xref: /linux/arch/riscv/mm/init.c (revision 2c1ed907520c50326b8f604907a8478b27881a2e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012 Regents of the University of California
4  * Copyright (C) 2019 Western Digital Corporation or its affiliates.
5  * Copyright (C) 2020 FORTH-ICS/CARV
6  *  Nick Kossifidis <mick@ics.forth.gr>
7  */
8 
9 #include <linux/init.h>
10 #include <linux/mm.h>
11 #include <linux/memblock.h>
12 #include <linux/initrd.h>
13 #include <linux/swap.h>
14 #include <linux/swiotlb.h>
15 #include <linux/sizes.h>
16 #include <linux/of_fdt.h>
17 #include <linux/of_reserved_mem.h>
18 #include <linux/libfdt.h>
19 #include <linux/set_memory.h>
20 #include <linux/dma-map-ops.h>
21 #include <linux/crash_dump.h>
22 #include <linux/hugetlb.h>
23 #ifdef CONFIG_RELOCATABLE
24 #include <linux/elf.h>
25 #endif
26 #include <linux/kfence.h>
27 #include <linux/execmem.h>
28 
29 #include <asm/fixmap.h>
30 #include <asm/io.h>
31 #include <asm/kasan.h>
32 #include <asm/numa.h>
33 #include <asm/pgtable.h>
34 #include <asm/sections.h>
35 #include <asm/soc.h>
36 #include <asm/sparsemem.h>
37 #include <asm/tlbflush.h>
38 
39 #include "../kernel/head.h"
40 
41 u64 new_vmalloc[NR_CPUS / sizeof(u64) + 1];
42 
43 struct kernel_mapping kernel_map __ro_after_init;
44 EXPORT_SYMBOL(kernel_map);
45 #ifdef CONFIG_XIP_KERNEL
46 #define kernel_map	(*(struct kernel_mapping *)XIP_FIXUP(&kernel_map))
47 #endif
48 
49 #ifdef CONFIG_64BIT
50 u64 satp_mode __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL) ? SATP_MODE_57 : SATP_MODE_39;
51 #else
52 u64 satp_mode __ro_after_init = SATP_MODE_32;
53 #endif
54 EXPORT_SYMBOL(satp_mode);
55 
56 #ifdef CONFIG_64BIT
57 bool pgtable_l4_enabled __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL);
58 bool pgtable_l5_enabled __ro_after_init = !IS_ENABLED(CONFIG_XIP_KERNEL);
59 EXPORT_SYMBOL(pgtable_l4_enabled);
60 EXPORT_SYMBOL(pgtable_l5_enabled);
61 #endif
62 
63 phys_addr_t phys_ram_base __ro_after_init;
64 EXPORT_SYMBOL(phys_ram_base);
65 
66 #ifdef CONFIG_SPARSEMEM_VMEMMAP
67 #define VMEMMAP_ADDR_ALIGN	(1ULL << SECTION_SIZE_BITS)
68 
69 unsigned long vmemmap_start_pfn __ro_after_init;
70 EXPORT_SYMBOL(vmemmap_start_pfn);
71 #endif
72 
73 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
74 							__page_aligned_bss;
75 EXPORT_SYMBOL(empty_zero_page);
76 
77 extern char _start[];
78 void *_dtb_early_va __initdata;
79 uintptr_t _dtb_early_pa __initdata;
80 
81 phys_addr_t dma32_phys_limit __initdata;
82 
zone_sizes_init(void)83 static void __init zone_sizes_init(void)
84 {
85 	unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
86 
87 #ifdef CONFIG_ZONE_DMA32
88 	max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
89 #endif
90 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
91 
92 	free_area_init(max_zone_pfns);
93 }
94 
95 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
96 
97 #define LOG2_SZ_1K  ilog2(SZ_1K)
98 #define LOG2_SZ_1M  ilog2(SZ_1M)
99 #define LOG2_SZ_1G  ilog2(SZ_1G)
100 #define LOG2_SZ_1T  ilog2(SZ_1T)
101 
print_mlk(char * name,unsigned long b,unsigned long t)102 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
103 {
104 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld kB)\n", name, b, t,
105 		  (((t) - (b)) >> LOG2_SZ_1K));
106 }
107 
print_mlm(char * name,unsigned long b,unsigned long t)108 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
109 {
110 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld MB)\n", name, b, t,
111 		  (((t) - (b)) >> LOG2_SZ_1M));
112 }
113 
print_mlg(char * name,unsigned long b,unsigned long t)114 static inline void print_mlg(char *name, unsigned long b, unsigned long t)
115 {
116 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld GB)\n", name, b, t,
117 		   (((t) - (b)) >> LOG2_SZ_1G));
118 }
119 
120 #ifdef CONFIG_64BIT
print_mlt(char * name,unsigned long b,unsigned long t)121 static inline void print_mlt(char *name, unsigned long b, unsigned long t)
122 {
123 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld TB)\n", name, b, t,
124 		   (((t) - (b)) >> LOG2_SZ_1T));
125 }
126 #else
127 #define print_mlt(n, b, t) do {} while (0)
128 #endif
129 
print_ml(char * name,unsigned long b,unsigned long t)130 static inline void print_ml(char *name, unsigned long b, unsigned long t)
131 {
132 	unsigned long diff = t - b;
133 
134 	if (IS_ENABLED(CONFIG_64BIT) && (diff >> LOG2_SZ_1T) >= 10)
135 		print_mlt(name, b, t);
136 	else if ((diff >> LOG2_SZ_1G) >= 10)
137 		print_mlg(name, b, t);
138 	else if ((diff >> LOG2_SZ_1M) >= 10)
139 		print_mlm(name, b, t);
140 	else
141 		print_mlk(name, b, t);
142 }
143 
print_vm_layout(void)144 static void __init print_vm_layout(void)
145 {
146 	pr_notice("Virtual kernel memory layout:\n");
147 	print_ml("fixmap", (unsigned long)FIXADDR_START,
148 		(unsigned long)FIXADDR_TOP);
149 	print_ml("pci io", (unsigned long)PCI_IO_START,
150 		(unsigned long)PCI_IO_END);
151 	print_ml("vmemmap", (unsigned long)VMEMMAP_START,
152 		(unsigned long)VMEMMAP_END);
153 	print_ml("vmalloc", (unsigned long)VMALLOC_START,
154 		(unsigned long)VMALLOC_END);
155 #ifdef CONFIG_64BIT
156 	print_ml("modules", (unsigned long)MODULES_VADDR,
157 		(unsigned long)MODULES_END);
158 #endif
159 	print_ml("lowmem", (unsigned long)PAGE_OFFSET,
160 		(unsigned long)high_memory);
161 	if (IS_ENABLED(CONFIG_64BIT)) {
162 #ifdef CONFIG_KASAN
163 		print_ml("kasan", KASAN_SHADOW_START, KASAN_SHADOW_END);
164 #endif
165 
166 		print_ml("kernel", (unsigned long)kernel_map.virt_addr,
167 			 (unsigned long)ADDRESS_SPACE_END);
168 	}
169 }
170 #else
print_vm_layout(void)171 static void print_vm_layout(void) { }
172 #endif /* CONFIG_DEBUG_VM */
173 
mem_init(void)174 void __init mem_init(void)
175 {
176 	bool swiotlb = max_pfn > PFN_DOWN(dma32_phys_limit);
177 #ifdef CONFIG_FLATMEM
178 	BUG_ON(!mem_map);
179 #endif /* CONFIG_FLATMEM */
180 
181 	if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && !swiotlb &&
182 	    dma_cache_alignment != 1) {
183 		/*
184 		 * If no bouncing needed for ZONE_DMA, allocate 1MB swiotlb
185 		 * buffer per 1GB of RAM for kmalloc() bouncing on
186 		 * non-coherent platforms.
187 		 */
188 		unsigned long size =
189 			DIV_ROUND_UP(memblock_phys_mem_size(), 1024);
190 		swiotlb_adjust_size(min(swiotlb_size_or_default(), size));
191 		swiotlb = true;
192 	}
193 
194 	swiotlb_init(swiotlb, SWIOTLB_VERBOSE);
195 	memblock_free_all();
196 
197 	print_vm_layout();
198 }
199 
200 /* Limit the memory size via mem. */
201 static phys_addr_t memory_limit;
202 #ifdef CONFIG_XIP_KERNEL
203 #define memory_limit	(*(phys_addr_t *)XIP_FIXUP(&memory_limit))
204 #endif /* CONFIG_XIP_KERNEL */
205 
early_mem(char * p)206 static int __init early_mem(char *p)
207 {
208 	u64 size;
209 
210 	if (!p)
211 		return 1;
212 
213 	size = memparse(p, &p) & PAGE_MASK;
214 	memory_limit = min_t(u64, size, memory_limit);
215 
216 	pr_notice("Memory limited to %lldMB\n", (u64)memory_limit >> 20);
217 
218 	return 0;
219 }
220 early_param("mem", early_mem);
221 
setup_bootmem(void)222 static void __init setup_bootmem(void)
223 {
224 	phys_addr_t vmlinux_end = __pa_symbol(&_end);
225 	phys_addr_t max_mapped_addr;
226 	phys_addr_t phys_ram_end, vmlinux_start;
227 
228 	if (IS_ENABLED(CONFIG_XIP_KERNEL))
229 		vmlinux_start = __pa_symbol(&_sdata);
230 	else
231 		vmlinux_start = __pa_symbol(&_start);
232 
233 	memblock_enforce_memory_limit(memory_limit);
234 
235 	/*
236 	 * Make sure we align the reservation on PMD_SIZE since we will
237 	 * map the kernel in the linear mapping as read-only: we do not want
238 	 * any allocation to happen between _end and the next pmd aligned page.
239 	 */
240 	if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_STRICT_KERNEL_RWX))
241 		vmlinux_end = (vmlinux_end + PMD_SIZE - 1) & PMD_MASK;
242 	/*
243 	 * Reserve from the start of the kernel to the end of the kernel
244 	 */
245 	memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
246 
247 	/*
248 	 * Make sure we align the start of the memory on a PMD boundary so that
249 	 * at worst, we map the linear mapping with PMD mappings.
250 	 */
251 	if (!IS_ENABLED(CONFIG_XIP_KERNEL)) {
252 		phys_ram_base = memblock_start_of_DRAM() & PMD_MASK;
253 #ifdef CONFIG_SPARSEMEM_VMEMMAP
254 		vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT;
255 #endif
256 	}
257 
258 	/*
259 	 * In 64-bit, any use of __va/__pa before this point is wrong as we
260 	 * did not know the start of DRAM before.
261 	 */
262 	if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU))
263 		kernel_map.va_pa_offset = PAGE_OFFSET - phys_ram_base;
264 
265 	/*
266 	 * The size of the linear page mapping may restrict the amount of
267 	 * usable RAM.
268 	 */
269 	if (IS_ENABLED(CONFIG_64BIT) && IS_ENABLED(CONFIG_MMU)) {
270 		max_mapped_addr = __pa(PAGE_OFFSET) + KERN_VIRT_SIZE;
271 		if (memblock_end_of_DRAM() > max_mapped_addr) {
272 			memblock_cap_memory_range(phys_ram_base,
273 						  max_mapped_addr - phys_ram_base);
274 			pr_warn("Physical memory overflows the linear mapping size: region above %pa removed",
275 				&max_mapped_addr);
276 		}
277 	}
278 
279 	/*
280 	 * Reserve physical address space that would be mapped to virtual
281 	 * addresses greater than (void *)(-PAGE_SIZE) because:
282 	 *  - This memory would overlap with ERR_PTR
283 	 *  - This memory belongs to high memory, which is not supported
284 	 *
285 	 * This is not applicable to 64-bit kernel, because virtual addresses
286 	 * after (void *)(-PAGE_SIZE) are not linearly mapped: they are
287 	 * occupied by kernel mapping. Also it is unrealistic for high memory
288 	 * to exist on 64-bit platforms.
289 	 */
290 	if (!IS_ENABLED(CONFIG_64BIT)) {
291 		max_mapped_addr = __va_to_pa_nodebug(-PAGE_SIZE);
292 		memblock_reserve(max_mapped_addr, (phys_addr_t)-max_mapped_addr);
293 	}
294 
295 	phys_ram_end = memblock_end_of_DRAM();
296 	min_low_pfn = PFN_UP(phys_ram_base);
297 	max_low_pfn = max_pfn = PFN_DOWN(phys_ram_end);
298 	high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
299 
300 	dma32_phys_limit = min(4UL * SZ_1G, (unsigned long)PFN_PHYS(max_low_pfn));
301 	set_max_mapnr(max_low_pfn - ARCH_PFN_OFFSET);
302 
303 	reserve_initrd_mem();
304 
305 	/*
306 	 * No allocation should be done before reserving the memory as defined
307 	 * in the device tree, otherwise the allocation could end up in a
308 	 * reserved region.
309 	 */
310 	early_init_fdt_scan_reserved_mem();
311 
312 	/*
313 	 * If DTB is built in, no need to reserve its memblock.
314 	 * Otherwise, do reserve it but avoid using
315 	 * early_init_fdt_reserve_self() since __pa() does
316 	 * not work for DTB pointers that are fixmap addresses
317 	 */
318 	if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
319 		memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
320 
321 	dma_contiguous_reserve(dma32_phys_limit);
322 	if (IS_ENABLED(CONFIG_64BIT))
323 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
324 }
325 
326 #ifdef CONFIG_MMU
327 struct pt_alloc_ops pt_ops __meminitdata;
328 
329 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
330 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
331 static pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
332 
333 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
334 
335 #ifdef CONFIG_XIP_KERNEL
336 #define pt_ops			(*(struct pt_alloc_ops *)XIP_FIXUP(&pt_ops))
337 #define trampoline_pg_dir      ((pgd_t *)XIP_FIXUP(trampoline_pg_dir))
338 #define fixmap_pte             ((pte_t *)XIP_FIXUP(fixmap_pte))
339 #define early_pg_dir           ((pgd_t *)XIP_FIXUP(early_pg_dir))
340 #endif /* CONFIG_XIP_KERNEL */
341 
342 static const pgprot_t protection_map[16] = {
343 	[VM_NONE]					= PAGE_NONE,
344 	[VM_READ]					= PAGE_READ,
345 	[VM_WRITE]					= PAGE_COPY,
346 	[VM_WRITE | VM_READ]				= PAGE_COPY,
347 	[VM_EXEC]					= PAGE_EXEC,
348 	[VM_EXEC | VM_READ]				= PAGE_READ_EXEC,
349 	[VM_EXEC | VM_WRITE]				= PAGE_COPY_EXEC,
350 	[VM_EXEC | VM_WRITE | VM_READ]			= PAGE_COPY_EXEC,
351 	[VM_SHARED]					= PAGE_NONE,
352 	[VM_SHARED | VM_READ]				= PAGE_READ,
353 	[VM_SHARED | VM_WRITE]				= PAGE_SHARED,
354 	[VM_SHARED | VM_WRITE | VM_READ]		= PAGE_SHARED,
355 	[VM_SHARED | VM_EXEC]				= PAGE_EXEC,
356 	[VM_SHARED | VM_EXEC | VM_READ]			= PAGE_READ_EXEC,
357 	[VM_SHARED | VM_EXEC | VM_WRITE]		= PAGE_SHARED_EXEC,
358 	[VM_SHARED | VM_EXEC | VM_WRITE | VM_READ]	= PAGE_SHARED_EXEC
359 };
360 DECLARE_VM_GET_PAGE_PROT
361 
__set_fixmap(enum fixed_addresses idx,phys_addr_t phys,pgprot_t prot)362 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
363 {
364 	unsigned long addr = __fix_to_virt(idx);
365 	pte_t *ptep;
366 
367 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
368 
369 	ptep = &fixmap_pte[pte_index(addr)];
370 
371 	if (pgprot_val(prot))
372 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
373 	else
374 		pte_clear(&init_mm, addr, ptep);
375 	local_flush_tlb_page(addr);
376 }
377 
get_pte_virt_early(phys_addr_t pa)378 static inline pte_t *__init get_pte_virt_early(phys_addr_t pa)
379 {
380 	return (pte_t *)((uintptr_t)pa);
381 }
382 
get_pte_virt_fixmap(phys_addr_t pa)383 static inline pte_t *__init get_pte_virt_fixmap(phys_addr_t pa)
384 {
385 	clear_fixmap(FIX_PTE);
386 	return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
387 }
388 
get_pte_virt_late(phys_addr_t pa)389 static inline pte_t *__meminit get_pte_virt_late(phys_addr_t pa)
390 {
391 	return (pte_t *) __va(pa);
392 }
393 
alloc_pte_early(uintptr_t va)394 static inline phys_addr_t __init alloc_pte_early(uintptr_t va)
395 {
396 	/*
397 	 * We only create PMD or PGD early mappings so we
398 	 * should never reach here with MMU disabled.
399 	 */
400 	BUG();
401 }
402 
alloc_pte_fixmap(uintptr_t va)403 static inline phys_addr_t __init alloc_pte_fixmap(uintptr_t va)
404 {
405 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
406 }
407 
alloc_pte_late(uintptr_t va)408 static phys_addr_t __meminit alloc_pte_late(uintptr_t va)
409 {
410 	struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
411 
412 	BUG_ON(!ptdesc || !pagetable_pte_ctor(ptdesc));
413 	return __pa((pte_t *)ptdesc_address(ptdesc));
414 }
415 
create_pte_mapping(pte_t * ptep,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)416 static void __meminit create_pte_mapping(pte_t *ptep, uintptr_t va, phys_addr_t pa, phys_addr_t sz,
417 					 pgprot_t prot)
418 {
419 	uintptr_t pte_idx = pte_index(va);
420 
421 	BUG_ON(sz != PAGE_SIZE);
422 
423 	if (pte_none(ptep[pte_idx]))
424 		ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
425 }
426 
427 #ifndef __PAGETABLE_PMD_FOLDED
428 
429 static pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
430 static pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
431 static pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
432 
433 #ifdef CONFIG_XIP_KERNEL
434 #define trampoline_pmd ((pmd_t *)XIP_FIXUP(trampoline_pmd))
435 #define fixmap_pmd     ((pmd_t *)XIP_FIXUP(fixmap_pmd))
436 #define early_pmd      ((pmd_t *)XIP_FIXUP(early_pmd))
437 #endif /* CONFIG_XIP_KERNEL */
438 
439 static p4d_t trampoline_p4d[PTRS_PER_P4D] __page_aligned_bss;
440 static p4d_t fixmap_p4d[PTRS_PER_P4D] __page_aligned_bss;
441 static p4d_t early_p4d[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE);
442 
443 #ifdef CONFIG_XIP_KERNEL
444 #define trampoline_p4d ((p4d_t *)XIP_FIXUP(trampoline_p4d))
445 #define fixmap_p4d     ((p4d_t *)XIP_FIXUP(fixmap_p4d))
446 #define early_p4d      ((p4d_t *)XIP_FIXUP(early_p4d))
447 #endif /* CONFIG_XIP_KERNEL */
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 
453 #ifdef CONFIG_XIP_KERNEL
454 #define trampoline_pud ((pud_t *)XIP_FIXUP(trampoline_pud))
455 #define fixmap_pud     ((pud_t *)XIP_FIXUP(fixmap_pud))
456 #define early_pud      ((pud_t *)XIP_FIXUP(early_pud))
457 #endif /* CONFIG_XIP_KERNEL */
458 
get_pmd_virt_early(phys_addr_t pa)459 static pmd_t *__init get_pmd_virt_early(phys_addr_t pa)
460 {
461 	/* Before MMU is enabled */
462 	return (pmd_t *)((uintptr_t)pa);
463 }
464 
get_pmd_virt_fixmap(phys_addr_t pa)465 static pmd_t *__init get_pmd_virt_fixmap(phys_addr_t pa)
466 {
467 	clear_fixmap(FIX_PMD);
468 	return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
469 }
470 
get_pmd_virt_late(phys_addr_t pa)471 static pmd_t *__meminit get_pmd_virt_late(phys_addr_t pa)
472 {
473 	return (pmd_t *) __va(pa);
474 }
475 
alloc_pmd_early(uintptr_t va)476 static phys_addr_t __init alloc_pmd_early(uintptr_t va)
477 {
478 	BUG_ON((va - kernel_map.virt_addr) >> PUD_SHIFT);
479 
480 	return (uintptr_t)early_pmd;
481 }
482 
alloc_pmd_fixmap(uintptr_t va)483 static phys_addr_t __init alloc_pmd_fixmap(uintptr_t va)
484 {
485 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
486 }
487 
alloc_pmd_late(uintptr_t va)488 static phys_addr_t __meminit alloc_pmd_late(uintptr_t va)
489 {
490 	struct ptdesc *ptdesc = pagetable_alloc(GFP_KERNEL & ~__GFP_HIGHMEM, 0);
491 
492 	BUG_ON(!ptdesc || !pagetable_pmd_ctor(ptdesc));
493 	return __pa((pmd_t *)ptdesc_address(ptdesc));
494 }
495 
create_pmd_mapping(pmd_t * pmdp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)496 static void __meminit create_pmd_mapping(pmd_t *pmdp,
497 					 uintptr_t va, phys_addr_t pa,
498 					 phys_addr_t sz, pgprot_t prot)
499 {
500 	pte_t *ptep;
501 	phys_addr_t pte_phys;
502 	uintptr_t pmd_idx = pmd_index(va);
503 
504 	if (sz == PMD_SIZE) {
505 		if (pmd_none(pmdp[pmd_idx]))
506 			pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
507 		return;
508 	}
509 
510 	if (pmd_none(pmdp[pmd_idx])) {
511 		pte_phys = pt_ops.alloc_pte(va);
512 		pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
513 		ptep = pt_ops.get_pte_virt(pte_phys);
514 		memset(ptep, 0, PAGE_SIZE);
515 	} else {
516 		pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
517 		ptep = pt_ops.get_pte_virt(pte_phys);
518 	}
519 
520 	create_pte_mapping(ptep, va, pa, sz, prot);
521 }
522 
get_pud_virt_early(phys_addr_t pa)523 static pud_t *__init get_pud_virt_early(phys_addr_t pa)
524 {
525 	return (pud_t *)((uintptr_t)pa);
526 }
527 
get_pud_virt_fixmap(phys_addr_t pa)528 static pud_t *__init get_pud_virt_fixmap(phys_addr_t pa)
529 {
530 	clear_fixmap(FIX_PUD);
531 	return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
532 }
533 
get_pud_virt_late(phys_addr_t pa)534 static pud_t *__meminit get_pud_virt_late(phys_addr_t pa)
535 {
536 	return (pud_t *)__va(pa);
537 }
538 
alloc_pud_early(uintptr_t va)539 static phys_addr_t __init alloc_pud_early(uintptr_t va)
540 {
541 	/* Only one PUD is available for early mapping */
542 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
543 
544 	return (uintptr_t)early_pud;
545 }
546 
alloc_pud_fixmap(uintptr_t va)547 static phys_addr_t __init alloc_pud_fixmap(uintptr_t va)
548 {
549 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
550 }
551 
alloc_pud_late(uintptr_t va)552 static phys_addr_t __meminit alloc_pud_late(uintptr_t va)
553 {
554 	unsigned long vaddr;
555 
556 	vaddr = __get_free_page(GFP_KERNEL);
557 	BUG_ON(!vaddr);
558 	return __pa(vaddr);
559 }
560 
get_p4d_virt_early(phys_addr_t pa)561 static p4d_t *__init get_p4d_virt_early(phys_addr_t pa)
562 {
563 	return (p4d_t *)((uintptr_t)pa);
564 }
565 
get_p4d_virt_fixmap(phys_addr_t pa)566 static p4d_t *__init get_p4d_virt_fixmap(phys_addr_t pa)
567 {
568 	clear_fixmap(FIX_P4D);
569 	return (p4d_t *)set_fixmap_offset(FIX_P4D, pa);
570 }
571 
get_p4d_virt_late(phys_addr_t pa)572 static p4d_t *__meminit get_p4d_virt_late(phys_addr_t pa)
573 {
574 	return (p4d_t *)__va(pa);
575 }
576 
alloc_p4d_early(uintptr_t va)577 static phys_addr_t __init alloc_p4d_early(uintptr_t va)
578 {
579 	/* Only one P4D is available for early mapping */
580 	BUG_ON((va - kernel_map.virt_addr) >> PGDIR_SHIFT);
581 
582 	return (uintptr_t)early_p4d;
583 }
584 
alloc_p4d_fixmap(uintptr_t va)585 static phys_addr_t __init alloc_p4d_fixmap(uintptr_t va)
586 {
587 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
588 }
589 
alloc_p4d_late(uintptr_t va)590 static phys_addr_t __meminit alloc_p4d_late(uintptr_t va)
591 {
592 	unsigned long vaddr;
593 
594 	vaddr = __get_free_page(GFP_KERNEL);
595 	BUG_ON(!vaddr);
596 	return __pa(vaddr);
597 }
598 
create_pud_mapping(pud_t * pudp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)599 static void __meminit create_pud_mapping(pud_t *pudp, uintptr_t va, phys_addr_t pa, phys_addr_t sz,
600 					 pgprot_t prot)
601 {
602 	pmd_t *nextp;
603 	phys_addr_t next_phys;
604 	uintptr_t pud_index = pud_index(va);
605 
606 	if (sz == PUD_SIZE) {
607 		if (pud_val(pudp[pud_index]) == 0)
608 			pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
609 		return;
610 	}
611 
612 	if (pud_val(pudp[pud_index]) == 0) {
613 		next_phys = pt_ops.alloc_pmd(va);
614 		pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
615 		nextp = pt_ops.get_pmd_virt(next_phys);
616 		memset(nextp, 0, PAGE_SIZE);
617 	} else {
618 		next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
619 		nextp = pt_ops.get_pmd_virt(next_phys);
620 	}
621 
622 	create_pmd_mapping(nextp, va, pa, sz, prot);
623 }
624 
create_p4d_mapping(p4d_t * p4dp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)625 static void __meminit create_p4d_mapping(p4d_t *p4dp, uintptr_t va, phys_addr_t pa, phys_addr_t sz,
626 					 pgprot_t prot)
627 {
628 	pud_t *nextp;
629 	phys_addr_t next_phys;
630 	uintptr_t p4d_index = p4d_index(va);
631 
632 	if (sz == P4D_SIZE) {
633 		if (p4d_val(p4dp[p4d_index]) == 0)
634 			p4dp[p4d_index] = pfn_p4d(PFN_DOWN(pa), prot);
635 		return;
636 	}
637 
638 	if (p4d_val(p4dp[p4d_index]) == 0) {
639 		next_phys = pt_ops.alloc_pud(va);
640 		p4dp[p4d_index] = pfn_p4d(PFN_DOWN(next_phys), PAGE_TABLE);
641 		nextp = pt_ops.get_pud_virt(next_phys);
642 		memset(nextp, 0, PAGE_SIZE);
643 	} else {
644 		next_phys = PFN_PHYS(_p4d_pfn(p4dp[p4d_index]));
645 		nextp = pt_ops.get_pud_virt(next_phys);
646 	}
647 
648 	create_pud_mapping(nextp, va, pa, sz, prot);
649 }
650 
651 #define pgd_next_t		p4d_t
652 #define alloc_pgd_next(__va)	(pgtable_l5_enabled ?			\
653 		pt_ops.alloc_p4d(__va) : (pgtable_l4_enabled ?		\
654 		pt_ops.alloc_pud(__va) : pt_ops.alloc_pmd(__va)))
655 #define get_pgd_next_virt(__pa)	(pgtable_l5_enabled ?			\
656 		pt_ops.get_p4d_virt(__pa) : (pgd_next_t *)(pgtable_l4_enabled ?	\
657 		pt_ops.get_pud_virt(__pa) : (pud_t *)pt_ops.get_pmd_virt(__pa)))
658 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
659 				(pgtable_l5_enabled ?			\
660 		create_p4d_mapping(__nextp, __va, __pa, __sz, __prot) : \
661 				(pgtable_l4_enabled ?			\
662 		create_pud_mapping((pud_t *)__nextp, __va, __pa, __sz, __prot) :	\
663 		create_pmd_mapping((pmd_t *)__nextp, __va, __pa, __sz, __prot)))
664 #define fixmap_pgd_next		(pgtable_l5_enabled ?			\
665 		(uintptr_t)fixmap_p4d : (pgtable_l4_enabled ?		\
666 		(uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd))
667 #define trampoline_pgd_next	(pgtable_l5_enabled ?			\
668 		(uintptr_t)trampoline_p4d : (pgtable_l4_enabled ?	\
669 		(uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd))
670 #else
671 #define pgd_next_t		pte_t
672 #define alloc_pgd_next(__va)	pt_ops.alloc_pte(__va)
673 #define get_pgd_next_virt(__pa)	pt_ops.get_pte_virt(__pa)
674 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
675 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
676 #define fixmap_pgd_next		((uintptr_t)fixmap_pte)
677 #define create_p4d_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
678 #define create_pud_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
679 #define create_pmd_mapping(__pmdp, __va, __pa, __sz, __prot) do {} while(0)
680 #endif /* __PAGETABLE_PMD_FOLDED */
681 
create_pgd_mapping(pgd_t * pgdp,uintptr_t va,phys_addr_t pa,phys_addr_t sz,pgprot_t prot)682 void __meminit create_pgd_mapping(pgd_t *pgdp, uintptr_t va, phys_addr_t pa, phys_addr_t sz,
683 				  pgprot_t prot)
684 {
685 	pgd_next_t *nextp;
686 	phys_addr_t next_phys;
687 	uintptr_t pgd_idx = pgd_index(va);
688 
689 	if (sz == PGDIR_SIZE) {
690 		if (pgd_val(pgdp[pgd_idx]) == 0)
691 			pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
692 		return;
693 	}
694 
695 	if (pgd_val(pgdp[pgd_idx]) == 0) {
696 		next_phys = alloc_pgd_next(va);
697 		pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
698 		nextp = get_pgd_next_virt(next_phys);
699 		memset(nextp, 0, PAGE_SIZE);
700 	} else {
701 		next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
702 		nextp = get_pgd_next_virt(next_phys);
703 	}
704 
705 	create_pgd_next_mapping(nextp, va, pa, sz, prot);
706 }
707 
best_map_size(phys_addr_t pa,uintptr_t va,phys_addr_t size)708 static uintptr_t __meminit best_map_size(phys_addr_t pa, uintptr_t va, phys_addr_t size)
709 {
710 	if (debug_pagealloc_enabled())
711 		return PAGE_SIZE;
712 
713 	if (pgtable_l5_enabled &&
714 	    !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
715 		return P4D_SIZE;
716 
717 	if (pgtable_l4_enabled &&
718 	    !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
719 		return PUD_SIZE;
720 
721 	if (IS_ENABLED(CONFIG_64BIT) &&
722 	    !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
723 		return PMD_SIZE;
724 
725 	return PAGE_SIZE;
726 }
727 
728 #ifdef CONFIG_XIP_KERNEL
729 #define phys_ram_base  (*(phys_addr_t *)XIP_FIXUP(&phys_ram_base))
730 extern char _xiprom[], _exiprom[], __data_loc;
731 
732 /* called from head.S with MMU off */
__copy_data(void)733 asmlinkage void __init __copy_data(void)
734 {
735 	void *from = (void *)(&__data_loc);
736 	void *to = (void *)CONFIG_PHYS_RAM_BASE;
737 	size_t sz = (size_t)((uintptr_t)(&_end) - (uintptr_t)(&_sdata));
738 
739 	memcpy(to, from, sz);
740 }
741 #endif
742 
743 #ifdef CONFIG_STRICT_KERNEL_RWX
pgprot_from_va(uintptr_t va)744 static __meminit pgprot_t pgprot_from_va(uintptr_t va)
745 {
746 	if (is_va_kernel_text(va))
747 		return PAGE_KERNEL_READ_EXEC;
748 
749 	/*
750 	 * In 64-bit kernel, the kernel mapping is outside the linear mapping so
751 	 * we must protect its linear mapping alias from being executed and
752 	 * written.
753 	 * And rodata section is marked readonly in mark_rodata_ro.
754 	 */
755 	if (IS_ENABLED(CONFIG_64BIT) && is_va_kernel_lm_alias_text(va))
756 		return PAGE_KERNEL_READ;
757 
758 	return PAGE_KERNEL;
759 }
760 
mark_rodata_ro(void)761 void mark_rodata_ro(void)
762 {
763 	set_kernel_memory(__start_rodata, _data, set_memory_ro);
764 	if (IS_ENABLED(CONFIG_64BIT))
765 		set_kernel_memory(lm_alias(__start_rodata), lm_alias(_data),
766 				  set_memory_ro);
767 }
768 #else
pgprot_from_va(uintptr_t va)769 static __meminit pgprot_t pgprot_from_va(uintptr_t va)
770 {
771 	if (IS_ENABLED(CONFIG_64BIT) && !is_kernel_mapping(va))
772 		return PAGE_KERNEL;
773 
774 	return PAGE_KERNEL_EXEC;
775 }
776 #endif /* CONFIG_STRICT_KERNEL_RWX */
777 
778 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
779 u64 __pi_set_satp_mode_from_cmdline(uintptr_t dtb_pa);
780 
disable_pgtable_l5(void)781 static void __init disable_pgtable_l5(void)
782 {
783 	pgtable_l5_enabled = false;
784 	kernel_map.page_offset = PAGE_OFFSET_L4;
785 	satp_mode = SATP_MODE_48;
786 }
787 
disable_pgtable_l4(void)788 static void __init disable_pgtable_l4(void)
789 {
790 	pgtable_l4_enabled = false;
791 	kernel_map.page_offset = PAGE_OFFSET_L3;
792 	satp_mode = SATP_MODE_39;
793 }
794 
print_no4lvl(char * p)795 static int __init print_no4lvl(char *p)
796 {
797 	pr_info("Disabled 4-level and 5-level paging");
798 	return 0;
799 }
800 early_param("no4lvl", print_no4lvl);
801 
print_no5lvl(char * p)802 static int __init print_no5lvl(char *p)
803 {
804 	pr_info("Disabled 5-level paging");
805 	return 0;
806 }
807 early_param("no5lvl", print_no5lvl);
808 
set_mmap_rnd_bits_max(void)809 static void __init set_mmap_rnd_bits_max(void)
810 {
811 	mmap_rnd_bits_max = MMAP_VA_BITS - PAGE_SHIFT - 3;
812 }
813 
814 /*
815  * There is a simple way to determine if 4-level is supported by the
816  * underlying hardware: establish 1:1 mapping in 4-level page table mode
817  * then read SATP to see if the configuration was taken into account
818  * meaning sv48 is supported.
819  */
set_satp_mode(uintptr_t dtb_pa)820 static __init void set_satp_mode(uintptr_t dtb_pa)
821 {
822 	u64 identity_satp, hw_satp;
823 	uintptr_t set_satp_mode_pmd = ((unsigned long)set_satp_mode) & PMD_MASK;
824 	u64 satp_mode_cmdline = __pi_set_satp_mode_from_cmdline(dtb_pa);
825 
826 	if (satp_mode_cmdline == SATP_MODE_57) {
827 		disable_pgtable_l5();
828 	} else if (satp_mode_cmdline == SATP_MODE_48) {
829 		disable_pgtable_l5();
830 		disable_pgtable_l4();
831 		return;
832 	}
833 
834 	create_p4d_mapping(early_p4d,
835 			set_satp_mode_pmd, (uintptr_t)early_pud,
836 			P4D_SIZE, PAGE_TABLE);
837 	create_pud_mapping(early_pud,
838 			   set_satp_mode_pmd, (uintptr_t)early_pmd,
839 			   PUD_SIZE, PAGE_TABLE);
840 	/* Handle the case where set_satp_mode straddles 2 PMDs */
841 	create_pmd_mapping(early_pmd,
842 			   set_satp_mode_pmd, set_satp_mode_pmd,
843 			   PMD_SIZE, PAGE_KERNEL_EXEC);
844 	create_pmd_mapping(early_pmd,
845 			   set_satp_mode_pmd + PMD_SIZE,
846 			   set_satp_mode_pmd + PMD_SIZE,
847 			   PMD_SIZE, PAGE_KERNEL_EXEC);
848 retry:
849 	create_pgd_mapping(early_pg_dir,
850 			   set_satp_mode_pmd,
851 			   pgtable_l5_enabled ?
852 				(uintptr_t)early_p4d : (uintptr_t)early_pud,
853 			   PGDIR_SIZE, PAGE_TABLE);
854 
855 	identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
856 
857 	local_flush_tlb_all();
858 	csr_write(CSR_SATP, identity_satp);
859 	hw_satp = csr_swap(CSR_SATP, 0ULL);
860 	local_flush_tlb_all();
861 
862 	if (hw_satp != identity_satp) {
863 		if (pgtable_l5_enabled) {
864 			disable_pgtable_l5();
865 			memset(early_pg_dir, 0, PAGE_SIZE);
866 			goto retry;
867 		}
868 		disable_pgtable_l4();
869 	}
870 
871 	memset(early_pg_dir, 0, PAGE_SIZE);
872 	memset(early_p4d, 0, PAGE_SIZE);
873 	memset(early_pud, 0, PAGE_SIZE);
874 	memset(early_pmd, 0, PAGE_SIZE);
875 }
876 #endif
877 
878 /*
879  * setup_vm() is called from head.S with MMU-off.
880  *
881  * Following requirements should be honoured for setup_vm() to work
882  * correctly:
883  * 1) It should use PC-relative addressing for accessing kernel symbols.
884  *    To achieve this we always use GCC cmodel=medany.
885  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
886  *    so disable compiler instrumentation when FTRACE is enabled.
887  *
888  * Currently, the above requirements are honoured by using custom CFLAGS
889  * for init.o in mm/Makefile.
890  */
891 
892 #ifndef __riscv_cmodel_medany
893 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
894 #endif
895 
896 #ifdef CONFIG_RELOCATABLE
897 extern unsigned long __rela_dyn_start, __rela_dyn_end;
898 
relocate_kernel(void)899 static void __init relocate_kernel(void)
900 {
901 	Elf64_Rela *rela = (Elf64_Rela *)&__rela_dyn_start;
902 	/*
903 	 * This holds the offset between the linked virtual address and the
904 	 * relocated virtual address.
905 	 */
906 	uintptr_t reloc_offset = kernel_map.virt_addr - KERNEL_LINK_ADDR;
907 	/*
908 	 * This holds the offset between kernel linked virtual address and
909 	 * physical address.
910 	 */
911 	uintptr_t va_kernel_link_pa_offset = KERNEL_LINK_ADDR - kernel_map.phys_addr;
912 
913 	for ( ; rela < (Elf64_Rela *)&__rela_dyn_end; rela++) {
914 		Elf64_Addr addr = (rela->r_offset - va_kernel_link_pa_offset);
915 		Elf64_Addr relocated_addr = rela->r_addend;
916 
917 		if (rela->r_info != R_RISCV_RELATIVE)
918 			continue;
919 
920 		/*
921 		 * Make sure to not relocate vdso symbols like rt_sigreturn
922 		 * which are linked from the address 0 in vmlinux since
923 		 * vdso symbol addresses are actually used as an offset from
924 		 * mm->context.vdso in VDSO_OFFSET macro.
925 		 */
926 		if (relocated_addr >= KERNEL_LINK_ADDR)
927 			relocated_addr += reloc_offset;
928 
929 		*(Elf64_Addr *)addr = relocated_addr;
930 	}
931 }
932 #endif /* CONFIG_RELOCATABLE */
933 
934 #ifdef CONFIG_XIP_KERNEL
create_kernel_page_table(pgd_t * pgdir,__always_unused bool early)935 static void __init create_kernel_page_table(pgd_t *pgdir,
936 					    __always_unused bool early)
937 {
938 	uintptr_t va, start_va, end_va;
939 
940 	/* Map the flash resident part */
941 	end_va = kernel_map.virt_addr + kernel_map.xiprom_sz;
942 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
943 		create_pgd_mapping(pgdir, va,
944 				   kernel_map.xiprom + (va - kernel_map.virt_addr),
945 				   PMD_SIZE, PAGE_KERNEL_EXEC);
946 
947 	/* Map the data in RAM */
948 	start_va = kernel_map.virt_addr + (uintptr_t)&_sdata - (uintptr_t)&_start;
949 	end_va = kernel_map.virt_addr + kernel_map.size;
950 	for (va = start_va; va < end_va; va += PMD_SIZE)
951 		create_pgd_mapping(pgdir, va,
952 				   kernel_map.phys_addr + (va - start_va),
953 				   PMD_SIZE, PAGE_KERNEL);
954 }
955 #else
create_kernel_page_table(pgd_t * pgdir,bool early)956 static void __init create_kernel_page_table(pgd_t *pgdir, bool early)
957 {
958 	uintptr_t va, end_va;
959 
960 	end_va = kernel_map.virt_addr + kernel_map.size;
961 	for (va = kernel_map.virt_addr; va < end_va; va += PMD_SIZE)
962 		create_pgd_mapping(pgdir, va,
963 				   kernel_map.phys_addr + (va - kernel_map.virt_addr),
964 				   PMD_SIZE,
965 				   early ?
966 					PAGE_KERNEL_EXEC : pgprot_from_va(va));
967 }
968 #endif
969 
970 /*
971  * Setup a 4MB mapping that encompasses the device tree: for 64-bit kernel,
972  * this means 2 PMD entries whereas for 32-bit kernel, this is only 1 PGDIR
973  * entry.
974  */
create_fdt_early_page_table(uintptr_t fix_fdt_va,uintptr_t dtb_pa)975 static void __init create_fdt_early_page_table(uintptr_t fix_fdt_va,
976 					       uintptr_t dtb_pa)
977 {
978 #ifndef CONFIG_BUILTIN_DTB
979 	uintptr_t pa = dtb_pa & ~(PMD_SIZE - 1);
980 
981 	/* Make sure the fdt fixmap address is always aligned on PMD size */
982 	BUILD_BUG_ON(FIX_FDT % (PMD_SIZE / PAGE_SIZE));
983 
984 	/* In 32-bit only, the fdt lies in its own PGD */
985 	if (!IS_ENABLED(CONFIG_64BIT)) {
986 		create_pgd_mapping(early_pg_dir, fix_fdt_va,
987 				   pa, MAX_FDT_SIZE, PAGE_KERNEL);
988 	} else {
989 		create_pmd_mapping(fixmap_pmd, fix_fdt_va,
990 				   pa, PMD_SIZE, PAGE_KERNEL);
991 		create_pmd_mapping(fixmap_pmd, fix_fdt_va + PMD_SIZE,
992 				   pa + PMD_SIZE, PMD_SIZE, PAGE_KERNEL);
993 	}
994 
995 	dtb_early_va = (void *)fix_fdt_va + (dtb_pa & (PMD_SIZE - 1));
996 #else
997 	/*
998 	 * For 64-bit kernel, __va can't be used since it would return a linear
999 	 * mapping address whereas dtb_early_va will be used before
1000 	 * setup_vm_final installs the linear mapping. For 32-bit kernel, as the
1001 	 * kernel is mapped in the linear mapping, that makes no difference.
1002 	 */
1003 	dtb_early_va = kernel_mapping_pa_to_va(dtb_pa);
1004 #endif
1005 
1006 	dtb_early_pa = dtb_pa;
1007 }
1008 
1009 /*
1010  * MMU is not enabled, the page tables are allocated directly using
1011  * early_pmd/pud/p4d and the address returned is the physical one.
1012  */
pt_ops_set_early(void)1013 static void __init pt_ops_set_early(void)
1014 {
1015 	pt_ops.alloc_pte = alloc_pte_early;
1016 	pt_ops.get_pte_virt = get_pte_virt_early;
1017 #ifndef __PAGETABLE_PMD_FOLDED
1018 	pt_ops.alloc_pmd = alloc_pmd_early;
1019 	pt_ops.get_pmd_virt = get_pmd_virt_early;
1020 	pt_ops.alloc_pud = alloc_pud_early;
1021 	pt_ops.get_pud_virt = get_pud_virt_early;
1022 	pt_ops.alloc_p4d = alloc_p4d_early;
1023 	pt_ops.get_p4d_virt = get_p4d_virt_early;
1024 #endif
1025 }
1026 
1027 /*
1028  * MMU is enabled but page table setup is not complete yet.
1029  * fixmap page table alloc functions must be used as a means to temporarily
1030  * map the allocated physical pages since the linear mapping does not exist yet.
1031  *
1032  * Note that this is called with MMU disabled, hence kernel_mapping_pa_to_va,
1033  * but it will be used as described above.
1034  */
pt_ops_set_fixmap(void)1035 static void __init pt_ops_set_fixmap(void)
1036 {
1037 	pt_ops.alloc_pte = kernel_mapping_pa_to_va(alloc_pte_fixmap);
1038 	pt_ops.get_pte_virt = kernel_mapping_pa_to_va(get_pte_virt_fixmap);
1039 #ifndef __PAGETABLE_PMD_FOLDED
1040 	pt_ops.alloc_pmd = kernel_mapping_pa_to_va(alloc_pmd_fixmap);
1041 	pt_ops.get_pmd_virt = kernel_mapping_pa_to_va(get_pmd_virt_fixmap);
1042 	pt_ops.alloc_pud = kernel_mapping_pa_to_va(alloc_pud_fixmap);
1043 	pt_ops.get_pud_virt = kernel_mapping_pa_to_va(get_pud_virt_fixmap);
1044 	pt_ops.alloc_p4d = kernel_mapping_pa_to_va(alloc_p4d_fixmap);
1045 	pt_ops.get_p4d_virt = kernel_mapping_pa_to_va(get_p4d_virt_fixmap);
1046 #endif
1047 }
1048 
1049 /*
1050  * MMU is enabled and page table setup is complete, so from now, we can use
1051  * generic page allocation functions to setup page table.
1052  */
pt_ops_set_late(void)1053 static void __init pt_ops_set_late(void)
1054 {
1055 	pt_ops.alloc_pte = alloc_pte_late;
1056 	pt_ops.get_pte_virt = get_pte_virt_late;
1057 #ifndef __PAGETABLE_PMD_FOLDED
1058 	pt_ops.alloc_pmd = alloc_pmd_late;
1059 	pt_ops.get_pmd_virt = get_pmd_virt_late;
1060 	pt_ops.alloc_pud = alloc_pud_late;
1061 	pt_ops.get_pud_virt = get_pud_virt_late;
1062 	pt_ops.alloc_p4d = alloc_p4d_late;
1063 	pt_ops.get_p4d_virt = get_p4d_virt_late;
1064 #endif
1065 }
1066 
1067 #ifdef CONFIG_RANDOMIZE_BASE
1068 extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa);
1069 extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa);
1070 extern u64 __init __pi_get_kaslr_seed_zkr(const uintptr_t dtb_pa);
1071 
print_nokaslr(char * p)1072 static int __init print_nokaslr(char *p)
1073 {
1074 	pr_info("Disabled KASLR");
1075 	return 0;
1076 }
1077 early_param("nokaslr", print_nokaslr);
1078 
kaslr_offset(void)1079 unsigned long kaslr_offset(void)
1080 {
1081 	return kernel_map.virt_offset;
1082 }
1083 #endif
1084 
setup_vm(uintptr_t dtb_pa)1085 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1086 {
1087 	pmd_t __maybe_unused fix_bmap_spmd, fix_bmap_epmd;
1088 
1089 #ifdef CONFIG_RANDOMIZE_BASE
1090 	if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) {
1091 		u64 kaslr_seed = __pi_get_kaslr_seed_zkr(dtb_pa);
1092 		u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
1093 		u32 nr_pos;
1094 
1095 		if (kaslr_seed == 0)
1096 			kaslr_seed = __pi_get_kaslr_seed(dtb_pa);
1097 		/*
1098 		 * Compute the number of positions available: we are limited
1099 		 * by the early page table that only has one PUD and we must
1100 		 * be aligned on PMD_SIZE.
1101 		 */
1102 		nr_pos = (PUD_SIZE - kernel_size) / PMD_SIZE;
1103 
1104 		kernel_map.virt_offset = (kaslr_seed % nr_pos) * PMD_SIZE;
1105 	}
1106 #endif
1107 
1108 	kernel_map.virt_addr = KERNEL_LINK_ADDR + kernel_map.virt_offset;
1109 
1110 #ifdef CONFIG_XIP_KERNEL
1111 #ifdef CONFIG_64BIT
1112 	kernel_map.page_offset = PAGE_OFFSET_L3;
1113 #else
1114 	kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1115 #endif
1116 	kernel_map.xiprom = (uintptr_t)CONFIG_XIP_PHYS_ADDR;
1117 	kernel_map.xiprom_sz = (uintptr_t)(&_exiprom) - (uintptr_t)(&_xiprom);
1118 
1119 	phys_ram_base = CONFIG_PHYS_RAM_BASE;
1120 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1121 	vmemmap_start_pfn = round_down(phys_ram_base, VMEMMAP_ADDR_ALIGN) >> PAGE_SHIFT;
1122 #endif
1123 	kernel_map.phys_addr = (uintptr_t)CONFIG_PHYS_RAM_BASE;
1124 	kernel_map.size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
1125 
1126 	kernel_map.va_kernel_xip_text_pa_offset = kernel_map.virt_addr - kernel_map.xiprom;
1127 	kernel_map.va_kernel_xip_data_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr
1128 						+ (uintptr_t)&_sdata - (uintptr_t)&_start;
1129 #else
1130 	kernel_map.page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
1131 	kernel_map.phys_addr = (uintptr_t)(&_start);
1132 	kernel_map.size = (uintptr_t)(&_end) - kernel_map.phys_addr;
1133 	kernel_map.va_kernel_pa_offset = kernel_map.virt_addr - kernel_map.phys_addr;
1134 #endif
1135 
1136 #if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
1137 	set_satp_mode(dtb_pa);
1138 	set_mmap_rnd_bits_max();
1139 #endif
1140 
1141 	/*
1142 	 * In 64-bit, we defer the setup of va_pa_offset to setup_bootmem,
1143 	 * where we have the system memory layout: this allows us to align
1144 	 * the physical and virtual mappings and then make use of PUD/P4D/PGD
1145 	 * for the linear mapping. This is only possible because the kernel
1146 	 * mapping lies outside the linear mapping.
1147 	 * In 32-bit however, as the kernel resides in the linear mapping,
1148 	 * setup_vm_final can not change the mapping established here,
1149 	 * otherwise the same kernel addresses would get mapped to different
1150 	 * physical addresses (if the start of dram is different from the
1151 	 * kernel physical address start).
1152 	 */
1153 	kernel_map.va_pa_offset = IS_ENABLED(CONFIG_64BIT) ?
1154 				0UL : PAGE_OFFSET - kernel_map.phys_addr;
1155 
1156 	memory_limit = KERN_VIRT_SIZE;
1157 
1158 	/* Sanity check alignment and size */
1159 	BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
1160 	BUG_ON((kernel_map.phys_addr % PMD_SIZE) != 0);
1161 
1162 #ifdef CONFIG_64BIT
1163 	/*
1164 	 * The last 4K bytes of the addressable memory can not be mapped because
1165 	 * of IS_ERR_VALUE macro.
1166 	 */
1167 	BUG_ON((kernel_map.virt_addr + kernel_map.size) > ADDRESS_SPACE_END - SZ_4K);
1168 #endif
1169 
1170 #ifdef CONFIG_RELOCATABLE
1171 	/*
1172 	 * Early page table uses only one PUD, which makes it possible
1173 	 * to map PUD_SIZE aligned on PUD_SIZE: if the relocation offset
1174 	 * makes the kernel cross over a PUD_SIZE boundary, raise a bug
1175 	 * since a part of the kernel would not get mapped.
1176 	 */
1177 	BUG_ON(PUD_SIZE - (kernel_map.virt_addr & (PUD_SIZE - 1)) < kernel_map.size);
1178 	relocate_kernel();
1179 #endif
1180 
1181 	apply_early_boot_alternatives();
1182 	pt_ops_set_early();
1183 
1184 	/* Setup early PGD for fixmap */
1185 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
1186 			   fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1187 
1188 #ifndef __PAGETABLE_PMD_FOLDED
1189 	/* Setup fixmap P4D and PUD */
1190 	if (pgtable_l5_enabled)
1191 		create_p4d_mapping(fixmap_p4d, FIXADDR_START,
1192 				   (uintptr_t)fixmap_pud, P4D_SIZE, PAGE_TABLE);
1193 	/* Setup fixmap PUD and PMD */
1194 	if (pgtable_l4_enabled)
1195 		create_pud_mapping(fixmap_pud, FIXADDR_START,
1196 				   (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
1197 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
1198 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
1199 	/* Setup trampoline PGD and PMD */
1200 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1201 			   trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
1202 	if (pgtable_l5_enabled)
1203 		create_p4d_mapping(trampoline_p4d, kernel_map.virt_addr,
1204 				   (uintptr_t)trampoline_pud, P4D_SIZE, PAGE_TABLE);
1205 	if (pgtable_l4_enabled)
1206 		create_pud_mapping(trampoline_pud, kernel_map.virt_addr,
1207 				   (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
1208 #ifdef CONFIG_XIP_KERNEL
1209 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1210 			   kernel_map.xiprom, PMD_SIZE, PAGE_KERNEL_EXEC);
1211 #else
1212 	create_pmd_mapping(trampoline_pmd, kernel_map.virt_addr,
1213 			   kernel_map.phys_addr, PMD_SIZE, PAGE_KERNEL_EXEC);
1214 #endif
1215 #else
1216 	/* Setup trampoline PGD */
1217 	create_pgd_mapping(trampoline_pg_dir, kernel_map.virt_addr,
1218 			   kernel_map.phys_addr, PGDIR_SIZE, PAGE_KERNEL_EXEC);
1219 #endif
1220 
1221 	/*
1222 	 * Setup early PGD covering entire kernel which will allow
1223 	 * us to reach paging_init(). We map all memory banks later
1224 	 * in setup_vm_final() below.
1225 	 */
1226 	create_kernel_page_table(early_pg_dir, true);
1227 
1228 	/* Setup early mapping for FDT early scan */
1229 	create_fdt_early_page_table(__fix_to_virt(FIX_FDT), dtb_pa);
1230 
1231 	/*
1232 	 * Bootime fixmap only can handle PMD_SIZE mapping. Thus, boot-ioremap
1233 	 * range can not span multiple pmds.
1234 	 */
1235 	BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1236 		     != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1237 
1238 #ifndef __PAGETABLE_PMD_FOLDED
1239 	/*
1240 	 * Early ioremap fixmap is already created as it lies within first 2MB
1241 	 * of fixmap region. We always map PMD_SIZE. Thus, both FIX_BTMAP_END
1242 	 * FIX_BTMAP_BEGIN should lie in the same pmd. Verify that and warn
1243 	 * the user if not.
1244 	 */
1245 	fix_bmap_spmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_BEGIN))];
1246 	fix_bmap_epmd = fixmap_pmd[pmd_index(__fix_to_virt(FIX_BTMAP_END))];
1247 	if (pmd_val(fix_bmap_spmd) != pmd_val(fix_bmap_epmd)) {
1248 		WARN_ON(1);
1249 		pr_warn("fixmap btmap start [%08lx] != end [%08lx]\n",
1250 			pmd_val(fix_bmap_spmd), pmd_val(fix_bmap_epmd));
1251 		pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1252 			fix_to_virt(FIX_BTMAP_BEGIN));
1253 		pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
1254 			fix_to_virt(FIX_BTMAP_END));
1255 
1256 		pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
1257 		pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
1258 	}
1259 #endif
1260 
1261 	pt_ops_set_fixmap();
1262 }
1263 
create_linear_mapping_range(phys_addr_t start,phys_addr_t end,uintptr_t fixed_map_size,const pgprot_t * pgprot)1264 static void __meminit create_linear_mapping_range(phys_addr_t start, phys_addr_t end,
1265 						  uintptr_t fixed_map_size, const pgprot_t *pgprot)
1266 {
1267 	phys_addr_t pa;
1268 	uintptr_t va, map_size;
1269 
1270 	for (pa = start; pa < end; pa += map_size) {
1271 		va = (uintptr_t)__va(pa);
1272 		map_size = fixed_map_size ? fixed_map_size :
1273 					    best_map_size(pa, va, end - pa);
1274 
1275 		create_pgd_mapping(swapper_pg_dir, va, pa, map_size,
1276 				   pgprot ? *pgprot : pgprot_from_va(va));
1277 	}
1278 }
1279 
create_linear_mapping_page_table(void)1280 static void __init create_linear_mapping_page_table(void)
1281 {
1282 	phys_addr_t start, end;
1283 	phys_addr_t kfence_pool __maybe_unused;
1284 	u64 i;
1285 
1286 #ifdef CONFIG_STRICT_KERNEL_RWX
1287 	phys_addr_t ktext_start = __pa_symbol(_start);
1288 	phys_addr_t ktext_size = __init_data_begin - _start;
1289 	phys_addr_t krodata_start = __pa_symbol(__start_rodata);
1290 	phys_addr_t krodata_size = _data - __start_rodata;
1291 
1292 	/* Isolate kernel text and rodata so they don't get mapped with a PUD */
1293 	memblock_mark_nomap(ktext_start,  ktext_size);
1294 	memblock_mark_nomap(krodata_start, krodata_size);
1295 #endif
1296 
1297 #ifdef CONFIG_KFENCE
1298 	/*
1299 	 *  kfence pool must be backed by PAGE_SIZE mappings, so allocate it
1300 	 *  before we setup the linear mapping so that we avoid using hugepages
1301 	 *  for this region.
1302 	 */
1303 	kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
1304 	BUG_ON(!kfence_pool);
1305 
1306 	memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
1307 	__kfence_pool = __va(kfence_pool);
1308 #endif
1309 
1310 	/* Map all memory banks in the linear mapping */
1311 	for_each_mem_range(i, &start, &end) {
1312 		if (start >= end)
1313 			break;
1314 		if (start <= __pa(PAGE_OFFSET) &&
1315 		    __pa(PAGE_OFFSET) < end)
1316 			start = __pa(PAGE_OFFSET);
1317 
1318 		create_linear_mapping_range(start, end, 0, NULL);
1319 	}
1320 
1321 #ifdef CONFIG_STRICT_KERNEL_RWX
1322 	create_linear_mapping_range(ktext_start, ktext_start + ktext_size, 0, NULL);
1323 	create_linear_mapping_range(krodata_start, krodata_start + krodata_size, 0, NULL);
1324 
1325 	memblock_clear_nomap(ktext_start,  ktext_size);
1326 	memblock_clear_nomap(krodata_start, krodata_size);
1327 #endif
1328 
1329 #ifdef CONFIG_KFENCE
1330 	create_linear_mapping_range(kfence_pool, kfence_pool + KFENCE_POOL_SIZE, PAGE_SIZE, NULL);
1331 
1332 	memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
1333 #endif
1334 }
1335 
setup_vm_final(void)1336 static void __init setup_vm_final(void)
1337 {
1338 	/* Setup swapper PGD for fixmap */
1339 #if !defined(CONFIG_64BIT)
1340 	/*
1341 	 * In 32-bit, the device tree lies in a pgd entry, so it must be copied
1342 	 * directly in swapper_pg_dir in addition to the pgd entry that points
1343 	 * to fixmap_pte.
1344 	 */
1345 	unsigned long idx = pgd_index(__fix_to_virt(FIX_FDT));
1346 
1347 	set_pgd(&swapper_pg_dir[idx], early_pg_dir[idx]);
1348 #endif
1349 	create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
1350 			   __pa_symbol(fixmap_pgd_next),
1351 			   PGDIR_SIZE, PAGE_TABLE);
1352 
1353 	/* Map the linear mapping */
1354 	create_linear_mapping_page_table();
1355 
1356 	/* Map the kernel */
1357 	if (IS_ENABLED(CONFIG_64BIT))
1358 		create_kernel_page_table(swapper_pg_dir, false);
1359 
1360 #ifdef CONFIG_KASAN
1361 	kasan_swapper_init();
1362 #endif
1363 
1364 	/* Clear fixmap PTE and PMD mappings */
1365 	clear_fixmap(FIX_PTE);
1366 	clear_fixmap(FIX_PMD);
1367 	clear_fixmap(FIX_PUD);
1368 	clear_fixmap(FIX_P4D);
1369 
1370 	/* Move to swapper page table */
1371 	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
1372 	local_flush_tlb_all();
1373 
1374 	pt_ops_set_late();
1375 }
1376 #else
setup_vm(uintptr_t dtb_pa)1377 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
1378 {
1379 	dtb_early_va = (void *)dtb_pa;
1380 	dtb_early_pa = dtb_pa;
1381 }
1382 
setup_vm_final(void)1383 static inline void setup_vm_final(void)
1384 {
1385 }
1386 #endif /* CONFIG_MMU */
1387 
1388 /*
1389  * reserve_crashkernel() - reserves memory for crash kernel
1390  *
1391  * This function reserves memory area given in "crashkernel=" kernel command
1392  * line parameter. The memory reserved is used by dump capture kernel when
1393  * primary kernel is crashing.
1394  */
arch_reserve_crashkernel(void)1395 static void __init arch_reserve_crashkernel(void)
1396 {
1397 	unsigned long long low_size = 0;
1398 	unsigned long long crash_base, crash_size;
1399 	char *cmdline = boot_command_line;
1400 	bool high = false;
1401 	int ret;
1402 
1403 	if (!IS_ENABLED(CONFIG_CRASH_RESERVE))
1404 		return;
1405 
1406 	ret = parse_crashkernel(cmdline, memblock_phys_mem_size(),
1407 				&crash_size, &crash_base,
1408 				&low_size, &high);
1409 	if (ret)
1410 		return;
1411 
1412 	reserve_crashkernel_generic(cmdline, crash_size, crash_base,
1413 				    low_size, high);
1414 }
1415 
paging_init(void)1416 void __init paging_init(void)
1417 {
1418 	setup_bootmem();
1419 	setup_vm_final();
1420 
1421 	/* Depend on that Linear Mapping is ready */
1422 	memblock_allow_resize();
1423 }
1424 
misc_mem_init(void)1425 void __init misc_mem_init(void)
1426 {
1427 	early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
1428 	arch_numa_init();
1429 	sparse_init();
1430 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1431 	/* The entire VMEMMAP region has been populated. Flush TLB for this region */
1432 	local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END);
1433 #endif
1434 	zone_sizes_init();
1435 	arch_reserve_crashkernel();
1436 	memblock_dump_all();
1437 }
1438 
1439 #ifdef CONFIG_SPARSEMEM_VMEMMAP
vmemmap_set_pmd(pmd_t * pmd,void * p,int node,unsigned long addr,unsigned long next)1440 void __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node,
1441 			       unsigned long addr, unsigned long next)
1442 {
1443 	pmd_set_huge(pmd, virt_to_phys(p), PAGE_KERNEL);
1444 }
1445 
vmemmap_check_pmd(pmd_t * pmdp,int node,unsigned long addr,unsigned long next)1446 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node,
1447 				unsigned long addr, unsigned long next)
1448 {
1449 	vmemmap_verify((pte_t *)pmdp, node, addr, next);
1450 	return 1;
1451 }
1452 
vmemmap_populate(unsigned long start,unsigned long end,int node,struct vmem_altmap * altmap)1453 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1454 			       struct vmem_altmap *altmap)
1455 {
1456 	/*
1457 	 * Note that SPARSEMEM_VMEMMAP is only selected for rv64 and that we
1458 	 * can't use hugepage mappings for 2-level page table because in case of
1459 	 * memory hotplug, we are not able to update all the page tables with
1460 	 * the new PMDs.
1461 	 */
1462 	return vmemmap_populate_hugepages(start, end, node, altmap);
1463 }
1464 #endif
1465 
1466 #if defined(CONFIG_MMU) && defined(CONFIG_64BIT)
1467 /*
1468  * Pre-allocates page-table pages for a specific area in the kernel
1469  * page-table. Only the level which needs to be synchronized between
1470  * all page-tables is allocated because the synchronization can be
1471  * expensive.
1472  */
preallocate_pgd_pages_range(unsigned long start,unsigned long end,const char * area)1473 static void __init preallocate_pgd_pages_range(unsigned long start, unsigned long end,
1474 					       const char *area)
1475 {
1476 	unsigned long addr;
1477 	const char *lvl;
1478 
1479 	for (addr = start; addr < end && addr >= start; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1480 		pgd_t *pgd = pgd_offset_k(addr);
1481 		p4d_t *p4d;
1482 		pud_t *pud;
1483 		pmd_t *pmd;
1484 
1485 		lvl = "p4d";
1486 		p4d = p4d_alloc(&init_mm, pgd, addr);
1487 		if (!p4d)
1488 			goto failed;
1489 
1490 		if (pgtable_l5_enabled)
1491 			continue;
1492 
1493 		lvl = "pud";
1494 		pud = pud_alloc(&init_mm, p4d, addr);
1495 		if (!pud)
1496 			goto failed;
1497 
1498 		if (pgtable_l4_enabled)
1499 			continue;
1500 
1501 		lvl = "pmd";
1502 		pmd = pmd_alloc(&init_mm, pud, addr);
1503 		if (!pmd)
1504 			goto failed;
1505 	}
1506 	return;
1507 
1508 failed:
1509 	/*
1510 	 * The pages have to be there now or they will be missing in
1511 	 * process page-tables later.
1512 	 */
1513 	panic("Failed to pre-allocate %s pages for %s area\n", lvl, area);
1514 }
1515 
1516 #define PAGE_END KASAN_SHADOW_START
1517 
pgtable_cache_init(void)1518 void __init pgtable_cache_init(void)
1519 {
1520 	preallocate_pgd_pages_range(VMALLOC_START, VMALLOC_END, "vmalloc");
1521 	if (IS_ENABLED(CONFIG_MODULES))
1522 		preallocate_pgd_pages_range(MODULES_VADDR, MODULES_END, "bpf/modules");
1523 	if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) {
1524 		preallocate_pgd_pages_range(VMEMMAP_START, VMEMMAP_END, "vmemmap");
1525 		preallocate_pgd_pages_range(PAGE_OFFSET, PAGE_END, "direct map");
1526 		if (IS_ENABLED(CONFIG_KASAN))
1527 			preallocate_pgd_pages_range(KASAN_SHADOW_START, KASAN_SHADOW_END, "kasan");
1528 	}
1529 }
1530 #endif
1531 
1532 #ifdef CONFIG_EXECMEM
1533 #ifdef CONFIG_MMU
1534 static struct execmem_info execmem_info __ro_after_init;
1535 
execmem_arch_setup(void)1536 struct execmem_info __init *execmem_arch_setup(void)
1537 {
1538 	execmem_info = (struct execmem_info){
1539 		.ranges = {
1540 			[EXECMEM_DEFAULT] = {
1541 				.start	= MODULES_VADDR,
1542 				.end	= MODULES_END,
1543 				.pgprot	= PAGE_KERNEL,
1544 				.alignment = 1,
1545 			},
1546 			[EXECMEM_KPROBES] = {
1547 				.start	= VMALLOC_START,
1548 				.end	= VMALLOC_END,
1549 				.pgprot	= PAGE_KERNEL_READ_EXEC,
1550 				.alignment = 1,
1551 			},
1552 			[EXECMEM_BPF] = {
1553 				.start	= BPF_JIT_REGION_START,
1554 				.end	= BPF_JIT_REGION_END,
1555 				.pgprot	= PAGE_KERNEL,
1556 				.alignment = PAGE_SIZE,
1557 			},
1558 		},
1559 	};
1560 
1561 	return &execmem_info;
1562 }
1563 #endif /* CONFIG_MMU */
1564 #endif /* CONFIG_EXECMEM */
1565 
1566 #ifdef CONFIG_MEMORY_HOTPLUG
free_pte_table(pte_t * pte_start,pmd_t * pmd)1567 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
1568 {
1569 	struct page *page = pmd_page(*pmd);
1570 	struct ptdesc *ptdesc = page_ptdesc(page);
1571 	pte_t *pte;
1572 	int i;
1573 
1574 	for (i = 0; i < PTRS_PER_PTE; i++) {
1575 		pte = pte_start + i;
1576 		if (!pte_none(*pte))
1577 			return;
1578 	}
1579 
1580 	pagetable_dtor(ptdesc);
1581 	if (PageReserved(page))
1582 		free_reserved_page(page);
1583 	else
1584 		pagetable_free(ptdesc);
1585 	pmd_clear(pmd);
1586 }
1587 
free_pmd_table(pmd_t * pmd_start,pud_t * pud,bool is_vmemmap)1588 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud, bool is_vmemmap)
1589 {
1590 	struct page *page = pud_page(*pud);
1591 	struct ptdesc *ptdesc = page_ptdesc(page);
1592 	pmd_t *pmd;
1593 	int i;
1594 
1595 	for (i = 0; i < PTRS_PER_PMD; i++) {
1596 		pmd = pmd_start + i;
1597 		if (!pmd_none(*pmd))
1598 			return;
1599 	}
1600 
1601 	if (!is_vmemmap)
1602 		pagetable_dtor(ptdesc);
1603 	if (PageReserved(page))
1604 		free_reserved_page(page);
1605 	else
1606 		pagetable_free(ptdesc);
1607 	pud_clear(pud);
1608 }
1609 
free_pud_table(pud_t * pud_start,p4d_t * p4d)1610 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d)
1611 {
1612 	struct page *page = p4d_page(*p4d);
1613 	pud_t *pud;
1614 	int i;
1615 
1616 	for (i = 0; i < PTRS_PER_PUD; i++) {
1617 		pud = pud_start + i;
1618 		if (!pud_none(*pud))
1619 			return;
1620 	}
1621 
1622 	if (PageReserved(page))
1623 		free_reserved_page(page);
1624 	else
1625 		free_pages((unsigned long)page_address(page), 0);
1626 	p4d_clear(p4d);
1627 }
1628 
free_vmemmap_storage(struct page * page,size_t size,struct vmem_altmap * altmap)1629 static void __meminit free_vmemmap_storage(struct page *page, size_t size,
1630 					   struct vmem_altmap *altmap)
1631 {
1632 	int order = get_order(size);
1633 
1634 	if (altmap) {
1635 		vmem_altmap_free(altmap, size >> PAGE_SHIFT);
1636 		return;
1637 	}
1638 
1639 	if (PageReserved(page)) {
1640 		unsigned int nr_pages = 1 << order;
1641 
1642 		while (nr_pages--)
1643 			free_reserved_page(page++);
1644 		return;
1645 	}
1646 
1647 	free_pages((unsigned long)page_address(page), order);
1648 }
1649 
remove_pte_mapping(pte_t * pte_base,unsigned long addr,unsigned long end,bool is_vmemmap,struct vmem_altmap * altmap)1650 static void __meminit remove_pte_mapping(pte_t *pte_base, unsigned long addr, unsigned long end,
1651 					 bool is_vmemmap, struct vmem_altmap *altmap)
1652 {
1653 	unsigned long next;
1654 	pte_t *ptep, pte;
1655 
1656 	for (; addr < end; addr = next) {
1657 		next = (addr + PAGE_SIZE) & PAGE_MASK;
1658 		if (next > end)
1659 			next = end;
1660 
1661 		ptep = pte_base + pte_index(addr);
1662 		pte = ptep_get(ptep);
1663 		if (!pte_present(*ptep))
1664 			continue;
1665 
1666 		pte_clear(&init_mm, addr, ptep);
1667 		if (is_vmemmap)
1668 			free_vmemmap_storage(pte_page(pte), PAGE_SIZE, altmap);
1669 	}
1670 }
1671 
remove_pmd_mapping(pmd_t * pmd_base,unsigned long addr,unsigned long end,bool is_vmemmap,struct vmem_altmap * altmap)1672 static void __meminit remove_pmd_mapping(pmd_t *pmd_base, unsigned long addr, unsigned long end,
1673 					 bool is_vmemmap, struct vmem_altmap *altmap)
1674 {
1675 	unsigned long next;
1676 	pte_t *pte_base;
1677 	pmd_t *pmdp, pmd;
1678 
1679 	for (; addr < end; addr = next) {
1680 		next = pmd_addr_end(addr, end);
1681 		pmdp = pmd_base + pmd_index(addr);
1682 		pmd = pmdp_get(pmdp);
1683 		if (!pmd_present(pmd))
1684 			continue;
1685 
1686 		if (pmd_leaf(pmd)) {
1687 			pmd_clear(pmdp);
1688 			if (is_vmemmap)
1689 				free_vmemmap_storage(pmd_page(pmd), PMD_SIZE, altmap);
1690 			continue;
1691 		}
1692 
1693 		pte_base = (pte_t *)pmd_page_vaddr(*pmdp);
1694 		remove_pte_mapping(pte_base, addr, next, is_vmemmap, altmap);
1695 		free_pte_table(pte_base, pmdp);
1696 	}
1697 }
1698 
remove_pud_mapping(pud_t * pud_base,unsigned long addr,unsigned long end,bool is_vmemmap,struct vmem_altmap * altmap)1699 static void __meminit remove_pud_mapping(pud_t *pud_base, unsigned long addr, unsigned long end,
1700 					 bool is_vmemmap, struct vmem_altmap *altmap)
1701 {
1702 	unsigned long next;
1703 	pud_t *pudp, pud;
1704 	pmd_t *pmd_base;
1705 
1706 	for (; addr < end; addr = next) {
1707 		next = pud_addr_end(addr, end);
1708 		pudp = pud_base + pud_index(addr);
1709 		pud = pudp_get(pudp);
1710 		if (!pud_present(pud))
1711 			continue;
1712 
1713 		if (pud_leaf(pud)) {
1714 			if (pgtable_l4_enabled) {
1715 				pud_clear(pudp);
1716 				if (is_vmemmap)
1717 					free_vmemmap_storage(pud_page(pud), PUD_SIZE, altmap);
1718 			}
1719 			continue;
1720 		}
1721 
1722 		pmd_base = pmd_offset(pudp, 0);
1723 		remove_pmd_mapping(pmd_base, addr, next, is_vmemmap, altmap);
1724 
1725 		if (pgtable_l4_enabled)
1726 			free_pmd_table(pmd_base, pudp, is_vmemmap);
1727 	}
1728 }
1729 
remove_p4d_mapping(p4d_t * p4d_base,unsigned long addr,unsigned long end,bool is_vmemmap,struct vmem_altmap * altmap)1730 static void __meminit remove_p4d_mapping(p4d_t *p4d_base, unsigned long addr, unsigned long end,
1731 					 bool is_vmemmap, struct vmem_altmap *altmap)
1732 {
1733 	unsigned long next;
1734 	p4d_t *p4dp, p4d;
1735 	pud_t *pud_base;
1736 
1737 	for (; addr < end; addr = next) {
1738 		next = p4d_addr_end(addr, end);
1739 		p4dp = p4d_base + p4d_index(addr);
1740 		p4d = p4dp_get(p4dp);
1741 		if (!p4d_present(p4d))
1742 			continue;
1743 
1744 		if (p4d_leaf(p4d)) {
1745 			if (pgtable_l5_enabled) {
1746 				p4d_clear(p4dp);
1747 				if (is_vmemmap)
1748 					free_vmemmap_storage(p4d_page(p4d), P4D_SIZE, altmap);
1749 			}
1750 			continue;
1751 		}
1752 
1753 		pud_base = pud_offset(p4dp, 0);
1754 		remove_pud_mapping(pud_base, addr, next, is_vmemmap, altmap);
1755 
1756 		if (pgtable_l5_enabled)
1757 			free_pud_table(pud_base, p4dp);
1758 	}
1759 }
1760 
remove_pgd_mapping(unsigned long va,unsigned long end,bool is_vmemmap,struct vmem_altmap * altmap)1761 static void __meminit remove_pgd_mapping(unsigned long va, unsigned long end, bool is_vmemmap,
1762 					 struct vmem_altmap *altmap)
1763 {
1764 	unsigned long addr, next;
1765 	p4d_t *p4d_base;
1766 	pgd_t *pgd;
1767 
1768 	for (addr = va; addr < end; addr = next) {
1769 		next = pgd_addr_end(addr, end);
1770 		pgd = pgd_offset_k(addr);
1771 
1772 		if (!pgd_present(*pgd))
1773 			continue;
1774 
1775 		if (pgd_leaf(*pgd))
1776 			continue;
1777 
1778 		p4d_base = p4d_offset(pgd, 0);
1779 		remove_p4d_mapping(p4d_base, addr, next, is_vmemmap, altmap);
1780 	}
1781 
1782 	flush_tlb_all();
1783 }
1784 
remove_linear_mapping(phys_addr_t start,u64 size)1785 static void __meminit remove_linear_mapping(phys_addr_t start, u64 size)
1786 {
1787 	unsigned long va = (unsigned long)__va(start);
1788 	unsigned long end = (unsigned long)__va(start + size);
1789 
1790 	remove_pgd_mapping(va, end, false, NULL);
1791 }
1792 
arch_get_mappable_range(void)1793 struct range arch_get_mappable_range(void)
1794 {
1795 	struct range mhp_range;
1796 
1797 	mhp_range.start = __pa(PAGE_OFFSET);
1798 	mhp_range.end = __pa(PAGE_END - 1);
1799 	return mhp_range;
1800 }
1801 
arch_add_memory(int nid,u64 start,u64 size,struct mhp_params * params)1802 int __ref arch_add_memory(int nid, u64 start, u64 size, struct mhp_params *params)
1803 {
1804 	int ret = 0;
1805 
1806 	create_linear_mapping_range(start, start + size, 0, &params->pgprot);
1807 	ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT, params);
1808 	if (ret) {
1809 		remove_linear_mapping(start, size);
1810 		goto out;
1811 	}
1812 
1813 	max_pfn = PFN_UP(start + size);
1814 	max_low_pfn = max_pfn;
1815 
1816  out:
1817 	flush_tlb_all();
1818 	return ret;
1819 }
1820 
arch_remove_memory(u64 start,u64 size,struct vmem_altmap * altmap)1821 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
1822 {
1823 	__remove_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT, altmap);
1824 	remove_linear_mapping(start, size);
1825 	flush_tlb_all();
1826 }
1827 
vmemmap_free(unsigned long start,unsigned long end,struct vmem_altmap * altmap)1828 void __ref vmemmap_free(unsigned long start, unsigned long end, struct vmem_altmap *altmap)
1829 {
1830 	remove_pgd_mapping(start, end, true, altmap);
1831 }
1832 #endif /* CONFIG_MEMORY_HOTPLUG */
1833