Skip to contentSkip to conversation

qla2xxx: Return EBUSY on fcport deletion

PATCH Superseded Daniel Wagner 3 participants 6 messages Oct 12, 2020 – Oct 14, 2020

Subsystems (2)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 qla2xxx: Return EBUSY on fcport deletion

When the fcport is about to be deleted we should return EBUSY instead of ENODEV. Only for EBUSY the request will be requeued in a multipath setup.

Also in case we have a valid qpair but the firmware has not yet started return EBUSY to avoid dropping the request.

---

v3: simplify test logic as suggested by Arun.
v2: rebased on mkp/staging

 drivers/scsi/qla2xxx/qla_nvme.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/scsi/qla2xxx/qla_nvme.c +4 -2
index 2cd9bd288910..1fa457a5736e 100644
@@ -555,10 +555,12 @@ static int qla_nvme_post_cmd(struct nvme_fc_local_port *lport,
 
 	fcport = qla_rport->fcport;
 
-	if (!qpair || !fcport || (qpair && !qpair->fw_started) ||
-	    (fcport && fcport->deleted))
+	if (!qpair || !fcport)
 		return -ENODEV;
 
+	if (!qpair->fw_started || fcport->deleted)
+		return -EBUSY;
+
 	vha = fcport->vha;
 
 	if (!(fcport->nvme_flag & NVME_FLAG_REGISTERED))
Signature · 1 line
2.16.4
Reviewed-by on v3 1/1
#1 lore raw
Quoted 27 lines · Daniel Wagner

----------------------------------------------------------------------
When the fcport is about to be deleted we should return EBUSY instead of ENODEV. Only for EBUSY the request will be requeued in a multipath setup.

Also in case we have a valid qpair but the firmware has not yet started return EBUSY to avoid dropping the request.

---

v3: simplify test logic as suggested by Arun.
v2: rebased on mkp/staging

 drivers/scsi/qla2xxx/qla_nvme.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/scsi/qla2xxx/qla_nvme.c +0 -1
index 2cd9bd288910..1fa457a5736e 100644
@@ -555,10 +555,12 @@ static int qla_nvme_post_cmd(struct nvme_fc_local_port *lport,
 
 	fcport = qla_rport->fcport;
 
-	if (!qpair || !fcport || (qpair && !qpair->fw_started) ||

drivers/scsi/qla2xxx/qla_nvme.c static int qla_nvme_post_cmd(struct nvme_fc_local_port *lport, · line 566

Quoted 10 lines
drivers/scsi/qla2xxx/qla_nvme.c +4 -1
-	    (fcport && fcport->deleted))
+	if (!qpair || !fcport)
 		return -ENODEV;
 
+	if (!qpair->fw_started || fcport->deleted)
+		return -EBUSY;
+
 	vha = fcport->vha;
 
 	if (!(fcport->nvme_flag & NVME_FLAG_REGISTERED))

Thanks Daniel.

#2 lore raw
In reply to Daniel Wagner
Quoted 11 lines · Daniel Wagner

When the fcport is about to be deleted we should return EBUSY instead of ENODEV. Only for EBUSY the request will be requeued in a multipath setup.

Also in case we have a valid qpair but the firmware has not yet started return EBUSY to avoid dropping the request.

---

v3: simplify test logic as suggested by Arun.

Not exactly a "simplification": there was a change of behaviour between v2 and v3. It seems the commit log no longer reflects the code.

Quoted 25 lines

v2: rebased on mkp/staging

 drivers/scsi/qla2xxx/qla_nvme.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/scsi/qla2xxx/qla_nvme.c +4 -2
index 2cd9bd288910..1fa457a5736e 100644
@@ -555,10 +555,12 @@ static int qla_nvme_post_cmd(struct nvme_fc_local_port *lport,
 
 	fcport = qla_rport->fcport;
 
-	if (!qpair || !fcport || (qpair && !qpair->fw_started) ||
-	    (fcport && fcport->deleted))
+	if (!qpair || !fcport)
 		return -ENODEV;
 
+	if (!qpair->fw_started || fcport->deleted)
+		return -EBUSY;
+
 	vha = fcport->vha;
 
 	if (!(fcport->nvme_flag & NVME_FLAG_REGISTERED))
#3 lore raw
Quoted 10 lines · Finn Thain
Quoted 7 lines · Daniel Wagner

When the fcport is about to be deleted we should return EBUSY instead of ENODEV. Only for EBUSY the request will be requeued in a multipath setup.

Also in case we have a valid qpair but the firmware has not yet started return EBUSY to avoid dropping the request.

Quoted 7 lines
Quoted 4 lines

---

v3: simplify test logic as suggested by Arun.

Not exactly a "simplification": there was a change of behaviour between v2 and v3. It seems the commit log no longer reflects the code.

How so? I am struggling to see how it could be a change in behavior. But then I sometimes fail at simple logic ;)

v2 and v3 will return ENODEV if qpair or fcport are invalid and for EBUSY one of the other condition needs be true. The difference between v2 and v3 should only be the order how tests are executed. The outcome should be the same.

#4 lore raw
Quoted 14 lines · Daniel Wagner
Quoted 13 lines · Finn Thain
Quoted 10 lines · Daniel Wagner

When the fcport is about to be deleted we should return EBUSY instead of ENODEV. Only for EBUSY the request will be requeued in a multipath setup.

Also in case we have a valid qpair but the firmware has not yet started return EBUSY to avoid dropping the request.

---

Quoted 9 lines
Quoted 5 lines
Quoted 1 line

v3: simplify test logic as suggested by Arun.

Not exactly a "simplification": there was a change of behaviour between v2 and v3. It seems the commit log no longer reflects the code.

How so? I am struggling to see how it could be a change in behavior. But then I sometimes fail at simple logic ;)

Me too, so I confirmed the result by executing the code snippets.

Quoted 5 lines

v2 and v3 will return ENODEV if qpair or fcport are invalid and for EBUSY one of the other condition needs be true. The difference between v2 and v3 should only be the order how tests are executed. The outcome should be the same.

Here's a truth table for v2:

qpair fw_started fcport deleted result
---------------------------------------

F       X       F       X       -ENODEV
F       F       T       F       -ENODEV
F       F       T       T       -EBUSY  *
F       T       T       F       -ENODEV
F       T       T       T       -EBUSY  *
T       F       F       X       -EBUSY  *
T       F       T       X       -EBUSY
T       T       F       X       -ENODEV
T       T       T       F       neither
T       T       T       T       -EBUSY

Here's a truth table for v3:

qpair fw_started fcport deleted result
---------------------------------------

F       X       F       X       -ENODEV
F       F       T       F       -ENODEV
F       F       T       T       -ENODEV *
F       T       T       F       -ENODEV
F       T       T       T       -ENODEV *
T       F       F       X       -ENODEV *
T       F       T       X       -EBUSY
T       T       F       X       -ENODEV
T       T       T       F       neither
T       T       T       T       -EBUSY

The asterisks mark the changed rows.

I don't know whether the changes in v3 are desirable or not, I was just pointing out that the commit log ("valid qpair but the firmware has not yet started return EBUSY") now seems to disagree with the code.

#5 lore raw

Hi Finn,

Quoted 7 lines · Finn Thain
Quoted 3 lines · Daniel Wagner

How so? I am struggling to see how it could be a change in behavior. But then I sometimes fail at simple logic ;)

Me too, so I confirmed the result by executing the code snippets.

Thanks for taking the time to explain it to me!

Quoted 3 lines

I don't know whether the changes in v3 are desirable or not, I was just pointing out that the commit log ("valid qpair but the firmware has not yet started return EBUSY") now seems to disagree with the code.

I see where I did my thinko. In v3 we have more ENODEV due to the fact that we test the pointers first (qpair, fpcort). If either of them is invalid we get the ENODEV. Actually, it's makes more sense and is likely to be 'more correct'.

Anyway, let me update the commit log.

Thanks a lot!
Daniel