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
24 #include <asm/fixmap.h>
25 #include <asm/tlbflush.h>
26 #include <asm/sections.h>
27 #include <asm/soc.h>
28 #include <asm/io.h>
29 #include <asm/ptdump.h>
30 #include <asm/numa.h>
31
32 #include "../kernel/head.h"
33
34 struct kernel_mapping kernel_map __ro_after_init;
35 EXPORT_SYMBOL(kernel_map);
36 #ifdef CONFIG_XIP_KERNEL
37 #define kernel_map (*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
38 #endif
39
40 phys_addr_t phys_ram_base __ro_after_init;
41 EXPORT_SYMBOL(phys_ram_base);
42
43 #ifdef CONFIG_XIP_KERNEL
44 extern char _xiprom[], _exiprom[];
45 #endif
46
47 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
48 __page_aligned_bss;
49 EXPORT_SYMBOL(empty_zero_page);
50
51 extern char _start[];
52 #define DTB_EARLY_BASE_VA PGDIR_SIZE
53 void *_dtb_early_va __initdata;
54 uintptr_t _dtb_early_pa __initdata;
55
56 struct pt_alloc_ops {
57 pte_t *(*get_pte_virt)(phys_addr_t pa);
58 phys_addr_t (*alloc_pte)(uintptr_t va);
59 #ifndef __PAGETABLE_PMD_FOLDED
60 pmd_t *(*get_pmd_virt)(phys_addr_t pa);
61 phys_addr_t (*alloc_pmd)(uintptr_t va);
62 #endif
63 };
64
65 static phys_addr_t dma32_phys_limit __initdata;
66
zone_sizes_init(void)67 static void __init zone_sizes_init(void)
68 {
69 unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
70
71 #ifdef CONFIG_ZONE_DMA32
72 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
73 #endif
74 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
75
76 free_area_init(max_zone_pfns);
77 }
78
79 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
print_mlk(char * name,unsigned long b,unsigned long t)80 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
81 {
82 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld kB)\n", name, b, t,
83 (((t) - (b)) >> 10));
84 }
85
print_mlm(char * name,unsigned long b,unsigned long t)86 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
87 {
88 pr_notice("%12s : 0x%08lx - 0x%08lx (%4ld MB)\n", name, b, t,
89 (((t) - (b)) >> 20));
90 }
91
print_vm_layout(void)92 static void __init print_vm_layout(void)
93 {
94 pr_notice("Virtual kernel memory layout:\n");
95 print_mlk("fixmap", (unsigned long)FIXADDR_START,
96 (unsigned long)FIXADDR_TOP);
97 print_mlm("pci io", (unsigned long)PCI_IO_START,
98 (unsigned long)PCI_IO_END);
99 print_mlm("vmemmap", (unsigned long)VMEMMAP_START,
100 (unsigned long)VMEMMAP_END);
101 print_mlm("vmalloc", (unsigned long)VMALLOC_START,
102 (unsigned long)VMALLOC_END);
103 print_mlm("lowmem", (unsigned long)PAGE_OFFSET,
104 (unsigned long)high_memory);
105 #ifdef CONFIG_64BIT
106 print_mlm("kernel", (unsigned long)KERNEL_LINK_ADDR,
107 (unsigned long)ADDRESS_SPACE_END);
108 #endif
109 }
110 #else
print_vm_layout(void)111 static void print_vm_layout(void) { }
112 #endif /* CONFIG_DEBUG_VM */
113
mem_init(void)114 void __init mem_init(void)
115 {
116 #ifdef CONFIG_FLATMEM
117 BUG_ON(!mem_map);
118 #endif /* CONFIG_FLATMEM */
119
120 #ifdef CONFIG_SWIOTLB
121 if (swiotlb_force == SWIOTLB_FORCE ||
122 max_pfn > PFN_DOWN(dma32_phys_limit))
123 swiotlb_init(1);
124 else
125 swiotlb_force = SWIOTLB_NO_FORCE;
126 #endif
127 high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
128 memblock_free_all();
129
130 print_vm_layout();
131 }
132
133 /*
134 * The default maximal physical memory size is -PAGE_OFFSET for 32-bit kernel,
135 * whereas for 64-bit kernel, the end of the virtual address space is occupied
136 * by the modules/BPF/kernel mappings which reduces the available size of the
137 * linear mapping.
138 * Limit the memory size via mem.
139 */
140 #ifdef CONFIG_64BIT
141 static phys_addr_t memory_limit = -PAGE_OFFSET - SZ_4G;
142 #else
143 static phys_addr_t memory_limit = -PAGE_OFFSET;
144 #endif
145
early_mem(char * p)146 static int __init early_mem(char *p)
147 {
148 u64 size;
149
150 if (!p)
151 return 1;
152
153 size = memparse(p, &p) & PAGE_MASK;
154 memory_limit = min_t(u64, size, memory_limit);
155
156 pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
157
158 return 0;
159 }
160 early_param("mem", early_mem);
161
setup_bootmem(void)162 static void __init setup_bootmem(void)
163 {
164 phys_addr_t vmlinux_end = __pa_symbol(&_end);
165 phys_addr_t vmlinux_start = __pa_symbol(&_start);
166 phys_addr_t __maybe_unused max_mapped_addr;
167 phys_addr_t phys_ram_end;
168
169 #ifdef CONFIG_XIP_KERNEL
170 vmlinux_start = __pa_symbol(&_sdata);
171 #endif
172
173 memblock_enforce_memory_limit(memory_limit);
174
175 /*
176 * Reserve from the start of the kernel to the end of the kernel
177 */
178 #if defined(CONFIG_64BIT) && defined(CONFIG_STRICT_KERNEL_RWX)
179 /*
180 * Make sure we align the reservation on PMD_SIZE since we will
181 * map the kernel in the linear mapping as read-only: we do not want
182 * any allocation to happen between _end and the next pmd aligned page.
183 */
184 vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
185 #endif
186 memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
187
188
189 phys_ram_end = memblock_end_of_DRAM();
190 #ifndef CONFIG_64BIT
191 #ifndef CONFIG_XIP_KERNEL
192 phys_ram_base = memblock_start_of_DRAM();
193 #endif
194 /*
195 * memblock allocator is not aware of the fact that last 4K bytes of
196 * the addressable memory can not be mapped because of IS_ERR_VALUE
197 * macro. Make sure that last 4k bytes are not usable by memblock
198 * if end of dram is equal to maximum addressable memory. For 64-bit
199 * kernel, this problem can't happen here as the end of the virtual
200 * address space is occupied by the kernel mapping then this check must
201 * be done as soon as the kernel mapping base address is determined.
202 */
203 max_mapped_addr = __pa(~(ulong)0);
204 if (max_mapped_addr == (phys_ram_end - 1))
205 memblock_set_current_limit(max_mapped_addr - 4096);
206 #endif
207
208 min_low_pfn = PFN_UP(phys_ram_base);
209 max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
210
211 dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
212 set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
213
214 reserve_initrd_mem();
215 /*
216 * If DTB is built in, no need to reserve its memblock.
217 * Otherwise, do reserve it but avoid using
218 * early_init_fdt_reserve_self() since __pa() does
219 * not work for DTB pointers that are fixmap addresses
220 */
221 if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
222 memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
223
224 early_init_fdt_scan_reserved_mem();
225 dma_contiguous_reserve(dma32_phys_limit);
226 if (IS_ENABLED(CONFIG_64BIT))
227 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
228 memblock_allow_resize();
229 }
230
231 #ifdef CONFIG_MMU
232 static struct pt_alloc_ops _pt_ops __initdata;
233
234 #ifdef CONFIG_XIP_KERNEL
235 #define pt_ops (*(struct pt_alloc_ops *)XIP_FIXUP(&_pt_ops))
236 #else
237 #define pt_ops _pt_ops
238 #endif
239
240 unsigned long riscv_pfn_base __ro_after_init;
241 EXPORT_SYMBOL(riscv_pfn_base);
242
243 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
244 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
245 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
246
247 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
248 static pmd_t __maybe_unused early_dtb_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
249
250 #ifdef CONFIG_XIP_KERNEL
251 #define trampoline_pg_dir ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
252 #define fixmap_pte ((pte_t *)XIP_FIXUP(fixmap_pte))
253 #define early_pg_dir ((pgd_t *)XIP_FIXUP(early_pg_dir))
254 #endif /* CONFIG_XIP_KERNEL */
255
__set_fixmap(enum fixed_addresses idx,phys_addr_t phys,pgprot_t prot)256 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
257 {
258 unsigned long addr = __fix_to_virt(idx);
259 pte_t *ptep;
260
261 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
262
263 ptep = &fixmap_pte[pte_index(addr)];
264
265 if (pgprot_val(prot))
266 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
267 else
268 pte_clear(&init_mm, addr, ptep);
269 local_flush_tlb_page(addr);
270 }
271
get_pte_virt_early(phys_addr_t pa)272 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
273 {
274 return (pte_t *)((uintptr_t)pa);
275 }
276
get_pte_virt_fixmap(phys_addr_t pa)277 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
278 {
279 clear_fixmap(FIX_PTE);
280 return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
281 }
282
get_pte_virt_late(phys_addr_t pa)283 static inline pte_t *__init get_pte_virt_late(phys_addr_t pa)
284 {
285 return (pte_t *) __va(pa);
286 }
287
alloc_pte_early(uintptr_t va)288 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
289 {
290 /*
291 * We only create PMD or PGD early mappings so we
292 * should never reach here with MMU disabled.
293 */
294 BUG();
295 }
296
alloc_pte_fixmap(uintptr_t va)297 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
298 {
299 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
300 }
301
alloc_pte_late(uintptr_t va)302 static phys_addr_t __init alloc_pte_late(uintptr_t va)
303 {
304 unsigned long vaddr;
305
306 vaddr = __get_free_page(GFP_KERNEL);
307 BUG_ON(!vaddr || !pgtable_pte_page_ctor(virt_to_page(vaddr)));
308
309 return __pa(vaddr);
310 }
311
create_pte_mapping(pte_t * ptep,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)312 static void __init create_pte_mapping(pte_t *ptep,
313 uintptr_t va, phys_addr_t pa,
314 phys_addr_t sz, pgprot_t prot)
315 {
316 uintptr_t pte_idx = pte_index(va);
317
318 BUG_ON(sz != PAGE_SIZE);
319
320 if (pte_none(ptep[pte_idx]))
321 ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
322 }
323
324 #ifndef __PAGETABLE_PMD_FOLDED
325
326 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
327 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
328 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
329
330 #ifdef CONFIG_XIP_KERNEL
331 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
332 #define fixmap_pmd ((pmd_t *)XIP_FIXUP(fixmap_pmd))
333 #define early_pmd ((pmd_t *)XIP_FIXUP(early_pmd))
334 #endif /* CONFIG_XIP_KERNEL */
335
get_pmd_virt_early(phys_addr_t pa)336 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
337 {
338 /* Before MMU is enabled */
339 return (pmd_t *)((uintptr_t)pa);
340 }
341
get_pmd_virt_fixmap(phys_addr_t pa)342 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
343 {
344 clear_fixmap(FIX_PMD);
345 return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
346 }
347
get_pmd_virt_late(phys_addr_t pa)348 static pmd_t *__init get_pmd_virt_late(phys_addr_t pa)
349 {
350 return (pmd_t *) __va(pa);
351 }
352
alloc_pmd_early(uintptr_t va)353 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
354 {
355 BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
356
357 return (uintptr_t)early_pmd;
358 }
359
alloc_pmd_fixmap(uintptr_t va)360 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
361 {
362 return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
363 }
364
alloc_pmd_late(uintptr_t va)365 static phys_addr_t __init alloc_pmd_late(uintptr_t va)
366 {
367 unsigned long vaddr;
368
369 vaddr = __get_free_page(GFP_KERNEL);
370 BUG_ON(!vaddr);
371 return __pa(vaddr);
372 }
373
create_pmd_mapping(pmd_t * pmdp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)374 static void __init create_pmd_mapping(pmd_t *pmdp,
375 uintptr_t va, phys_addr_t pa,
376 phys_addr_t sz, pgprot_t prot)
377 {
378 pte_t *ptep;
379 phys_addr_t pte_phys;
380 uintptr_t pmd_idx = pmd_index(va);
381
382 if (sz == PMD_SIZE) {
383 if (pmd_none(pmdp[pmd_idx]))
384 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
385 return;
386 }
387
388 if (pmd_none(pmdp[pmd_idx])) {
389 pte_phys = pt_ops.alloc_pte(va);
390 pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
391 ptep = pt_ops.get_pte_virt(pte_phys);
392 memset(ptep, 0, PAGE_SIZE);
393 } else {
394 pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
395 ptep = pt_ops.get_pte_virt(pte_phys);
396 }
397
398 create_pte_mapping(ptep, va, pa, sz, prot);
399 }
400
401 #define pgd_next_t pmd_t
402 #define alloc_pgd_next(__va) pt_ops.alloc_pmd(__va)
403 #define get_pgd_next_virt(__pa) pt_ops.get_pmd_virt(__pa)
404 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
405 create_pmd_mapping(__nextp, __va, __pa, __sz, __prot)
406 #define fixmap_pgd_next fixmap_pmd
407 #else
408 #define pgd_next_t pte_t
409 #define alloc_pgd_next(__va) pt_ops.alloc_pte(__va)
410 #define get_pgd_next_virt(__pa) pt_ops.get_pte_virt(__pa)
411 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot) \
412 create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
413 #define fixmap_pgd_next fixmap_pte
414 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot)
415 #endif
416
create_pgd_mapping(pgd_t * pgdp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)417 void __init create_pgd_mapping(pgd_t *pgdp,
418 uintptr_t va, phys_addr_t pa,
419 phys_addr_t sz, pgprot_t prot)
420 {
421 pgd_next_t *nextp;
422 phys_addr_t next_phys;
423 uintptr_t pgd_idx = pgd_index(va);
424
425 if (sz == PGDIR_SIZE) {
426 if (pgd_val(pgdp[pgd_idx]) == 0)
427 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
428 return;
429 }
430
431 if (pgd_val(pgdp[pgd_idx]) == 0) {
432 next_phys = alloc_pgd_next(va);
433 pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
434 nextp = get_pgd_next_virt(next_phys);
435 memset(nextp, 0, PAGE_SIZE);
436 } else {
437 next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
438 nextp = get_pgd_next_virt(next_phys);
439 }
440
441 create_pgd_next_mapping(nextp, va, pa, sz, prot);
442 }
443
best_map_size(phys_addr_t base,phys_addr_t size)444 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
445 {
446 /* Upgrade to PMD_SIZE mappings whenever possible */
447 if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
448 return PAGE_SIZE;
449
450 return PMD_SIZE;
451 }
452
453 #ifdef CONFIG_XIP_KERNEL
454 /* called from head.S with MMU off */
__copy_data(void)455 asmlinkage void __init __copy_data(void)
456 {
457 void *from = (void *)(&_sdata);
458 void *end = (void *)(&_end);
459 void *to = (void *)CONFIG_PHYS_RAM_BASE;
460 size_t sz = (size_t)(end - from + 1);
461
462 memcpy(to, from, sz);
463 }
464 #endif
465
466 #ifdef CONFIG_STRICT_KERNEL_RWX
pgprot_from_va(uintptr_t va)467 static __init pgprot_t pgprot_from_va(uintptr_t va)
468 {
469 if (is_va_kernel_text(va))
470 return PAGE_KERNEL_READ_EXEC;
471
472 /*
473 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
474 * we must protect its linear mapping alias from being executed and
475 * written.
476 * And rodata section is marked readonly in mark_rodata_ro.
477 */
478 if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
479 return PAGE_KERNEL_READ;
480
481 return PAGE_KERNEL;
482 }
483
mark_rodata_ro(void)484 void mark_rodata_ro(void)
485 {
486 set_kernel_memory(__start_rodata, _data, set_memory_ro);
487 if (IS_ENABLED(CONFIG_64BIT))
488 set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
489 set_memory_ro);
490
491 debug_checkwx();
492 }
493 #else
pgprot_from_va(uintptr_t va)494 static __init pgprot_t pgprot_from_va(uintptr_t va)
495 {
496 if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
497 return PAGE_KERNEL;
498
499 return PAGE_KERNEL_EXEC;
500 }
501 #endif /* CONFIG_STRICT_KERNEL_RWX */
502
503 /*
504 * setup_vm() is called from head.S with MMU-off.
505 *
506 * Following requirements should be honoured for setup_vm() to work
507 * correctly:
508 * 1) It should use PC-relative addressing for accessing kernel symbols.
509 * To achieve this we always use GCC cmodel=medany.
510 * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
511 * so disable compiler instrumentation when FTRACE is enabled.
512 *
513 * Currently, the above requirements are honoured by using custom CFLAGS
514 * for init.o in mm/Makefile.
515 */
516
517 #ifndef __riscv_cmodel_medany
518 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
519 #endif
520
521 #ifdef CONFIG_XIP_KERNEL
create_kernel_page_table(pgd_t * pgdir,__always_unused bool early)522 static void __init create_kernel_page_table(pgd_t *pgdir,
523 __always_unused bool early)
524 {
525 uintptr_t va, end_va;
526
527 /* Map the flash resident part */
528 end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
529 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
530 create_pgd_mapping(pgdir, va,
531 kernel_map.xiprom + (va - kernel_map.virt_addr),
532 PMD_SIZE, PAGE_KERNEL_EXEC);
533
534 /* Map the data in RAM */
535 end_va = kernel_map.virt_addr + XIP_OFFSET + kernel_map.size;
536 for (va = kernel_map.virt_addr + XIP_OFFSET; va < end_va; va += PMD_SIZE)
537 create_pgd_mapping(pgdir, va,
538 kernel_map.phys_addr + (va - (kernel_map.virt_addr + XIP_OFFSET)),
539 PMD_SIZE, PAGE_KERNEL);
540 }
541 #else
create_kernel_page_table(pgd_t * pgdir,bool early)542 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
543 {
544 uintptr_t va, end_va;
545
546 end_va = kernel_map.virt_addr + kernel_map.size;
547 for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
548 create_pgd_mapping(pgdir, va,
549 kernel_map.phys_addr + (va - kernel_map.virt_addr),
550 PMD_SIZE,
551 early ?
552 PAGE_KERNEL_EXEC : pgprot_from_va(va));
553 }
554 #endif
555
556 /*
557 * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
558 * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
559 * entry.
560 */
create_fdt_early_page_table(pgd_t * pgdir,uintptr_t dtb_pa)561 static void __init create_fdt_early_page_table(pgd_t *pgdir, uintptr_t dtb_pa)
562 {
563 #ifndef CONFIG_BUILTIN_DTB
564 uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
565
566 create_pgd_mapping(early_pg_dir, DTB_EARLY_BASE_VA,
567 IS_ENABLED(CONFIG_64BIT) ? (uintptr_t)early_dtb_pmd : pa,
568 PGDIR_SIZE,
569 IS_ENABLED(CONFIG_64BIT) ? PAGE_TABLE : PAGE_KERNEL);
570
571 if (IS_ENABLED(CONFIG_64BIT)) {
572 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA,
573 pa, PMD_SIZE, PAGE_KERNEL);
574 create_pmd_mapping(early_dtb_pmd, DTB_EARLY_BASE_VA + PMD_SIZE,
575 pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
576 }
577
578 dtb_early_va = (void *)DTB_EARLY_BASE_VA + (dtb_pa & (PMD_SIZE - 1));
579 #else
580 /*
581 * For 64-bit kernel, __va can't be used since it would return a linear
582 * mapping address whereas dtb_early_va will be used before
583 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
584 * kernel is mapped in the linear mapping, that makes no difference.
585 */
586 dtb_early_va = kernel_mapping_pa_to_va(XIP_FIXUP(dtb_pa));
587 #endif
588
589 dtb_early_pa = dtb_pa;
590 }
591
setup_vm(uintptr_t dtb_pa)592 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
593 {
594 pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
595
596 kernel_map.virt_addr = KERNEL_LINK_ADDR;
597
598 #ifdef CONFIG_XIP_KERNEL
599 kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
600 kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
601
602 phys_ram_base = CONFIG_PHYS_RAM_BASE;
603 kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
604 kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_sdata);
605
606 kernel_map.va_kernel_xip_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
607 #else
608 kernel_map.phys_addr = (uintptr_t)(&_start);
609 kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
610 #endif
611 kernel_map.va_pa_offset = PAGE_OFFSET - kernel_map.phys_addr;
612 kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
613
614 riscv_pfn_base = PFN_DOWN(kernel_map.phys_addr);
615
616 /* Sanity check alignment and size */
617 BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
618 BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
619
620 #ifdef CONFIG_64BIT
621 /*
622 * The last 4K bytes of the addressable memory can not be mapped because
623 * of IS_ERR_VALUE macro.
624 */
625 BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
626 #endif
627
628 pt_ops.alloc_pte = alloc_pte_early;
629 pt_ops.get_pte_virt = get_pte_virt_early;
630 #ifndef __PAGETABLE_PMD_FOLDED
631 pt_ops.alloc_pmd = alloc_pmd_early;
632 pt_ops.get_pmd_virt = get_pmd_virt_early;
633 #endif
634 /* Setup early PGD for fixmap */
635 create_pgd_mapping(early_pg_dir, FIXADDR_START,
636 (uintptr_t)fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
637
638 #ifndef __PAGETABLE_PMD_FOLDED
639 /* Setup fixmap PMD */
640 create_pmd_mapping(fixmap_pmd, FIXADDR_START,
641 (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
642 /* Setup trampoline PGD and PMD */
643 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
644 (uintptr_t)trampoline_pmd, PGDIR_SIZE, PAGE_TABLE);
645 #ifdef CONFIG_XIP_KERNEL
646 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
647 kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
648 #else
649 create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
650 kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
651 #endif
652 #else
653 /* Setup trampoline PGD */
654 create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
655 kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
656 #endif
657
658 /*
659 * Setup early PGD covering entire kernel which will allow
660 * us to reach paging_init(). We map all memory banks later
661 * in setup_vm_final() below.
662 */
663 create_kernel_page_table(early_pg_dir, true);
664
665 /* Setup early mapping for FDT early scan */
666 create_fdt_early_page_table(early_pg_dir, dtb_pa);
667
668 /*
669 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
670 * range can not span multiple pmds.
671 */
672 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
673 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
674
675 #ifndef __PAGETABLE_PMD_FOLDED
676 /*
677 * Early ioremap fixmap is already created as it lies within first 2MB
678 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
679 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
680 * the user if not.
681 */
682 fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
683 fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
684 if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
685 WARN_ON(1);
686 pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
687 pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
688 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
689 fix_to_virt(FIX_BTMAP_BEGIN));
690 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
691 fix_to_virt(FIX_BTMAP_END));
692
693 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
694 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
695 }
696 #endif
697 }
698
setup_vm_final(void)699 static void __init setup_vm_final(void)
700 {
701 uintptr_t va, map_size;
702 phys_addr_t pa, start, end;
703 u64 i;
704
705 /**
706 * MMU is enabled at this point. But page table setup is not complete yet.
707 * fixmap page table alloc functions should be used at this point
708 */
709 pt_ops.alloc_pte = alloc_pte_fixmap;
710 pt_ops.get_pte_virt = get_pte_virt_fixmap;
711 #ifndef __PAGETABLE_PMD_FOLDED
712 pt_ops.alloc_pmd = alloc_pmd_fixmap;
713 pt_ops.get_pmd_virt = get_pmd_virt_fixmap;
714 #endif
715 /* Setup swapper PGD for fixmap */
716 create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
717 __pa_symbol(fixmap_pgd_next),
718 PGDIR_SIZE, PAGE_TABLE);
719
720 /* Map all memory banks in the linear mapping */
721 for_each_mem_range(i, &start, &end) {
722 if (start >= end)
723 break;
724 if (start <= __pa(PAGE_OFFSET) &&
725 __pa(PAGE_OFFSET) < end)
726 start = __pa(PAGE_OFFSET);
727 if (end >= __pa(PAGE_OFFSET) + memory_limit)
728 end = __pa(PAGE_OFFSET) + memory_limit;
729
730 map_size = best_map_size(start, end - start);
731 for (pa = start; pa < end; pa += map_size) {
732 va = (uintptr_t)__va(pa);
733
734 create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
735 pgprot_from_va(va));
736 }
737 }
738
739 #ifdef CONFIG_64BIT
740 /* Map the kernel */
741 create_kernel_page_table(swapper_pg_dir, false);
742 #endif
743
744 /* Clear fixmap PTE and PMD mappings */
745 clear_fixmap(FIX_PTE);
746 clear_fixmap(FIX_PMD);
747
748 /* Move to swapper page table */
749 csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE);
750 local_flush_tlb_all();
751
752 /* generic page allocation functions must be used to setup page table */
753 pt_ops.alloc_pte = alloc_pte_late;
754 pt_ops.get_pte_virt = get_pte_virt_late;
755 #ifndef __PAGETABLE_PMD_FOLDED
756 pt_ops.alloc_pmd = alloc_pmd_late;
757 pt_ops.get_pmd_virt = get_pmd_virt_late;
758 #endif
759 }
760 #else
setup_vm(uintptr_t dtb_pa)761 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
762 {
763 dtb_early_va = (void *)dtb_pa;
764 dtb_early_pa = dtb_pa;
765 }
766
setup_vm_final(void)767 static inline void setup_vm_final(void)
768 {
769 }
770 #endif /* CONFIG_MMU */
771
772 #ifdef CONFIG_KEXEC_CORE
773 /*
774 * reserve_crashkernel() - reserves memory for crash kernel
775 *
776 * This function reserves memory area given in "crashkernel=" kernel command
777 * line parameter. The memory reserved is used by dump capture kernel when
778 * primary kernel is crashing.
779 */
reserve_crashkernel(void)780 static void __init reserve_crashkernel(void)
781 {
782 unsigned long long crash_base = 0;
783 unsigned long long crash_size = 0;
784 unsigned long search_start = memblock_start_of_DRAM();
785 unsigned long search_end = memblock_end_of_DRAM();
786
787 int ret = 0;
788
789 /*
790 * Don't reserve a region for a crash kernel on a crash kernel
791 * since it doesn't make much sense and we have limited memory
792 * resources.
793 */
794 #ifdef CONFIG_CRASH_DUMP
795 if (is_kdump_kernel()) {
796 pr_info("crashkernel: ignoring reservation request\n");
797 return;
798 }
799 #endif
800
801 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
802 &crash_size, &crash_base);
803 if (ret || !crash_size)
804 return;
805
806 crash_size = PAGE_ALIGN(crash_size);
807
808 if (crash_base) {
809 search_start = crash_base;
810 search_end = crash_base + crash_size;
811 }
812
813 /*
814 * Current riscv boot protocol requires 2MB alignment for
815 * RV64 and 4MB alignment for RV32 (hugepage size)
816 */
817 crash_base = memblock_phys_alloc_range(crash_size, PMD_SIZE,
818 search_start, search_end);
819 if (crash_base == 0) {
820 pr_warn("crashkernel: couldn't allocate %lldKB\n",
821 crash_size >> 10);
822 return;
823 }
824
825 pr_info("crashkernel: reserved 0x%016llx - 0x%016llx (%lld MB)\n",
826 crash_base, crash_base + crash_size, crash_size >> 20);
827
828 crashk_res.start = crash_base;
829 crashk_res.end = crash_base + crash_size - 1;
830 }
831 #endif /* CONFIG_KEXEC_CORE */
832
paging_init(void)833 void __init paging_init(void)
834 {
835 setup_bootmem();
836 setup_vm_final();
837 }
838
misc_mem_init(void)839 void __init misc_mem_init(void)
840 {
841 early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
842 arch_numa_init();
843 sparse_init();
844 zone_sizes_init();
845 #ifdef CONFIG_KEXEC_CORE
846 reserve_crashkernel();
847 #endif
848 memblock_dump_all();
849 }
850
851 #ifdef CONFIG_SPARSEMEM_VMEMMAP
vmemmap_populate(unsigned long start,unsigned long end,int node,struct vmem_altmap * altmap)852 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
853 struct vmem_altmap *altmap)
854 {
855 return vmemmap_populate_basepages(start, end, node, NULL);
856 }
857 #endif
858
