diff options
| author | Mike Christie <michael.christie@oracle.com> | 2025-04-23 22:26:32 -0500 |
|---|---|---|
| committer | Martin K. Petersen <martin.petersen@oracle.com> | 2025-04-28 21:47:55 -0400 |
| commit | 9cf2317b795d6cde0fccb8744b5a080a9586020e (patch) | |
| tree | 3081c6a4f985736babdceda2ff8596ab000d5752 /include/target | |
| parent | 0af2f6be1b4281385b618cb86ad946eded089ac8 (diff) | |
scsi: target: Move I/O path stats to per CPU
The atomic use in the main I/O path is causing perf issues when using
higher performance backend devices and multiple queues. This moves the
stats to per CPU. Combined with the next patch that moves the
non_ordered/delayed_cmd_count to per CPU, IOPS by up to 33% for 8K IOS
when using 4 or more queues.
Signed-off-by: Mike Christie <michael.christie@oracle.com>
Link: https://lore.kernel.org/r/20250424032741.16216-2-michael.christie@oracle.com
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Diffstat (limited to 'include/target')
| -rw-r--r-- | include/target/target_core_base.h | 20 |
1 files changed, 14 insertions, 6 deletions
diff --git a/include/target/target_core_base.h b/include/target/target_core_base.h index 97099a5e3f6c..05e3673607b8 100644 --- a/include/target/target_core_base.h +++ b/include/target/target_core_base.h @@ -669,15 +669,19 @@ struct se_lun_acl { struct se_ml_stat_grps ml_stat_grps; }; +struct se_dev_entry_io_stats { + u32 total_cmds; + u32 read_bytes; + u32 write_bytes; +}; + struct se_dev_entry { u64 mapped_lun; u64 pr_res_key; u64 creation_time; bool lun_access_ro; u32 attach_count; - atomic_long_t total_cmds; - atomic_long_t read_bytes; - atomic_long_t write_bytes; + struct se_dev_entry_io_stats __percpu *stats; /* Used for PR SPEC_I_PT=1 and REGISTER_AND_MOVE */ struct kref pr_kref; struct completion pr_comp; @@ -800,6 +804,12 @@ struct se_device_queue { struct se_cmd_queue sq; }; +struct se_dev_io_stats { + u32 total_cmds; + u32 read_bytes; + u32 write_bytes; +}; + struct se_device { /* Used for SAM Task Attribute ordering */ u32 dev_cur_ordered_id; @@ -821,9 +831,7 @@ struct se_device { atomic_long_t num_resets; atomic_long_t aborts_complete; atomic_long_t aborts_no_task; - atomic_long_t num_cmds; - atomic_long_t read_bytes; - atomic_long_t write_bytes; + struct se_dev_io_stats __percpu *stats; /* Active commands on this virtual SE device */ atomic_t non_ordered; bool ordered_sync_in_progress; |