~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/x86/mm/pgtable.c

Version: ~ [ linux-5.19-rc3 ] ~ [ linux-5.18.5 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.48 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.123 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.199 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.248 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.284 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.319 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/pgtable.h>
  7 #include <asm/tlb.h>
  8 #include <asm/fixmap.h>
  9 #include <asm/mtrr.h>
 10 
 11 #define PGALLOC_GFP (GFP_KERNEL_ACCOUNT | __GFP_ZERO)
 12 
 13 #ifdef CONFIG_HIGHPTE
 14 #define PGALLOC_USER_GFP __GFP_HIGHMEM
 15 #else
 16 #define PGALLOC_USER_GFP 0
 17 #endif
 18 
 19 gfp_t __userpte_alloc_gfp = PGALLOC_GFP | PGALLOC_USER_GFP;
 20 
 21 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
 22 {
 23         return (pte_t *)__get_free_page(PGALLOC_GFP & ~__GFP_ACCOUNT);
 24 }
 25 
 26 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
 27 {
 28         struct page *pte;
 29 
 30         pte = alloc_pages(__userpte_alloc_gfp, 0);
 31         if (!pte)
 32                 return NULL;
 33         if (!pgtable_page_ctor(pte)) {
 34                 __free_page(pte);
 35                 return NULL;
 36         }
 37         return pte;
 38 }
 39 
 40 static int __init setup_userpte(char *arg)
 41 {
 42         if (!arg)
 43                 return -EINVAL;
 44 
 45         /*
 46          * "userpte=nohigh" disables allocation of user pagetables in
 47          * high memory.
 48          */
 49         if (strcmp(arg, "nohigh") == 0)
 50                 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
 51         else
 52                 return -EINVAL;
 53         return 0;
 54 }
 55 early_param("userpte", setup_userpte);
 56 
 57 void ___pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
 58 {
 59         pgtable_page_dtor(pte);
 60         paravirt_release_pte(page_to_pfn(pte));
 61         tlb_remove_table(tlb, pte);
 62 }
 63 
 64 #if CONFIG_PGTABLE_LEVELS > 2
 65 void ___pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
 66 {
 67         struct page *page = virt_to_page(pmd);
 68         paravirt_release_pmd(__pa(pmd) >> PAGE_SHIFT);
 69         /*
 70          * NOTE! For PAE, any changes to the top page-directory-pointer-table
 71          * entries need a full cr3 reload to flush.
 72          */
 73 #ifdef CONFIG_X86_PAE
 74         tlb->need_flush_all = 1;
 75 #endif
 76         pgtable_pmd_page_dtor(page);
 77         tlb_remove_table(tlb, page);
 78 }
 79 
 80 #if CONFIG_PGTABLE_LEVELS > 3
 81 void ___pud_free_tlb(struct mmu_gather *tlb, pud_t *pud)
 82 {
 83         paravirt_release_pud(__pa(pud) >> PAGE_SHIFT);
 84         tlb_remove_table(tlb, virt_to_page(pud));
 85 }
 86 
 87 #if CONFIG_PGTABLE_LEVELS > 4
 88 void ___p4d_free_tlb(struct mmu_gather *tlb, p4d_t *p4d)
 89 {
 90         paravirt_release_p4d(__pa(p4d) >> PAGE_SHIFT);
 91         tlb_remove_table(tlb, virt_to_page(p4d));
 92 }
 93 #endif  /* CONFIG_PGTABLE_LEVELS > 4 */
 94 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
 95 #endif  /* CONFIG_PGTABLE_LEVELS > 2 */
 96 
 97 static inline void pgd_list_add(pgd_t *pgd)
 98 {
 99         struct page *page = virt_to_page(pgd);
100 
101         list_add(&page->lru, &pgd_list);
102 }
103 
104 static inline void pgd_list_del(pgd_t *pgd)
105 {
106         struct page *page = virt_to_page(pgd);
107 
108         list_del(&page->lru);
109 }
110 
111 #define UNSHARED_PTRS_PER_PGD                           \
112         (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD)
113 
114 
115 static void pgd_set_mm(pgd_t *pgd, struct mm_struct *mm)
116 {
117         BUILD_BUG_ON(sizeof(virt_to_page(pgd)->index) < sizeof(mm));
118         virt_to_page(pgd)->index = (pgoff_t)mm;
119 }
120 
121 struct mm_struct *pgd_page_get_mm(struct page *page)
122 {
123         return (struct mm_struct *)page->index;
124 }
125 
126 static void pgd_ctor(struct mm_struct *mm, pgd_t *pgd)
127 {
128         /* If the pgd points to a shared pagetable level (either the
129            ptes in non-PAE, or shared PMD in PAE), then just copy the
130            references from swapper_pg_dir. */
131         if (CONFIG_PGTABLE_LEVELS == 2 ||
132             (CONFIG_PGTABLE_LEVELS == 3 && SHARED_KERNEL_PMD) ||
133             CONFIG_PGTABLE_LEVELS >= 4) {
134                 clone_pgd_range(pgd + KERNEL_PGD_BOUNDARY,
135                                 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
136                                 KERNEL_PGD_PTRS);
137         }
138 
139         /* list required to sync kernel mapping updates */
140         if (!SHARED_KERNEL_PMD) {
141                 pgd_set_mm(pgd, mm);
142                 pgd_list_add(pgd);
143         }
144 }
145 
146 static void pgd_dtor(pgd_t *pgd)
147 {
148         if (SHARED_KERNEL_PMD)
149                 return;
150 
151         spin_lock(&pgd_lock);
152         pgd_list_del(pgd);
153         spin_unlock(&pgd_lock);
154 }
155 
156 /*
157  * List of all pgd's needed for non-PAE so it can invalidate entries
158  * in both cached and uncached pgd's; not needed for PAE since the
159  * kernel pmd is shared. If PAE were not to share the pmd a similar
160  * tactic would be needed. This is essentially codepath-based locking
161  * against pageattr.c; it is the unique case in which a valid change
162  * of kernel pagetables can't be lazily synchronized by vmalloc faults.
163  * vmalloc faults work because attached pagetables are never freed.
164  * -- nyc
165  */
166 
167 #ifdef CONFIG_X86_PAE
168 /*
169  * In PAE mode, we need to do a cr3 reload (=tlb flush) when
170  * updating the top-level pagetable entries to guarantee the
171  * processor notices the update.  Since this is expensive, and
172  * all 4 top-level entries are used almost immediately in a
173  * new process's life, we just pre-populate them here.
174  *
175  * Also, if we're in a paravirt environment where the kernel pmd is
176  * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
177  * and initialize the kernel pmds here.
178  */
179 #define PREALLOCATED_PMDS       UNSHARED_PTRS_PER_PGD
180 
181 void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
182 {
183         paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT);
184 
185         /* Note: almost everything apart from _PAGE_PRESENT is
186            reserved at the pmd (PDPT) level. */
187         set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
188 
189         /*
190          * According to Intel App note "TLBs, Paging-Structure Caches,
191          * and Their Invalidation", April 2007, document 317080-001,
192          * section 8.1: in PAE mode we explicitly have to flush the
193          * TLB via cr3 if the top-level pgd is changed...
194          */
195         flush_tlb_mm(mm);
196 }
197 #else  /* !CONFIG_X86_PAE */
198 
199 /* No need to prepopulate any pagetable entries in non-PAE modes. */
200 #define PREALLOCATED_PMDS       0
201 
202 #endif  /* CONFIG_X86_PAE */
203 
204 static void free_pmds(struct mm_struct *mm, pmd_t *pmds[])
205 {
206         int i;
207 
208         for(i = 0; i < PREALLOCATED_PMDS; i++)
209                 if (pmds[i]) {
210                         pgtable_pmd_page_dtor(virt_to_page(pmds[i]));
211                         free_page((unsigned long)pmds[i]);
212                         mm_dec_nr_pmds(mm);
213                 }
214 }
215 
216 static int preallocate_pmds(struct mm_struct *mm, pmd_t *pmds[])
217 {
218         int i;
219         bool failed = false;
220         gfp_t gfp = PGALLOC_GFP;
221 
222         if (mm == &init_mm)
223                 gfp &= ~__GFP_ACCOUNT;
224 
225         for(i = 0; i < PREALLOCATED_PMDS; i++) {
226                 pmd_t *pmd = (pmd_t *)__get_free_page(gfp);
227                 if (!pmd)
228                         failed = true;
229                 if (pmd && !pgtable_pmd_page_ctor(virt_to_page(pmd))) {
230                         free_page((unsigned long)pmd);
231                         pmd = NULL;
232                         failed = true;
233                 }
234                 if (pmd)
235                         mm_inc_nr_pmds(mm);
236                 pmds[i] = pmd;
237         }
238 
239         if (failed) {
240                 free_pmds(mm, pmds);
241                 return -ENOMEM;
242         }
243 
244         return 0;
245 }
246 
247 /*
248  * Mop up any pmd pages which may still be attached to the pgd.
249  * Normally they will be freed by munmap/exit_mmap, but any pmd we
250  * preallocate which never got a corresponding vma will need to be
251  * freed manually.
252  */
253 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
254 {
255         int i;
256 
257         for(i = 0; i < PREALLOCATED_PMDS; i++) {
258                 pgd_t pgd = pgdp[i];
259 
260                 if (pgd_val(pgd) != 0) {
261                         pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
262 
263                         pgdp[i] = native_make_pgd(0);
264 
265                         paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
266                         pmd_free(mm, pmd);
267                         mm_dec_nr_pmds(mm);
268                 }
269         }
270 }
271 
272 static void pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmds[])
273 {
274         p4d_t *p4d;
275         pud_t *pud;
276         int i;
277 
278         if (PREALLOCATED_PMDS == 0) /* Work around gcc-3.4.x bug */
279                 return;
280 
281         p4d = p4d_offset(pgd, 0);
282         pud = pud_offset(p4d, 0);
283 
284         for (i = 0; i < PREALLOCATED_PMDS; i++, pud++) {
285                 pmd_t *pmd = pmds[i];
286 
287                 if (i >= KERNEL_PGD_BOUNDARY)
288                         memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
289                                sizeof(pmd_t) * PTRS_PER_PMD);
290 
291                 pud_populate(mm, pud, pmd);
292         }
293 }
294 
295 /*
296  * Xen paravirt assumes pgd table should be in one page. 64 bit kernel also
297  * assumes that pgd should be in one page.
298  *
299  * But kernel with PAE paging that is not running as a Xen domain
300  * only needs to allocate 32 bytes for pgd instead of one page.
301  */
302 #ifdef CONFIG_X86_PAE
303 
304 #include <linux/slab.h>
305 
306 #define PGD_SIZE        (PTRS_PER_PGD * sizeof(pgd_t))
307 #define PGD_ALIGN       32
308 
309 static struct kmem_cache *pgd_cache;
310 
311 static int __init pgd_cache_init(void)
312 {
313         /*
314          * When PAE kernel is running as a Xen domain, it does not use
315          * shared kernel pmd. And this requires a whole page for pgd.
316          */
317         if (!SHARED_KERNEL_PMD)
318                 return 0;
319 
320         /*
321          * when PAE kernel is not running as a Xen domain, it uses
322          * shared kernel pmd. Shared kernel pmd does not require a whole
323          * page for pgd. We are able to just allocate a 32-byte for pgd.
324          * During boot time, we create a 32-byte slab for pgd table allocation.
325          */
326         pgd_cache = kmem_cache_create("pgd_cache", PGD_SIZE, PGD_ALIGN,
327                                       SLAB_PANIC, NULL);
328         if (!pgd_cache)
329                 return -ENOMEM;
330 
331         return 0;
332 }
333 core_initcall(pgd_cache_init);
334 
335 static inline pgd_t *_pgd_alloc(void)
336 {
337         /*
338          * If no SHARED_KERNEL_PMD, PAE kernel is running as a Xen domain.
339          * We allocate one page for pgd.
340          */
341         if (!SHARED_KERNEL_PMD)
342                 return (pgd_t *)__get_free_page(PGALLOC_GFP);
343 
344         /*
345          * Now PAE kernel is not running as a Xen domain. We can allocate
346          * a 32-byte slab for pgd to save memory space.
347          */
348         return kmem_cache_alloc(pgd_cache, PGALLOC_GFP);
349 }
350 
351 static inline void _pgd_free(pgd_t *pgd)
352 {
353         if (!SHARED_KERNEL_PMD)
354                 free_page((unsigned long)pgd);
355         else
356                 kmem_cache_free(pgd_cache, pgd);
357 }
358 #else
359 
360 static inline pgd_t *_pgd_alloc(void)
361 {
362         return (pgd_t *)__get_free_pages(PGALLOC_GFP, PGD_ALLOCATION_ORDER);
363 }
364 
365 static inline void _pgd_free(pgd_t *pgd)
366 {
367         free_pages((unsigned long)pgd, PGD_ALLOCATION_ORDER);
368 }
369 #endif /* CONFIG_X86_PAE */
370 
371 pgd_t *pgd_alloc(struct mm_struct *mm)
372 {
373         pgd_t *pgd;
374         pmd_t *pmds[PREALLOCATED_PMDS];
375 
376         pgd = _pgd_alloc();
377 
378         if (pgd == NULL)
379                 goto out;
380 
381         mm->pgd = pgd;
382 
383         if (preallocate_pmds(mm, pmds) != 0)
384                 goto out_free_pgd;
385 
386         if (paravirt_pgd_alloc(mm) != 0)
387                 goto out_free_pmds;
388 
389         /*
390          * Make sure that pre-populating the pmds is atomic with
391          * respect to anything walking the pgd_list, so that they
392          * never see a partially populated pgd.
393          */
394         spin_lock(&pgd_lock);
395 
396         pgd_ctor(mm, pgd);
397         pgd_prepopulate_pmd(mm, pgd, pmds);
398 
399         spin_unlock(&pgd_lock);
400 
401         return pgd;
402 
403 out_free_pmds:
404         free_pmds(mm, pmds);
405 out_free_pgd:
406         _pgd_free(pgd);
407 out:
408         return NULL;
409 }
410 
411 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
412 {
413         pgd_mop_up_pmds(mm, pgd);
414         pgd_dtor(pgd);
415         paravirt_pgd_free(mm, pgd);
416         _pgd_free(pgd);
417 }
418 
419 /*
420  * Used to set accessed or dirty bits in the page table entries
421  * on other architectures. On x86, the accessed and dirty bits
422  * are tracked by hardware. However, do_wp_page calls this function
423  * to also make the pte writeable at the same time the dirty bit is
424  * set. In that case we do actually need to write the PTE.
425  */
426 int ptep_set_access_flags(struct vm_area_struct *vma,
427                           unsigned long address, pte_t *ptep,
428                           pte_t entry, int dirty)
429 {
430         int changed = !pte_same(*ptep, entry);
431 
432         if (changed && dirty)
433                 *ptep = entry;
434 
435         return changed;
436 }
437 
438 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
439 int pmdp_set_access_flags(struct vm_area_struct *vma,
440                           unsigned long address, pmd_t *pmdp,
441                           pmd_t entry, int dirty)
442 {
443         int changed = !pmd_same(*pmdp, entry);
444 
445         VM_BUG_ON(address & ~HPAGE_PMD_MASK);
446 
447         if (changed && dirty) {
448                 *pmdp = entry;
449                 /*
450                  * We had a write-protection fault here and changed the pmd
451                  * to to more permissive. No need to flush the TLB for that,
452                  * #PF is architecturally guaranteed to do that and in the
453                  * worst-case we'll generate a spurious fault.
454                  */
455         }
456 
457         return changed;
458 }
459 
460 int pudp_set_access_flags(struct vm_area_struct *vma, unsigned long address,
461                           pud_t *pudp, pud_t entry, int dirty)
462 {
463         int changed = !pud_same(*pudp, entry);
464 
465         VM_BUG_ON(address & ~HPAGE_PUD_MASK);
466 
467         if (changed && dirty) {
468                 *pudp = entry;
469                 /*
470                  * We had a write-protection fault here and changed the pud
471                  * to to more permissive. No need to flush the TLB for that,
472                  * #PF is architecturally guaranteed to do that and in the
473                  * worst-case we'll generate a spurious fault.
474                  */
475         }
476 
477         return changed;
478 }
479 #endif
480 
481 int ptep_test_and_clear_young(struct vm_area_struct *vma,
482                               unsigned long addr, pte_t *ptep)
483 {
484         int ret = 0;
485 
486         if (pte_young(*ptep))
487                 ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
488                                          (unsigned long *) &ptep->pte);
489 
490         return ret;
491 }
492 
493 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
494 int pmdp_test_and_clear_young(struct vm_area_struct *vma,
495                               unsigned long addr, pmd_t *pmdp)
496 {
497         int ret = 0;
498 
499         if (pmd_young(*pmdp))
500                 ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
501                                          (unsigned long *)pmdp);
502 
503         return ret;
504 }
505 int pudp_test_and_clear_young(struct vm_area_struct *vma,
506                               unsigned long addr, pud_t *pudp)
507 {
508         int ret = 0;
509 
510         if (pud_young(*pudp))
511                 ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
512                                          (unsigned long *)pudp);
513 
514         return ret;
515 }
516 #endif
517 
518 int ptep_clear_flush_young(struct vm_area_struct *vma,
519                            unsigned long address, pte_t *ptep)
520 {
521         /*
522          * On x86 CPUs, clearing the accessed bit without a TLB flush
523          * doesn't cause data corruption. [ It could cause incorrect
524          * page aging and the (mistaken) reclaim of hot pages, but the
525          * chance of that should be relatively low. ]
526          *
527          * So as a performance optimization don't flush the TLB when
528          * clearing the accessed bit, it will eventually be flushed by
529          * a context switch or a VM operation anyway. [ In the rare
530          * event of it not getting flushed for a long time the delay
531          * shouldn't really matter because there's no real memory
532          * pressure for swapout to react to. ]
533          */
534         return ptep_test_and_clear_young(vma, address, ptep);
535 }
536 
537 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
538 int pmdp_clear_flush_young(struct vm_area_struct *vma,
539                            unsigned long address, pmd_t *pmdp)
540 {
541         int young;
542 
543         VM_BUG_ON(address & ~HPAGE_PMD_MASK);
544 
545         young = pmdp_test_and_clear_young(vma, address, pmdp);
546         if (young)
547                 flush_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
548 
549         return young;
550 }
551 #endif
552 
553 /**
554  * reserve_top_address - reserves a hole in the top of kernel address space
555  * @reserve - size of hole to reserve
556  *
557  * Can be used to relocate the fixmap area and poke a hole in the top
558  * of kernel address space to make room for a hypervisor.
559  */
560 void __init reserve_top_address(unsigned long reserve)
561 {
562 #ifdef CONFIG_X86_32
563         BUG_ON(fixmaps_set > 0);
564         __FIXADDR_TOP = round_down(-reserve, 1 << PMD_SHIFT) - PAGE_SIZE;
565         printk(KERN_INFO "Reserving virtual address space above 0x%08lx (rounded to 0x%08lx)\n",
566                -reserve, __FIXADDR_TOP + PAGE_SIZE);
567 #endif
568 }
569 
570 int fixmaps_set;
571 
572 void __native_set_fixmap(enum fixed_addresses idx, pte_t pte)
573 {
574         unsigned long address = __fix_to_virt(idx);
575 
576         if (idx >= __end_of_fixed_addresses) {
577                 BUG();
578                 return;
579         }
580         set_pte_vaddr(address, pte);
581         fixmaps_set++;
582 }
583 
584 void native_set_fixmap(enum fixed_addresses idx, phys_addr_t phys,
585                        pgprot_t flags)
586 {
587         __native_set_fixmap(idx, pfn_pte(phys >> PAGE_SHIFT, flags));
588 }
589 
590 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
591 #ifdef CONFIG_X86_5LEVEL
592 /**
593  * p4d_set_huge - setup kernel P4D mapping
594  *
595  * No 512GB pages yet -- always return 0
596  */
597 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
598 {
599         return 0;
600 }
601 
602 /**
603  * p4d_clear_huge - clear kernel P4D mapping when it is set
604  *
605  * No 512GB pages yet -- always return 0
606  */
607 int p4d_clear_huge(p4d_t *p4d)
608 {
609         return 0;
610 }
611 #endif
612 
613 /**
614  * pud_set_huge - setup kernel PUD mapping
615  *
616  * MTRRs can override PAT memory types with 4KiB granularity. Therefore, this
617  * function sets up a huge page only if any of the following conditions are met:
618  *
619  * - MTRRs are disabled, or
620  *
621  * - MTRRs are enabled and the range is completely covered by a single MTRR, or
622  *
623  * - MTRRs are enabled and the corresponding MTRR memory type is WB, which
624  *   has no effect on the requested PAT memory type.
625  *
626  * Callers should try to decrease page size (1GB -> 2MB -> 4K) if the bigger
627  * page mapping attempt fails.
628  *
629  * Returns 1 on success and 0 on failure.
630  */
631 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
632 {
633         u8 mtrr, uniform;
634 
635         mtrr = mtrr_type_lookup(addr, addr + PUD_SIZE, &uniform);
636         if ((mtrr != MTRR_TYPE_INVALID) && (!uniform) &&
637             (mtrr != MTRR_TYPE_WRBACK))
638                 return 0;
639 
640         /* Bail out if we are we on a populated non-leaf entry: */
641         if (pud_present(*pud) && !pud_huge(*pud))
642                 return 0;
643 
644         prot = pgprot_4k_2_large(prot);
645 
646         set_pte((pte_t *)pud, pfn_pte(
647                 (u64)addr >> PAGE_SHIFT,
648                 __pgprot(pgprot_val(prot) | _PAGE_PSE)));
649 
650         return 1;
651 }
652 
653 /**
654  * pmd_set_huge - setup kernel PMD mapping
655  *
656  * See text over pud_set_huge() above.
657  *
658  * Returns 1 on success and 0 on failure.
659  */
660 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
661 {
662         u8 mtrr, uniform;
663 
664         mtrr = mtrr_type_lookup(addr, addr + PMD_SIZE, &uniform);
665         if ((mtrr != MTRR_TYPE_INVALID) && (!uniform) &&
666             (mtrr != MTRR_TYPE_WRBACK)) {
667                 pr_warn_once("%s: Cannot satisfy [mem %#010llx-%#010llx] with a huge-page mapping due to MTRR override.\n",
668                              __func__, addr, addr + PMD_SIZE);
669                 return 0;
670         }
671 
672         /* Bail out if we are we on a populated non-leaf entry: */
673         if (pmd_present(*pmd) && !pmd_huge(*pmd))
674                 return 0;
675 
676         prot = pgprot_4k_2_large(prot);
677 
678         set_pte((pte_t *)pmd, pfn_pte(
679                 (u64)addr >> PAGE_SHIFT,
680                 __pgprot(pgprot_val(prot) | _PAGE_PSE)));
681 
682         return 1;
683 }
684 
685 /**
686  * pud_clear_huge - clear kernel PUD mapping when it is set
687  *
688  * Returns 1 on success and 0 on failure (no PUD map is found).
689  */
690 int pud_clear_huge(pud_t *pud)
691 {
692         if (pud_large(*pud)) {
693                 pud_clear(pud);
694                 return 1;
695         }
696 
697         return 0;
698 }
699 
700 /**
701  * pmd_clear_huge - clear kernel PMD mapping when it is set
702  *
703  * Returns 1 on success and 0 on failure (no PMD map is found).
704  */
705 int pmd_clear_huge(pmd_t *pmd)
706 {
707         if (pmd_large(*pmd)) {
708                 pmd_clear(pmd);
709                 return 1;
710         }
711 
712         return 0;
713 }
714 
715 /**
716  * pud_free_pmd_page - Clear pud entry and free pmd page.
717  * @pud: Pointer to a PUD.
718  *
719  * Context: The pud range has been unmaped and TLB purged.
720  * Return: 1 if clearing the entry succeeded. 0 otherwise.
721  */
722 int pud_free_pmd_page(pud_t *pud)
723 {
724         pmd_t *pmd;
725         int i;
726 
727         if (pud_none(*pud))
728                 return 1;
729 
730         pmd = (pmd_t *)pud_page_vaddr(*pud);
731 
732         for (i = 0; i < PTRS_PER_PMD; i++)
733                 if (!pmd_free_pte_page(&pmd[i]))
734                         return 0;
735 
736         pud_clear(pud);
737         free_page((unsigned long)pmd);
738 
739         return 1;
740 }
741 
742 /**
743  * pmd_free_pte_page - Clear pmd entry and free pte page.
744  * @pmd: Pointer to a PMD.
745  *
746  * Context: The pmd range has been unmaped and TLB purged.
747  * Return: 1 if clearing the entry succeeded. 0 otherwise.
748  */
749 int pmd_free_pte_page(pmd_t *pmd)
750 {
751         pte_t *pte;
752 
753         if (pmd_none(*pmd))
754                 return 1;
755 
756         pte = (pte_t *)pmd_page_vaddr(*pmd);
757         pmd_clear(pmd);
758         free_page((unsigned long)pte);
759 
760         return 1;
761 }
762 #endif  /* CONFIG_HAVE_ARCH_HUGE_VMAP */
763 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp