Lines Matching +full:tlb +full:- +full:split
1 // SPDX-License-Identifier: GPL-2.0
13 #include <linux/page-isolation.h>
26 #include <linux/backing-dev.h>
33 #include <asm/tlb.h>
38 struct mmu_gather *tlb; member
43 * Any behaviour which results in changes to the vma->vm_flags needs to
66 * We can potentially split a vm area into separate
73 struct mm_struct *mm = vma->vm_mm; in madvise_behavior()
76 unsigned long new_flags = vma->vm_flags; in madvise_behavior()
92 if (vma->vm_flags & VM_IO) { in madvise_behavior()
93 error = -EINVAL; in madvise_behavior()
100 if (vma->vm_file || vma->vm_flags & VM_SHARED) { in madvise_behavior()
101 error = -EINVAL; in madvise_behavior()
114 error = -EINVAL; in madvise_behavior()
133 if (new_flags == vma->vm_flags) { in madvise_behavior()
138 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); in madvise_behavior()
139 *prev = vma_merge(mm, *prev, start, end, new_flags, vma->anon_vma, in madvise_behavior()
140 vma->vm_file, pgoff, vma_policy(vma), in madvise_behavior()
141 vma->vm_userfaultfd_ctx); in madvise_behavior()
149 if (start != vma->vm_start) { in madvise_behavior()
150 if (unlikely(mm->map_count >= sysctl_max_map_count)) { in madvise_behavior()
151 error = -ENOMEM; in madvise_behavior()
159 if (end != vma->vm_end) { in madvise_behavior()
160 if (unlikely(mm->map_count >= sysctl_max_map_count)) { in madvise_behavior()
161 error = -ENOMEM; in madvise_behavior()
173 vma->vm_flags = new_flags; in madvise_behavior()
180 if (error == -ENOMEM) in madvise_behavior()
181 error = -EAGAIN; in madvise_behavior()
191 struct vm_area_struct *vma = walk->private; in swapin_walk_pmd_entry()
203 orig_pte = pte_offset_map_lock(vma->vm_mm, pmd, start, &ptl); in swapin_walk_pmd_entry()
204 pte = *(orig_pte + ((index - start) / PAGE_SIZE)); in swapin_walk_pmd_entry()
230 XA_STATE(xas, &mapping->i_pages, linear_page_index(vma, start)); in force_shm_swapin_readahead()
231 pgoff_t end_index = linear_page_index(vma, end + PAGE_SIZE - 1); in force_shm_swapin_readahead()
264 struct mm_struct *mm = vma->vm_mm; in madvise_willneed()
265 struct file *file = vma->vm_file; in madvise_willneed()
271 walk_page_range(vma->vm_mm, start, end, &swapin_walk_ops, vma); in madvise_willneed()
276 if (shmem_mapping(file->f_mapping)) { in madvise_willneed()
278 file->f_mapping); in madvise_willneed()
283 return -EBADF; in madvise_willneed()
299 offset = (loff_t)(start - vma->vm_start) in madvise_willneed()
300 + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); in madvise_willneed()
302 vfs_fadvise(file, offset, end - start, POSIX_FADV_WILLNEED); in madvise_willneed()
312 struct madvise_walk_private *private = walk->private; in madvise_cold_or_pageout_pte_range()
313 struct mmu_gather *tlb = private->tlb; in madvise_cold_or_pageout_pte_range() local
314 bool pageout = private->pageout; in madvise_cold_or_pageout_pte_range()
315 struct mm_struct *mm = tlb->mm; in madvise_cold_or_pageout_pte_range()
316 struct vm_area_struct *vma = walk->vma; in madvise_cold_or_pageout_pte_range()
323 return -EINTR; in madvise_cold_or_pageout_pte_range()
330 tlb_change_page_size(tlb, HPAGE_PMD_SIZE); in madvise_cold_or_pageout_pte_range()
351 if (next - addr != HPAGE_PMD_SIZE) { in madvise_cold_or_pageout_pte_range()
370 tlb_remove_pmd_tlb_entry(tlb, pmd, addr); in madvise_cold_or_pageout_pte_range()
380 list_add(&page->lru, &page_list); in madvise_cold_or_pageout_pte_range()
395 tlb_change_page_size(tlb, PAGE_SIZE); in madvise_cold_or_pageout_pte_range()
396 orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); in madvise_cold_or_pageout_pte_range()
413 * Creating a THP page is expensive so split it only if we in madvise_cold_or_pageout_pte_range()
414 * are sure it's worth. Split it if we are only owner. in madvise_cold_or_pageout_pte_range()
434 pte--; in madvise_cold_or_pageout_pte_range()
435 addr -= PAGE_SIZE; in madvise_cold_or_pageout_pte_range()
447 tlb->fullmm); in madvise_cold_or_pageout_pte_range()
450 tlb_remove_tlb_entry(tlb, pte, addr); in madvise_cold_or_pageout_pte_range()
456 * As a side effect, it makes confuse idle-page tracking in madvise_cold_or_pageout_pte_range()
466 list_add(&page->lru, &page_list); in madvise_cold_or_pageout_pte_range()
485 static void madvise_cold_page_range(struct mmu_gather *tlb, in madvise_cold_page_range() argument
491 .tlb = tlb, in madvise_cold_page_range()
494 tlb_start_vma(tlb, vma); in madvise_cold_page_range()
495 walk_page_range(vma->vm_mm, addr, end, &cold_walk_ops, &walk_private); in madvise_cold_page_range()
496 tlb_end_vma(tlb, vma); in madvise_cold_page_range()
503 struct mm_struct *mm = vma->vm_mm; in madvise_cold()
504 struct mmu_gather tlb; in madvise_cold() local
508 return -EINVAL; in madvise_cold()
511 tlb_gather_mmu(&tlb, mm); in madvise_cold()
512 madvise_cold_page_range(&tlb, vma, start_addr, end_addr); in madvise_cold()
513 tlb_finish_mmu(&tlb); in madvise_cold()
518 static void madvise_pageout_page_range(struct mmu_gather *tlb, in madvise_pageout_page_range() argument
524 .tlb = tlb, in madvise_pageout_page_range()
527 tlb_start_vma(tlb, vma); in madvise_pageout_page_range()
528 walk_page_range(vma->vm_mm, addr, end, &cold_walk_ops, &walk_private); in madvise_pageout_page_range()
529 tlb_end_vma(tlb, vma); in madvise_pageout_page_range()
536 if (!vma->vm_file) in can_do_pageout()
539 * paging out pagecache only for non-anonymous mappings that correspond in can_do_pageout()
541 * otherwise we'd be including shared non-exclusive mappings, which in can_do_pageout()
545 file_inode(vma->vm_file)) || in can_do_pageout()
546 file_permission(vma->vm_file, MAY_WRITE) == 0; in can_do_pageout()
553 struct mm_struct *mm = vma->vm_mm; in madvise_pageout()
554 struct mmu_gather tlb; in madvise_pageout() local
558 return -EINVAL; in madvise_pageout()
564 tlb_gather_mmu(&tlb, mm); in madvise_pageout()
565 madvise_pageout_page_range(&tlb, vma, start_addr, end_addr); in madvise_pageout()
566 tlb_finish_mmu(&tlb); in madvise_pageout()
575 struct mmu_gather *tlb = walk->private; in madvise_free_pte_range() local
576 struct mm_struct *mm = tlb->mm; in madvise_free_pte_range()
577 struct vm_area_struct *vma = walk->vma; in madvise_free_pte_range()
586 if (madvise_free_huge_pmd(tlb, vma, pmd, addr, next)) in madvise_free_pte_range()
592 tlb_change_page_size(tlb, PAGE_SIZE); in madvise_free_pte_range()
603 * prevent swap-in which is more expensive rather than in madvise_free_pte_range()
612 nr_swap--; in madvise_free_pte_range()
614 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); in madvise_free_pte_range()
624 * is owned by only this process, split it and in madvise_free_pte_range()
645 pte--; in madvise_free_pte_range()
646 addr -= PAGE_SIZE; in madvise_free_pte_range()
675 * Some of architecture(ex, PPC) don't update TLB in madvise_free_pte_range()
681 tlb->fullmm); in madvise_free_pte_range()
686 tlb_remove_tlb_entry(tlb, pte, addr); in madvise_free_pte_range()
692 if (current->mm == mm) in madvise_free_pte_range()
711 struct mm_struct *mm = vma->vm_mm; in madvise_free_single_vma()
713 struct mmu_gather tlb; in madvise_free_single_vma() local
717 return -EINVAL; in madvise_free_single_vma()
719 range.start = max(vma->vm_start, start_addr); in madvise_free_single_vma()
720 if (range.start >= vma->vm_end) in madvise_free_single_vma()
721 return -EINVAL; in madvise_free_single_vma()
722 range.end = min(vma->vm_end, end_addr); in madvise_free_single_vma()
723 if (range.end <= vma->vm_start) in madvise_free_single_vma()
724 return -EINVAL; in madvise_free_single_vma()
729 tlb_gather_mmu(&tlb, mm); in madvise_free_single_vma()
733 tlb_start_vma(&tlb, vma); in madvise_free_single_vma()
734 walk_page_range(vma->vm_mm, range.start, range.end, in madvise_free_single_vma()
735 &madvise_free_walk_ops, &tlb); in madvise_free_single_vma()
736 tlb_end_vma(&tlb, vma); in madvise_free_single_vma()
738 tlb_finish_mmu(&tlb); in madvise_free_single_vma()
765 zap_page_range(vma, start, end - start); in madvise_dontneed_single_vma()
774 struct mm_struct *mm = vma->vm_mm; in madvise_dontneed_free()
778 return -EINVAL; in madvise_dontneed_free()
786 return -ENOMEM; in madvise_dontneed_free()
787 if (start < vma->vm_start) { in madvise_dontneed_free()
790 * with the lowest vma->vm_start where start in madvise_dontneed_free()
791 * is also < vma->vm_end. If start < in madvise_dontneed_free()
792 * vma->vm_start it means an hole materialized in madvise_dontneed_free()
797 return -ENOMEM; in madvise_dontneed_free()
800 return -EINVAL; in madvise_dontneed_free()
801 if (end > vma->vm_end) { in madvise_dontneed_free()
803 * Don't fail if end > vma->vm_end. If the old in madvise_dontneed_free()
804 * vma was split while the mmap_lock was in madvise_dontneed_free()
811 * end-vma->vm_end range, but the manager can in madvise_dontneed_free()
814 end = vma->vm_end; in madvise_dontneed_free()
824 return -EINVAL; in madvise_dontneed_free()
833 struct mm_struct *mm = vma->vm_mm; in madvise_populate()
843 * our VMA might have been split. in madvise_populate()
845 if (!vma || start >= vma->vm_end) { in madvise_populate()
847 if (!vma || start < vma->vm_start) in madvise_populate()
848 return -ENOMEM; in madvise_populate()
851 tmp_end = min_t(unsigned long, end, vma->vm_end); in madvise_populate()
863 case -EINTR: in madvise_populate()
864 return -EINTR; in madvise_populate()
865 case -EINVAL: /* Incompatible mappings / permissions. */ in madvise_populate()
866 return -EINVAL; in madvise_populate()
867 case -EHWPOISON: in madvise_populate()
868 return -EHWPOISON; in madvise_populate()
869 case -EFAULT: /* VM_FAULT_SIGBUS or VM_FAULT_SIGSEGV */ in madvise_populate()
870 return -EFAULT; in madvise_populate()
875 case -ENOMEM: in madvise_populate()
876 return -ENOMEM; in madvise_populate()
895 struct mm_struct *mm = vma->vm_mm; in madvise_remove()
899 if (vma->vm_flags & VM_LOCKED) in madvise_remove()
900 return -EINVAL; in madvise_remove()
902 f = vma->vm_file; in madvise_remove()
904 if (!f || !f->f_mapping || !f->f_mapping->host) { in madvise_remove()
905 return -EINVAL; in madvise_remove()
908 if ((vma->vm_flags & (VM_SHARED|VM_WRITE)) != (VM_SHARED|VM_WRITE)) in madvise_remove()
909 return -EACCES; in madvise_remove()
911 offset = (loff_t)(start - vma->vm_start) in madvise_remove()
912 + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); in madvise_remove()
927 offset, end - start); in madvise_remove()
943 return -EPERM; in madvise_inject_error()
1063 * use appropriate read-ahead and caching techniques. The information
1068 * MADV_NORMAL - the default behavior is to read clusters. This
1069 * results in some read-ahead and read-behind.
1070 * MADV_RANDOM - the system should read the minimum amount of data
1071 * on any access, since it is unlikely that the appli-
1073 * MADV_SEQUENTIAL - pages in the given range will probably be accessed
1076 * MADV_WILLNEED - the application is notifying the system to read
1078 * MADV_DONTNEED - the application is finished with the given range,
1080 * MADV_FREE - the application marks pages in the given range as lazy free,
1082 * MADV_REMOVE - the application wants to free up the given range of
1084 * MADV_DONTFORK - omit this area from child's address space when forking:
1086 * MADV_DOFORK - cancel MADV_DONTFORK: no longer omit this area when forking.
1087 * MADV_WIPEONFORK - present the child process with zero-filled memory in this
1089 * MADV_KEEPONFORK - undo the effect of MADV_WIPEONFORK
1090 * MADV_HWPOISON - trigger memory error handler as if the given memory range
1092 * MADV_SOFT_OFFLINE - try to soft-offline the given range of memory.
1093 * MADV_MERGEABLE - the application recommends that KSM try to merge pages in
1095 * MADV_UNMERGEABLE- cancel MADV_MERGEABLE: no longer merge pages with others.
1096 * MADV_HUGEPAGE - the application wants to back the given range by transparent
1099 * MADV_NOHUGEPAGE - mark the given range as not worth being backed by
1102 * MADV_DONTDUMP - the application wants to prevent pages in the given range
1104 * MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump.
1105 * MADV_COLD - the application is not expected to use this memory soon,
1108 * MADV_PAGEOUT - the application is not expected to use this memory soon,
1110 * MADV_POPULATE_READ - populate (prefault) page tables readable by
1112 * MADV_POPULATE_WRITE - populate (prefault) page tables writable by
1116 * zero - success
1117 * -EINVAL - start + len < 0, start is not page-aligned,
1120 * or the specified address range includes file, Huge TLB,
1122 * -ENOMEM - addresses in the specified range are not currently
1124 * -EIO - an I/O error occurred while paging in data.
1125 * -EBADF - map exists, but area maps something that isn't a file.
1126 * -EAGAIN - a kernel resource was temporarily unavailable.
1133 int error = -EINVAL; in do_madvise()
1147 /* Check to see whether len was rounded up from small -ve to zero */ in do_madvise()
1167 return -EINTR; in do_madvise()
1174 * ranges, just ignore them, but return -ENOMEM at the end. in do_madvise()
1175 * - different from the way of handling in mlock etc. in do_madvise()
1178 if (vma && start > vma->vm_start) in do_madvise()
1184 error = -ENOMEM; in do_madvise()
1188 /* Here start < (end|vma->vm_end). */ in do_madvise()
1189 if (start < vma->vm_start) { in do_madvise()
1190 unmapped_error = -ENOMEM; in do_madvise()
1191 start = vma->vm_start; in do_madvise()
1196 /* Here vma->vm_start <= start < (end|vma->vm_end) */ in do_madvise()
1197 tmp = vma->vm_end; in do_madvise()
1201 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */ in do_madvise()
1206 if (prev && start < prev->vm_end) in do_madvise()
1207 start = prev->vm_end; in do_madvise()
1212 vma = prev->vm_next; in do_madvise()
1228 return do_madvise(current->mm, start, len_in, behavior); in SYSCALL_DEFINE3()
1245 ret = -EINVAL; in SYSCALL_DEFINE5()
1261 ret = -ESRCH; in SYSCALL_DEFINE5()
1266 ret = -EINVAL; in SYSCALL_DEFINE5()
1273 ret = IS_ERR(mm) ? PTR_ERR(mm) : -ESRCH; in SYSCALL_DEFINE5()
1279 * only non-destructive hints are currently supported. in SYSCALL_DEFINE5()
1282 ret = -EPERM; in SYSCALL_DEFINE5()
1298 ret = total_len - iov_iter_count(&iter); in SYSCALL_DEFINE5()