[PATCH 1/1] target:separate tx/rx cmd_puds

[PATCH 1/1] target:separate tx/rx cmd_puds

[Zhang Zhuoyu]

Separate tx/rx cmd_pdus in order to distinguish LUN read/write IOPS.

Signed-off-by: Zhang Zhuoyu <zhangzhuoyu@cmss.chinamobile.com>
---
 drivers/target/target_core_stat.c      | 26 ++++++++++++++++++++++----
 drivers/target/target_core_transport.c | 11 ++++++++++-
 include/target/target_core_base.h      |  3 ++-
 3 files changed, 34 insertions(+), 6 deletions(-)

diff --git a/drivers/target/target_core_stat.c b/drivers/target/target_core_stat.c
index f0db91e..9099494 100644
--- a/drivers/target/target_core_stat.c
+++ b/drivers/target/target_core_stat.c
@@ -636,7 +636,7 @@ static ssize_t target_stat_tgt_port_port_index_show(struct config_item *item,
 	return ret;
 }
 
-static ssize_t target_stat_tgt_port_in_cmds_show(struct config_item *item,
+static ssize_t target_stat_tgt_port_tx_cmds_show(struct config_item *item,
 		char *page)
 {
 	struct se_lun *lun = to_stat_tgt_port(item);
@@ -647,7 +647,23 @@ static ssize_t target_stat_tgt_port_in_cmds_show(struct config_item *item,
 	dev = rcu_dereference(lun->lun_se_dev);
 	if (dev)
 		ret = snprintf(page, PAGE_SIZE, "%lu\n",
-			       atomic_long_read(&lun->lun_stats.cmd_pdus));
+			       atomic_long_read(&lun->lun_stats.tx_cmd_pdus));
+	rcu_read_unlock();
+	return ret;
+}
+
+static ssize_t target_stat_tgt_port_rx_cmds_show(struct config_item *item,
+		char *page)
+{
+	struct se_lun *lun = to_stat_tgt_port(item);
+	struct se_device *dev;
+	ssize_t ret = -ENODEV;
+
+	rcu_read_lock();
+	dev = rcu_dereference(lun->lun_se_dev);
+	if (dev)
+		ret = snprintf(page, PAGE_SIZE, "%lu\n",
+			       atomic_long_read(&lun->lun_stats.rx_cmd_pdus));
 	rcu_read_unlock();
 	return ret;
 }
@@ -706,7 +722,8 @@ static ssize_t target_stat_tgt_port_hs_in_cmds_show(struct config_item *item,
 CONFIGFS_ATTR_RO(target_stat_tgt_port_, indx);
 CONFIGFS_ATTR_RO(target_stat_tgt_port_, name);
 CONFIGFS_ATTR_RO(target_stat_tgt_port_, port_index);
-CONFIGFS_ATTR_RO(target_stat_tgt_port_, in_cmds);
+CONFIGFS_ATTR_RO(target_stat_tgt_port_, tx_cmds);
+CONFIGFS_ATTR_RO(target_stat_tgt_port_, rx_cmds);
 CONFIGFS_ATTR_RO(target_stat_tgt_port_, write_mbytes);
 CONFIGFS_ATTR_RO(target_stat_tgt_port_, read_mbytes);
 CONFIGFS_ATTR_RO(target_stat_tgt_port_, hs_in_cmds);
@@ -717,7 +734,8 @@ static ssize_t target_stat_tgt_port_hs_in_cmds_show(struct config_item *item,
 	&target_stat_tgt_port_attr_indx,
 	&target_stat_tgt_port_attr_name,
 	&target_stat_tgt_port_attr_port_index,
-	&target_stat_tgt_port_attr_in_cmds,
+	&target_stat_tgt_port_attr_tx_cmds,
+	&target_stat_tgt_port_attr_rx_cmds,
 	&target_stat_tgt_port_attr_write_mbytes,
 	&target_stat_tgt_port_attr_read_mbytes,
 	&target_stat_tgt_port_attr_hs_in_cmds,
diff --git a/drivers/target/target_core_transport.c b/drivers/target/target_core_transport.c
index 74b646f..79cb96f 100644
--- a/drivers/target/target_core_transport.c
+++ b/drivers/target/target_core_transport.c
@@ -1356,7 +1356,16 @@ void transport_init_se_cmd(
 		return ret;
 
 	cmd->se_cmd_flags |= SCF_SUPPORTED_SAM_OPCODE;
-	atomic_long_inc(&cmd->se_lun->lun_stats.cmd_pdus);
+	switch (cmd->data_direction) {
+	case DMA_FROM_DEVICE:
+		atomic_long_inc(&cmd->se_lun->lun_stats.tx_cmd_pdus);
+		break;
+	case DMA_TO_DEVICE:
+		atomic_long_inc(&cmd->se_lun->lun_stats.rx_cmd_pdus);
+		break;
+	default:
+		break;
+	}
 	return 0;
 }
 EXPORT_SYMBOL(target_setup_cmd_from_cdb);
diff --git a/include/target/target_core_base.h b/include/target/target_core_base.h
index 9f9f590..6a0a2a7 100644
--- a/include/target/target_core_base.h
+++ b/include/target/target_core_base.h
@@ -698,7 +698,8 @@ struct se_port_stat_grps {
 };
 
 struct scsi_port_stats {
-	atomic_long_t	cmd_pdus;
+	atomic_long_t	tx_cmd_pdus;
+	atomic_long_t	rx_cmd_pdus;
 	atomic_long_t	tx_data_octets;
 	atomic_long_t	rx_data_octets;
 };
-- 
1.8.3.1

Re: [PATCH 1/1] target:separate tx/rx cmd_puds

[David Disseldorp]

Hi,

The commit summary has a typo (cmd_puds). That said, this change
isn't iSCSI specific, so using "pdu" here doesn't make much sense IMO.

On Wed, 21 Mar 2018 17:52:43 +0800, Zhang Zhuoyu wrote:

> Separate tx/rx cmd_pdus in order to distinguish LUN read/write IOPS.
> 
> Signed-off-by: Zhang Zhuoyu <zhangzhuoyu@cmss.chinamobile.com>
> ---
>  drivers/target/target_core_stat.c      | 26 ++++++++++++++++++++++----
>  drivers/target/target_core_transport.c | 11 ++++++++++-
>  include/target/target_core_base.h      |  3 ++-
>  3 files changed, 34 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/target/target_core_stat.c b/drivers/target/target_core_stat.c
> index f0db91e..9099494 100644
> --- a/drivers/target/target_core_stat.c
> +++ b/drivers/target/target_core_stat.c
> @@ -706,7 +722,8 @@ static ssize_t target_stat_tgt_port_hs_in_cmds_show(struct config_item *item,
>  CONFIGFS_ATTR_RO(target_stat_tgt_port_, indx);
>  CONFIGFS_ATTR_RO(target_stat_tgt_port_, name);
>  CONFIGFS_ATTR_RO(target_stat_tgt_port_, port_index);
> -CONFIGFS_ATTR_RO(target_stat_tgt_port_, in_cmds);
> +CONFIGFS_ATTR_RO(target_stat_tgt_port_, tx_cmds);
> +CONFIGFS_ATTR_RO(target_stat_tgt_port_, rx_cmds);

I don't think the in_cmds metric should be deleted here. It could be
calculated on the fly via tx_cmds + rx_cmds + nodata_cmds.

Cheers, David

Re: [PATCH 1/1] target:separate tx/rx cmd_puds

[David Disseldorp]

On Thu, 5 Apr 2018 13:12:12 +0200, David Disseldorp wrote:

> > -CONFIGFS_ATTR_RO(target_stat_tgt_port_, in_cmds);
> > +CONFIGFS_ATTR_RO(target_stat_tgt_port_, tx_cmds);
> > +CONFIGFS_ATTR_RO(target_stat_tgt_port_, rx_cmds);  
> 
> I don't think the in_cmds metric should be deleted here. It could be
> calculated on the fly via tx_cmds + rx_cmds + nodata_cmds.

@Zhang Zhuoyu: How about something like the following?
https://git.samba.org/?p=ddiss/linux.git;a=commitdiff;h=73723ccf433424721830797d70cfb88d4596e0fc

...this keeps the in_cmds metric, and renames tx/rx_cmds read/write_cmds
respectively. read/write_cmds is still a bit ambiguous, as it refers to
the command data direction rather than SCSI READ/WRITE CDBs, but IMO
it's clearer, and more consistent with the read/write_mbytes metrics.

Cheers, David

RE: [PATCH 1/1] target:separate tx/rx cmd_puds

[Zhang Zhuoyu]

> -----Original Message-----
> From: David Disseldorp [mailto:ddiss@suse.de]
> Sent: Friday, April 6, 2018 8:10 PM
> To: Zhang Zhuoyu <zhangzhuoyu@cmss.chinamobile.com>
> Cc: nab@linux-iscsi.org; linux-scsi@vger.kernel.org; target-
> devel@vger.kernel.org; linux-kernel@vger.kernel.org
> Subject: Re: [PATCH 1/1] target:separate tx/rx cmd_puds
> 
> On Thu, 5 Apr 2018 13:12:12 +0200, David Disseldorp wrote:
> 
> > > -CONFIGFS_ATTR_RO(target_stat_tgt_port_, in_cmds);
> > > +CONFIGFS_ATTR_RO(target_stat_tgt_port_, tx_cmds);
> > > +CONFIGFS_ATTR_RO(target_stat_tgt_port_, rx_cmds);
> >
> > I don't think the in_cmds metric should be deleted here. It could be
> > calculated on the fly via tx_cmds + rx_cmds + nodata_cmds.
> 
> @Zhang Zhuoyu: How about something like the following?
> https://git.samba.org/?p=ddiss/linux.git;a=commitdiff;h=73723ccf433424721
> 830797d70cfb88d4596e0fc
> 

Mmm... This patch is much better. 
Looks good to me.

Zhuoyu

> ...this keeps the in_cmds metric, and renames tx/rx_cmds read/write_cmds
> respectively. read/write_cmds is still a bit ambiguous, as it refers to
the
> command data direction rather than SCSI READ/WRITE CDBs, but IMO it's
> clearer, and more consistent with the read/write_mbytes metrics.
> 
> Cheers, David