[dm-devel] [dm-5.19 PATCH 21/21] dm: improve abnormal bio processing
Mike Snitzer
snitzer at redhat.com
Fri Apr 22 13:13:37 UTC 2022
On Thu, Apr 21 2022 at 12:06P -0400,
Shinichiro Kawasaki <shinichiro.kawasaki at wdc.com> wrote:
> On Apr 17, 2022 / 22:27, Mike Snitzer wrote:
> > Read/write/flush are the most common operations, optimize switch in
> > is_abnormal_io() for those cases. Follows same pattern established in
> > block perf-wip commit ("block: optimise blk_may_split for normal rw")
> >
> > Also, push is_abnormal_io() check and blk_queue_split() down from
> > dm_submit_bio() to dm_split_and_process_bio() and set new
> > 'is_abnormal_io' flag in clone_info. Optimize __split_and_process_bio
> > and __process_abnormal_io by leveraging ci.is_abnormal_io flag.
> >
> > Signed-off-by: Mike Snitzer <snitzer at kernel.org>
> > ---
> > drivers/md/dm.c | 60 +++++++++++++++++++++++++++++----------------------------
> > 1 file changed, 31 insertions(+), 29 deletions(-)
> >
> > diff --git a/drivers/md/dm.c b/drivers/md/dm.c
> > index 3b87d151ef88..b9c30dfe0f2a 100644
> > --- a/drivers/md/dm.c
> > +++ b/drivers/md/dm.c
> > @@ -84,7 +84,8 @@ struct clone_info {
> > struct dm_io *io;
> > sector_t sector;
>
> (snip)
>
> > ci->sector = bio->bi_iter.bi_sector;
> > ci->sector_count = bio_sectors(bio);
> > @@ -1645,6 +1647,13 @@ static void dm_split_and_process_bio(struct mapped_device *md,
> > __send_empty_flush(&ci);
> > /* dm_io_complete submits any data associated with flush */
> > goto out;
> > + } else if (unlikely(is_abnormal_io(bio))) {
> > + /*
> > + * Use blk_queue_split() for abnormal IO (e.g. discard, etc)
> > + * otherwise associated queue_limits won't be imposed.
> > + */
> > + blk_queue_split(&bio);
> > + ci.is_abnormal_io = true;
> > }
> >
> > error = __split_and_process_bio(&ci);
>
> Hi Mike,
>
> The hunk above triggered a WARNING [1] which is observed at mkfs.xfs for dm-
> zoned devices. With the hunk, the blk_queue_split(&bio) for abnormal I/O is
> called after init_clone_info(). I think it made the cloned bio different from
> the split bio. I suggest to move the "if (unlikely(is_abnormal_io(bio)))" block
> at the beginning of dm_split_and_process_bio(), so that blk_queue_split() is
> called before init_clone_info(). I created a patch for such change [2], and
> confirmed the WARNING goes away.
>
> > @@ -1698,13 +1707,6 @@ static void dm_submit_bio(struct bio *bio)
> > goto out;
> > }
> >
> > - /*
> > - * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
> > - * otherwise associated queue_limits won't be imposed.
> > - */
> > - if (unlikely(is_abnormal_io(bio)))
> > - blk_queue_split(&bio);
> > -
> > dm_split_and_process_bio(md, map, bio);
> > out:
> > dm_put_live_table_bio(md, srcu_idx, bio);
> > --
> > 2.15.0
> >
> > --
> > dm-devel mailing list
> > dm-devel at redhat.com
> > https://listman.redhat.com/mailman/listinfo/dm-devel
> >
>
> [1]
>
> WARNING: CPU: 3 PID: 79434 at block/bio.c:1629 bio_trim+0x10a/0x150
> Modules linked in: dm_zoned null_blk f2fs crc32_generic dm_flakey iscsi_target_mod tcm_loop target_core_pscsi target_core_file target_core_iblock xt_CHECKSUM xt_MASQUERADE xt_conntrack ipt_REJECT nf_nat_tftp nf_conntrack_tftp bridge stp llc nft_objref nf_conntrack_netbios_ns nf_conntrack_broadcast nft_fib_inet nft_fib_ipv4 nft_fib_ipv6 nft_fib nft_reject_inet nf_reject_ipv4 nf_reject_ipv6 nft_reject nft_ct nft_chain_nat nf_tables ebtable_nat ebtable_broute ip6table_nat ip6table_mangle ip6table_raw ip6table_security iptable_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 iptable_mangle iptable_raw rfkill iptable_security ip_set target_core_user nfnetlink target_core_mod ebtable_filter ebtables ip6table_filter ip6_tables iptable_filter qrtr sunrpc intel_rapl_msr intel_rapl_common x86_pkg_temp_thermal intel_powerclamp coretemp kvm_intel at24 iTCO_wdt intel_pmc_bxt iTCO_vendor_support btrfs kvm irqbypass rapl raid6_pq intel_cstate zstd_compress joydev
> intel_uncore i2c_i801 pcspkr i2c_smbus intel_pch_thermal lpc_ich ses enclosure ie31200_edac video zram ip_tables crct10dif_pclmul crc32_pclmul crc32c_intel ast ghash_clmulni_intel drm_vram_helper drm_kms_helper drm_ttm_helper ttm igb drm mpt3sas e1000e dca i2c_algo_bit raid_class scsi_transport_sas fuse [last unloaded: zonefs]
> CPU: 3 PID: 79434 Comm: mkfs.xfs Kdump: loaded Not tainted 5.18.0-rc3+ #2
> Hardware name: Supermicro X10SLL-F/X10SLL-F, BIOS 3.0 04/24/2015
> RIP: 0010:bio_trim+0x10a/0x150
> Code: 8d 7c 24 68 48 89 fa 48 c1 ea 03 80 3c 02 00 75 58 49 83 7c 24 68 00 74 13 48 83 c4 10 4c 89 e7 5d 41 5c 41 5d e9 f6 aa 0f 00 <0f> 0b 48 83 c4 10 5d 41 5c 41 5d c3 4c 89 ef 48 89 54 24 08 48 89
> RSP: 0018:ffff8881977d7928 EFLAGS: 00010206
> RAX: 0000000008000000 RBX: ffff88852b000000 RCX: 0000000000040000
> RDX: 00000000003c0000 RSI: 0000000000040000 RDI: ffff8883215a3d00
> RBP: 0000000000400000 R08: 0000000000000001 R09: ffff8881124b9057
> R10: ffffed102249720a R11: 0000000000000001 R12: ffff8883215a3d00
> R13: ffff8883215a3d28 R14: ffff8881124b9098 R15: ffff8881124b9054
> FS: 00007ff3a993dbc0(0000) GS:ffff8886ecf80000(0000) knlGS:0000000000000000
> CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 000055a1bf241900 CR3: 0000000177074002 CR4: 00000000001706e0
> Call Trace:
> <TASK>
> dm_submit_bio+0x5fa/0x13b0
> ? dm_dax_direct_access+0x1c0/0x1c0
> ? lock_release+0x3a7/0x6c0
> ? lock_downgrade+0x6a0/0x6a0
> __submit_bio+0x1c2/0x2c0
> ? __bio_queue_enter+0x570/0x570
> submit_bio_noacct_nocheck+0x2f7/0x810
> ? should_fail_request+0x70/0x70
> ? rcu_read_lock_sched_held+0x3f/0x70
> ? __lock_acquire+0x1591/0x5030
> submit_bio+0x10a/0x270
> ? submit_bio_noacct+0x15c0/0x15c0
> submit_bio_wait+0xf2/0x1d0
> ? submit_bio_wait_endio+0x40/0x40
> ? lock_release+0x6c0/0x6c0
> blkdev_issue_discard+0xb5/0x110
> ? __blkdev_issue_discard+0x590/0x590
> ? truncate_bdev_range+0x15d/0x240
> blkdev_common_ioctl+0x853/0x1670
> ? blkdev_bszset+0x160/0x160
> ? __ia32_sys_readlinkat+0x87/0xf0
> ? __ia32_compat_sys_newlstat+0x70/0x70
> ? count_memcg_events.constprop.0+0x44/0x50
> blkdev_ioctl+0x23b/0x690
> ? blkdev_common_ioctl+0x1670/0x1670
> ? security_file_ioctl+0x50/0x90
> __x64_sys_ioctl+0x127/0x190
> do_syscall_64+0x3b/0x90
> entry_SYSCALL_64_after_hwframe+0x44/0xae
> RIP: 0033:0x7ff3a970731b
> Code: ff ff ff 85 c0 79 9b 49 c7 c4 ff ff ff ff 5b 5d 4c 89 e0 41 5c c3 66 0f 1f 84 00 00 00 00 00 f3 0f 1e fa b8 10 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d dd 2a 0f 00 f7 d8 64 89 01 48
> RSP: 002b:00007ffd79c8db68 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
> RAX: ffffffffffffffda RBX: 00000000e8000000 RCX: 00007ff3a970731b
> RDX: 00007ffd79c8db80 RSI: 0000000000001277 RDI: 0000000000000004
> RBP: 0000000000000004 R08: 0000000000000000 R09: 00007ffd79c8d997
> R10: 0000000000000000 R11: 0000000000000246 R12: 00007ffd79c8db80
> R13: 0000000000000001 R14: 0000000000000000 R15: 0000000080000000
> </TASK>
> irq event stamp: 10213
> hardirqs last enabled at (10223): [<ffffffffaa36e02e>] __up_console_sem+0x5e/0x70
> hardirqs last disabled at (10232): [<ffffffffaa36e013>] __up_console_sem+0x43/0x70
> softirqs last enabled at (10190): [<ffffffffaa1faf06>] __irq_exit_rcu+0x1c6/0x260
> softirqs last disabled at (10185): [<ffffffffaa1faf06>] __irq_exit_rcu+0x1c6/0x260
> ---[ end trace 0000000000000000 ]---
>
> [2]
>
> diff --git a/drivers/md/dm.c b/drivers/md/dm.c
> index 7e3b5bdcf520..50382c98d7b3 100644
> --- a/drivers/md/dm.c
> +++ b/drivers/md/dm.c
> @@ -1639,6 +1639,15 @@ static void dm_split_and_process_bio(struct mapped_device *md,
> struct dm_io *io;
> blk_status_t error = BLK_STS_OK;
>
> + if (unlikely(is_abnormal_io(bio))) {
> + /*
> + * Use blk_queue_split() for abnormal IO (e.g. discard, etc)
> + * otherwise associated queue_limits won't be imposed.
> + */
> + blk_queue_split(&bio);
> + ci.is_abnormal_io = true;
> + }
> +
> init_clone_info(&ci, md, map, bio);
> io = ci.io;
>
> @@ -1646,13 +1655,6 @@ static void dm_split_and_process_bio(struct mapped_device *md,
> __send_empty_flush(&ci);
> /* dm_io_complete submits any data associated with flush */
> goto out;
> - } else if (unlikely(is_abnormal_io(bio))) {
> - /*
> - * Use blk_queue_split() for abnormal IO (e.g. discard, etc)
> - * otherwise associated queue_limits won't be imposed.
> - */
> - blk_queue_split(&bio);
> - ci.is_abnormal_io = true;
> }
>
> error = __split_and_process_bio(&ci);
>
> --
> Best Regards,
> Shin'ichiro Kawasaki
>
Thanks for your report, I'll sort it out but your patch doesn't seem right.
init_clone_info() will reset ci.is_abnormal_io so the bio won't be
processed properly.
Mike
More information about the dm-devel
mailing list