page-io.c
swap io
read swap 是一个不同的行为
#0 folio_wait_bit_common (folio=0xffffea000100c700, bit_nr=0, state=258, behavior=SHARED) at mm/filemap.c:1220
#1 0xffffffff812e37b1 in folio_wait_bit_killable (bit_nr=0, folio=0xffffea000100c700) at mm/filemap.c:1447
#2 folio_wait_locked_killable (folio=0xffffea000100c700) at ./include/linux/pagemap.h:1028
#3 folio_wait_locked_killable (folio=0xffffea000100c700) at ./include/linux/pagemap.h:1024
#4 __folio_lock_or_retry (folio=folio@entry=0xffffea000100c700, mm=mm@entry=0xffff8881401cc740, flags=flags@entry=2644) at mm/filemap.c:1722
#5 0xffffffff81325984 in folio_lock_or_retry (flags=2644, mm=<optimized out>, folio=0xffffea000100c700) at ./include/linux/pagemap.h:1001
#6 do_swap_page (vmf=vmf@entry=0xffffc900404f3df8) at mm/memory.c:3815
#7 0xffffffff8132a27e in handle_pte_fault (vmf=0xffffc900404f3df8) at mm/memory.c:4935
#8 __handle_mm_fault (vma=vma@entry=0xffff888118dc2428, address=address@entry=139787038240768, flags=flags@entry=596) at mm/memory.c:5073
#9 0xffffffff8132ae24 in handle_mm_fault (vma=0xffff888118dc2428, address=address@entry=139787038240768, flags=<optimized out>, flags@entry=596, regs=regs@entry=0xffffc900404f3f58) at mm/memory.c:5219
#10 0xffffffff8110d947 in do_user_addr_fault (regs=regs@entry=0xffffc900404f3f58, error_code=error_code@entry=4, address=address@entry=139787038240768) at arch/x86/mm/fault.c:1428
#11 0xffffffff82184606 in handle_page_fault (address=139787038240768, error_code=4, regs=0xffffc900404f3f58) at arch/x86/mm/fault.c:1519
#12 exc_page_fault (regs=0xffffc900404f3f58, error_code=4) at arch/x86/mm/fault.c:1575
#13 0xffffffff82201286 in asm_exc_page_fault () at ./arch/x86/include/asm/idtentry.h:570
#0 folio_wake_bit (folio=0xffffea00010051c0, bit_nr=bit_nr@entry=0) at mm/filemap.c:1130
#1 0xffffffff812ddbf6 in folio_unlock (folio=<optimized out>) at mm/filemap.c:1527
#2 0xffffffff81357945 in end_swap_bio_read (bio=0xffff88803a2ffe00) at mm/page_io.c:71
#3 0xffffffff816ccd4c in req_bio_endio (error=0 '\000', nbytes=4096, bio=0xffff88803a2ffe00, rq=0xffff88801b73e180) at block/blk-mq.c:794
#4 blk_update_request (req=req@entry=0xffff88801b73e180, error=error@entry=0 '\000', nr_bytes=4096, nr_bytes@entry=8192) at block/blk-mq.c:926
#5 0xffffffff81acd5a2 in scsi_end_request (req=req@entry=0xffff88801b73e180, error=error@entry=0 '\000', bytes=bytes@entry=8192) at drivers/scsi/scsi_lib.c:539
#6 0xffffffff81ace089 in scsi_io_completion (cmd=0xffff88801b73e290, good_bytes=8192) at drivers/scsi/scsi_lib.c:977
#7 0xffffffff816c99f8 in blk_complete_reqs (list=<optimized out>) at block/blk-mq.c:1131
#8 0xffffffff82196e34 in __do_softirq () at kernel/softirq.c:571
#9 0xffffffff811318aa in invoke_softirq () at kernel/softirq.c:445
#10 __irq_exit_rcu () at kernel/softirq.c:650
#11 0xffffffff82183fa6 in sysvec_apic_timer_interrupt (regs=0xffffffff82c03df8) at arch/x86/kernel/apic/apic.c:1107
如果出现错误 do_swap_page 可以走到这个位置上去，
if (unlikely(!folio_test_uptodate(folio))) {
ret = VM_FAULT_SIGBUS;
goto out_nomap;
}
如果 end_swap_bio_read 失败了:
static void end_swap_bio_read(struct bio *bio)
{
struct page *page = bio_first_page_all(bio);
struct task_struct *waiter = bio->bi_private;
if (bio->bi_status) {
SetPageError(page);
ClearPageUptodate(page);
pr_alert_ratelimited("Read-error on swap-device (%u:%u:%llu)\n",
MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
(unsigned long long)bio->bi_iter.bi_sector);
goto out;
}
SetPageUptodate(page);
out:
unlock_page(page);
WRITE_ONCE(bio->bi_private, NULL);
bio_put(bio);
if (waiter) {
blk_wake_io_task(waiter);
put_task_struct(waiter);
}
}
如果是采用 fault inject 机制注入错误，submit_bio_noacct 中直接跳转到最后，开始执行 bio_endio ， 进而调用到 end_swap_bio_read
如果是 scsi debug 机制的:
[ 49.130061] Hardware name: Martins3 Inc Hacking Alpine, BIOS 12 2022-2-2
[ 49.130373] Call Trace:
[ 49.130483] <IRQ>
[ 49.130578] dump_stack_lvl+0x38/0x4c
[ 49.130724] end_swap_bio_read.cold+0x35/0x3a
[ 49.130901] blk_update_request+0xfc/0x410
[ 49.131060] scsi_end_request+0x22/0x1b0
[ 49.131215] scsi_io_completion+0x4db/0x880
[ 49.131378] blk_complete_reqs+0x3b/0x50
[ 49.131532] __do_softirq+0xf7/0x301
[ 49.131672] __irq_exit_rcu+0xca/0x110
[ 49.131817] sysvec_apic_timer_interrupt+0xa6/0xd0
[ 49.132003] </IRQ>
所以，正常来说，只要是注入错误，都是要立刻挂掉的。
write swap
submit_bio_noacct 的位置本身就是一个异步的操作。
#0 swap_writepage (page=0xffffea00054bddc0, wbc=0xffffc9003f1ab8e0) at mm/page_io.c:184
#1 0xffffffff812f67d2 in pageout (folio=folio@entry=0xffffea00054bddc0, mapping=mapping@entry=0xffff88814743b600, plug=plug@entry=0xffffc9003f1ab9a8) at mm/vmscan.c:1298
#2 0xffffffff812f7c22 in shrink_folio_list (folio_list=folio_list@entry=0xffffc9003f1aba90, pgdat=pgdat@entry=0xffff88823fff9000, sc=sc@entry=0xffffc9003f1abc68, stat=stat@entry=0xffffc9003f1abb18, ignore_references=ignore_references@entry=false) at mm/vmscan.c:1947
#3 0xffffffff812f9944 in shrink_inactive_list (lru=LRU_INACTIVE_ANON, sc=0xffffc9003f1abc68, lruvec=0xffff8881022f2800, nr_to_scan=<optimized out>) at mm/vmscan.c:2526
#4 shrink_list (sc=0xffffc9003f1abc68, lruvec=0xffff8881022f2800, nr_to_scan=<optimized out>, lru=LRU_INACTIVE_ANON) at mm/vmscan.c:2767
#5 shrink_lruvec (lruvec=lruvec@entry=0xffff8881022f2800, sc=sc@entry=0xffffc9003f1abc68) at mm/vmscan.c:5951
#6 0xffffffff812fa20e in shrink_node_memcgs (sc=0xffffc9003f1abc68, pgdat=0xffff88823fff9000) at mm/vmscan.c:6138
#7 shrink_node (pgdat=pgdat@entry=0xffff88823fff9000, sc=sc@entry=0xffffc9003f1abc68) at mm/vmscan.c:6169
#8 0xffffffff812fb450 in shrink_zones (sc=0xffffc9003f1abc68, zonelist=<optimized out>) at mm/vmscan.c:6407
#9 do_try_to_free_pages (zonelist=zonelist@entry=0xffff88823fffab00, sc=sc@entry=0xffffc9003f1abc68) at mm/vmscan.c:6469
#10 0xffffffff812fc367 in try_to_free_mem_cgroup_pages (memcg=memcg@entry=0xffff8881022f8000, nr_pages=nr_pages@entry=1, gfp_mask=gfp_mask@entry=3264, reclaim_options=reclaim_options@entry=2, nodemask=nodemask@entry=0x0 <fixed_percpu_data>) at mm/vmscan.c:6786
#11 0xffffffff813a0bfa in try_charge_memcg (memcg=memcg@entry=0xffff8881022f8000, gfp_mask=gfp_mask@entry=3264, nr_pages=1) at mm/memcontrol.c:2687
#12 0xffffffff813a1b6d in try_charge (nr_pages=1, gfp_mask=3264, memcg=0xffff8881022f8000) at mm/memcontrol.c:2830
#13 charge_memcg (folio=folio@entry=0xffffea00052ceb00, memcg=memcg@entry=0xffff8881022f8000, gfp=gfp@entry=3264) at mm/memcontrol.c:6947
#14 0xffffffff813a3468 in __mem_cgroup_charge (folio=0xffffea00052ceb00, mm=<optimized out>, gfp=gfp@entry=3264) at mm/memcontrol.c:6968
#15 0xffffffff8132a69b in mem_cgroup_charge (gfp=3264, mm=<optimized out>, folio=<optimized out>) at ./include/linux/memcontrol.h:671
#16 do_anonymous_page (vmf=0xffffc9003f1abdf8) at mm/memory.c:4078
#17 handle_pte_fault (vmf=0xffffc9003f1abdf8) at mm/memory.c:4929
#18 __handle_mm_fault (vma=vma@entry=0xffff88812842de40, address=address@entry=139790768721920, flags=flags@entry=597) at mm/memory.c:5073
#19 0xffffffff8132b104 in handle_mm_fault (vma=0xffff88812842de40, address=address@entry=139790768721920, flags=<optimized out>, flags@entry=597, regs=regs@entry=0xffffc9003f1abf58) at mm/memory.c:5219
#20 0xffffffff8110d937 in do_user_addr_fault (regs=regs@entry=0xffffc9003f1abf58, error_code=error_code@entry=6, address=address@entry=139790768721920) at arch/x86/mm/fault.c:1428
#21 0xffffffff8218d606 in handle_page_fault (address=139790768721920, error_code=6, regs=0xffffc9003f1abf58) at arch/x86/mm/fault.c:1519
#22 exc_page_fault (regs=0xffffc9003f1abf58, error_code=6) at arch/x86/mm/fault.c:1575
#23 0xffffffff82201286 in asm_exc_page_fault () at ./arch/x86/include/asm/idtentry.h:570
最后，应该无论如何都不会卡到 blk wbt 上啊发，中间都在干什么哇？
- 似乎如果是普通的 readwrite 是立刻死掉的。
然后马上进入到 end_swap_bio_write 中，
- 最后经过超级长的时间，程序才会被 kill 掉。
是同步的吗？
- swap_writepage
__swap_writepage- bdev_write_page
- set_page_writeback
- submit_bio
- end_swap_bio_write
- end_page_writeback
- folio_wake(folio, PG_writeback); // 这个到底是在唤醒谁？
- folio_wait_writeback : 等待者
- folio_wait_writeback_killable
换一个角度思考: page-io.md 中，直接看看，有没有什么等待之类的：
#0 swap_writepage (page=0xffffea00054bddc0, wbc=0xffffc9003f1ab8e0) at mm/page_io.c:184
#1 0xffffffff812f67d2 in pageout (folio=folio@entry=0xffffea00054bddc0, mapping=mapping@entry=0xffff88814743b600, plug=plug@entry=0xffffc9003f1ab9a8) at mm/vmscan.c:1298
#2 0xffffffff812f7c22 in shrink_folio_list (folio_list=folio_list@entry=0xffffc9003f1aba90, pgdat=pgdat@entry=0xffff88823fff9000, sc=sc@entry=0xffffc9003f1abc68, stat=stat@entry=0xffffc9003f1abb18, ignore_references=ignore_references@entry=false) at mm/vmscan.c:1947
#3 0xffffffff812f9944 in shrink_inactive_list (lru=LRU_INACTIVE_ANON, sc=0xffffc9003f1abc68, lruvec=0xffff8881022f2800, nr_to_scan=<optimized out>) at mm/vmscan.c:2526
#4 shrink_list (sc=0xffffc9003f1abc68, lruvec=0xffff8881022f2800, nr_to_scan=<optimized out>, lru=LRU_INACTIVE_ANON) at mm/vmscan.c:2767
#5 shrink_lruvec (lruvec=lruvec@entry=0xffff8881022f2800, sc=sc@entry=0xffffc9003f1abc68) at mm/vmscan.c:5951
#6 0xffffffff812fa20e in shrink_node_memcgs (sc=0xffffc9003f1abc68, pgdat=0xffff88823fff9000) at mm/vmscan.c:6138
#7 shrink_node (pgdat=pgdat@entry=0xffff88823fff9000, sc=sc@entry=0xffffc9003f1abc68) at mm/vmscan.c:6169
#8 0xffffffff812fb450 in shrink_zones (sc=0xffffc9003f1abc68, zonelist=<optimized out>) at mm/vmscan.c:6407
#9 do_try_to_free_pages (zonelist=zonelist@entry=0xffff88823fffab00, sc=sc@entry=0xffffc9003f1abc68) at mm/vmscan.c:6469
#10 0xffffffff812fc367 in try_to_free_mem_cgroup_pages (memcg=memcg@entry=0xffff8881022f8000, nr_pages=nr_pages@entry=1, gfp_mask=gfp_mask@entry=3264, reclaim_options=reclaim_options@entry=2, nodemask=nodemask@entry=0x0 <fixed_percpu_data>) at mm/vmscan.c:6786
#11 0xffffffff813a0bfa in try_charge_memcg (memcg=memcg@entry=0xffff8881022f8000, gfp_mask=gfp_mask@entry=3264, nr_pages=1) at mm/memcontrol.c:2687
#12 0xffffffff813a1b6d in try_charge (nr_pages=1, gfp_mask=3264, memcg=0xffff8881022f8000) at mm/memcontrol.c:2830
#13 charge_memcg (folio=folio@entry=0xffffea00052ceb00, memcg=memcg@entry=0xffff8881022f8000, gfp=gfp@entry=3264) at mm/memcontrol.c:6947
#14 0xffffffff813a3468 in __mem_cgroup_charge (folio=0xffffea00052ceb00, mm=<optimized out>, gfp=gfp@entry=3264) at mm/memcontrol.c:6968
#15 0xffffffff8132a69b in mem_cgroup_charge (gfp=3264, mm=<optimized out>, folio=<optimized out>) at ./include/linux/memcontrol.h:671
#16 do_anonymous_page (vmf=0xffffc9003f1abdf8) at mm/memory.c:4078
#17 handle_pte_fault (vmf=0xffffc9003f1abdf8) at mm/memory.c:4929
#18 __handle_mm_fault (vma=vma@entry=0xffff88812842de40, address=address@entry=139790768721920, flags=flags@entry=597) at mm/memory.c:5073
#19 0xffffffff8132b104 in handle_mm_fault (vma=0xffff88812842de40, address=address@entry=139790768721920, flags=<optimized out>, flags@entry=597, regs=regs@entry=0xffffc9003f1abf58) at mm/memory.c:5219
#20 0xffffffff8110d937 in do_user_addr_fault (regs=regs@entry=0xffffc9003f1abf58, error_code=error_code@entry=6, address=address@entry=139790768721920) at arch/x86/mm/fault.c:1428
#21 0xffffffff8218d606 in handle_page_fault (address=139790768721920, error_code=6, regs=0xffffc9003f1abf58) at arch/x86/mm/fault.c:1519
#22 exc_page_fault (regs=0xffffc9003f1abf58, error_code=6) at arch/x86/mm/fault.c:1575
#23 0xffffffff82201286 in asm_exc_page_fault () at ./arch/x86/include/asm/idtentry.h:570
分析普通的 io 的如何处理错误
完成 write 之后，是如何将 page 释放掉的?
因为一直在推迟 write 的，一直都是正常的，只有等到 sync 之后才会开始 write 的。
取决于是否为同步 IO，如果是，那么问题反映非常及时。
如果是 buffer io 的话 ，注入错误之后，需要 umount 一次，否则错误一直出现。
肯定是有问题的，如果同步 IO 出现错误是立刻的，那么在
[ ] wbt 延长了错误的出现时间而已，应该，总是会出现错误的
pageout 调用 address_space_operations::writepage，但是文件系统都是不注册这个 hook 的
- 等价于，dirty file 的 flush 是谁在做？
如何释放被 swap 的页
- end_swap_bio_write
- end_page_writeback : 这里的时候 folio 还是存在一个 reference count 的!
- 忽然意识到，这个 jb 玩意儿实际上是在 swap cache 中存在一个 referenced
- end_page_writeback : 这里的时候 folio 还是存在一个 reference count 的!
怎么有 hugepage 啊？
#0 free_swap_cache (page=page@entry=0xffffea00087ab740) at mm/swap_state.c:281
#1 0xffffffff8135961d in free_page_and_swap_cache (page=0xffffea00087ab740) at mm/swap_state.c:297
#2 0xffffffff8138f56b in __split_huge_page (end=18446744073709551615, list=0xffffea00087a8000, page=0x0 <fixed_percpu_data>) at mm/huge_memory.c:2613
#3 split_huge_page_to_list (page=page@entry=0xffffea00087a8000, list=list@entry=0xffffc90042bf7b90) at mm/huge_memory.c:2778
#4 0xffffffff812f801a in split_folio_to_list (list=0xffffc90042bf7b90, folio=0xffffea00087a8000) at ./include/linux/huge_mm.h:444
#5 shrink_folio_list (folio_list=folio_list@entry=0xffffc90042bf7b90, pgdat=pgdat@entry=0xffff88833fffc000, sc=sc@entry=0xffffc90042bf7d68, stat=stat@entry=0xffffc90042bf7c18, ignore_references=ignore_references@entry=false) at mm/vmscan.c:1856
#6 0xffffffff812f9944 in shrink_inactive_list (lru=LRU_INACTIVE_ANON, sc=0xffffc90042bf7d68, lruvec=0xffff8881036bd000, nr_to_scan=<optimized out>) at mm/vmscan.c:2526
#7 shrink_list (sc=0xffffc90042bf7d68, lruvec=0xffff8881036bd000, nr_to_scan=<optimized out>, lru=LRU_INACTIVE_ANON) at mm/vmscan.c:2767
#8 shrink_lruvec (lruvec=lruvec@entry=0xffff8881036bd000, sc=sc@entry=0xffffc90042bf7d68) at mm/vmscan.c:5951
#9 0xffffffff812fa20e in shrink_node_memcgs (sc=0xffffc90042bf7d68, pgdat=0xffff88833fffc000) at mm/vmscan.c:6138
#10 shrink_node (pgdat=pgdat@entry=0xffff88833fffc000, sc=sc@entry=0xffffc90042bf7d68) at mm/vmscan.c:6169
#11 0xffffffff812fb450 in shrink_zones (sc=0xffffc90042bf7d68, zonelist=<optimized out>) at mm/vmscan.c:6407
#12 do_try_to_free_pages (zonelist=zonelist@entry=0xffff8884bfff3700, sc=sc@entry=0xffffc90042bf7d68) at mm/vmscan.c:6469
#13 0xffffffff812fc367 in try_to_free_mem_cgroup_pages (memcg=memcg@entry=0xffff888358a97000, nr_pages=<optimized out>, gfp_mask=gfp_mask@entry=3264, reclaim_options=reclaim_options@entry=2, nodemask=nodemask@entry=0x0 <fixed_percpu_data>) at mm/vmscan.c:6786
#14 0xffffffff8139ecb2 in memory_max_write (of=<optimized out>, buf=<optimized out>, nbytes=5, off=<optimized out>) at mm/memcontrol.c:6481
#15 0xffffffff8146bbf4 in kernfs_fop_write_iter (iocb=0xffffc90042bf7ea0, iter=<optimized out>) at fs/kernfs/file.c:334
#16 0xffffffff813c634d in call_write_iter (iter=0x7f898e5ff, kio=0xffffea00087ab740, file=0xffff88810438f800) at ./include/linux/fs.h:2186
#17 new_sync_write (ppos=0xffffc90042bf7f08, len=5, buf=0x18f49f0 "1000m", filp=0xffff88810438f800) at fs/read_write.c:491
#18 vfs_write (file=file@entry=0xffff88810438f800, buf=buf@entry=0x18f49f0 "1000m", count=count@entry=5, pos=pos@entry=0xffffc90042bf7f08) at fs/read_write.c:584
#19 0xffffffff813c679e in ksys_write (fd=<optimized out>, buf=0x18f49f0 "1000m", count=5) at fs/read_write.c:637
#20 0xffffffff82188c1c in do_syscall_x64 (nr=<optimized out>, regs=0xffffc90042bf7f58) at arch/x86/entry/common.c:50
#21 do_syscall_64 (regs=0xffffc90042bf7f58, nr=<optimized out>) at arch/x86/entry/common.c:80
#22 0xffffffff822000ae in entry_SYSCALL_64 () at arch/x86/entry/entry_64.S:120
- add_to_swap_cache ?
如果 swap 正在 writeback 的过程中，对于该 page 进行了一次写，如何？
folio_mark_dirty 中会调用 folio_clear_reclaim，该 page 重新为 dirty 的，并且进入到 lru 的队列尾部。
如果 swap 正在 writeback 的过程中，对于该 page 进行了一次读，如何？
感觉这会导致，page 进入到 lru 队列尾。但是这会产生一个 swap cache 吗?
[ ] SetPageError 为什么是没有结果的?
static __always_inline bool folio_test_error(struct folio *folio) {
return test_bit(PG_error, folio_flags(folio, FOLIO_PF_NO_TAIL));
}
static __always_inline int PageError(struct page *page) {
return test_bit(PG_error, &PF_NO_TAIL(page, 0)->flags);
}
static __always_inline void folio_set_error(struct folio *folio) {
set_bit(PG_error, folio_flags(folio, FOLIO_PF_NO_TAIL));
}
static __always_inline void SetPageError(struct page *page) {
set_bit(PG_error, &PF_NO_TAIL(page, 1)->flags);
}
static __always_inline void folio_clear_error(struct folio *folio) {
clear_bit(PG_error, folio_flags(folio, FOLIO_PF_NO_TAIL));
}
static __always_inline void ClearPageError(struct page *page) {
clear_bit(PG_error, &PF_NO_TAIL(page, 1)->flags);
}
static __always_inline bool folio_test_clear_error(struct folio *folio) {
return test_and_clear_bit(PG_error, folio_flags(folio, FOLIO_PF_NO_TAIL));
}
static __always_inline int TestClearPageError(struct page *page) {
return test_and_clear_bit(PG_error, &PF_NO_TAIL(page, 1)->flags);
}
几个 Test 的使用者就是如下的: 感觉根本没有人使用这个东西！
mm/migrate.c
539: if (folio_test_error(folio))
fs/gfs2/lops.c
477: if (folio_test_error(folio))
fs/f2fs/node.c
2082: if (TestClearPageError(page))
mm/migrate.c 这个代码有意义吗?
if (folio_test_error(folio))
folio_set_error(newfolio);
统计的位置思考
同时观察三个位置
- sio_read_complete 完全没有调用的
@[
swap_readpage_bdev_sync+1
swap_readpage+157
do_swap_page+2179
__handle_mm_fault+2017
handle_mm_fault+341
do_user_addr_fault+515
exc_page_fault+109
asm_exc_page_fault+38
]: 1134
@[
swap_readpage+5
do_swap_page+2179
__handle_mm_fault+2017
handle_mm_fault+341
do_user_addr_fault+515
exc_page_fault+109
asm_exc_page_fault+38
]: 642472
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。