xref: /openbmc/linux/arch/arm64/mm/mmu.c (revision 11b4fa8b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Based on arch/arm/mm/mmu.c
4  *
5  * Copyright (C) 1995-2005 Russell King
6  * Copyright (C) 2012 ARM Ltd.
7  */
8 
9 #include <linux/cache.h>
10 #include <linux/export.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/init.h>
14 #include <linux/ioport.h>
15 #include <linux/kexec.h>
16 #include <linux/libfdt.h>
17 #include <linux/mman.h>
18 #include <linux/nodemask.h>
19 #include <linux/memblock.h>
20 #include <linux/memremap.h>
21 #include <linux/memory.h>
22 #include <linux/fs.h>
23 #include <linux/io.h>
24 #include <linux/mm.h>
25 #include <linux/vmalloc.h>
26 #include <linux/set_memory.h>
27 #include <linux/kfence.h>
28 
29 #include <asm/barrier.h>
30 #include <asm/cputype.h>
31 #include <asm/fixmap.h>
32 #include <asm/kasan.h>
33 #include <asm/kernel-pgtable.h>
34 #include <asm/sections.h>
35 #include <asm/setup.h>
36 #include <linux/sizes.h>
37 #include <asm/tlb.h>
38 #include <asm/mmu_context.h>
39 #include <asm/ptdump.h>
40 #include <asm/tlbflush.h>
41 #include <asm/pgalloc.h>
42 #include <asm/kfence.h>
43 
44 #define NO_BLOCK_MAPPINGS	BIT(0)
45 #define NO_CONT_MAPPINGS	BIT(1)
46 #define NO_EXEC_MAPPINGS	BIT(2)	/* assumes FEAT_HPDS is not used */
47 
48 int idmap_t0sz __ro_after_init;
49 
50 #if VA_BITS > 48
51 u64 vabits_actual __ro_after_init = VA_BITS_MIN;
52 EXPORT_SYMBOL(vabits_actual);
53 #endif
54 
55 u64 kimage_vaddr __ro_after_init = (u64)&_text;
56 EXPORT_SYMBOL(kimage_vaddr);
57 
58 u64 kimage_voffset __ro_after_init;
59 EXPORT_SYMBOL(kimage_voffset);
60 
61 u32 __boot_cpu_mode[] = { BOOT_CPU_MODE_EL2, BOOT_CPU_MODE_EL1 };
62 
63 /*
64  * The booting CPU updates the failed status @__early_cpu_boot_status,
65  * with MMU turned off.
66  */
67 long __section(".mmuoff.data.write") __early_cpu_boot_status;
68 
69 /*
70  * Empty_zero_page is a special page that is used for zero-initialized data
71  * and COW.
72  */
73 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
74 EXPORT_SYMBOL(empty_zero_page);
75 
76 static DEFINE_SPINLOCK(swapper_pgdir_lock);
77 static DEFINE_MUTEX(fixmap_lock);
78 
set_swapper_pgd(pgd_t * pgdp,pgd_t pgd)79 void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
80 {
81 	pgd_t *fixmap_pgdp;
82 
83 	spin_lock(&swapper_pgdir_lock);
84 	fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp));
85 	WRITE_ONCE(*fixmap_pgdp, pgd);
86 	/*
87 	 * We need dsb(ishst) here to ensure the page-table-walker sees
88 	 * our new entry before set_p?d() returns. The fixmap's
89 	 * flush_tlb_kernel_range() via clear_fixmap() does this for us.
90 	 */
91 	pgd_clear_fixmap();
92 	spin_unlock(&swapper_pgdir_lock);
93 }
94 
phys_mem_access_prot(struct file * file,unsigned long pfn,unsigned long size,pgprot_t vma_prot)95 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
96 			      unsigned long size, pgprot_t vma_prot)
97 {
98 	if (!pfn_is_map_memory(pfn))
99 		return pgprot_noncached(vma_prot);
100 	else if (file->f_flags & O_SYNC)
101 		return pgprot_writecombine(vma_prot);
102 	return vma_prot;
103 }
104 EXPORT_SYMBOL(phys_mem_access_prot);
105 
early_pgtable_alloc(int shift)106 static phys_addr_t __init early_pgtable_alloc(int shift)
107 {
108 	phys_addr_t phys;
109 	void *ptr;
110 
111 	phys = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0,
112 					 MEMBLOCK_ALLOC_NOLEAKTRACE);
113 	if (!phys)
114 		panic("Failed to allocate page table page\n");
115 
116 	/*
117 	 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
118 	 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
119 	 * any level of table.
120 	 */
121 	ptr = pte_set_fixmap(phys);
122 
123 	memset(ptr, 0, PAGE_SIZE);
124 
125 	/*
126 	 * Implicit barriers also ensure the zeroed page is visible to the page
127 	 * table walker
128 	 */
129 	pte_clear_fixmap();
130 
131 	return phys;
132 }
133 
pgattr_change_is_safe(u64 old,u64 new)134 bool pgattr_change_is_safe(u64 old, u64 new)
135 {
136 	/*
137 	 * The following mapping attributes may be updated in live
138 	 * kernel mappings without the need for break-before-make.
139 	 */
140 	pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG;
141 
142 	/* creating or taking down mappings is always safe */
143 	if (!pte_valid(__pte(old)) || !pte_valid(__pte(new)))
144 		return true;
145 
146 	/* A live entry's pfn should not change */
147 	if (pte_pfn(__pte(old)) != pte_pfn(__pte(new)))
148 		return false;
149 
150 	/* live contiguous mappings may not be manipulated at all */
151 	if ((old | new) & PTE_CONT)
152 		return false;
153 
154 	/* Transitioning from Non-Global to Global is unsafe */
155 	if (old & ~new & PTE_NG)
156 		return false;
157 
158 	/*
159 	 * Changing the memory type between Normal and Normal-Tagged is safe
160 	 * since Tagged is considered a permission attribute from the
161 	 * mismatched attribute aliases perspective.
162 	 */
163 	if (((old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
164 	     (old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)) &&
165 	    ((new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
166 	     (new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)))
167 		mask |= PTE_ATTRINDX_MASK;
168 
169 	return ((old ^ new) & ~mask) == 0;
170 }
171 
init_pte(pmd_t * pmdp,unsigned long addr,unsigned long end,phys_addr_t phys,pgprot_t prot)172 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
173 		     phys_addr_t phys, pgprot_t prot)
174 {
175 	pte_t *ptep;
176 
177 	ptep = pte_set_fixmap_offset(pmdp, addr);
178 	do {
179 		pte_t old_pte = READ_ONCE(*ptep);
180 
181 		set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
182 
183 		/*
184 		 * After the PTE entry has been populated once, we
185 		 * only allow updates to the permission attributes.
186 		 */
187 		BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
188 					      READ_ONCE(pte_val(*ptep))));
189 
190 		phys += PAGE_SIZE;
191 	} while (ptep++, addr += PAGE_SIZE, addr != end);
192 
193 	pte_clear_fixmap();
194 }
195 
alloc_init_cont_pte(pmd_t * pmdp,unsigned long addr,unsigned long end,phys_addr_t phys,pgprot_t prot,phys_addr_t (* pgtable_alloc)(int),int flags)196 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
197 				unsigned long end, phys_addr_t phys,
198 				pgprot_t prot,
199 				phys_addr_t (*pgtable_alloc)(int),
200 				int flags)
201 {
202 	unsigned long next;
203 	pmd_t pmd = READ_ONCE(*pmdp);
204 
205 	BUG_ON(pmd_sect(pmd));
206 	if (pmd_none(pmd)) {
207 		pmdval_t pmdval = PMD_TYPE_TABLE | PMD_TABLE_UXN;
208 		phys_addr_t pte_phys;
209 
210 		if (flags & NO_EXEC_MAPPINGS)
211 			pmdval |= PMD_TABLE_PXN;
212 		BUG_ON(!pgtable_alloc);
213 		pte_phys = pgtable_alloc(PAGE_SHIFT);
214 		__pmd_populate(pmdp, pte_phys, pmdval);
215 		pmd = READ_ONCE(*pmdp);
216 	}
217 	BUG_ON(pmd_bad(pmd));
218 
219 	do {
220 		pgprot_t __prot = prot;
221 
222 		next = pte_cont_addr_end(addr, end);
223 
224 		/* use a contiguous mapping if the range is suitably aligned */
225 		if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
226 		    (flags & NO_CONT_MAPPINGS) == 0)
227 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
228 
229 		init_pte(pmdp, addr, next, phys, __prot);
230 
231 		phys += next - addr;
232 	} while (addr = next, addr != end);
233 }
234 
init_pmd(pud_t * pudp,unsigned long addr,unsigned long end,phys_addr_t phys,pgprot_t prot,phys_addr_t (* pgtable_alloc)(int),int flags)235 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
236 		     phys_addr_t phys, pgprot_t prot,
237 		     phys_addr_t (*pgtable_alloc)(int), int flags)
238 {
239 	unsigned long next;
240 	pmd_t *pmdp;
241 
242 	pmdp = pmd_set_fixmap_offset(pudp, addr);
243 	do {
244 		pmd_t old_pmd = READ_ONCE(*pmdp);
245 
246 		next = pmd_addr_end(addr, end);
247 
248 		/* try section mapping first */
249 		if (((addr | next | phys) & ~PMD_MASK) == 0 &&
250 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
251 			pmd_set_huge(pmdp, phys, prot);
252 
253 			/*
254 			 * After the PMD entry has been populated once, we
255 			 * only allow updates to the permission attributes.
256 			 */
257 			BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
258 						      READ_ONCE(pmd_val(*pmdp))));
259 		} else {
260 			alloc_init_cont_pte(pmdp, addr, next, phys, prot,
261 					    pgtable_alloc, flags);
262 
263 			BUG_ON(pmd_val(old_pmd) != 0 &&
264 			       pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
265 		}
266 		phys += next - addr;
267 	} while (pmdp++, addr = next, addr != end);
268 
269 	pmd_clear_fixmap();
270 }
271 
alloc_init_cont_pmd(pud_t * pudp,unsigned long addr,unsigned long end,phys_addr_t phys,pgprot_t prot,phys_addr_t (* pgtable_alloc)(int),int flags)272 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
273 				unsigned long end, phys_addr_t phys,
274 				pgprot_t prot,
275 				phys_addr_t (*pgtable_alloc)(int), int flags)
276 {
277 	unsigned long next;
278 	pud_t pud = READ_ONCE(*pudp);
279 
280 	/*
281 	 * Check for initial section mappings in the pgd/pud.
282 	 */
283 	BUG_ON(pud_sect(pud));
284 	if (pud_none(pud)) {
285 		pudval_t pudval = PUD_TYPE_TABLE | PUD_TABLE_UXN;
286 		phys_addr_t pmd_phys;
287 
288 		if (flags & NO_EXEC_MAPPINGS)
289 			pudval |= PUD_TABLE_PXN;
290 		BUG_ON(!pgtable_alloc);
291 		pmd_phys = pgtable_alloc(PMD_SHIFT);
292 		__pud_populate(pudp, pmd_phys, pudval);
293 		pud = READ_ONCE(*pudp);
294 	}
295 	BUG_ON(pud_bad(pud));
296 
297 	do {
298 		pgprot_t __prot = prot;
299 
300 		next = pmd_cont_addr_end(addr, end);
301 
302 		/* use a contiguous mapping if the range is suitably aligned */
303 		if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
304 		    (flags & NO_CONT_MAPPINGS) == 0)
305 			__prot = __pgprot(pgprot_val(prot) | PTE_CONT);
306 
307 		init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
308 
309 		phys += next - addr;
310 	} while (addr = next, addr != end);
311 }
312 
alloc_init_pud(pgd_t * pgdp,unsigned long addr,unsigned long end,phys_addr_t phys,pgprot_t prot,phys_addr_t (* pgtable_alloc)(int),int flags)313 static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
314 			   phys_addr_t phys, pgprot_t prot,
315 			   phys_addr_t (*pgtable_alloc)(int),
316 			   int flags)
317 {
318 	unsigned long next;
319 	pud_t *pudp;
320 	p4d_t *p4dp = p4d_offset(pgdp, addr);
321 	p4d_t p4d = READ_ONCE(*p4dp);
322 
323 	if (p4d_none(p4d)) {
324 		p4dval_t p4dval = P4D_TYPE_TABLE | P4D_TABLE_UXN;
325 		phys_addr_t pud_phys;
326 
327 		if (flags & NO_EXEC_MAPPINGS)
328 			p4dval |= P4D_TABLE_PXN;
329 		BUG_ON(!pgtable_alloc);
330 		pud_phys = pgtable_alloc(PUD_SHIFT);
331 		__p4d_populate(p4dp, pud_phys, p4dval);
332 		p4d = READ_ONCE(*p4dp);
333 	}
334 	BUG_ON(p4d_bad(p4d));
335 
336 	pudp = pud_set_fixmap_offset(p4dp, addr);
337 	do {
338 		pud_t old_pud = READ_ONCE(*pudp);
339 
340 		next = pud_addr_end(addr, end);
341 
342 		/*
343 		 * For 4K granule only, attempt to put down a 1GB block
344 		 */
345 		if (pud_sect_supported() &&
346 		   ((addr | next | phys) & ~PUD_MASK) == 0 &&
347 		    (flags & NO_BLOCK_MAPPINGS) == 0) {
348 			pud_set_huge(pudp, phys, prot);
349 
350 			/*
351 			 * After the PUD entry has been populated once, we
352 			 * only allow updates to the permission attributes.
353 			 */
354 			BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
355 						      READ_ONCE(pud_val(*pudp))));
356 		} else {
357 			alloc_init_cont_pmd(pudp, addr, next, phys, prot,
358 					    pgtable_alloc, flags);
359 
360 			BUG_ON(pud_val(old_pud) != 0 &&
361 			       pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
362 		}
363 		phys += next - addr;
364 	} while (pudp++, addr = next, addr != end);
365 
366 	pud_clear_fixmap();
367 }
368 
__create_pgd_mapping_locked(pgd_t * pgdir,phys_addr_t phys,unsigned long virt,phys_addr_t size,pgprot_t prot,phys_addr_t (* pgtable_alloc)(int),int flags)369 static void __create_pgd_mapping_locked(pgd_t *pgdir, phys_addr_t phys,
370 					unsigned long virt, phys_addr_t size,
371 					pgprot_t prot,
372 					phys_addr_t (*pgtable_alloc)(int),
373 					int flags)
374 {
375 	unsigned long addr, end, next;
376 	pgd_t *pgdp = pgd_offset_pgd(pgdir, virt);
377 
378 	/*
379 	 * If the virtual and physical address don't have the same offset
380 	 * within a page, we cannot map the region as the caller expects.
381 	 */
382 	if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
383 		return;
384 
385 	phys &= PAGE_MASK;
386 	addr = virt & PAGE_MASK;
387 	end = PAGE_ALIGN(virt + size);
388 
389 	do {
390 		next = pgd_addr_end(addr, end);
391 		alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
392 			       flags);
393 		phys += next - addr;
394 	} while (pgdp++, addr = next, addr != end);
395 }
396 
__create_pgd_mapping(pgd_t * pgdir,phys_addr_t phys,unsigned long virt,phys_addr_t size,pgprot_t prot,phys_addr_t (* pgtable_alloc)(int),int flags)397 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
398 				 unsigned long virt, phys_addr_t size,
399 				 pgprot_t prot,
400 				 phys_addr_t (*pgtable_alloc)(int),
401 				 int flags)
402 {
403 	mutex_lock(&fixmap_lock);
404 	__create_pgd_mapping_locked(pgdir, phys, virt, size, prot,
405 				    pgtable_alloc, flags);
406 	mutex_unlock(&fixmap_lock);
407 }
408 
409 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
410 extern __alias(__create_pgd_mapping_locked)
411 void create_kpti_ng_temp_pgd(pgd_t *pgdir, phys_addr_t phys, unsigned long virt,
412 			     phys_addr_t size, pgprot_t prot,
413 			     phys_addr_t (*pgtable_alloc)(int), int flags);
414 #endif
415 
__pgd_pgtable_alloc(int shift)416 static phys_addr_t __pgd_pgtable_alloc(int shift)
417 {
418 	void *ptr = (void *)__get_free_page(GFP_PGTABLE_KERNEL);
419 	BUG_ON(!ptr);
420 
421 	/* Ensure the zeroed page is visible to the page table walker */
422 	dsb(ishst);
423 	return __pa(ptr);
424 }
425 
pgd_pgtable_alloc(int shift)426 static phys_addr_t pgd_pgtable_alloc(int shift)
427 {
428 	phys_addr_t pa = __pgd_pgtable_alloc(shift);
429 	struct ptdesc *ptdesc = page_ptdesc(phys_to_page(pa));
430 
431 	/*
432 	 * Call proper page table ctor in case later we need to
433 	 * call core mm functions like apply_to_page_range() on
434 	 * this pre-allocated page table.
435 	 *
436 	 * We don't select ARCH_ENABLE_SPLIT_PMD_PTLOCK if pmd is
437 	 * folded, and if so pagetable_pte_ctor() becomes nop.
438 	 */
439 	if (shift == PAGE_SHIFT)
440 		BUG_ON(!pagetable_pte_ctor(ptdesc));
441 	else if (shift == PMD_SHIFT)
442 		BUG_ON(!pagetable_pmd_ctor(ptdesc));
443 
444 	return pa;
445 }
446 
447 /*
448  * This function can only be used to modify existing table entries,
449  * without allocating new levels of table. Note that this permits the
450  * creation of new section or page entries.
451  */
create_mapping_noalloc(phys_addr_t phys,unsigned long virt,phys_addr_t size,pgprot_t prot)452 void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
453 				   phys_addr_t size, pgprot_t prot)
454 {
455 	if (virt < PAGE_OFFSET) {
456 		pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
457 			&phys, virt);
458 		return;
459 	}
460 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
461 			     NO_CONT_MAPPINGS);
462 }
463 
create_pgd_mapping(struct mm_struct * mm,phys_addr_t phys,unsigned long virt,phys_addr_t size,pgprot_t prot,bool page_mappings_only)464 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
465 			       unsigned long virt, phys_addr_t size,
466 			       pgprot_t prot, bool page_mappings_only)
467 {
468 	int flags = 0;
469 
470 	BUG_ON(mm == &init_mm);
471 
472 	if (page_mappings_only)
473 		flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
474 
475 	__create_pgd_mapping(mm->pgd, phys, virt, size, prot,
476 			     pgd_pgtable_alloc, flags);
477 }
478 
update_mapping_prot(phys_addr_t phys,unsigned long virt,phys_addr_t size,pgprot_t prot)479 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
480 				phys_addr_t size, pgprot_t prot)
481 {
482 	if (virt < PAGE_OFFSET) {
483 		pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
484 			&phys, virt);
485 		return;
486 	}
487 
488 	__create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
489 			     NO_CONT_MAPPINGS);
490 
491 	/* flush the TLBs after updating live kernel mappings */
492 	flush_tlb_kernel_range(virt, virt + size);
493 }
494 
__map_memblock(pgd_t * pgdp,phys_addr_t start,phys_addr_t end,pgprot_t prot,int flags)495 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
496 				  phys_addr_t end, pgprot_t prot, int flags)
497 {
498 	__create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
499 			     prot, early_pgtable_alloc, flags);
500 }
501 
mark_linear_text_alias_ro(void)502 void __init mark_linear_text_alias_ro(void)
503 {
504 	/*
505 	 * Remove the write permissions from the linear alias of .text/.rodata
506 	 */
507 	update_mapping_prot(__pa_symbol(_stext), (unsigned long)lm_alias(_stext),
508 			    (unsigned long)__init_begin - (unsigned long)_stext,
509 			    PAGE_KERNEL_RO);
510 }
511 
512 #ifdef CONFIG_KFENCE
513 
514 bool __ro_after_init kfence_early_init = !!CONFIG_KFENCE_SAMPLE_INTERVAL;
515 
516 /* early_param() will be parsed before map_mem() below. */
parse_kfence_early_init(char * arg)517 static int __init parse_kfence_early_init(char *arg)
518 {
519 	int val;
520 
521 	if (get_option(&arg, &val))
522 		kfence_early_init = !!val;
523 	return 0;
524 }
525 early_param("kfence.sample_interval", parse_kfence_early_init);
526 
arm64_kfence_alloc_pool(void)527 static phys_addr_t __init arm64_kfence_alloc_pool(void)
528 {
529 	phys_addr_t kfence_pool;
530 
531 	if (!kfence_early_init)
532 		return 0;
533 
534 	kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
535 	if (!kfence_pool) {
536 		pr_err("failed to allocate kfence pool\n");
537 		kfence_early_init = false;
538 		return 0;
539 	}
540 
541 	/* Temporarily mark as NOMAP. */
542 	memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
543 
544 	return kfence_pool;
545 }
546 
arm64_kfence_map_pool(phys_addr_t kfence_pool,pgd_t * pgdp)547 static void __init arm64_kfence_map_pool(phys_addr_t kfence_pool, pgd_t *pgdp)
548 {
549 	if (!kfence_pool)
550 		return;
551 
552 	/* KFENCE pool needs page-level mapping. */
553 	__map_memblock(pgdp, kfence_pool, kfence_pool + KFENCE_POOL_SIZE,
554 			pgprot_tagged(PAGE_KERNEL),
555 			NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
556 	memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
557 	__kfence_pool = phys_to_virt(kfence_pool);
558 }
559 #else /* CONFIG_KFENCE */
560 
arm64_kfence_alloc_pool(void)561 static inline phys_addr_t arm64_kfence_alloc_pool(void) { return 0; }
arm64_kfence_map_pool(phys_addr_t kfence_pool,pgd_t * pgdp)562 static inline void arm64_kfence_map_pool(phys_addr_t kfence_pool, pgd_t *pgdp) { }
563 
564 #endif /* CONFIG_KFENCE */
565 
map_mem(pgd_t * pgdp)566 static void __init map_mem(pgd_t *pgdp)
567 {
568 	static const u64 direct_map_end = _PAGE_END(VA_BITS_MIN);
569 	phys_addr_t kernel_start = __pa_symbol(_stext);
570 	phys_addr_t kernel_end = __pa_symbol(__init_begin);
571 	phys_addr_t start, end;
572 	phys_addr_t early_kfence_pool;
573 	int flags = NO_EXEC_MAPPINGS;
574 	u64 i;
575 
576 	/*
577 	 * Setting hierarchical PXNTable attributes on table entries covering
578 	 * the linear region is only possible if it is guaranteed that no table
579 	 * entries at any level are being shared between the linear region and
580 	 * the vmalloc region. Check whether this is true for the PGD level, in
581 	 * which case it is guaranteed to be true for all other levels as well.
582 	 */
583 	BUILD_BUG_ON(pgd_index(direct_map_end - 1) == pgd_index(direct_map_end));
584 
585 	early_kfence_pool = arm64_kfence_alloc_pool();
586 
587 	if (can_set_direct_map())
588 		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
589 
590 	/*
591 	 * Take care not to create a writable alias for the
592 	 * read-only text and rodata sections of the kernel image.
593 	 * So temporarily mark them as NOMAP to skip mappings in
594 	 * the following for-loop
595 	 */
596 	memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
597 
598 	/* map all the memory banks */
599 	for_each_mem_range(i, &start, &end) {
600 		if (start >= end)
601 			break;
602 		/*
603 		 * The linear map must allow allocation tags reading/writing
604 		 * if MTE is present. Otherwise, it has the same attributes as
605 		 * PAGE_KERNEL.
606 		 */
607 		__map_memblock(pgdp, start, end, pgprot_tagged(PAGE_KERNEL),
608 			       flags);
609 	}
610 
611 	/*
612 	 * Map the linear alias of the [_stext, __init_begin) interval
613 	 * as non-executable now, and remove the write permission in
614 	 * mark_linear_text_alias_ro() below (which will be called after
615 	 * alternative patching has completed). This makes the contents
616 	 * of the region accessible to subsystems such as hibernate,
617 	 * but protects it from inadvertent modification or execution.
618 	 * Note that contiguous mappings cannot be remapped in this way,
619 	 * so we should avoid them here.
620 	 */
621 	__map_memblock(pgdp, kernel_start, kernel_end,
622 		       PAGE_KERNEL, NO_CONT_MAPPINGS);
623 	memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
624 	arm64_kfence_map_pool(early_kfence_pool, pgdp);
625 }
626 
mark_rodata_ro(void)627 void mark_rodata_ro(void)
628 {
629 	unsigned long section_size;
630 
631 	/*
632 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
633 	 * to cover NOTES and EXCEPTION_TABLE.
634 	 */
635 	section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
636 	update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
637 			    section_size, PAGE_KERNEL_RO);
638 
639 	debug_checkwx();
640 }
641 
map_kernel_segment(pgd_t * pgdp,void * va_start,void * va_end,pgprot_t prot,struct vm_struct * vma,int flags,unsigned long vm_flags)642 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
643 				      pgprot_t prot, struct vm_struct *vma,
644 				      int flags, unsigned long vm_flags)
645 {
646 	phys_addr_t pa_start = __pa_symbol(va_start);
647 	unsigned long size = va_end - va_start;
648 
649 	BUG_ON(!PAGE_ALIGNED(pa_start));
650 	BUG_ON(!PAGE_ALIGNED(size));
651 
652 	__create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
653 			     early_pgtable_alloc, flags);
654 
655 	if (!(vm_flags & VM_NO_GUARD))
656 		size += PAGE_SIZE;
657 
658 	vma->addr	= va_start;
659 	vma->phys_addr	= pa_start;
660 	vma->size	= size;
661 	vma->flags	= VM_MAP | vm_flags;
662 	vma->caller	= __builtin_return_address(0);
663 
664 	vm_area_add_early(vma);
665 }
666 
kernel_exec_prot(void)667 static pgprot_t kernel_exec_prot(void)
668 {
669 	return rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
670 }
671 
672 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
map_entry_trampoline(void)673 static int __init map_entry_trampoline(void)
674 {
675 	int i;
676 
677 	pgprot_t prot = kernel_exec_prot();
678 	phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
679 
680 	/* The trampoline is always mapped and can therefore be global */
681 	pgprot_val(prot) &= ~PTE_NG;
682 
683 	/* Map only the text into the trampoline page table */
684 	memset(tramp_pg_dir, 0, PGD_SIZE);
685 	__create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS,
686 			     entry_tramp_text_size(), prot,
687 			     __pgd_pgtable_alloc, NO_BLOCK_MAPPINGS);
688 
689 	/* Map both the text and data into the kernel page table */
690 	for (i = 0; i < DIV_ROUND_UP(entry_tramp_text_size(), PAGE_SIZE); i++)
691 		__set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
692 			     pa_start + i * PAGE_SIZE, prot);
693 
694 	if (IS_ENABLED(CONFIG_RELOCATABLE))
695 		__set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
696 			     pa_start + i * PAGE_SIZE, PAGE_KERNEL_RO);
697 
698 	return 0;
699 }
700 core_initcall(map_entry_trampoline);
701 #endif
702 
703 /*
704  * Open coded check for BTI, only for use to determine configuration
705  * for early mappings for before the cpufeature code has run.
706  */
arm64_early_this_cpu_has_bti(void)707 static bool arm64_early_this_cpu_has_bti(void)
708 {
709 	u64 pfr1;
710 
711 	if (!IS_ENABLED(CONFIG_ARM64_BTI_KERNEL))
712 		return false;
713 
714 	pfr1 = __read_sysreg_by_encoding(SYS_ID_AA64PFR1_EL1);
715 	return cpuid_feature_extract_unsigned_field(pfr1,
716 						    ID_AA64PFR1_EL1_BT_SHIFT);
717 }
718 
719 /*
720  * Create fine-grained mappings for the kernel.
721  */
map_kernel(pgd_t * pgdp)722 static void __init map_kernel(pgd_t *pgdp)
723 {
724 	static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
725 				vmlinux_initdata, vmlinux_data;
726 
727 	/*
728 	 * External debuggers may need to write directly to the text
729 	 * mapping to install SW breakpoints. Allow this (only) when
730 	 * explicitly requested with rodata=off.
731 	 */
732 	pgprot_t text_prot = kernel_exec_prot();
733 
734 	/*
735 	 * If we have a CPU that supports BTI and a kernel built for
736 	 * BTI then mark the kernel executable text as guarded pages
737 	 * now so we don't have to rewrite the page tables later.
738 	 */
739 	if (arm64_early_this_cpu_has_bti())
740 		text_prot = __pgprot_modify(text_prot, PTE_GP, PTE_GP);
741 
742 	/*
743 	 * Only rodata will be remapped with different permissions later on,
744 	 * all other segments are allowed to use contiguous mappings.
745 	 */
746 	map_kernel_segment(pgdp, _stext, _etext, text_prot, &vmlinux_text, 0,
747 			   VM_NO_GUARD);
748 	map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
749 			   &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
750 	map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
751 			   &vmlinux_inittext, 0, VM_NO_GUARD);
752 	map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
753 			   &vmlinux_initdata, 0, VM_NO_GUARD);
754 	map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
755 
756 	fixmap_copy(pgdp);
757 	kasan_copy_shadow(pgdp);
758 }
759 
create_idmap(void)760 static void __init create_idmap(void)
761 {
762 	u64 start = __pa_symbol(__idmap_text_start);
763 	u64 size = __pa_symbol(__idmap_text_end) - start;
764 	pgd_t *pgd = idmap_pg_dir;
765 	u64 pgd_phys;
766 
767 	/* check if we need an additional level of translation */
768 	if (VA_BITS < 48 && idmap_t0sz < (64 - VA_BITS_MIN)) {
769 		pgd_phys = early_pgtable_alloc(PAGE_SHIFT);
770 		set_pgd(&idmap_pg_dir[start >> VA_BITS],
771 			__pgd(pgd_phys | P4D_TYPE_TABLE));
772 		pgd = __va(pgd_phys);
773 	}
774 	__create_pgd_mapping(pgd, start, start, size, PAGE_KERNEL_ROX,
775 			     early_pgtable_alloc, 0);
776 
777 	if (IS_ENABLED(CONFIG_UNMAP_KERNEL_AT_EL0)) {
778 		extern u32 __idmap_kpti_flag;
779 		u64 pa = __pa_symbol(&__idmap_kpti_flag);
780 
781 		/*
782 		 * The KPTI G-to-nG conversion code needs a read-write mapping
783 		 * of its synchronization flag in the ID map.
784 		 */
785 		__create_pgd_mapping(pgd, pa, pa, sizeof(u32), PAGE_KERNEL,
786 				     early_pgtable_alloc, 0);
787 	}
788 }
789 
paging_init(void)790 void __init paging_init(void)
791 {
792 	pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir));
793 	extern pgd_t init_idmap_pg_dir[];
794 
795 	idmap_t0sz = 63UL - __fls(__pa_symbol(_end) | GENMASK(VA_BITS_MIN - 1, 0));
796 
797 	map_kernel(pgdp);
798 	map_mem(pgdp);
799 
800 	pgd_clear_fixmap();
801 
802 	cpu_replace_ttbr1(lm_alias(swapper_pg_dir), init_idmap_pg_dir);
803 	init_mm.pgd = swapper_pg_dir;
804 
805 	memblock_phys_free(__pa_symbol(init_pg_dir),
806 			   __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir));
807 
808 	memblock_allow_resize();
809 
810 	create_idmap();
811 }
812 
813 #ifdef CONFIG_MEMORY_HOTPLUG
free_hotplug_page_range(struct page * page,size_t size,struct vmem_altmap * altmap)814 static void free_hotplug_page_range(struct page *page, size_t size,
815 				    struct vmem_altmap *altmap)
816 {
817 	if (altmap) {
818 		vmem_altmap_free(altmap, size >> PAGE_SHIFT);
819 	} else {
820 		WARN_ON(PageReserved(page));
821 		free_pages((unsigned long)page_address(page), get_order(size));
822 	}
823 }
824 
free_hotplug_pgtable_page(struct page * page)825 static void free_hotplug_pgtable_page(struct page *page)
826 {
827 	free_hotplug_page_range(page, PAGE_SIZE, NULL);
828 }
829 
pgtable_range_aligned(unsigned long start,unsigned long end,unsigned long floor,unsigned long ceiling,unsigned long mask)830 static bool pgtable_range_aligned(unsigned long start, unsigned long end,
831 				  unsigned long floor, unsigned long ceiling,
832 				  unsigned long mask)
833 {
834 	start &= mask;
835 	if (start < floor)
836 		return false;
837 
838 	if (ceiling) {
839 		ceiling &= mask;
840 		if (!ceiling)
841 			return false;
842 	}
843 
844 	if (end - 1 > ceiling - 1)
845 		return false;
846 	return true;
847 }
848 
unmap_hotplug_pte_range(pmd_t * pmdp,unsigned long addr,unsigned long end,bool free_mapped,struct vmem_altmap * altmap)849 static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr,
850 				    unsigned long end, bool free_mapped,
851 				    struct vmem_altmap *altmap)
852 {
853 	pte_t *ptep, pte;
854 
855 	do {
856 		ptep = pte_offset_kernel(pmdp, addr);
857 		pte = READ_ONCE(*ptep);
858 		if (pte_none(pte))
859 			continue;
860 
861 		WARN_ON(!pte_present(pte));
862 		pte_clear(&init_mm, addr, ptep);
863 		flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
864 		if (free_mapped)
865 			free_hotplug_page_range(pte_page(pte),
866 						PAGE_SIZE, altmap);
867 	} while (addr += PAGE_SIZE, addr < end);
868 }
869 
unmap_hotplug_pmd_range(pud_t * pudp,unsigned long addr,unsigned long end,bool free_mapped,struct vmem_altmap * altmap)870 static void unmap_hotplug_pmd_range(pud_t *pudp, unsigned long addr,
871 				    unsigned long end, bool free_mapped,
872 				    struct vmem_altmap *altmap)
873 {
874 	unsigned long next;
875 	pmd_t *pmdp, pmd;
876 
877 	do {
878 		next = pmd_addr_end(addr, end);
879 		pmdp = pmd_offset(pudp, addr);
880 		pmd = READ_ONCE(*pmdp);
881 		if (pmd_none(pmd))
882 			continue;
883 
884 		WARN_ON(!pmd_present(pmd));
885 		if (pmd_sect(pmd)) {
886 			pmd_clear(pmdp);
887 
888 			/*
889 			 * One TLBI should be sufficient here as the PMD_SIZE
890 			 * range is mapped with a single block entry.
891 			 */
892 			flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
893 			if (free_mapped)
894 				free_hotplug_page_range(pmd_page(pmd),
895 							PMD_SIZE, altmap);
896 			continue;
897 		}
898 		WARN_ON(!pmd_table(pmd));
899 		unmap_hotplug_pte_range(pmdp, addr, next, free_mapped, altmap);
900 	} while (addr = next, addr < end);
901 }
902 
unmap_hotplug_pud_range(p4d_t * p4dp,unsigned long addr,unsigned long end,bool free_mapped,struct vmem_altmap * altmap)903 static void unmap_hotplug_pud_range(p4d_t *p4dp, unsigned long addr,
904 				    unsigned long end, bool free_mapped,
905 				    struct vmem_altmap *altmap)
906 {
907 	unsigned long next;
908 	pud_t *pudp, pud;
909 
910 	do {
911 		next = pud_addr_end(addr, end);
912 		pudp = pud_offset(p4dp, addr);
913 		pud = READ_ONCE(*pudp);
914 		if (pud_none(pud))
915 			continue;
916 
917 		WARN_ON(!pud_present(pud));
918 		if (pud_sect(pud)) {
919 			pud_clear(pudp);
920 
921 			/*
922 			 * One TLBI should be sufficient here as the PUD_SIZE
923 			 * range is mapped with a single block entry.
924 			 */
925 			flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
926 			if (free_mapped)
927 				free_hotplug_page_range(pud_page(pud),
928 							PUD_SIZE, altmap);
929 			continue;
930 		}
931 		WARN_ON(!pud_table(pud));
932 		unmap_hotplug_pmd_range(pudp, addr, next, free_mapped, altmap);
933 	} while (addr = next, addr < end);
934 }
935 
unmap_hotplug_p4d_range(pgd_t * pgdp,unsigned long addr,unsigned long end,bool free_mapped,struct vmem_altmap * altmap)936 static void unmap_hotplug_p4d_range(pgd_t *pgdp, unsigned long addr,
937 				    unsigned long end, bool free_mapped,
938 				    struct vmem_altmap *altmap)
939 {
940 	unsigned long next;
941 	p4d_t *p4dp, p4d;
942 
943 	do {
944 		next = p4d_addr_end(addr, end);
945 		p4dp = p4d_offset(pgdp, addr);
946 		p4d = READ_ONCE(*p4dp);
947 		if (p4d_none(p4d))
948 			continue;
949 
950 		WARN_ON(!p4d_present(p4d));
951 		unmap_hotplug_pud_range(p4dp, addr, next, free_mapped, altmap);
952 	} while (addr = next, addr < end);
953 }
954 
unmap_hotplug_range(unsigned long addr,unsigned long end,bool free_mapped,struct vmem_altmap * altmap)955 static void unmap_hotplug_range(unsigned long addr, unsigned long end,
956 				bool free_mapped, struct vmem_altmap *altmap)
957 {
958 	unsigned long next;
959 	pgd_t *pgdp, pgd;
960 
961 	/*
962 	 * altmap can only be used as vmemmap mapping backing memory.
963 	 * In case the backing memory itself is not being freed, then
964 	 * altmap is irrelevant. Warn about this inconsistency when
965 	 * encountered.
966 	 */
967 	WARN_ON(!free_mapped && altmap);
968 
969 	do {
970 		next = pgd_addr_end(addr, end);
971 		pgdp = pgd_offset_k(addr);
972 		pgd = READ_ONCE(*pgdp);
973 		if (pgd_none(pgd))
974 			continue;
975 
976 		WARN_ON(!pgd_present(pgd));
977 		unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped, altmap);
978 	} while (addr = next, addr < end);
979 }
980 
free_empty_pte_table(pmd_t * pmdp,unsigned long addr,unsigned long end,unsigned long floor,unsigned long ceiling)981 static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr,
982 				 unsigned long end, unsigned long floor,
983 				 unsigned long ceiling)
984 {
985 	pte_t *ptep, pte;
986 	unsigned long i, start = addr;
987 
988 	do {
989 		ptep = pte_offset_kernel(pmdp, addr);
990 		pte = READ_ONCE(*ptep);
991 
992 		/*
993 		 * This is just a sanity check here which verifies that
994 		 * pte clearing has been done by earlier unmap loops.
995 		 */
996 		WARN_ON(!pte_none(pte));
997 	} while (addr += PAGE_SIZE, addr < end);
998 
999 	if (!pgtable_range_aligned(start, end, floor, ceiling, PMD_MASK))
1000 		return;
1001 
1002 	/*
1003 	 * Check whether we can free the pte page if the rest of the
1004 	 * entries are empty. Overlap with other regions have been
1005 	 * handled by the floor/ceiling check.
1006 	 */
1007 	ptep = pte_offset_kernel(pmdp, 0UL);
1008 	for (i = 0; i < PTRS_PER_PTE; i++) {
1009 		if (!pte_none(READ_ONCE(ptep[i])))
1010 			return;
1011 	}
1012 
1013 	pmd_clear(pmdp);
1014 	__flush_tlb_kernel_pgtable(start);
1015 	free_hotplug_pgtable_page(virt_to_page(ptep));
1016 }
1017 
free_empty_pmd_table(pud_t * pudp,unsigned long addr,unsigned long end,unsigned long floor,unsigned long ceiling)1018 static void free_empty_pmd_table(pud_t *pudp, unsigned long addr,
1019 				 unsigned long end, unsigned long floor,
1020 				 unsigned long ceiling)
1021 {
1022 	pmd_t *pmdp, pmd;
1023 	unsigned long i, next, start = addr;
1024 
1025 	do {
1026 		next = pmd_addr_end(addr, end);
1027 		pmdp = pmd_offset(pudp, addr);
1028 		pmd = READ_ONCE(*pmdp);
1029 		if (pmd_none(pmd))
1030 			continue;
1031 
1032 		WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd));
1033 		free_empty_pte_table(pmdp, addr, next, floor, ceiling);
1034 	} while (addr = next, addr < end);
1035 
1036 	if (CONFIG_PGTABLE_LEVELS <= 2)
1037 		return;
1038 
1039 	if (!pgtable_range_aligned(start, end, floor, ceiling, PUD_MASK))
1040 		return;
1041 
1042 	/*
1043 	 * Check whether we can free the pmd page if the rest of the
1044 	 * entries are empty. Overlap with other regions have been
1045 	 * handled by the floor/ceiling check.
1046 	 */
1047 	pmdp = pmd_offset(pudp, 0UL);
1048 	for (i = 0; i < PTRS_PER_PMD; i++) {
1049 		if (!pmd_none(READ_ONCE(pmdp[i])))
1050 			return;
1051 	}
1052 
1053 	pud_clear(pudp);
1054 	__flush_tlb_kernel_pgtable(start);
1055 	free_hotplug_pgtable_page(virt_to_page(pmdp));
1056 }
1057 
free_empty_pud_table(p4d_t * p4dp,unsigned long addr,unsigned long end,unsigned long floor,unsigned long ceiling)1058 static void free_empty_pud_table(p4d_t *p4dp, unsigned long addr,
1059 				 unsigned long end, unsigned long floor,
1060 				 unsigned long ceiling)
1061 {
1062 	pud_t *pudp, pud;
1063 	unsigned long i, next, start = addr;
1064 
1065 	do {
1066 		next = pud_addr_end(addr, end);
1067 		pudp = pud_offset(p4dp, addr);
1068 		pud = READ_ONCE(*pudp);
1069 		if (pud_none(pud))
1070 			continue;
1071 
1072 		WARN_ON(!pud_present(pud) || !pud_table(pud) || pud_sect(pud));
1073 		free_empty_pmd_table(pudp, addr, next, floor, ceiling);
1074 	} while (addr = next, addr < end);
1075 
1076 	if (CONFIG_PGTABLE_LEVELS <= 3)
1077 		return;
1078 
1079 	if (!pgtable_range_aligned(start, end, floor, ceiling, PGDIR_MASK))
1080 		return;
1081 
1082 	/*
1083 	 * Check whether we can free the pud page if the rest of the
1084 	 * entries are empty. Overlap with other regions have been
1085 	 * handled by the floor/ceiling check.
1086 	 */
1087 	pudp = pud_offset(p4dp, 0UL);
1088 	for (i = 0; i < PTRS_PER_PUD; i++) {
1089 		if (!pud_none(READ_ONCE(pudp[i])))
1090 			return;
1091 	}
1092 
1093 	p4d_clear(p4dp);
1094 	__flush_tlb_kernel_pgtable(start);
1095 	free_hotplug_pgtable_page(virt_to_page(pudp));
1096 }
1097 
free_empty_p4d_table(pgd_t * pgdp,unsigned long addr,unsigned long end,unsigned long floor,unsigned long ceiling)1098 static void free_empty_p4d_table(pgd_t *pgdp, unsigned long addr,
1099 				 unsigned long end, unsigned long floor,
1100 				 unsigned long ceiling)
1101 {
1102 	unsigned long next;
1103 	p4d_t *p4dp, p4d;
1104 
1105 	do {
1106 		next = p4d_addr_end(addr, end);
1107 		p4dp = p4d_offset(pgdp, addr);
1108 		p4d = READ_ONCE(*p4dp);
1109 		if (p4d_none(p4d))
1110 			continue;
1111 
1112 		WARN_ON(!p4d_present(p4d));
1113 		free_empty_pud_table(p4dp, addr, next, floor, ceiling);
1114 	} while (addr = next, addr < end);
1115 }
1116 
free_empty_tables(unsigned long addr,unsigned long end,unsigned long floor,unsigned long ceiling)1117 static void free_empty_tables(unsigned long addr, unsigned long end,
1118 			      unsigned long floor, unsigned long ceiling)
1119 {
1120 	unsigned long next;
1121 	pgd_t *pgdp, pgd;
1122 
1123 	do {
1124 		next = pgd_addr_end(addr, end);
1125 		pgdp = pgd_offset_k(addr);
1126 		pgd = READ_ONCE(*pgdp);
1127 		if (pgd_none(pgd))
1128 			continue;
1129 
1130 		WARN_ON(!pgd_present(pgd));
1131 		free_empty_p4d_table(pgdp, addr, next, floor, ceiling);
1132 	} while (addr = next, addr < end);
1133 }
1134 #endif
1135 
vmemmap_set_pmd(pmd_t * pmdp,void * p,int node,unsigned long addr,unsigned long next)1136 void __meminit vmemmap_set_pmd(pmd_t *pmdp, void *p, int node,
1137 			       unsigned long addr, unsigned long next)
1138 {
1139 	pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
1140 }
1141 
vmemmap_check_pmd(pmd_t * pmdp,int node,unsigned long addr,unsigned long next)1142 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node,
1143 				unsigned long addr, unsigned long next)
1144 {
1145 	vmemmap_verify((pte_t *)pmdp, node, addr, next);
1146 	return 1;
1147 }
1148 
vmemmap_populate(unsigned long start,unsigned long end,int node,struct vmem_altmap * altmap)1149 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1150 		struct vmem_altmap *altmap)
1151 {
1152 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1153 
1154 	if (!IS_ENABLED(CONFIG_ARM64_4K_PAGES))
1155 		return vmemmap_populate_basepages(start, end, node, altmap);
1156 	else
1157 		return vmemmap_populate_hugepages(start, end, node, altmap);
1158 }
1159 
1160 #ifdef CONFIG_MEMORY_HOTPLUG
vmemmap_free(unsigned long start,unsigned long end,struct vmem_altmap * altmap)1161 void vmemmap_free(unsigned long start, unsigned long end,
1162 		struct vmem_altmap *altmap)
1163 {
1164 	WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1165 
1166 	unmap_hotplug_range(start, end, true, altmap);
1167 	free_empty_tables(start, end, VMEMMAP_START, VMEMMAP_END);
1168 }
1169 #endif /* CONFIG_MEMORY_HOTPLUG */
1170 
pud_set_huge(pud_t * pudp,phys_addr_t phys,pgprot_t prot)1171 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
1172 {
1173 	pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot));
1174 
1175 	/* Only allow permission changes for now */
1176 	if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
1177 				   pud_val(new_pud)))
1178 		return 0;
1179 
1180 	VM_BUG_ON(phys & ~PUD_MASK);
1181 	set_pud(pudp, new_pud);
1182 	return 1;
1183 }
1184 
pmd_set_huge(pmd_t * pmdp,phys_addr_t phys,pgprot_t prot)1185 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
1186 {
1187 	pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), mk_pmd_sect_prot(prot));
1188 
1189 	/* Only allow permission changes for now */
1190 	if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
1191 				   pmd_val(new_pmd)))
1192 		return 0;
1193 
1194 	VM_BUG_ON(phys & ~PMD_MASK);
1195 	set_pmd(pmdp, new_pmd);
1196 	return 1;
1197 }
1198 
pud_clear_huge(pud_t * pudp)1199 int pud_clear_huge(pud_t *pudp)
1200 {
1201 	if (!pud_sect(READ_ONCE(*pudp)))
1202 		return 0;
1203 	pud_clear(pudp);
1204 	return 1;
1205 }
1206 
pmd_clear_huge(pmd_t * pmdp)1207 int pmd_clear_huge(pmd_t *pmdp)
1208 {
1209 	if (!pmd_sect(READ_ONCE(*pmdp)))
1210 		return 0;
1211 	pmd_clear(pmdp);
1212 	return 1;
1213 }
1214 
pmd_free_pte_page(pmd_t * pmdp,unsigned long addr)1215 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
1216 {
1217 	pte_t *table;
1218 	pmd_t pmd;
1219 
1220 	pmd = READ_ONCE(*pmdp);
1221 
1222 	if (!pmd_table(pmd)) {
1223 		VM_WARN_ON(1);
1224 		return 1;
1225 	}
1226 
1227 	table = pte_offset_kernel(pmdp, addr);
1228 	pmd_clear(pmdp);
1229 	__flush_tlb_kernel_pgtable(addr);
1230 	pte_free_kernel(NULL, table);
1231 	return 1;
1232 }
1233 
pud_free_pmd_page(pud_t * pudp,unsigned long addr)1234 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1235 {
1236 	pmd_t *table;
1237 	pmd_t *pmdp;
1238 	pud_t pud;
1239 	unsigned long next, end;
1240 
1241 	pud = READ_ONCE(*pudp);
1242 
1243 	if (!pud_table(pud)) {
1244 		VM_WARN_ON(1);
1245 		return 1;
1246 	}
1247 
1248 	table = pmd_offset(pudp, addr);
1249 	pmdp = table;
1250 	next = addr;
1251 	end = addr + PUD_SIZE;
1252 	do {
1253 		pmd_free_pte_page(pmdp, next);
1254 	} while (pmdp++, next += PMD_SIZE, next != end);
1255 
1256 	pud_clear(pudp);
1257 	__flush_tlb_kernel_pgtable(addr);
1258 	pmd_free(NULL, table);
1259 	return 1;
1260 }
1261 
1262 #ifdef CONFIG_MEMORY_HOTPLUG
__remove_pgd_mapping(pgd_t * pgdir,unsigned long start,u64 size)1263 static void __remove_pgd_mapping(pgd_t *pgdir, unsigned long start, u64 size)
1264 {
1265 	unsigned long end = start + size;
1266 
1267 	WARN_ON(pgdir != init_mm.pgd);
1268 	WARN_ON((start < PAGE_OFFSET) || (end > PAGE_END));
1269 
1270 	unmap_hotplug_range(start, end, false, NULL);
1271 	free_empty_tables(start, end, PAGE_OFFSET, PAGE_END);
1272 }
1273 
arch_get_mappable_range(void)1274 struct range arch_get_mappable_range(void)
1275 {
1276 	struct range mhp_range;
1277 	u64 start_linear_pa = __pa(_PAGE_OFFSET(vabits_actual));
1278 	u64 end_linear_pa = __pa(PAGE_END - 1);
1279 
1280 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
1281 		/*
1282 		 * Check for a wrap, it is possible because of randomized linear
1283 		 * mapping the start physical address is actually bigger than
1284 		 * the end physical address. In this case set start to zero
1285 		 * because [0, end_linear_pa] range must still be able to cover
1286 		 * all addressable physical addresses.
1287 		 */
1288 		if (start_linear_pa > end_linear_pa)
1289 			start_linear_pa = 0;
1290 	}
1291 
1292 	WARN_ON(start_linear_pa > end_linear_pa);
1293 
1294 	/*
1295 	 * Linear mapping region is the range [PAGE_OFFSET..(PAGE_END - 1)]
1296 	 * accommodating both its ends but excluding PAGE_END. Max physical
1297 	 * range which can be mapped inside this linear mapping range, must
1298 	 * also be derived from its end points.
1299 	 */
1300 	mhp_range.start = start_linear_pa;
1301 	mhp_range.end =  end_linear_pa;
1302 
1303 	return mhp_range;
1304 }
1305 
arch_add_memory(int nid,u64 start,u64 size,struct mhp_params * params)1306 int arch_add_memory(int nid, u64 start, u64 size,
1307 		    struct mhp_params *params)
1308 {
1309 	int ret, flags = NO_EXEC_MAPPINGS;
1310 
1311 	VM_BUG_ON(!mhp_range_allowed(start, size, true));
1312 
1313 	if (can_set_direct_map())
1314 		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1315 
1316 	__create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1317 			     size, params->pgprot, __pgd_pgtable_alloc,
1318 			     flags);
1319 
1320 	memblock_clear_nomap(start, size);
1321 
1322 	ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1323 			   params);
1324 	if (ret)
1325 		__remove_pgd_mapping(swapper_pg_dir,
1326 				     __phys_to_virt(start), size);
1327 	else {
1328 		max_pfn = PFN_UP(start + size);
1329 		max_low_pfn = max_pfn;
1330 	}
1331 
1332 	return ret;
1333 }
1334 
arch_remove_memory(u64 start,u64 size,struct vmem_altmap * altmap)1335 void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
1336 {
1337 	unsigned long start_pfn = start >> PAGE_SHIFT;
1338 	unsigned long nr_pages = size >> PAGE_SHIFT;
1339 
1340 	__remove_pages(start_pfn, nr_pages, altmap);
1341 	__remove_pgd_mapping(swapper_pg_dir, __phys_to_virt(start), size);
1342 }
1343 
1344 /*
1345  * This memory hotplug notifier helps prevent boot memory from being
1346  * inadvertently removed as it blocks pfn range offlining process in
1347  * __offline_pages(). Hence this prevents both offlining as well as
1348  * removal process for boot memory which is initially always online.
1349  * In future if and when boot memory could be removed, this notifier
1350  * should be dropped and free_hotplug_page_range() should handle any
1351  * reserved pages allocated during boot.
1352  */
prevent_bootmem_remove_notifier(struct notifier_block * nb,unsigned long action,void * data)1353 static int prevent_bootmem_remove_notifier(struct notifier_block *nb,
1354 					   unsigned long action, void *data)
1355 {
1356 	struct mem_section *ms;
1357 	struct memory_notify *arg = data;
1358 	unsigned long end_pfn = arg->start_pfn + arg->nr_pages;
1359 	unsigned long pfn = arg->start_pfn;
1360 
1361 	if ((action != MEM_GOING_OFFLINE) && (action != MEM_OFFLINE))
1362 		return NOTIFY_OK;
1363 
1364 	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1365 		unsigned long start = PFN_PHYS(pfn);
1366 		unsigned long end = start + (1UL << PA_SECTION_SHIFT);
1367 
1368 		ms = __pfn_to_section(pfn);
1369 		if (!early_section(ms))
1370 			continue;
1371 
1372 		if (action == MEM_GOING_OFFLINE) {
1373 			/*
1374 			 * Boot memory removal is not supported. Prevent
1375 			 * it via blocking any attempted offline request
1376 			 * for the boot memory and just report it.
1377 			 */
1378 			pr_warn("Boot memory [%lx %lx] offlining attempted\n", start, end);
1379 			return NOTIFY_BAD;
1380 		} else if (action == MEM_OFFLINE) {
1381 			/*
1382 			 * This should have never happened. Boot memory
1383 			 * offlining should have been prevented by this
1384 			 * very notifier. Probably some memory removal
1385 			 * procedure might have changed which would then
1386 			 * require further debug.
1387 			 */
1388 			pr_err("Boot memory [%lx %lx] offlined\n", start, end);
1389 
1390 			/*
1391 			 * Core memory hotplug does not process a return
1392 			 * code from the notifier for MEM_OFFLINE events.
1393 			 * The error condition has been reported. Return
1394 			 * from here as if ignored.
1395 			 */
1396 			return NOTIFY_DONE;
1397 		}
1398 	}
1399 	return NOTIFY_OK;
1400 }
1401 
1402 static struct notifier_block prevent_bootmem_remove_nb = {
1403 	.notifier_call = prevent_bootmem_remove_notifier,
1404 };
1405 
1406 /*
1407  * This ensures that boot memory sections on the platform are online
1408  * from early boot. Memory sections could not be prevented from being
1409  * offlined, unless for some reason they are not online to begin with.
1410  * This helps validate the basic assumption on which the above memory
1411  * event notifier works to prevent boot memory section offlining and
1412  * its possible removal.
1413  */
validate_bootmem_online(void)1414 static void validate_bootmem_online(void)
1415 {
1416 	phys_addr_t start, end, addr;
1417 	struct mem_section *ms;
1418 	u64 i;
1419 
1420 	/*
1421 	 * Scanning across all memblock might be expensive
1422 	 * on some big memory systems. Hence enable this
1423 	 * validation only with DEBUG_VM.
1424 	 */
1425 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
1426 		return;
1427 
1428 	for_each_mem_range(i, &start, &end) {
1429 		for (addr = start; addr < end; addr += (1UL << PA_SECTION_SHIFT)) {
1430 			ms = __pfn_to_section(PHYS_PFN(addr));
1431 
1432 			/*
1433 			 * All memory ranges in the system at this point
1434 			 * should have been marked as early sections.
1435 			 */
1436 			WARN_ON(!early_section(ms));
1437 
1438 			/*
1439 			 * Memory notifier mechanism here to prevent boot
1440 			 * memory offlining depends on the fact that each
1441 			 * early section memory on the system is initially
1442 			 * online. Otherwise a given memory section which
1443 			 * is already offline will be overlooked and can
1444 			 * be removed completely. Call out such sections.
1445 			 */
1446 			if (!online_section(ms))
1447 				pr_err("Boot memory [%llx %llx] is offline, can be removed\n",
1448 					addr, addr + (1UL << PA_SECTION_SHIFT));
1449 		}
1450 	}
1451 }
1452 
prevent_bootmem_remove_init(void)1453 static int __init prevent_bootmem_remove_init(void)
1454 {
1455 	int ret = 0;
1456 
1457 	if (!IS_ENABLED(CONFIG_MEMORY_HOTREMOVE))
1458 		return ret;
1459 
1460 	validate_bootmem_online();
1461 	ret = register_memory_notifier(&prevent_bootmem_remove_nb);
1462 	if (ret)
1463 		pr_err("%s: Notifier registration failed %d\n", __func__, ret);
1464 
1465 	return ret;
1466 }
1467 early_initcall(prevent_bootmem_remove_init);
1468 #endif
1469 
ptep_modify_prot_start(struct vm_area_struct * vma,unsigned long addr,pte_t * ptep)1470 pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
1471 {
1472 	if (IS_ENABLED(CONFIG_ARM64_ERRATUM_2645198) &&
1473 	    cpus_have_const_cap(ARM64_WORKAROUND_2645198)) {
1474 		/*
1475 		 * Break-before-make (BBM) is required for all user space mappings
1476 		 * when the permission changes from executable to non-executable
1477 		 * in cases where cpu is affected with errata #2645198.
1478 		 */
1479 		if (pte_user_exec(READ_ONCE(*ptep)))
1480 			return ptep_clear_flush(vma, addr, ptep);
1481 	}
1482 	return ptep_get_and_clear(vma->vm_mm, addr, ptep);
1483 }
1484 
ptep_modify_prot_commit(struct vm_area_struct * vma,unsigned long addr,pte_t * ptep,pte_t old_pte,pte_t pte)1485 void ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep,
1486 			     pte_t old_pte, pte_t pte)
1487 {
1488 	set_pte_at(vma->vm_mm, addr, ptep, pte);
1489 }
1490