1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/mm.h>
3 #include <linux/gfp.h>
4 #include <linux/hugetlb.h>
5 #include <asm/pgalloc.h>
6 #include <asm/tlb.h>
7 #include <asm/fixmap.h>
8 #include <asm/mtrr.h>
9 
10 #ifdef CONFIG_DYNAMIC_PHYSICAL_MASK
11 phys_addr_t physical_mask __ro_after_init = (1ULL << __PHYSICAL_MASK_SHIFT) - 1;
12 EXPORT_SYMBOL(physical_mask);
13 #endif
14 
15 #ifdef CONFIG_HIGHPTE
16 #define PGTABLE_HIGHMEM __GFP_HIGHMEM
17 #else
18 #define PGTABLE_HIGHMEM 0
19 #endif
20 
21 #ifndef CONFIG_PARAVIRT
22 static inline
paravirt_tlb_remove_table(struct mmu_gather * tlb,void * table)23 void paravirt_tlb_remove_table(struct mmu_gather *tlb, void *table)
24 {
25 	tlb_remove_page(tlb, table);
26 }
27 #endif
28 
29 gfp_t __userpte_alloc_gfp = GFP_PGTABLE_USER | PGTABLE_HIGHMEM;
30 
pte_alloc_one(struct mm_struct * mm)31 pgtable_t pte_alloc_one(struct mm_struct *mm)
32 {
33 	return __pte_alloc_one(mm, __userpte_alloc_gfp);
34 }
35 
setup_userpte(char * arg)36 static int __init setup_userpte(char *arg)
37 {
38 	if (!arg)
39 		return -EINVAL;
40 
41 	/*
42 	 * "userpte=nohigh" disables allocation of user pagetables in
43 	 * high memory.
44 	 */
45 	if (strcmp(arg, "nohigh") == 0)
46 		__userpte_alloc_gfp &= ~__GFP_HIGHMEM;
47 	else
48 		return -EINVAL;
49 	return 0;
50 }
51 early_param("userpte", setup_userpte);
52 
___pte_free_tlb(struct mmu_gather * tlb,struct page * pte)53 void ___pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
54 {
55 	pgtable_pte_page_dtor(pte);
56 	paravirt_release_pte(page_to_pfn(pte));
57 	paravirt_tlb_remove_table(tlb, pte);
58 }
59 
60 #if CONFIG_PGTABLE_LEVELS > 2
___pmd_free_tlb(struct mmu_gather * tlb,pmd_t * pmd)61 void ___pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
62 {
63 	struct page *page = virt_to_page(pmd);
64 	paravirt_release_pmd(__pa(pmd) >> PAGE_SHIFT);
65 	/*
66 	 * NOTE! For PAE, any changes to the top page-directory-pointer-table
67 	 * entries need a full cr3 reload to flush.
68 	 */
69 #ifdef CONFIG_X86_PAE
70 	tlb->need_flush_all = 1;
71 #endif
72 	pgtable_pmd_page_dtor(page);
73 	paravirt_tlb_remove_table(tlb, page);
74 }
75 
76 #if CONFIG_PGTABLE_LEVELS > 3
___pud_free_tlb(struct mmu_gather * tlb,pud_t * pud)77 void ___pud_free_tlb(struct mmu_gather *tlb, pud_t *pud)
78 {
79 	paravirt_release_pud(__pa(pud) >> PAGE_SHIFT);
80 	paravirt_tlb_remove_table(tlb, virt_to_page(pud));
81 }
82 
83 #if CONFIG_PGTABLE_LEVELS > 4
___p4d_free_tlb(struct mmu_gather * tlb,p4d_t * p4d)84 void ___p4d_free_tlb(struct mmu_gather *tlb, p4d_t *p4d)
85 {
86 	paravirt_release_p4d(__pa(p4d) >> PAGE_SHIFT);
87 	paravirt_tlb_remove_table(tlb, virt_to_page(p4d));
88 }
89 #endif	/* CONFIG_PGTABLE_LEVELS > 4 */
90 #endif	/* CONFIG_PGTABLE_LEVELS > 3 */
91 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
92 
pgd_list_add(pgd_t * pgd)93 static inline void pgd_list_add(pgd_t *pgd)
94 {
95 	struct page *page = virt_to_page(pgd);
96 
97 	list_add(&page->lru, &pgd_list);
98 }
99 
pgd_list_del(pgd_t * pgd)100 static inline void pgd_list_del(pgd_t *pgd)
101 {
102 	struct page *page = virt_to_page(pgd);
103 
104 	list_del(&page->lru);
105 }
106 
107 #define UNSHARED_PTRS_PER_PGD				\
108 	(SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD)
109 #define MAX_UNSHARED_PTRS_PER_PGD			\
110 	max_t(size_t, KERNEL_PGD_BOUNDARY, PTRS_PER_PGD)
111 
112 
pgd_set_mm(pgd_t * pgd,struct mm_struct * mm)113 static void pgd_set_mm(pgd_t *pgd, struct mm_struct *mm)
114 {
115 	virt_to_page(pgd)->pt_mm = mm;
116 }
117 
pgd_page_get_mm(struct page * page)118 struct mm_struct *pgd_page_get_mm(struct page *page)
119 {
120 	return page->pt_mm;
121 }
122 
pgd_ctor(struct mm_struct * mm,pgd_t * pgd)123 static void pgd_ctor(struct mm_struct *mm, pgd_t *pgd)
124 {
125 	/* If the pgd points to a shared pagetable level (either the
126 	   ptes in non-PAE, or shared PMD in PAE), then just copy the
127 	   references from swapper_pg_dir. */
128 	if (CONFIG_PGTABLE_LEVELS == 2 ||
129 	    (CONFIG_PGTABLE_LEVELS == 3 && SHARED_KERNEL_PMD) ||
130 	    CONFIG_PGTABLE_LEVELS >= 4) {
131 		clone_pgd_range(pgd + KERNEL_PGD_BOUNDARY,
132 				swapper_pg_dir + KERNEL_PGD_BOUNDARY,
133 				KERNEL_PGD_PTRS);
134 	}
135 
136 	/* list required to sync kernel mapping updates */
137 	if (!SHARED_KERNEL_PMD) {
138 		pgd_set_mm(pgd, mm);
139 		pgd_list_add(pgd);
140 	}
141 }
142 
pgd_dtor(pgd_t * pgd)143 static void pgd_dtor(pgd_t *pgd)
144 {
145 	if (SHARED_KERNEL_PMD)
146 		return;
147 
148 	spin_lock(&pgd_lock);
149 	pgd_list_del(pgd);
150 	spin_unlock(&pgd_lock);
151 }
152 
153 /*
154  * List of all pgd's needed for non-PAE so it can invalidate entries
155  * in both cached and uncached pgd's; not needed for PAE since the
156  * kernel pmd is shared. If PAE were not to share the pmd a similar
157  * tactic would be needed. This is essentially codepath-based locking
158  * against pageattr.c; it is the unique case in which a valid change
159  * of kernel pagetables can't be lazily synchronized by vmalloc faults.
160  * vmalloc faults work because attached pagetables are never freed.
161  * -- nyc
162  */
163 
164 #ifdef CONFIG_X86_PAE
165 /*
166  * In PAE mode, we need to do a cr3 reload (=tlb flush) when
167  * updating the top-level pagetable entries to guarantee the
168  * processor notices the update.  Since this is expensive, and
169  * all 4 top-level entries are used almost immediately in a
170  * new process's life, we just pre-populate them here.
171  *
172  * Also, if we're in a paravirt environment where the kernel pmd is
173  * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
174  * and initialize the kernel pmds here.
175  */
176 #define PREALLOCATED_PMDS	UNSHARED_PTRS_PER_PGD
177 #define MAX_PREALLOCATED_PMDS	MAX_UNSHARED_PTRS_PER_PGD
178 
179 /*
180  * We allocate separate PMDs for the kernel part of the user page-table
181  * when PTI is enabled. We need them to map the per-process LDT into the
182  * user-space page-table.
183  */
184 #define PREALLOCATED_USER_PMDS	 (boot_cpu_has(X86_FEATURE_PTI) ? \
185 					KERNEL_PGD_PTRS : 0)
186 #define MAX_PREALLOCATED_USER_PMDS KERNEL_PGD_PTRS
187 
pud_populate(struct mm_struct * mm,pud_t * pudp,pmd_t * pmd)188 void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
189 {
190 	paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT);
191 
192 	/* Note: almost everything apart from _PAGE_PRESENT is
193 	   reserved at the pmd (PDPT) level. */
194 	set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
195 
196 	/*
197 	 * According to Intel App note "TLBs, Paging-Structure Caches,
198 	 * and Their Invalidation", April 2007, document 317080-001,
199 	 * section 8.1: in PAE mode we explicitly have to flush the
200 	 * TLB via cr3 if the top-level pgd is changed...
201 	 */
202 	flush_tlb_mm(mm);
203 }
204 #else  /* !CONFIG_X86_PAE */
205 
206 /* No need to prepopulate any pagetable entries in non-PAE modes. */
207 #define PREALLOCATED_PMDS	0
208 #define MAX_PREALLOCATED_PMDS	0
209 #define PREALLOCATED_USER_PMDS	 0
210 #define MAX_PREALLOCATED_USER_PMDS 0
211 #endif	/* CONFIG_X86_PAE */
212 
free_pmds(struct mm_struct * mm,pmd_t * pmds[],int count)213 static void free_pmds(struct mm_struct *mm, pmd_t *pmds[], int count)
214 {
215 	int i;
216 
217 	for (i = 0; i < count; i++)
218 		if (pmds[i]) {
219 			pgtable_pmd_page_dtor(virt_to_page(pmds[i]));
220 			free_page((unsigned long)pmds[i]);
221 			mm_dec_nr_pmds(mm);
222 		}
223 }
224 
preallocate_pmds(struct mm_struct * mm,pmd_t * pmds[],int count)225 static int preallocate_pmds(struct mm_struct *mm, pmd_t *pmds[], int count)
226 {
227 	int i;
228 	bool failed = false;
229 	gfp_t gfp = GFP_PGTABLE_USER;
230 
231 	if (mm == &init_mm)
232 		gfp &= ~__GFP_ACCOUNT;
233 
234 	for (i = 0; i < count; i++) {
235 		pmd_t *pmd = (pmd_t *)__get_free_page(gfp);
236 		if (!pmd)
237 			failed = true;
238 		if (pmd && !pgtable_pmd_page_ctor(virt_to_page(pmd))) {
239 			free_page((unsigned long)pmd);
240 			pmd = NULL;
241 			failed = true;
242 		}
243 		if (pmd)
244 			mm_inc_nr_pmds(mm);
245 		pmds[i] = pmd;
246 	}
247 
248 	if (failed) {
249 		free_pmds(mm, pmds, count);
250 		return -ENOMEM;
251 	}
252 
253 	return 0;
254 }
255 
256 /*
257  * Mop up any pmd pages which may still be attached to the pgd.
258  * Normally they will be freed by munmap/exit_mmap, but any pmd we
259  * preallocate which never got a corresponding vma will need to be
260  * freed manually.
261  */
mop_up_one_pmd(struct mm_struct * mm,pgd_t * pgdp)262 static void mop_up_one_pmd(struct mm_struct *mm, pgd_t *pgdp)
263 {
264 	pgd_t pgd = *pgdp;
265 
266 	if (pgd_val(pgd) != 0) {
267 		pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
268 
269 		pgd_clear(pgdp);
270 
271 		paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
272 		pmd_free(mm, pmd);
273 		mm_dec_nr_pmds(mm);
274 	}
275 }
276 
pgd_mop_up_pmds(struct mm_struct * mm,pgd_t * pgdp)277 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
278 {
279 	int i;
280 
281 	for (i = 0; i < PREALLOCATED_PMDS; i++)
282 		mop_up_one_pmd(mm, &pgdp[i]);
283 
284 #ifdef CONFIG_PAGE_TABLE_ISOLATION
285 
286 	if (!boot_cpu_has(X86_FEATURE_PTI))
287 		return;
288 
289 	pgdp = kernel_to_user_pgdp(pgdp);
290 
291 	for (i = 0; i < PREALLOCATED_USER_PMDS; i++)
292 		mop_up_one_pmd(mm, &pgdp[i + KERNEL_PGD_BOUNDARY]);
293 #endif
294 }
295 
pgd_prepopulate_pmd(struct mm_struct * mm,pgd_t * pgd,pmd_t * pmds[])296 static void pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmds[])
297 {
298 	p4d_t *p4d;
299 	pud_t *pud;
300 	int i;
301 
302 	if (PREALLOCATED_PMDS == 0) /* Work around gcc-3.4.x bug */
303 		return;
304 
305 	p4d = p4d_offset(pgd, 0);
306 	pud = pud_offset(p4d, 0);
307 
308 	for (i = 0; i < PREALLOCATED_PMDS; i++, pud++) {
309 		pmd_t *pmd = pmds[i];
310 
311 		if (i >= KERNEL_PGD_BOUNDARY)
312 			memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
313 			       sizeof(pmd_t) * PTRS_PER_PMD);
314 
315 		pud_populate(mm, pud, pmd);
316 	}
317 }
318 
319 #ifdef CONFIG_PAGE_TABLE_ISOLATION
pgd_prepopulate_user_pmd(struct mm_struct * mm,pgd_t * k_pgd,pmd_t * pmds[])320 static void pgd_prepopulate_user_pmd(struct mm_struct *mm,
321 				     pgd_t *k_pgd, pmd_t *pmds[])
322 {
323 	pgd_t *s_pgd = kernel_to_user_pgdp(swapper_pg_dir);
324 	pgd_t *u_pgd = kernel_to_user_pgdp(k_pgd);
325 	p4d_t *u_p4d;
326 	pud_t *u_pud;
327 	int i;
328 
329 	u_p4d = p4d_offset(u_pgd, 0);
330 	u_pud = pud_offset(u_p4d, 0);
331 
332 	s_pgd += KERNEL_PGD_BOUNDARY;
333 	u_pud += KERNEL_PGD_BOUNDARY;
334 
335 	for (i = 0; i < PREALLOCATED_USER_PMDS; i++, u_pud++, s_pgd++) {
336 		pmd_t *pmd = pmds[i];
337 
338 		memcpy(pmd, (pmd_t *)pgd_page_vaddr(*s_pgd),
339 		       sizeof(pmd_t) * PTRS_PER_PMD);
340 
341 		pud_populate(mm, u_pud, pmd);
342 	}
343 
344 }
345 #else
pgd_prepopulate_user_pmd(struct mm_struct * mm,pgd_t * k_pgd,pmd_t * pmds[])346 static void pgd_prepopulate_user_pmd(struct mm_struct *mm,
347 				     pgd_t *k_pgd, pmd_t *pmds[])
348 {
349 }
350 #endif
351 /*
352  * Xen paravirt assumes pgd table should be in one page. 64 bit kernel also
353  * assumes that pgd should be in one page.
354  *
355  * But kernel with PAE paging that is not running as a Xen domain
356  * only needs to allocate 32 bytes for pgd instead of one page.
357  */
358 #ifdef CONFIG_X86_PAE
359 
360 #include <linux/slab.h>
361 
362 #define PGD_SIZE	(PTRS_PER_PGD * sizeof(pgd_t))
363 #define PGD_ALIGN	32
364 
365 static struct kmem_cache *pgd_cache;
366 
pgtable_cache_init(void)367 void __init pgtable_cache_init(void)
368 {
369 	/*
370 	 * When PAE kernel is running as a Xen domain, it does not use
371 	 * shared kernel pmd. And this requires a whole page for pgd.
372 	 */
373 	if (!SHARED_KERNEL_PMD)
374 		return;
375 
376 	/*
377 	 * when PAE kernel is not running as a Xen domain, it uses
378 	 * shared kernel pmd. Shared kernel pmd does not require a whole
379 	 * page for pgd. We are able to just allocate a 32-byte for pgd.
380 	 * During boot time, we create a 32-byte slab for pgd table allocation.
381 	 */
382 	pgd_cache = kmem_cache_create("pgd_cache", PGD_SIZE, PGD_ALIGN,
383 				      SLAB_PANIC, NULL);
384 }
385 
_pgd_alloc(void)386 static inline pgd_t *_pgd_alloc(void)
387 {
388 	/*
389 	 * If no SHARED_KERNEL_PMD, PAE kernel is running as a Xen domain.
390 	 * We allocate one page for pgd.
391 	 */
392 	if (!SHARED_KERNEL_PMD)
393 		return (pgd_t *)__get_free_pages(GFP_PGTABLE_USER,
394 						 PGD_ALLOCATION_ORDER);
395 
396 	/*
397 	 * Now PAE kernel is not running as a Xen domain. We can allocate
398 	 * a 32-byte slab for pgd to save memory space.
399 	 */
400 	return kmem_cache_alloc(pgd_cache, GFP_PGTABLE_USER);
401 }
402 
_pgd_free(pgd_t * pgd)403 static inline void _pgd_free(pgd_t *pgd)
404 {
405 	if (!SHARED_KERNEL_PMD)
406 		free_pages((unsigned long)pgd, PGD_ALLOCATION_ORDER);
407 	else
408 		kmem_cache_free(pgd_cache, pgd);
409 }
410 #else
411 
_pgd_alloc(void)412 static inline pgd_t *_pgd_alloc(void)
413 {
414 	return (pgd_t *)__get_free_pages(GFP_PGTABLE_USER,
415 					 PGD_ALLOCATION_ORDER);
416 }
417 
_pgd_free(pgd_t * pgd)418 static inline void _pgd_free(pgd_t *pgd)
419 {
420 	free_pages((unsigned long)pgd, PGD_ALLOCATION_ORDER);
421 }
422 #endif /* CONFIG_X86_PAE */
423 
pgd_alloc(struct mm_struct * mm)424 pgd_t *pgd_alloc(struct mm_struct *mm)
425 {
426 	pgd_t *pgd;
427 	pmd_t *u_pmds[MAX_PREALLOCATED_USER_PMDS];
428 	pmd_t *pmds[MAX_PREALLOCATED_PMDS];
429 
430 	pgd = _pgd_alloc();
431 
432 	if (pgd == NULL)
433 		goto out;
434 
435 	mm->pgd = pgd;
436 
437 	if (preallocate_pmds(mm, pmds, PREALLOCATED_PMDS) != 0)
438 		goto out_free_pgd;
439 
440 	if (preallocate_pmds(mm, u_pmds, PREALLOCATED_USER_PMDS) != 0)
441 		goto out_free_pmds;
442 
443 	if (paravirt_pgd_alloc(mm) != 0)
444 		goto out_free_user_pmds;
445 
446 	/*
447 	 * Make sure that pre-populating the pmds is atomic with
448 	 * respect to anything walking the pgd_list, so that they
449 	 * never see a partially populated pgd.
450 	 */
451 	spin_lock(&pgd_lock);
452 
453 	pgd_ctor(mm, pgd);
454 	pgd_prepopulate_pmd(mm, pgd, pmds);
455 	pgd_prepopulate_user_pmd(mm, pgd, u_pmds);
456 
457 	spin_unlock(&pgd_lock);
458 
459 	return pgd;
460 
461 out_free_user_pmds:
462 	free_pmds(mm, u_pmds, PREALLOCATED_USER_PMDS);
463 out_free_pmds:
464 	free_pmds(mm, pmds, PREALLOCATED_PMDS);
465 out_free_pgd:
466 	_pgd_free(pgd);
467 out:
468 	return NULL;
469 }
470 
pgd_free(struct mm_struct * mm,pgd_t * pgd)471 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
472 {
473 	pgd_mop_up_pmds(mm, pgd);
474 	pgd_dtor(pgd);
475 	paravirt_pgd_free(mm, pgd);
476 	_pgd_free(pgd);
477 }
478 
479 /*
480  * Used to set accessed or dirty bits in the page table entries
481  * on other architectures. On x86, the accessed and dirty bits
482  * are tracked by hardware. However, do_wp_page calls this function
483  * to also make the pte writeable at the same time the dirty bit is
484  * set. In that case we do actually need to write the PTE.
485  */
ptep_set_access_flags(struct vm_area_struct * vma,unsigned long address,pte_t * ptep,pte_t entry,int dirty)486 int ptep_set_access_flags(struct vm_area_struct *vma,
487 			  unsigned long address, pte_t *ptep,
488 			  pte_t entry, int dirty)
489 {
490 	int changed = !pte_same(*ptep, entry);
491 
492 	if (changed && dirty)
493 		set_pte(ptep, entry);
494 
495 	return changed;
496 }
497 
498 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmdp_set_access_flags(struct vm_area_struct * vma,unsigned long address,pmd_t * pmdp,pmd_t entry,int dirty)499 int pmdp_set_access_flags(struct vm_area_struct *vma,
500 			  unsigned long address, pmd_t *pmdp,
501 			  pmd_t entry, int dirty)
502 {
503 	int changed = !pmd_same(*pmdp, entry);
504 
505 	VM_BUG_ON(address & ~HPAGE_PMD_MASK);
506 
507 	if (changed && dirty) {
508 		set_pmd(pmdp, entry);
509 		/*
510 		 * We had a write-protection fault here and changed the pmd
511 		 * to to more permissive. No need to flush the TLB for that,
512 		 * #PF is architecturally guaranteed to do that and in the
513 		 * worst-case we'll generate a spurious fault.
514 		 */
515 	}
516 
517 	return changed;
518 }
519 
pudp_set_access_flags(struct vm_area_struct * vma,unsigned long address,pud_t * pudp,pud_t entry,int dirty)520 int pudp_set_access_flags(struct vm_area_struct *vma, unsigned long address,
521 			  pud_t *pudp, pud_t entry, int dirty)
522 {
523 	int changed = !pud_same(*pudp, entry);
524 
525 	VM_BUG_ON(address & ~HPAGE_PUD_MASK);
526 
527 	if (changed && dirty) {
528 		set_pud(pudp, entry);
529 		/*
530 		 * We had a write-protection fault here and changed the pud
531 		 * to to more permissive. No need to flush the TLB for that,
532 		 * #PF is architecturally guaranteed to do that and in the
533 		 * worst-case we'll generate a spurious fault.
534 		 */
535 	}
536 
537 	return changed;
538 }
539 #endif
540 
ptep_test_and_clear_young(struct vm_area_struct * vma,unsigned long addr,pte_t * ptep)541 int ptep_test_and_clear_young(struct vm_area_struct *vma,
542 			      unsigned long addr, pte_t *ptep)
543 {
544 	int ret = 0;
545 
546 	if (pte_young(*ptep))
547 		ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
548 					 (unsigned long *) &ptep->pte);
549 
550 	return ret;
551 }
552 
553 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmdp_test_and_clear_young(struct vm_area_struct * vma,unsigned long addr,pmd_t * pmdp)554 int pmdp_test_and_clear_young(struct vm_area_struct *vma,
555 			      unsigned long addr, pmd_t *pmdp)
556 {
557 	int ret = 0;
558 
559 	if (pmd_young(*pmdp))
560 		ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
561 					 (unsigned long *)pmdp);
562 
563 	return ret;
564 }
pudp_test_and_clear_young(struct vm_area_struct * vma,unsigned long addr,pud_t * pudp)565 int pudp_test_and_clear_young(struct vm_area_struct *vma,
566 			      unsigned long addr, pud_t *pudp)
567 {
568 	int ret = 0;
569 
570 	if (pud_young(*pudp))
571 		ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
572 					 (unsigned long *)pudp);
573 
574 	return ret;
575 }
576 #endif
577 
ptep_clear_flush_young(struct vm_area_struct * vma,unsigned long address,pte_t * ptep)578 int ptep_clear_flush_young(struct vm_area_struct *vma,
579 			   unsigned long address, pte_t *ptep)
580 {
581 	/*
582 	 * On x86 CPUs, clearing the accessed bit without a TLB flush
583 	 * doesn't cause data corruption. [ It could cause incorrect
584 	 * page aging and the (mistaken) reclaim of hot pages, but the
585 	 * chance of that should be relatively low. ]
586 	 *
587 	 * So as a performance optimization don't flush the TLB when
588 	 * clearing the accessed bit, it will eventually be flushed by
589 	 * a context switch or a VM operation anyway. [ In the rare
590 	 * event of it not getting flushed for a long time the delay
591 	 * shouldn't really matter because there's no real memory
592 	 * pressure for swapout to react to. ]
593 	 */
594 	return ptep_test_and_clear_young(vma, address, ptep);
595 }
596 
597 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pmdp_clear_flush_young(struct vm_area_struct * vma,unsigned long address,pmd_t * pmdp)598 int pmdp_clear_flush_young(struct vm_area_struct *vma,
599 			   unsigned long address, pmd_t *pmdp)
600 {
601 	int young;
602 
603 	VM_BUG_ON(address & ~HPAGE_PMD_MASK);
604 
605 	young = pmdp_test_and_clear_young(vma, address, pmdp);
606 	if (young)
607 		flush_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
608 
609 	return young;
610 }
611 
pmdp_invalidate_ad(struct vm_area_struct * vma,unsigned long address,pmd_t * pmdp)612 pmd_t pmdp_invalidate_ad(struct vm_area_struct *vma, unsigned long address,
613 			 pmd_t *pmdp)
614 {
615 	/*
616 	 * No flush is necessary. Once an invalid PTE is established, the PTE's
617 	 * access and dirty bits cannot be updated.
618 	 */
619 	return pmdp_establish(vma, address, pmdp, pmd_mkinvalid(*pmdp));
620 }
621 #endif
622 
623 /**
624  * reserve_top_address - reserves a hole in the top of kernel address space
625  * @reserve - size of hole to reserve
626  *
627  * Can be used to relocate the fixmap area and poke a hole in the top
628  * of kernel address space to make room for a hypervisor.
629  */
reserve_top_address(unsigned long reserve)630 void __init reserve_top_address(unsigned long reserve)
631 {
632 #ifdef CONFIG_X86_32
633 	BUG_ON(fixmaps_set > 0);
634 	__FIXADDR_TOP = round_down(-reserve, 1 << PMD_SHIFT) - PAGE_SIZE;
635 	printk(KERN_INFO "Reserving virtual address space above 0x%08lx (rounded to 0x%08lx)\n",
636 	       -reserve, __FIXADDR_TOP + PAGE_SIZE);
637 #endif
638 }
639 
640 int fixmaps_set;
641 
__native_set_fixmap(enum fixed_addresses idx,pte_t pte)642 void __native_set_fixmap(enum fixed_addresses idx, pte_t pte)
643 {
644 	unsigned long address = __fix_to_virt(idx);
645 
646 #ifdef CONFIG_X86_64
647        /*
648 	* Ensure that the static initial page tables are covering the
649 	* fixmap completely.
650 	*/
651 	BUILD_BUG_ON(__end_of_permanent_fixed_addresses >
652 		     (FIXMAP_PMD_NUM * PTRS_PER_PTE));
653 #endif
654 
655 	if (idx >= __end_of_fixed_addresses) {
656 		BUG();
657 		return;
658 	}
659 	set_pte_vaddr(address, pte);
660 	fixmaps_set++;
661 }
662 
native_set_fixmap(unsigned idx,phys_addr_t phys,pgprot_t flags)663 void native_set_fixmap(unsigned /* enum fixed_addresses */ idx,
664 		       phys_addr_t phys, pgprot_t flags)
665 {
666 	/* Sanitize 'prot' against any unsupported bits: */
667 	pgprot_val(flags) &= __default_kernel_pte_mask;
668 
669 	__native_set_fixmap(idx, pfn_pte(phys >> PAGE_SHIFT, flags));
670 }
671 
672 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
673 #ifdef CONFIG_X86_5LEVEL
674 /**
675  * p4d_set_huge - setup kernel P4D mapping
676  *
677  * No 512GB pages yet -- always return 0
678  */
p4d_set_huge(p4d_t * p4d,phys_addr_t addr,pgprot_t prot)679 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
680 {
681 	return 0;
682 }
683 
684 /**
685  * p4d_clear_huge - clear kernel P4D mapping when it is set
686  *
687  * No 512GB pages yet -- always return 0
688  */
p4d_clear_huge(p4d_t * p4d)689 void p4d_clear_huge(p4d_t *p4d)
690 {
691 }
692 #endif
693 
694 /**
695  * pud_set_huge - setup kernel PUD mapping
696  *
697  * MTRRs can override PAT memory types with 4KiB granularity. Therefore, this
698  * function sets up a huge page only if any of the following conditions are met:
699  *
700  * - MTRRs are disabled, or
701  *
702  * - MTRRs are enabled and the range is completely covered by a single MTRR, or
703  *
704  * - MTRRs are enabled and the corresponding MTRR memory type is WB, which
705  *   has no effect on the requested PAT memory type.
706  *
707  * Callers should try to decrease page size (1GB -> 2MB -> 4K) if the bigger
708  * page mapping attempt fails.
709  *
710  * Returns 1 on success and 0 on failure.
711  */
pud_set_huge(pud_t * pud,phys_addr_t addr,pgprot_t prot)712 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
713 {
714 	u8 mtrr, uniform;
715 
716 	mtrr = mtrr_type_lookup(addr, addr + PUD_SIZE, &uniform);
717 	if ((mtrr != MTRR_TYPE_INVALID) && (!uniform) &&
718 	    (mtrr != MTRR_TYPE_WRBACK))
719 		return 0;
720 
721 	/* Bail out if we are we on a populated non-leaf entry: */
722 	if (pud_present(*pud) && !pud_huge(*pud))
723 		return 0;
724 
725 	set_pte((pte_t *)pud, pfn_pte(
726 		(u64)addr >> PAGE_SHIFT,
727 		__pgprot(protval_4k_2_large(pgprot_val(prot)) | _PAGE_PSE)));
728 
729 	return 1;
730 }
731 
732 /**
733  * pmd_set_huge - setup kernel PMD mapping
734  *
735  * See text over pud_set_huge() above.
736  *
737  * Returns 1 on success and 0 on failure.
738  */
pmd_set_huge(pmd_t * pmd,phys_addr_t addr,pgprot_t prot)739 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
740 {
741 	u8 mtrr, uniform;
742 
743 	mtrr = mtrr_type_lookup(addr, addr + PMD_SIZE, &uniform);
744 	if ((mtrr != MTRR_TYPE_INVALID) && (!uniform) &&
745 	    (mtrr != MTRR_TYPE_WRBACK)) {
746 		pr_warn_once("%s: Cannot satisfy [mem %#010llx-%#010llx] with a huge-page mapping due to MTRR override.\n",
747 			     __func__, addr, addr + PMD_SIZE);
748 		return 0;
749 	}
750 
751 	/* Bail out if we are we on a populated non-leaf entry: */
752 	if (pmd_present(*pmd) && !pmd_huge(*pmd))
753 		return 0;
754 
755 	set_pte((pte_t *)pmd, pfn_pte(
756 		(u64)addr >> PAGE_SHIFT,
757 		__pgprot(protval_4k_2_large(pgprot_val(prot)) | _PAGE_PSE)));
758 
759 	return 1;
760 }
761 
762 /**
763  * pud_clear_huge - clear kernel PUD mapping when it is set
764  *
765  * Returns 1 on success and 0 on failure (no PUD map is found).
766  */
pud_clear_huge(pud_t * pud)767 int pud_clear_huge(pud_t *pud)
768 {
769 	if (pud_large(*pud)) {
770 		pud_clear(pud);
771 		return 1;
772 	}
773 
774 	return 0;
775 }
776 
777 /**
778  * pmd_clear_huge - clear kernel PMD mapping when it is set
779  *
780  * Returns 1 on success and 0 on failure (no PMD map is found).
781  */
pmd_clear_huge(pmd_t * pmd)782 int pmd_clear_huge(pmd_t *pmd)
783 {
784 	if (pmd_large(*pmd)) {
785 		pmd_clear(pmd);
786 		return 1;
787 	}
788 
789 	return 0;
790 }
791 
792 #ifdef CONFIG_X86_64
793 /**
794  * pud_free_pmd_page - Clear pud entry and free pmd page.
795  * @pud: Pointer to a PUD.
796  * @addr: Virtual address associated with pud.
797  *
798  * Context: The pud range has been unmapped and TLB purged.
799  * Return: 1 if clearing the entry succeeded. 0 otherwise.
800  *
801  * NOTE: Callers must allow a single page allocation.
802  */
pud_free_pmd_page(pud_t * pud,unsigned long addr)803 int pud_free_pmd_page(pud_t *pud, unsigned long addr)
804 {
805 	pmd_t *pmd, *pmd_sv;
806 	pte_t *pte;
807 	int i;
808 
809 	pmd = pud_pgtable(*pud);
810 	pmd_sv = (pmd_t *)__get_free_page(GFP_KERNEL);
811 	if (!pmd_sv)
812 		return 0;
813 
814 	for (i = 0; i < PTRS_PER_PMD; i++) {
815 		pmd_sv[i] = pmd[i];
816 		if (!pmd_none(pmd[i]))
817 			pmd_clear(&pmd[i]);
818 	}
819 
820 	pud_clear(pud);
821 
822 	/* INVLPG to clear all paging-structure caches */
823 	flush_tlb_kernel_range(addr, addr + PAGE_SIZE-1);
824 
825 	for (i = 0; i < PTRS_PER_PMD; i++) {
826 		if (!pmd_none(pmd_sv[i])) {
827 			pte = (pte_t *)pmd_page_vaddr(pmd_sv[i]);
828 			free_page((unsigned long)pte);
829 		}
830 	}
831 
832 	free_page((unsigned long)pmd_sv);
833 
834 	pgtable_pmd_page_dtor(virt_to_page(pmd));
835 	free_page((unsigned long)pmd);
836 
837 	return 1;
838 }
839 
840 /**
841  * pmd_free_pte_page - Clear pmd entry and free pte page.
842  * @pmd: Pointer to a PMD.
843  * @addr: Virtual address associated with pmd.
844  *
845  * Context: The pmd range has been unmapped and TLB purged.
846  * Return: 1 if clearing the entry succeeded. 0 otherwise.
847  */
pmd_free_pte_page(pmd_t * pmd,unsigned long addr)848 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
849 {
850 	pte_t *pte;
851 
852 	pte = (pte_t *)pmd_page_vaddr(*pmd);
853 	pmd_clear(pmd);
854 
855 	/* INVLPG to clear all paging-structure caches */
856 	flush_tlb_kernel_range(addr, addr + PAGE_SIZE-1);
857 
858 	free_page((unsigned long)pte);
859 
860 	return 1;
861 }
862 
863 #else /* !CONFIG_X86_64 */
864 
865 /*
866  * Disable free page handling on x86-PAE. This assures that ioremap()
867  * does not update sync'd pmd entries. See vmalloc_sync_one().
868  */
pmd_free_pte_page(pmd_t * pmd,unsigned long addr)869 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
870 {
871 	return pmd_none(*pmd);
872 }
873 
874 #endif /* CONFIG_X86_64 */
875 #endif	/* CONFIG_HAVE_ARCH_HUGE_VMAP */
876