Ming - Performance drop is resolved on my setup, but may be some stability of the kernel is caused due to this patch set. I have not tried without patch set, but in case you can figure out if below crash is due to this patch set, I can try potential fix as well. I am seeing kernel panic if I use below three settings in my fio script. iodepth_batch=8 iodepth_batch_complete_min=4 iodepth_batch_complete_max=16 Here is panic detail - [65237.831029] ------------[ cut here ]------------ [65237.831031] list_add corruption. prev->next should be next (ffffbb1d0a41fdf8), but was ffff94d0f0c57240. (prev=ffff94d0f0c57240). [65237.831046] WARNING: CPU: 18 PID: 13897 at lib/list_debug.c:28 __list_add_valid+0x6a/0x70 [65237.831046] Modules linked in: mpt3sas raid_class ses enclosure scsi_transport_sas megaraid_sas(OE) xt_CHECKSUM ipt_MASQUERADE tun ip6t_rpfilter ipt_REJECT nf_reject_ipv4 ip6t_REJECT nf_reject_ipv6 xt_conntrack ip_set nfnetlink ebtable_nat ebtable_broute bridge stp llc ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6 ip6table_mangle ip6table_security ip6table_raw iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw ebtable_filter ebtables ip6table_filter ip6_tables iptable_filter intel_rapl skx_edac nfit libnvdimm x86_pkg_temp_thermal intel_powerclamp coretemp kvm_intel kvm irqbypass crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcbc aesni_intel snd_hda_codec_realtek snd_hda_codec_generic snd_hda_intel crypto_simd cryptd [65237.831079] snd_hda_codec snd_hda_core glue_helper snd_hwdep iTCO_wdt snd_seq iTCO_vendor_support snd_seq_device pcspkr snd_pcm sg i2c_i801 joydev wmi mei_me ipmi_si mei snd_timer acpi_power_meter ioatdma snd acpi_pad ipmi_devintf ipmi_msghandler soundcore lpc_ich nfsd auth_rpcgss nfs_acl lockd grace sunrpc ip_tables xfs libcrc32c sd_mod ast i2c_algo_bit drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops ttm drm ixgbe ahci libahci crc32c_intel libata mdio i2c_core dca dm_mirror dm_region_hash dm_log dm_mod [65237.831109] CPU: 18 PID: 13897 Comm: fio Kdump: loaded Tainted: G OE 4.18.0-rc1+ #1 [65237.831110] Hardware name: Supermicro Super Server/X11DPG-QT, BIOS 1.0 06/22/2017 [65237.831112] RIP: 0010:__list_add_valid+0x6a/0x70 [65237.831112] Code: 31 c0 48 c7 c7 90 08 ab a0 e8 22 b4 cd ff 0f 0b 31 c0 c3 48 89 d1 48 c7 c7 40 08 ab a0 48 89 f2 48 89 c6 31 c0 e8 06 b4 cd ff <0f> 0b 31 c0 c3 90 48 8b 07 48 b9 00 01 00 00 00 00 ad de 48 8b 57 [65237.831130] RSP: 0018:ffffbb1d0a41fdd8 EFLAGS: 00010286 [65237.831131] RAX: 0000000000000000 RBX: ffff94d0ebf90000 RCX: 0000000000000006 [65237.831132] RDX: 0000000000000000 RSI: 0000000000000086 RDI: ffff94d91ec16970 [65237.831132] RBP: ffffdb1cff59e980 R08: 0000000000000000 R09: 0000000000000663 [65237.831133] R10: 0000000000000004 R11: 0000000000000000 R12: 0000000000000001 [65237.831134] R13: 0000000000000000 R14: ffff94d0f0c57240 R15: ffff94d0f0f04c40 [65237.831135] FS: 00007fad10d02740(0000) GS:ffff94d91ec00000(0000) knlGS:0000000000000000 [65237.831135] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [65237.831136] CR2: 00007fad10c2a004 CR3: 000000085600e006 CR4: 00000000007606e0 [65237.831137] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [65237.831138] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 [65237.831138] PKRU: 55555554 [65237.831138] Call Trace: [65237.831145] blk_mq_flush_plug_list+0x12d/0x260 [65237.831150] blk_flush_plug_list+0xe7/0x240 [65237.831152] blk_finish_plug+0x27/0x40 [65237.831156] __x64_sys_io_submit+0xc9/0x180 [65237.831162] do_syscall_64+0x5b/0x180 [65237.831166] entry_SYSCALL_64_after_hwframe+0x44/0xa9 Kashyap > -----Original Message----- > From: Ming Lei [mailto:ming.lei@xxxxxxxxxx] > Sent: Thursday, June 28, 2018 8:49 AM > To: Jens Axboe > Cc: linux-block@xxxxxxxxxxxxxxx; Ming Lei; Kashyap Desai; Laurence Oberman; > Omar Sandoval; Christoph Hellwig; Bart Van Assche; Hannes Reinecke > Subject: [PATCH 0/3] blk-mq: improve IO perf in case of none io sched > > Hi, > > The 1st 2 patch improves ctx->lock uses, and it is observed that IOPS > may be improved by ~5% in rand IO test on MegaRaid SAS run by Kashyap. > > The 3rd patch fixes rand IO performance regression on MegaRaid SAS > test, still reported by Kashyap. > > Ming Lei (3): > blk-mq: use list_splice_tail() to insert requests > blk-mq: only attempt to merge bio if there is rq in sw queue > blk-mq: dequeue request one by one from sw queue iff hctx is busy > > block/blk-mq-sched.c | 14 ++++---------- > block/blk-mq.c | 38 ++++++++++++++++++++++++++++++-------- > include/linux/blk-mq.h | 1 + > 3 files changed, 35 insertions(+), 18 deletions(-) > > Cc: Kashyap Desai <kashyap.desai@xxxxxxxxxxxx> > Cc: Laurence Oberman <loberman@xxxxxxxxxx> > Cc: Omar Sandoval <osandov@xxxxxx> > Cc: Christoph Hellwig <hch@xxxxxx> > Cc: Bart Van Assche <bart.vanassche@xxxxxxx> > Cc: Hannes Reinecke <hare@xxxxxxx> > > -- > 2.9.5