic page allocation functions to setup page table.
1003 */
pt_ops_set_late(void)1004 static void __init pt_ops_set_late(void)
1005 {
1006 pt_ops.alloc_pte = alloc_pte_late;
1007 pt_ops.get_pte_virt = get_pte_virt_late;
1008 #ifndef __PAGETABLE_PMD_FOLDED
1009 pt_ops.alloc_pmd = alloc_pmd_late;
1010 pt_ops.get_pmd_virt = get_pmd_virt_late;
1011 pt_ops.alloc_pud = alloc_pud_late;
1012 pt_ops.get_pud_virt = get_pud_virt_late;
1013 pt_ops.alloc_p4d = alloc_p4d_late;
1014 pt_ops.get_p4d_virt = get_p4d_virt_late;
1015 #endif
1016 }
1017
1018 #ifdef CONFIG_RANDOMIZE_BASE
1019 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa);
1020 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa);
1021 extern u64 __init __pi_get_kaslr_seed_zkr(const uintptr_t dtb_pa);
1022
print_nokaslr(char * p)1023 static int __init print_nokaslr(char *p)
1024 {
1025 pr_info("Disabled KASLR");
1026 return 0;
1027 }
1028 early_param("nokaslr", print_nokaslr);
1029 #endif
1030
setup_vm(uintptr_t dtb_pa)1031 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1032 {
1033 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1034
1035 #ifdef CONFIG_RANDOMIZE_BASE
1036 if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) {
1037 u64 kaslr_seed = __pi_get_kaslr_seed_zkr(dtb_pa);
1038 u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
1039 u32 nr_pos;
1040
1041 if (kaslr_seed == 0)
1042 kaslr_seed = __pi_get_kaslr_seed(dtb_pa);
1043 /*
1044 * Compute the number of positions available: we are limited
1045 * by the early page table that only has one PUD and we must
1046 * be aligned on PMD_SIZE.
1047 */
1048 nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE;
1049
1050 kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE;
1051 }
1052 #endif
1053
1054 kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset;
1055
1056 kernel_map.phys_addr = (uintptr_t)(&_start);
1057 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1058 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1059
1060 #if defined(CONFIG_64BIT)
1061 set_satp_mode(dtb_pa);
1062 set_mmap_rnd_bits_max();
1063 #endif
1064
1065 /*
1066 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1067 * where we have the system memory layout: this allows us to align
1068 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1069 * for the linear mapping. This is only possible because the kernel
1070 * mapping lies outside the linear mapping.
1071 * In 32-bit however, as the kernel resides in the linear mapping,
1072 * setup_vm_final can not change the mapping established here,
1073 * otherwise the same kernel addresses would get mapped to different
1074 * physical addresses (if the start of dram is different from the
1075 * kernel physical address start).
1076 */
1077 kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1078 0UL : PAGE_OFFSET - kernel_map.phys_addr;
1079
1080 memory_limit = KERN_VIRT_SIZE;
1081
1082 /* Sanity check alignment and size */
1083 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1084 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1085
1086 #ifdef CONFIG_64BIT
1087 /*
1088 * The last 4K bytes of the addressable memory can not be mapped because
1089 * of IS_ERR_VALUE macro.
1090 */
1091 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1092 #endif
1093
1094 #ifdef CONFIG_RELOCATABLE
1095 /*
1096 * Early page table uses only one PUD, which makes it possible
1097 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1098 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1099 * since a part of the kernel would not get mapped.
1100 */
1101 if (IS_ENABLED(CONFIG_64BIT))
1102 BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1103 relocate_kernel();
1104 #endif
1105
1106 apply_early_boot_alternatives();
1107 pt_ops_set_early();
1108
1109 /* Setup early PGD for fixmap */
1110 create_pgd_mapping(early_pg_dir, FIXADDR_START,
1111 fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1112
1113 #ifndef __PAGETABLE_PMD_FOLDED
1114 /* Setup fixmap P4D and PUD */
1115 if (pgtable_l5_enabled)
1116 create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1117 (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1118 /* Setup fixmap PUD and PMD */
1119 if (pgtable_l4_enabled)
1120 create_pud_mapping(fixmap_pud, FIXADDR_START,
1121 (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1122 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1123 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1124 /* Setup trampoline PGD and PMD */
1125 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1126 trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1127 if (pgtable_l5_enabled)
1128 create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1129 (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1130 if (pgtable_l4_enabled)
1131 create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1132 (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1133 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1134 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1135 #else
1136 /* Setup trampoline PGD */
1137 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1138 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1139 #endif
1140
1141 /*
1142 * Setup early PGD covering entire kernel which will allow
1143 * us to reach paging_init(). We map all memory banks later
1144 * in setup_vm_final() below.
1145 */
1146 create_kernel_page_table(early_pg_dir, true);
1147
1148 /* Setup early mapping for FDT early scan */
1149 create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1150
1151 /*
1152 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1153 * range can not span multiple pmds.
1154 */
1155 BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1156 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1157
1158 #ifndef __PAGETABLE_PMD_FOLDED
1159 /*
1160 * Early ioremap fixmap is already created as it lies within first 2MB
1161 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1162 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1163 * the user if not.
1164 */
1165 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1166 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1167 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1168 WARN_ON(1);
1169 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1170 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1171 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1172 fix_to_virt(FIX_BTMAP_BEGIN));
1173 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1174 fix_to_virt(FIX_BTMAP_END));
1175
1176 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1177 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1178 }
1179 #endif
1180
1181 pt_ops_set_fixmap();
1182 }
1183
create_linear_mapping_range(phys_addr_t start,phys_addr_t end,uintptr_t fixed_map_size,const pgprot_t * pgprot)1184 static void __meminit create_linear_mapping_range(phys_addr_t start, phys_addr_t end,
1185 uintptr_t fixed_map_size, const pgprot_t *pgprot)
1186 {
1187 phys_addr_t pa;
1188 uintptr_t va, map_size;
1189
1190 for (pa = start; pa < end; pa += map_size) {
1191 va = (uintptr_t)__va(pa);
1192 map_size = fixed_map_size ? fixed_map_size :
1193 best_map_size(pa, va, end - pa);
1194
1195 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1196 pgprot ? *pgprot : pgprot_from_va(va));
1197 }
1198 }
1199
create_linear_mapping_page_table(void)1200 static void __init create_linear_mapping_page_table(void)
1201 {
1202 phys_addr_t start, end;
1203 phys_addr_t kfence_pool __maybe_unused;
1204 u64 i;
1205
1206 #ifdef CONFIG_STRICT_KERNEL_RWX
1207 phys_addr_t ktext_start = __pa_symbol(_start);
1208 phys_addr_t ktext_size = __init_data_begin - _start;
1209 phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1210 phys_addr_t krodata_size = _data - __start_rodata;
1211
1212 /* Isolate kernel text and rodata so they don't get mapped with a PUD */
1213 memblock_mark_nomap(ktext_start, ktext_size);
1214 memblock_mark_nomap(krodata_start, krodata_size);
1215 #endif
1216
1217 #ifdef CONFIG_KFENCE
1218 /*
1219 * kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1220 * before we setup the linear mapping so that we avoid using hugepages
1221 * for this region.
1222 */
1223 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1224 BUG_ON(!kfence_pool);
1225
1226 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1227 __kfence_pool = __va(kfence_pool);
1228 #endif
1229
1230 /* Map all memory banks in the linear mapping */
1231 for_each_mem_range(i, &start, &end) {
1232 if (start >= end)
1233 break;
1234 if (start <= __pa(PAGE_OFFSET) &&
1235 __pa(PAGE_OFFSET) < end)
1236 start = __pa(PAGE_OFFSET);
1237
1238 create_linear_mapping_range(start, end, 0, NULL);
1239 }
1240
1241 #ifdef CONFIG_STRICT_KERNEL_RWX
1242 create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0, NULL);
1243 create_linear_mapping_range(krodata_start, krodata_start + krodata_size, 0, NULL);
1244
1245 memblock_clear_nomap(ktext_start, ktext_size);
1246 memblock_clear_nomap(krodata_start, krodata_size);
1247 #endif
1248
1249 #ifdef CONFIG_KFENCE
1250 create_linear_mapping_range(kfence_pool, kfence_pool + KFENCE_POOL_SIZE, PAGE_SIZE, NULL);
1251
1252 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1253 #endif
1254 }
1255
setup_vm_final(void)1256 static void __init setup_vm_final(void)
1257 {
1258 /* Setup swapper PGD for fixmap */
1259 #if !defined(CONFIG_64BIT)
1260 /*
1261 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1262 * directly in swapper_pg_dir in addition to the pgd entry that points
1263 * to fixmap_pte.
1264 */
1265 unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1266
1267 set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1268 #endif
1269 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1270 __pa_symbol(fixmap_pgd_next),
1271 PGDIR_SIZE, PAGE_TABLE);
1272
1273 /* Map the linear mapping */
1274 create_linear_mapping_page_table();
1275
1276 /* Map the kernel */
1277 if (IS_ENABLED(CONFIG_64BIT))
1278 create_kernel_page_table(swapper_pg_dir, false);
1279
1280 #ifdef CONFIG_KASAN
1281 kasan_swapper_init();
1282 #endif
1283
1284 /* Clear fixmap PTE and PMD mappings */
1285 clear_fixmap(FIX_PTE);
1286 clear_fixmap(FIX_PMD);
1287 clear_fixmap(FIX_PUD);
1288 clear_fixmap(FIX_P4D);
1289
1290 /* Move to swapper page table */
1291 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1292 local_flush_tlb_all();
1293
1294 pt_ops_set_late();
1295 }
1296 #else
setup_vm(uintptr_t dtb_pa)1297 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1298 {
1299 dtb_early_va = (void *)dtb_pa;
1300 dtb_early_pa = dtb_pa;
1301
1302 #ifdef CONFIG_RELOCATABLE
1303 kernel_map.virt_addr = (uintptr_t)_start;
1304 kernel_map.phys_addr = (uintptr_t)_start;
1305 relocate_kernel();
1306 #endif
1307 }
1308
setup_vm_final(void)1309 static inline void setup_vm_final(void)
1310 {
1311 }
1312 #endif /* CONFIG_MMU */
1313
1314 /*
1315 * reserve_crashkernel() - reserves memory for crash kernel
1316 *
1317 * This function reserves memory area given in "crashkernel=" kernel command
1318 * line parameter. The memory reserved is used by dump capture kernel when
1319 * primary kernel is crashing.
1320 */
arch_reserve_crashkernel(void)1321 static void __init arch_reserve_crashkernel(void)
1322 {
1323 unsigned long long low_size = 0;
1324 unsigned long long crash_base, crash_size;
1325 bool high = false;
1326 int ret;
1327
1328 if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
1329 return;
1330
1331 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
1332 &crash_size, &crash_base,
1333 &low_size, NULL, &high);
1334 if (ret)
1335 return;
1336
1337 reserve_crashkernel_generic(crash_size, crash_base, low_size, high);
1338 }
1339
paging_init(void)1340 void __init paging_init(void)
1341 {
1342 setup_bootmem();
1343 setup_vm_final();
1344
1345 /* Depend on that Linear Mapping is ready */
1346 memblock_allow_resize();
1347 }
1348
misc_mem_init(void)1349 void __init misc_mem_init(void)
1350 {
1351 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1352 arch_numa_init();
1353 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1354 /* The entire VMEMMAP region has been populated. Flush TLB for this region */
1355 local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END);
1356 #endif
1357 arch_reserve_crashkernel();
1358 memblock_dump_all();
1359 }
1360
1361 #ifdef CONFIG_SPARSEMEM_VMEMMAP
vmemmap_set_pmd(pmd_t * pmd,void * p,int node,unsigned long addr,unsigned long next)1362 void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node,
1363 unsigned long addr, unsigned long next)
1364 {
1365 pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL);
1366 }
1367
vmemmap_check_pmd(pmd_t * pmdp,int node,unsigned long addr,unsigned long next)1368 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node,
1369 unsigned long addr, unsigned long next)
1370 {
1371 vmemmap_verify((pte_t *)pmdp, node, addr, next);
1372 return 1;
1373 }
1374
vmemmap_populate(unsigned long start,unsigned long end,int node,struct vmem_altmap * altmap)1375 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1376 struct vmem_altmap *altmap)
1377 {
1378 WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1379
1380 /*
1381 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we
1382 * can't use hugepage mappings for 2-level page table because in case of
1383 * memory hotplug, we are not able to update all the page tables with
1384 * the new PMDs.
1385 */
1386 return vmemmap_populate_hugepages(start, end, node, altmap);
1387 }
1388 #endif
1389
1390 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1391 /*
1392 * Pre-allocates page-table pages for a specific area in the kernel
1393 * page-table. Only the level which needs to be synchronized between
1394 * all page-tables is allocated because the synchronization can be
1395 * expensive.
1396 */
preallocate_pgd_pages_range(unsigned long start,unsigned long end,const char * area)1397 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1398 const char *area)
1399 {
1400 unsigned long addr;
1401 const char *lvl;
1402
1403 for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1404 pgd_t *pgd = pgd_offset_k(addr);
1405 p4d_t *p4d;
1406 pud_t *pud;
1407 pmd_t *pmd;
1408
1409 lvl = "p4d";
1410 p4d = p4d_alloc(&init_mm, pgd, addr);
1411 if (!p4d)
1412 goto failed;
1413
1414 if (pgtable_l5_enabled)
1415 continue;
1416
1417 lvl = "pud";
1418 pud = pud_alloc(&init_mm, p4d, addr);
1419 if (!pud)
1420 goto failed;
1421
1422 if (pgtable_l4_enabled)
1423 continue;
1424
1425 lvl = "pmd";
1426 pmd = pmd_alloc(&init_mm, pud, addr);
1427 if (!pmd)
1428 goto failed;
1429 }
1430 return;
1431
1432 failed:
1433 /*
1434 * The pages have to be there now or they will be missing in
1435 * process page-tables later.
1436 */
1437 panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1438 }
1439
1440 #define PAGE_END KASAN_SHADOW_START
1441
pgtable_cache_init(void)1442 void __init pgtable_cache_init(void)
1443 {
1444 preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1445 if (IS_ENABLED(CONFIG_MODULES))
1446 preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
1447 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) {
1448 preallocate_pgd_pages_range(VMEMMAP_START, VMEMMAP_END, "vmemmap");
1449 preallocate_pgd_pages_range(PAGE_OFFSET, PAGE_END, "direct map");
1450 if (IS_ENABLED(CONFIG_KASAN))
1451 preallocate_pgd_pages_range(KASAN_SHADOW_START, KASAN_SHADOW_END, "kasan");
1452 }
1453 }
1454 #endif
1455
1456 #ifdef CONFIG_EXECMEM
1457 #ifdef CONFIG_MMU
1458 static struct execmem_info execmem_info __ro_after_init;
1459
execmem_arch_setup(void)1460 struct execmem_info __init *execmem_arch_setup(void)
1461 {
1462 execmem_info = (struct execmem_info){
1463 .ranges = {
1464 [EXECMEM_DEFAULT] = {
1465 .start = MODULES_VADDR,
1466 .end = MODULES_END,
1467 .pgprot = PAGE_KERNEL,
1468 .alignment = 1,
1469 },
1470 [EXECMEM_KPROBES] = {
1471 .start = VMALLOC_START,
1472 .end = VMALLOC_END,
1473 .pgprot = PAGE_KERNEL_READ_EXEC,
1474 .alignment = 1,
1475 },
1476 [EXECMEM_BPF] = {
1477 .start = BPF_JIT_REGION_START,
1478 .end = BPF_JIT_REGION_END,
1479 .pgprot = PAGE_KERNEL,
1480 .alignment = PAGE_SIZE,
1481 },
1482 },
1483 };
1484
1485 return &execmem_info;
1486 }
1487 #endif /* CONFIG_MMU */
1488 #endif /* CONFIG_EXECMEM */
1489
1490 #ifdef CONFIG_MEMORY_HOTPLUG
free_pte_table(pte_t * pte_start,pmd_t * pmd)1491 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
1492 {
1493 struct page *page = pmd_page(*pmd);
1494 struct ptdesc *ptdesc = page_ptdesc(page);
1495 pte_t *pte;
1496 int i;
1497
1498 for (i = 0; i < PTRS_PER_PTE; i++) {
1499 pte = pte_start + i;
1500 if (!pte_none(*pte))
1501 return;
1502 }
1503
1504 pagetable_dtor(ptdesc);
1505 if (PageReserved(page))
1506 free_reserved_page(page);
1507 else
1508 pagetable_free(ptdesc);
1509 pmd_clear(pmd);
1510 }
1511
free_pmd_table(pmd_t * pmd_start,pud_t * pud,bool is_vmemmap)1512 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud, bool is_vmemmap)
1513 {
1514 struct page *page = pud_page(*pud);
1515 struct ptdesc *ptdesc = page_ptdesc(page);
1516 pmd_t *pmd;
1517 int i;
1518
1519 for (i = 0; i < PTRS_PER_PMD; i++) {
1520 pmd = pmd_start + i;
1521 if (!pmd_none(*pmd))
1522 return;
1523 }
1524
1525 if (!is_vmemmap)
1526 pagetable_dtor(ptdesc);
1527 if (PageReserved(page))
1528 free_reserved_page(page);
1529 else
1530 pagetable_free(ptdesc);
1531 pud_clear(pud);
1532 }
1533
free_pud_table(pud_t * pud_start,p4d_t * p4d)1534 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d)
1535 {
1536 struct page *page = p4d_page(*p4d);
1537 pud_t *pud;
1538 int i;
1539
1540 for (i = 0; i < PTRS_PER_PUD; i++) {
1541 pud = pud_start + i;
1542 if (!pud_none(*pud))
1543 return;
1544 }
1545
1546 if (PageReserved(page))
1547 free_reserved_page(page);
1548 else
1549 __free_pages(page, 0);
1550 p4d_clear(p4d);
1551 }
1552
free_vmemmap_storage(struct page * page,size_t size,struct vmem_altmap * altmap)1553 static void __meminit free_vmemmap_storage(struct page *page, size_t size,
1554 struct vmem_altmap *altmap)
1555 {
1556 int order = get_order(size);
1557
1558 if (altmap) {
1559 vmem_altmap_free(altmap, size >> PAGE_SHIFT);
1560 return;
1561 }
1562
1563 if (PageReserved(page)) {
1564 unsigned int nr_pages = 1 << order;
1565
1566 while (nr_pages--)
1567 free_reserved_page(page++);
1568 return;
1569 }
1570
1571 __free_pages(page, order);
1572 }
1573
remove_pte_mapping(pte_t * pte_base,unsigned long addr,unsigned long end,bool is_vmemmap,struct vmem_altmap * altmap)1574 static void __meminit remove_pte_mapping(pte_t *pte_base, unsigned long addr, unsigned long end,
1575 bool is_vmemmap, struct vmem_altmap *altmap)
1576 {
1577 unsigned long next;
1578 pte_t *ptep, pte;
1579
1580 for (; addr < end; addr = next) {
1581 next = (addr + PAGE_SIZE) & PAGE_MASK;
1582 if (next > end)
1583 next = end;
1584
1585 ptep = pte_base + pte_index(addr);
1586 pte = ptep_get(ptep);
1587 if (!pte_present(*ptep))
1588 continue;
1589
1590 pte_clear(&init_mm, addr, ptep);
1591 if (is_vmemmap)
1592 free_vmemmap_storage(pte_page(pte), PAGE_SIZE, altmap);
1593 }
1594 }
1595
remove_pmd_mapping(pmd_t * pmd_base,unsigned long addr,unsigned long end,bool is_vmemmap,struct vmem_altmap * altmap)1596 static void __meminit remove_pmd_mapping(pmd_t *pmd_base, unsigned long addr, unsigned long end,
1597 bool is_vmemmap, struct vmem_altmap *altmap)
1598 {
1599 unsigned long next;
1600 pte_t *pte_base;
1601 pmd_t *pmdp, pmd;
1602
1603 for (; addr < end; addr = next) {
1604 next = pmd_addr_end(addr, end);
1605 pmdp = pmd_base + pmd_index(addr);
1606 pmd = pmdp_get(pmdp);
1607 if (!pmd_present(pmd))
1608 continue;
1609
1610 if (pmd_leaf(pmd)) {
1611 pmd_clear(pmdp);
1612 if (is_vmemmap)
1613 free_vmemmap_storage(pmd_page(pmd), PMD_SIZE, altmap);
1614 continue;
1615 }
1616
1617 pte_base = (pte_t *)pmd_page_vaddr(*pmdp);
1618 remove_pte_mapping(pte_base, addr, next, is_vmemmap, altmap);
1619 free_pte_table(pte_base, pmdp);
1620 }
1621 }
1622
remove_pud_mapping(pud_t * pud_base,unsigned long addr,unsigned long end,bool is_vmemmap,struct vmem_altmap * altmap)1623 static void __meminit remove_pud_mapping(pud_t *pud_base, unsigned long addr, unsigned long end,
1624 bool is_vmemmap, struct vmem_altmap *altmap)
1625 {
1626 unsigned long next;
1627 pud_t *pudp, pud;
1628 pmd_t *pmd_base;
1629
1630 for (; addr < end; addr = next) {
1631 next = pud_addr_end(addr, end);
1632 pudp = pud_base + pud_index(addr);
1633 pud = pudp_get(pudp);
1634 if (!pud_present(pud))
1635 continue;
1636
1637 if (pud_leaf(pud)) {
1638 if (pgtable_l4_enabled) {
1639 pud_clear(pudp);
1640 if (is_vmemmap)
1641 free_vmemmap_storage(pud_page(pud), PUD_SIZE, altmap);
1642 }
1643 continue;
1644 }
1645
1646 pmd_base = pmd_offset(pudp, 0);
1647 remove_pmd_mapping(pmd_base, addr, next, is_vmemmap, altmap);
1648
1649 if (pgtable_l4_enabled)
1650 free_pmd_table(pmd_base, pudp, is_vmemmap);
1651 }
1652 }
1653
remove_p4d_mapping(p4d_t * p4d_base,unsigned long addr,unsigned long end,bool is_vmemmap,struct vmem_altmap * altmap)1654 static void __meminit remove_p4d_mapping(p4d_t *p4d_base, unsigned long addr, unsigned long end,
1655 bool is_vmemmap, struct vmem_altmap *altmap)
1656 {
1657 unsigned long next;
1658 p4d_t *p4dp, p4d;
1659 pud_t *pud_base;
1660
1661 for (; addr < end; addr = next) {
1662 next = p4d_addr_end(addr, end);
1663 p4dp = p4d_base + p4d_index(addr);
1664 p4d = p4dp_get(p4dp);
1665 if (!p4d_present(p4d))
1666 continue;
1667
1668 if (p4d_leaf(p4d)) {
1669 if (pgtable_l5_enabled) {
1670 p4d_clear(p4dp);
1671 if (is_vmemmap)
1672 free_vmemmap_storage(p4d_page(p4d), P4D_SIZE, altmap);
1673 }
1674 continue;
1675 }
1676
1677 pud_base = pud_offset(p4dp, 0);
1678 remove_pud_mapping(pud_base, addr, next, is_vmemmap, altmap);
1679
1680 if (pgtable_l5_enabled)
1681 free_pud_table(pud_base, p4dp);
1682 }
1683 }
1684
remove_pgd_mapping(unsigned long va,unsigned long end,bool is_vmemmap,struct vmem_altmap * altmap)1685 static void __meminit remove_pgd_mapping(unsigned long va, unsigned long end, bool is_vmemmap,
1686 struct vmem_altmap *altmap)
1687 {
1688 unsigned long addr, next;
1689 p4d_t *p4d_base;
1690 pgd_t *pgd;
1691
1692 for (addr = va; addr < end; addr = next) {
1693 next = pgd_addr_end(addr, end);
1694 pgd = pgd_offset_k(addr);
1695
1696 if (!pgd_present(*pgd))
1697 continue;
1698
1699 if (pgd_leaf(*pgd))
1700 continue;
1701
1702 p4d_base = p4d_offset(pgd, 0);
1703 remove_p4d_mapping(p4d_base, addr, next, is_vmemmap, altmap);
1704 }
1705
1706 flush_tlb_all();
1707 }
1708
remove_linear_mapping(phys_addr_t start,u64 size)1709 static void __meminit remove_linear_mapping(phys_addr_t start, u64 size)
1710 {
1711 unsigned long va = (unsigned long)__va(start);
1712 unsigned long end = (unsigned long)__va(start + size);
1713
1714 remove_pgd_mapping(va, end, false, NULL);
1715 }
1716
arch_get_mappable_range(void)1717 struct range arch_get_mappable_range(void)
1718 {
1719 struct range mhp_range;
1720
1721 mhp_range.start = __pa(PAGE_OFFSET);
1722 mhp_range.end = __pa(PAGE_END - 1);
1723 return mhp_range;
1724 }
1725
arch_add_memory(int nid,u64 start,u64 size,struct mhp_params * params)1726 int __ref arch_add_memory(int nid, u64 start, u64 size, struct mhp_params *params)
1727 {
1728 int ret = 0;
1729
1730 create_linear_mapping_range(start, start + size, 0, ¶ms->pgprot);
1731 ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT, params);
1732 if (ret) {
1733 remove_linear_mapping(start, size);
1734 goto out;
1735 }
1736
1737 max_pfn = PFN_UP(start + size);
1738 max_low_pfn = max_pfn;
1739
1740 out:
1741 flush_tlb_all();
1742 return ret;
1743 }
1744
arch_remove_memory(u64 start,u64 size,struct vmem_altmap * altmap)1745 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
1746 {
1747 __remove_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT, altmap);
1748 remove_linear_mapping(start, size);
1749 flush_tlb_all();
1750 }
1751
vmemmap_free(unsigned long start,unsigned long end,struct vmem_altmap * altmap)1752 void __ref vmemmap_free(unsigned long start, unsigned long end, struct vmem_altmap *altmap)
1753 {
1754 remove_pgd_mapping(start, end, true, altmap);
1755 }
1756 #endif /* CONFIG_MEMORY_HOTPLUG */
1757