Page Table Flags
一共都有那些 flags ?
pte ，还可以 huge 吗?
static inline pte_t pte_mkhuge(pte_t pte)
{
return pte_set_flags(pte, _PAGE_PSE);
}
为什么要这么实现?
pte_t pte_mkwrite(pte_t pte, struct vm_area_struct *vma)
{
if (vma->vm_flags & VM_SHADOW_STACK) // 先不看这个 security 的东西了
return pte_mkwrite_shstk(pte);
pte = pte_mkwrite_novma(pte); // 什么叫做 novma ?
return pte_clear_saveddirty(pte); // 多此一举吗?
}
看看 arch/x86/include/asm/pgtable.h
static inline pte_t pte_mksaveddirty(pte_t pte)
{
pteval_t v = native_pte_val(pte);
v = mksaveddirty_shift(v);
return native_make_pte(v);
}
pte_mkdirty
这个容易理解:
/*
* Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
* servicing faults for write access. In the normal case, do always want
* pte_mkwrite. But get_user_pages can cause write faults for mappings
* that do not have writing enabled, when used by access_process_vm.
*/
static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
{
if (likely(vma->vm_flags & VM_WRITE))
pte = pte_mkwrite(pte, vma);
return pte;
}
pte_mkdirty 的调用位置主要是在 page fault 当分配新的 page 的时候， 立刻标记上 dirty
static inline pte_t pte_mkdirty(pte_t pte)
{
pte = pte_set_flags(pte, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
return pte_mksaveddirty(pte);
}
要记住，之所以可以 pte_mkdirty ，是由于
@[
pte_mkdirty+101
do_swap_page+1740
handle_mm_fault+2151
do_user_addr_fault+780
exc_page_fault+117
asm_exc_page_fault+38
copy_fpstate_to_sigframe+426
get_sigframe+442
x64_setup_rt_frame+121
arch_do_signal_or_restart+306
syscall_exit_to_user_mode+85
do_syscall_64+250
entry_SYSCALL_64_after_hwframe+119
]: 1
@[
wp_page_reuse+203
do_wp_page+1672
handle_mm_fault+2657
do_user_addr_fault+780
exc_page_fault+117
asm_exc_page_fault+38
filldir64+208
ext4_readdir+2428
iterate_dir+123
__se_sys_getdents64+105
do_syscall_64+237
entry_SYSCALL_64_after_hwframe+119
]: 2
pte_mkclean : clean 的位置
- folio_clear_dirty_for_io
- folio_mkclean : 这里需要反向映射，将映射该 page 的所有的 pte 都 clean 一下
任何时候，都不会给 anonymous page 来 mkclean 吧 应该是的，使用 stress-ng 制作 swap ，只能观测到
🧀 sudo bpftrace -e "tracepoint:kmem:hi { @[kstack] = count(); }"
Attaching 1 probe...
^C
@[
xueshi+73
page_vma_mkclean_one+231
page_mkclean_one+142
rmap_walk_file+307
folio_mkclean+184
folio_clear_dirty_for_io+93
mpage_prepare_extent_to_map+704
ext4_do_writepages+865
ext4_normal_submit_inode_data_buffers+238
jbd2_journal_commit_transaction+1242
kjournald2+176
kthread+248
ret_from_fork+55
ret_from_fork_asm+26
]: 8
folio_mark_dirty 对比分析
这调用位置，只能说，只能说，这些调用位置都是预期之外的:
@[
folio_mark_dirty+5
fault_dirty_shared_page+74 // 被 do_shared_fault 和 wp_page_shared 无条件调用
do_pte_missing+1014
handle_mm_fault+2183
__get_user_pages+1100
get_user_pages_unlocked+264
hva_to_pfn+264
gfn_to_page+14
kvm_alloc_apic_access_page+101
vmx_vcpu_create+565
kvm_arch_vcpu_create+471
kvm_vm_ioctl_create_vcpu+348
kvm_vm_ioctl+1804
__se_sys_ioctl+110
do_syscall_64+237
entry_SYSCALL_64_after_hwframe+119
]: 1
@[
folio_mark_dirty+5
filemap_page_mkwrite+349
do_wp_page+2066
handle_mm_fault+2657
do_user_addr_fault+504
exc_page_fault+117
asm_exc_page_fault+38
]: 1
@[
folio_mark_dirty+5 // 为什么 unmap 的也需要 ?
unmap_page_range+2762
unmap_vmas+241
exit_mmap+498
__mmput+67
exit_mm+223
do_exit+512
do_group_exit+140
__x64_sys_exit_group+23
__pfx_ia32_emulation_override_cmdline+0
do_syscall_64+237
entry_SYSCALL_64_after_hwframe+119
]: 2
@[
folio_mark_dirty+5
unmap_page_range+4535
unmap_vmas+241
unmap_region+338
do_vmi_align_munmap+946
do_vmi_munmap+263
__vm_munmap+199
__x64_sys_munmap+27
do_syscall_64+237
entry_SYSCALL_64_after_hwframe+119
]: 3
@[
folio_mark_dirty+5
shmem_swapin_folio+1047
shmem_get_folio_gfp+151
shmem_fault+134
__do_fault+67
do_pte_missing+354
handle_mm_fault+2183
do_user_addr_fault+504
exc_page_fault+117
asm_exc_page_fault+38
]: 4
@[
folio_mark_dirty+5
fault_dirty_shared_page+74
do_pte_missing+1014
handle_mm_fault+2183
do_user_addr_fault+780
exc_page_fault+117
asm_exc_page_fault+38
]: 5
@[
folio_mark_dirty+5 // .. swap 进来，那么立刻标记下
shmem_swapin_folio+1047
shmem_get_folio_gfp+151
shmem_fault+134
__do_fault+67
do_pte_missing+831
handle_mm_fault+2183
do_user_addr_fault+504
exc_page_fault+117
asm_exc_page_fault+38
]: 7
@[
folio_mark_dirty+5
unmap_page_range+2762
unmap_vmas+241
exit_mmap+498
__mmput+67
exit_mm+223
do_exit+290
do_group_exit+140
__x64_sys_exit_group+23
__pfx_ia32_emulation_override_cmdline+0
do_syscall_64+237
entry_SYSCALL_64_after_hwframe+119
]: 8
@[
folio_mark_dirty+5
unmap_page_range+2762
unmap_vmas+241
exit_mmap+498
__mmput+67
exit_mm+223
do_exit+512
do_group_exit+120
get_signal+1710
arch_do_signal_or_restart+142
irqentry_exit_to_user_mode+75
asm_sysvec_reschedule_ipi+26
]: 16
@[
folio_mark_dirty+5
shmem_write_end+367
generic_perform_write+425
shmem_file_write_iter+107
vfs_write+935
ksys_write+114
do_syscall_64+237
entry_SYSCALL_64_after_hwframe+119
]: 18
@[
folio_mark_dirty+5
unmap_page_range+2762
unmap_vmas+241
exit_mmap+498
__mmput+67
exit_mm+223
do_exit+512
vhost_task_fn+285
ret_from_fork+55
ret_from_fork_asm+26
]: 31
@[
folio_mark_dirty+5
unmap_page_range+2762
unmap_vmas+241
unmap_region+338
do_vmi_align_munmap+946
do_vmi_munmap+263
__vm_munmap+199
__x64_sys_munmap+27
do_syscall_64+237
entry_SYSCALL_64_after_hwframe+119
]: 37
@[
folio_mark_dirty+5
bio_set_pages_dirty+224
iomap_dio_bio_iter+786
__iomap_dio_rw+680
iomap_dio_rw+18
ext4_file_read_iter+234
vfs_read+750
__x64_sys_pread64+109
do_syscall_64+237
entry_SYSCALL_64_after_hwframe+119
]: 37
@[
folio_mark_dirty+5
fault_dirty_shared_page+74
do_pte_missing+1014
handle_mm_fault+2183
do_user_addr_fault+504
exc_page_fault+117
asm_exc_page_fault+38
]: 49
@[
folio_mark_dirty+5
unmap_page_range+2762
unmap_vmas+241
exit_mmap+498
__mmput+67
exit_mm+223
do_exit+512
do_group_exit+120
get_signal+1710
arch_do_signal_or_restart+142
syscall_exit_to_user_mode+85
do_syscall_64+250
entry_SYSCALL_64_after_hwframe+119
]: 82
@[
folio_mark_dirty+5
block_page_mkwrite+344
ext4_page_mkwrite+851
do_wp_page+2066
handle_mm_fault+2657
do_user_addr_fault+504
exc_page_fault+117
asm_exc_page_fault+38
]: 132
@[
folio_mark_dirty+5
fault_dirty_shared_page+74
do_wp_page+2124
handle_mm_fault+2657
do_user_addr_fault+504
exc_page_fault+117
asm_exc_page_fault+38
]: 133
@[
folio_mark_dirty+5
try_to_migrate_one+906
rmap_walk_anon+319
try_to_migrate+214
migrate_pages_batch+944
migrate_pages+1911
compact_zone+3555
compact_node+195
kcompactd+1397
kthread+248
ret_from_fork+55
ret_from_fork_asm+26
]: 1945
@[
folio_mark_dirty+5 // 都 dio ，为什么还需要 mark ? 看看 bio_set_pages_dirty 的 comment
bio_set_pages_dirty+224
iomap_dio_bio_iter+786
__iomap_dio_rw+680
iomap_dio_rw+18
ext4_file_read_iter+234
aio_read+456
io_submit_one+1608
__se_sys_io_submit+189
do_syscall_64+237
entry_SYSCALL_64_after_hwframe+119
]: 11920
从 pte dirty 到 folio dirty 的传播
几乎都发生到 rmap 中，非常合理
/* Set the dirty flag on the folio now the pte is gone. */
if (pte_dirty(pteval))
folio_mark_dirty(folio);
PG_dirty 的作用
案例 1 : 从 block_dirty_folio 中可以看到，检查了有这个 flag ，如果是 newly dirty ，那么才去 到 address space 中去标记这些东西:
if (newly_dirty)
__folio_mark_dirty(folio, mapping, 1);
本站所有文章转发 CSDN 将按侵权追究法律责任，其它情况随意。