[PATCH RFC 0/6] libata/scsi/libsas: Allocate SCSI device earlier for ata port probe

[PATCH RFC 0/6] libata/scsi/libsas: Allocate SCSI device earlier for ata port probe

[John Garry]

Currently for libata the SCSI device (sdev) associated with an ata_device
is allocated when the port probe has completed.

It's useful to have the SCSI device and its associated request queue
available earlier for the port probe. Specifically if we have the
request queue available, then we can:
- Easily put ATA qc in SCSI cmnd priv data
- Send ATA internal commands on SCSI device request queue for [0]. The
  current solution there is to use the shost sdev request queue, which
  isn't great.
  
This series changes the ata port probe to alloc the sdev in the
ata_device revalidation, and then just do a SCSI starget scan afterwards.

Why an RFC?
1. IPR  driver needs to be fixed up - it does not use ATA EH port probe
   Mail [1] needs following up
2. SATA PMP support needs verification, but I don't have a setup
3. This series needs to be merged into or go after [0]

Patch 1/6 could be merged now.

[0] https://lore.kernel.org/linux-ide/1654770559-101375-1-git-send-email-john.garry@huawei.com/
[1] https://lore.kernel.org/linux-ide/369448ed-f89a-c2db-1850-91450d8b5998@opensource.wdc.com/

Any comments welcome - please have a look.

Based on v6.0-rc4 and tested for QEMU AHCI and libsas.

John Garry (6):
  scsi: core: Use SCSI_SCAN_RESCAN in  __scsi_add_device()
  scsi: scsi_transport_sas: Allocate end device target id in the rphy
    alloc
  scsi: core: Add scsi_get_dev()
  ata: libata-scsi: Add ata_scsi_setup_sdev()
  scsi: libsas: Add sas_ata_setup_device()
  ata: libata-scsi: Allocate sdev early in port probe

 drivers/ata/libata-eh.c           |  4 +++
 drivers/ata/libata-scsi.c         | 45 +++++++++++++++++++++----------
 drivers/ata/libata.h              |  1 +
 drivers/scsi/libsas/sas_ata.c     | 20 ++++++++++++++
 drivers/scsi/scsi_scan.c          | 28 ++++++++++++++++++-
 drivers/scsi/scsi_transport_sas.c | 25 +++++++++++------
 include/linux/libata.h            |  2 ++
 include/scsi/scsi_host.h          |  3 +++
 8 files changed, 105 insertions(+), 23 deletions(-)

-- 
2.35.3

[PATCH 1/6] scsi: core: Use SCSI_SCAN_RESCAN in __scsi_add_device()

[John Garry]

Instead of using hardcoded '1' as the __scsi_add_device() ->
scsi_probe_and_add_lun() rescan arg, use proper macro SCSI_SCAN_RESCAN.

Signed-off-by: John Garry <john.garry@huawei.com>
---
 drivers/scsi/scsi_scan.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c
index ac6059702d13..3759b1a77504 100644
--- a/drivers/scsi/scsi_scan.c
+++ b/drivers/scsi/scsi_scan.c
@@ -1588,7 +1588,8 @@ struct scsi_device *__scsi_add_device(struct Scsi_Host *shost, uint channel,
 		scsi_complete_async_scans();
 
 	if (scsi_host_scan_allowed(shost) && scsi_autopm_get_host(shost) == 0) {
-		scsi_probe_and_add_lun(starget, lun, NULL, &sdev, 1, hostdata);
+		scsi_probe_and_add_lun(starget, lun, NULL, &sdev,
+				       SCSI_SCAN_RESCAN, hostdata);
 		scsi_autopm_put_host(shost);
 	}
 	mutex_unlock(&shost->scan_mutex);
-- 
2.35.3

[PATCH RFC 2/6] scsi: scsi_transport_sas: Allocate end device target id in the rphy alloc

[John Garry]

Currently the per-end device target id is allocated when adding the rphy,
which is when we execute the scan of the target.

However it will be useful to have the target id allocated earlier when
allocating the rphy for the end device. For libata we want to move to a
scheme of allocating the sdev early in the probe process and then later
executing the scan (for that target). As such, users of would libata would
require that the target id allocated earlier here (before the scan).

Signed-off-by: John Garry <john.garry@huawei.com>
---
 drivers/scsi/scsi_transport_sas.c | 25 +++++++++++++++++--------
 1 file changed, 17 insertions(+), 8 deletions(-)

diff --git a/drivers/scsi/scsi_transport_sas.c b/drivers/scsi/scsi_transport_sas.c
index 2f88c61216ee..56d325665bc7 100644
--- a/drivers/scsi/scsi_transport_sas.c
+++ b/drivers/scsi/scsi_transport_sas.c
@@ -1433,6 +1433,7 @@ static void sas_rphy_initialize(struct sas_rphy *rphy)
 struct sas_rphy *sas_end_device_alloc(struct sas_port *parent)
 {
 	struct Scsi_Host *shost = dev_to_shost(&parent->dev);
+	struct sas_host_attrs *sas_host = to_sas_host_attrs(shost);
 	struct sas_end_device *rdev;
 
 	rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
@@ -1455,6 +1456,10 @@ struct sas_rphy *sas_end_device_alloc(struct sas_port *parent)
 	sas_rphy_initialize(&rdev->rphy);
 	transport_setup_device(&rdev->rphy.dev);
 
+	mutex_lock(&sas_host->lock);
+	rdev->rphy.scsi_target_id = sas_host->next_target_id++;
+	mutex_unlock(&sas_host->lock);
+
 	return &rdev->rphy;
 }
 EXPORT_SYMBOL(sas_end_device_alloc);
@@ -1500,6 +1505,16 @@ struct sas_rphy *sas_expander_alloc(struct sas_port *parent,
 }
 EXPORT_SYMBOL(sas_expander_alloc);
 
+static bool sas_rphy_end_device_valid_tproto(struct sas_rphy *rphy)
+{
+	struct sas_identify *identify = &rphy->identify;
+
+	if (identify->target_port_protocols &
+	    (SAS_PROTOCOL_SSP | SAS_PROTOCOL_STP | SAS_PROTOCOL_SATA))
+		return true;
+	return false;
+}
+
 /**
  * sas_rphy_add  -  add a SAS remote PHY to the device hierarchy
  * @rphy:	The remote PHY to be added
@@ -1529,16 +1544,10 @@ int sas_rphy_add(struct sas_rphy *rphy)
 
 	mutex_lock(&sas_host->lock);
 	list_add_tail(&rphy->list, &sas_host->rphy_list);
-	if (identify->device_type == SAS_END_DEVICE &&
-	    (identify->target_port_protocols &
-	     (SAS_PROTOCOL_SSP | SAS_PROTOCOL_STP | SAS_PROTOCOL_SATA)))
-		rphy->scsi_target_id = sas_host->next_target_id++;
-	else if (identify->device_type == SAS_END_DEVICE)
-		rphy->scsi_target_id = -1;
 	mutex_unlock(&sas_host->lock);
 
 	if (identify->device_type == SAS_END_DEVICE &&
-	    rphy->scsi_target_id != -1) {
+	    sas_rphy_end_device_valid_tproto(rphy)) {
 		int lun;
 
 		if (identify->target_port_protocols & SAS_PROTOCOL_SSP)
@@ -1667,7 +1676,7 @@ static int sas_user_scan(struct Scsi_Host *shost, uint channel,
 	mutex_lock(&sas_host->lock);
 	list_for_each_entry(rphy, &sas_host->rphy_list, list) {
 		if (rphy->identify.device_type != SAS_END_DEVICE ||
-		    rphy->scsi_target_id == -1)
+		    !sas_rphy_end_device_valid_tproto(rphy))
 			continue;
 
 		if ((channel == SCAN_WILD_CARD || channel == 0) &&
-- 
2.35.3

[PATCH RFC 3/6] scsi: core: Add scsi_get_dev()

[John Garry]

Add a function which allows use to alloc a sdev with configurable
device parent, and channel:id:lun.

This is useful for separating adding a scsi device into separate alloc
and scan steps.

Signed-off-by: John Garry <john.garry@huawei.com>
---
 drivers/scsi/scsi_scan.c | 25 +++++++++++++++++++++++++
 include/scsi/scsi_host.h |  3 +++
 2 files changed, 28 insertions(+)

diff --git a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c
index 3759b1a77504..fd15ddac01b6 100644
--- a/drivers/scsi/scsi_scan.c
+++ b/drivers/scsi/scsi_scan.c
@@ -1988,3 +1988,28 @@ void scsi_forget_host(struct Scsi_Host *shost)
 	spin_unlock_irqrestore(shost->host_lock, flags);
 }
 
+struct scsi_device *scsi_get_dev(struct device *parent, int channel, uint id, u64 lun)
+{
+	struct Scsi_Host *shost = dev_to_shost(parent);
+	struct scsi_device *sdev = NULL;
+	struct scsi_target *starget;
+
+	mutex_lock(&shost->scan_mutex);
+	if (!scsi_host_scan_allowed(shost))
+		goto out;
+
+	starget = scsi_alloc_target(parent, 0, id);
+	if (!starget)
+		goto out;
+
+	sdev = scsi_alloc_sdev(starget, 0, NULL);
+	if (sdev)
+		sdev->borken = 0;
+	else
+		scsi_target_reap(starget);
+	put_device(&starget->dev);
+ out:
+	mutex_unlock(&shost->scan_mutex);
+	return sdev;
+}
+EXPORT_SYMBOL(scsi_get_dev);
diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h
index aa7b7496c93a..5142c7df7647 100644
--- a/include/scsi/scsi_host.h
+++ b/include/scsi/scsi_host.h
@@ -794,7 +794,10 @@ void scsi_host_busy_iter(struct Scsi_Host *,
 
 struct class_container;
 
+extern struct scsi_device *scsi_get_dev(struct device *parent, int channel, uint id, u64 lun);
+
 /*
+
  * DIF defines the exchange of protection information between
  * initiator and SBC block device.
  *
-- 
2.35.3

[PATCH RFC 4/6] ata: libata-scsi: Add ata_scsi_setup_sdev()

[John Garry]

Add a function to setup the sdev associated with an ata_device.

Currently in libata to create the sdev we call __scsi_add_device() to
create the sdev and execute the scan. However if we want to move to a
2-step process where we allocate the sdev early in the port probe, then
we need a separate function just to allocate the sdev.

We add a ata_port_operations callback .setup_scsi_device for when the
driver needs a custom setup. This is essentially for libsas, which does
not use the same options for sdev parent and id.

Signed-off-by: John Garry <john.garry@huawei.com>
---
 drivers/ata/libata-scsi.c | 25 +++++++++++++++++++++++++
 drivers/ata/libata.h      |  1 +
 include/linux/libata.h    |  2 ++
 3 files changed, 28 insertions(+)

diff --git a/drivers/ata/libata-scsi.c b/drivers/ata/libata-scsi.c
index 29e2f55c6faa..88f39ab10a92 100644
--- a/drivers/ata/libata-scsi.c
+++ b/drivers/ata/libata-scsi.c
@@ -4257,6 +4257,31 @@ static void ata_scsi_assign_ofnode(struct ata_device *dev, struct ata_port *ap)
 }
 #endif
 
+int ata_scsi_setup_sdev(struct ata_device *dev)
+{
+	u64 lun = 0;
+	int channel = 0;
+	uint id = 0;
+	struct ata_link *link = dev->link;
+	struct ata_port *ap = link->ap;
+	struct Scsi_Host *shost = ap->scsi_host;
+	struct device *parent = &shost->shost_gendev;
+	struct scsi_device *sdev;
+
+	if (ap->ops->setup_scsi_device)
+		return ap->ops->setup_scsi_device(dev);
+
+	if (ata_is_host_link(link))
+		id = dev->devno;
+	else
+		channel = link->pmp;
+	sdev = scsi_get_dev(parent, channel, id, lun);
+	if (!sdev)
+		return -ENODEV;
+	dev->sdev = sdev;
+	return 0;
+}
+
 void ata_scsi_scan_host(struct ata_port *ap, int sync)
 {
 	int tries = 5;
diff --git a/drivers/ata/libata.h b/drivers/ata/libata.h
index 98bc8649c63f..ca6c3e0f81ea 100644
--- a/drivers/ata/libata.h
+++ b/drivers/ata/libata.h
@@ -113,6 +113,7 @@ extern struct ata_device *ata_scsi_find_dev(struct ata_port *ap,
 					    const struct scsi_device *scsidev);
 extern int ata_scsi_add_hosts(struct ata_host *host,
 			      struct scsi_host_template *sht);
+extern int ata_scsi_setup_sdev(struct ata_device *dev);
 extern void ata_scsi_scan_host(struct ata_port *ap, int sync);
 extern int ata_scsi_offline_dev(struct ata_device *dev);
 extern void ata_scsi_set_sense(struct ata_device *dev,
diff --git a/include/linux/libata.h b/include/linux/libata.h
index 698032e5ef2d..67aa1d7d909b 100644
--- a/include/linux/libata.h
+++ b/include/linux/libata.h
@@ -968,6 +968,8 @@ struct ata_port_operations {
 	void (*phy_reset)(struct ata_port *ap);
 	void (*eng_timeout)(struct ata_port *ap);
 
+	int (*setup_scsi_device)(struct ata_device *dev);
+
 	/*
 	 * ->inherits must be the last field and all the preceding
 	 * fields must be pointers.
-- 
2.35.3

[PATCH RFC 5/6] scsi: libsas: Add sas_ata_setup_device()

[John Garry]

Add a function which does a custom sdev alloc - essentially replicates
what we do in sas_rphy_add() in terms of sdev parent and id.

Signed-off-by: John Garry <john.garry@huawei.com>
---
 drivers/scsi/libsas/sas_ata.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)

diff --git a/drivers/scsi/libsas/sas_ata.c b/drivers/scsi/libsas/sas_ata.c
index d35c9296f738..6d3535a4bfbb 100644
--- a/drivers/scsi/libsas/sas_ata.c
+++ b/drivers/scsi/libsas/sas_ata.c
@@ -523,6 +523,25 @@ static int sas_ata_prereset(struct ata_link *link, unsigned long deadline)
 	return res;
 }
 
+static int sas_ata_setup_device(struct ata_device *dev)
+{
+	u64 lun = 0;
+	int channel = 0;
+	struct ata_link *link = dev->link;
+	struct ata_port *ap = link->ap;
+	struct scsi_device *sdev;
+	struct domain_device *ddev = ap->private_data;
+	struct sas_rphy *rphy = ddev->rphy;
+	struct device *parent = &rphy->dev;
+	uint id = rphy->scsi_target_id;
+
+	sdev = scsi_get_dev(parent, channel, id, lun);
+	if (!sdev)
+		return -ENODEV;
+	dev->sdev = sdev;
+	return 0;
+}
+
 static struct ata_port_operations sas_sata_ops = {
 	.prereset		= sas_ata_prereset,
 	.hardreset		= sas_ata_hard_reset,
@@ -537,6 +556,7 @@ static struct ata_port_operations sas_sata_ops = {
 	.set_dmamode		= sas_ata_set_dmamode,
 	.sched_eh		= sas_ata_sched_eh,
 	.end_eh			= sas_ata_end_eh,
+	.setup_scsi_device	= sas_ata_setup_device,
 };
 
 static struct ata_port_info sata_port_info = {
-- 
2.35.3

[PATCH RFC 6/6] ata: libata-scsi: Allocate sdev early in port probe

[John Garry]

Currently the per-ata device sdev is allocated as part of the scsi target
scan, which is after the ata port probe.

However it is useful to have the sdev available in the port probe. As an
example of an advantage, if the request queue is available in the probe
(which it would be if the sdev is available), then it is possible to use
a SCSI cmnd for ATA internal commands. The benefit of this is then we can
put the ATA qc structure in the SCSI cmnd private data. It will also be
useful if we want to send ATA internal commands as requests.

Signed-off-by: John Garry <john.garry@huawei.com>
---
 drivers/ata/libata-eh.c   |  4 ++++
 drivers/ata/libata-scsi.c | 20 ++++++--------------
 2 files changed, 10 insertions(+), 14 deletions(-)

diff --git a/drivers/ata/libata-eh.c b/drivers/ata/libata-eh.c
index 7c128c89b454..35375873a464 100644
--- a/drivers/ata/libata-eh.c
+++ b/drivers/ata/libata-eh.c
@@ -2971,6 +2971,10 @@ static int ata_eh_revalidate_and_attach(struct ata_link *link,
 			 */
 			dev->class = ehc->classes[dev->devno];
 
+			rc = ata_scsi_setup_sdev(dev);
+			if (rc)
+				goto err;
+
 			if (dev->class == ATA_DEV_PMP)
 				rc = sata_pmp_attach(dev);
 			else
diff --git a/drivers/ata/libata-scsi.c b/drivers/ata/libata-scsi.c
index 88f39ab10a92..0e26860a82e2 100644
--- a/drivers/ata/libata-scsi.c
+++ b/drivers/ata/libata-scsi.c
@@ -1111,7 +1111,9 @@ int ata_scsi_dev_config(struct scsi_device *sdev, struct ata_device *dev)
 	if (dev->flags & ATA_DFLAG_TRUSTED)
 		sdev->security_supported = 1;
 
-	dev->sdev = sdev;
+	/* Put extra reference which we get when allocating the starget initially */
+	scsi_target_reap(scsi_target(sdev));
+
 	return 0;
 }
 
@@ -4279,6 +4281,7 @@ int ata_scsi_setup_sdev(struct ata_device *dev)
 	if (!sdev)
 		return -ENODEV;
 	dev->sdev = sdev;
+	ata_scsi_assign_ofnode(dev, ap);
 	return 0;
 }
 
@@ -4292,26 +4295,15 @@ void ata_scsi_scan_host(struct ata_port *ap, int sync)
  repeat:
 	ata_for_each_link(link, ap, EDGE) {
 		ata_for_each_dev(dev, link, ENABLED) {
-			struct scsi_device *sdev;
+			struct Scsi_Host *shost = ap->scsi_host;
 			int channel = 0, id = 0;
 
-			if (dev->sdev)
-				continue;
-
 			if (ata_is_host_link(link))
 				id = dev->devno;
 			else
 				channel = link->pmp;
 
-			sdev = __scsi_add_device(ap->scsi_host, channel, id, 0,
-						 NULL);
-			if (!IS_ERR(sdev)) {
-				dev->sdev = sdev;
-				ata_scsi_assign_ofnode(dev, ap);
-				scsi_device_put(sdev);
-			} else {
-				dev->sdev = NULL;
-			}
+			scsi_scan_target(&shost->shost_gendev, channel, id, 0, SCSI_SCAN_INITIAL);
 		}
 	}
 
-- 
2.35.3

Re: [PATCH RFC 3/6] scsi: core: Add scsi_get_dev()

[Damien Le Moal]

On 9/20/22 19:27, John Garry wrote:
> Add a function which allows use to alloc a sdev with configurable
> device parent, and channel:id:lun.
> 
> This is useful for separating adding a scsi device into separate alloc
> and scan steps.
> 
> Signed-off-by: John Garry <john.garry@huawei.com>
> ---
>   drivers/scsi/scsi_scan.c | 25 +++++++++++++++++++++++++
>   include/scsi/scsi_host.h |  3 +++
>   2 files changed, 28 insertions(+)
> 
> diff --git a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c
> index 3759b1a77504..fd15ddac01b6 100644
> --- a/drivers/scsi/scsi_scan.c
> +++ b/drivers/scsi/scsi_scan.c
> @@ -1988,3 +1988,28 @@ void scsi_forget_host(struct Scsi_Host *shost)
>   	spin_unlock_irqrestore(shost->host_lock, flags);
>   }
>   
> +struct scsi_device *scsi_get_dev(struct device *parent, int channel, uint id, u64 lun)
> +{
> +	struct Scsi_Host *shost = dev_to_shost(parent);
> +	struct scsi_device *sdev = NULL;
> +	struct scsi_target *starget;
> +
> +	mutex_lock(&shost->scan_mutex);
> +	if (!scsi_host_scan_allowed(shost))
> +		goto out;
> +
> +	starget = scsi_alloc_target(parent, 0, id);
> +	if (!starget)
> +		goto out;
> +
> +	sdev = scsi_alloc_sdev(starget, 0, NULL);
> +	if (sdev)
> +		sdev->borken = 0;
> +	else
> +		scsi_target_reap(starget);
> +	put_device(&starget->dev);
> + out:
> +	mutex_unlock(&shost->scan_mutex);
> +	return sdev;
> +}
> +EXPORT_SYMBOL(scsi_get_dev);
> diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h
> index aa7b7496c93a..5142c7df7647 100644
> --- a/include/scsi/scsi_host.h
> +++ b/include/scsi/scsi_host.h
> @@ -794,7 +794,10 @@ void scsi_host_busy_iter(struct Scsi_Host *,
>   
>   struct class_container;
>   
> +extern struct scsi_device *scsi_get_dev(struct device *parent, int channel, uint id, u64 lun);

You should not need extern here, and long line...

> +
>   /*
> +

white line change.

>    * DIF defines the exchange of protection information between
>    * initiator and SBC block device.
>    *

-- 
Damien Le Moal
Western Digital Research

Re: [PATCH RFC 2/6] scsi: scsi_transport_sas: Allocate end device target id in the rphy alloc

[Damien Le Moal]

On 9/20/22 19:27, John Garry wrote:
> Currently the per-end device target id is allocated when adding the rphy,
> which is when we execute the scan of the target.
> 
> However it will be useful to have the target id allocated earlier when
> allocating the rphy for the end device. For libata we want to move to a
> scheme of allocating the sdev early in the probe process and then later
> executing the scan (for that target). As such, users of would libata would
> require that the target id allocated earlier here (before the scan).
> 
> Signed-off-by: John Garry <john.garry@huawei.com>
> ---
>   drivers/scsi/scsi_transport_sas.c | 25 +++++++++++++++++--------
>   1 file changed, 17 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/scsi/scsi_transport_sas.c b/drivers/scsi/scsi_transport_sas.c
> index 2f88c61216ee..56d325665bc7 100644
> --- a/drivers/scsi/scsi_transport_sas.c
> +++ b/drivers/scsi/scsi_transport_sas.c
> @@ -1433,6 +1433,7 @@ static void sas_rphy_initialize(struct sas_rphy *rphy)
>   struct sas_rphy *sas_end_device_alloc(struct sas_port *parent)
>   {
>   	struct Scsi_Host *shost = dev_to_shost(&parent->dev);
> +	struct sas_host_attrs *sas_host = to_sas_host_attrs(shost);
>   	struct sas_end_device *rdev;
>   
>   	rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
> @@ -1455,6 +1456,10 @@ struct sas_rphy *sas_end_device_alloc(struct sas_port *parent)
>   	sas_rphy_initialize(&rdev->rphy);
>   	transport_setup_device(&rdev->rphy.dev);
>   
> +	mutex_lock(&sas_host->lock);
> +	rdev->rphy.scsi_target_id = sas_host->next_target_id++;
> +	mutex_unlock(&sas_host->lock);
> +
>   	return &rdev->rphy;
>   }
>   EXPORT_SYMBOL(sas_end_device_alloc);
> @@ -1500,6 +1505,16 @@ struct sas_rphy *sas_expander_alloc(struct sas_port *parent,
>   }
>   EXPORT_SYMBOL(sas_expander_alloc);
>   
> +static bool sas_rphy_end_device_valid_tproto(struct sas_rphy *rphy)
> +{
> +	struct sas_identify *identify = &rphy->identify;
> +
> +	if (identify->target_port_protocols &
> +	    (SAS_PROTOCOL_SSP | SAS_PROTOCOL_STP | SAS_PROTOCOL_SATA))
> +		return true;
> +	return false;

You could just do:

return identify->target_port_protocols &
	(SAS_PROTOCOL_SSP | SAS_PROTOCOL_STP | SAS_PROTOCOL_SATA))

> +}
> +
>   /**
>    * sas_rphy_add  -  add a SAS remote PHY to the device hierarchy
>    * @rphy:	The remote PHY to be added
> @@ -1529,16 +1544,10 @@ int sas_rphy_add(struct sas_rphy *rphy)
>   
>   	mutex_lock(&sas_host->lock);
>   	list_add_tail(&rphy->list, &sas_host->rphy_list);
> -	if (identify->device_type == SAS_END_DEVICE &&
> -	    (identify->target_port_protocols &
> -	     (SAS_PROTOCOL_SSP | SAS_PROTOCOL_STP | SAS_PROTOCOL_SATA)))
> -		rphy->scsi_target_id = sas_host->next_target_id++;
> -	else if (identify->device_type == SAS_END_DEVICE)
> -		rphy->scsi_target_id = -1;
>   	mutex_unlock(&sas_host->lock);
>   
>   	if (identify->device_type == SAS_END_DEVICE &&

You could move this check inside sas_rphy_end_device_valid_tproto(),
no ?

> -	    rphy->scsi_target_id != -1) {
> +	    sas_rphy_end_device_valid_tproto(rphy)) {
>   		int lun;
>   
>   		if (identify->target_port_protocols & SAS_PROTOCOL_SSP)
> @@ -1667,7 +1676,7 @@ static int sas_user_scan(struct Scsi_Host *shost, uint channel,
>   	mutex_lock(&sas_host->lock);
>   	list_for_each_entry(rphy, &sas_host->rphy_list, list) {
>   		if (rphy->identify.device_type != SAS_END_DEVICE ||
> -		    rphy->scsi_target_id == -1)
> +		    !sas_rphy_end_device_valid_tproto(rphy))
>   			continue;
>   
>   		if ((channel == SCAN_WILD_CARD || channel == 0) &&

-- 
Damien Le Moal
Western Digital Research

Re: [PATCH 1/6] scsi: core: Use SCSI_SCAN_RESCAN in __scsi_add_device()

[Damien Le Moal]

On 9/20/22 19:27, John Garry wrote:
> Instead of using hardcoded '1' as the __scsi_add_device() ->
> scsi_probe_and_add_lun() rescan arg, use proper macro SCSI_SCAN_RESCAN.
> 
> Signed-off-by: John Garry <john.garry@huawei.com>
> ---
>   drivers/scsi/scsi_scan.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c
> index ac6059702d13..3759b1a77504 100644
> --- a/drivers/scsi/scsi_scan.c
> +++ b/drivers/scsi/scsi_scan.c
> @@ -1588,7 +1588,8 @@ struct scsi_device *__scsi_add_device(struct Scsi_Host *shost, uint channel,
>   		scsi_complete_async_scans();
>   
>   	if (scsi_host_scan_allowed(shost) && scsi_autopm_get_host(shost) == 0) {
> -		scsi_probe_and_add_lun(starget, lun, NULL, &sdev, 1, hostdata);
> +		scsi_probe_and_add_lun(starget, lun, NULL, &sdev,
> +				       SCSI_SCAN_RESCAN, hostdata);
>   		scsi_autopm_put_host(shost);
>   	}
>   	mutex_unlock(&shost->scan_mutex);

Looks good.

Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>

-- 
Damien Le Moal
Western Digital Research

Re: [PATCH RFC 0/6] libata/scsi/libsas: Allocate SCSI device earlier for ata port probe

[Damien Le Moal]

On 9/20/22 19:27, John Garry wrote:
> Currently for libata the SCSI device (sdev) associated with an ata_device
> is allocated when the port probe has completed.
> 
> It's useful to have the SCSI device and its associated request queue
> available earlier for the port probe. Specifically if we have the
> request queue available, then we can:
> - Easily put ATA qc in SCSI cmnd priv data
> - Send ATA internal commands on SCSI device request queue for [0]. The
>    current solution there is to use the shost sdev request queue, which
>    isn't great.
>    
> This series changes the ata port probe to alloc the sdev in the
> ata_device revalidation, and then just do a SCSI starget scan afterwards.
> 
> Why an RFC?
> 1. IPR  driver needs to be fixed up - it does not use ATA EH port probe
>     Mail [1] needs following up

Yes. If IPR could be converted to ata error_handler, a lot of code can 
be simplified in libata too.

> 2. SATA PMP support needs verification, but I don't have a setup

Port multiplier behind a sas HBA will be challenging to setup :)
I can try, but I will need to open up one of my servers and hook a small 
PMP box to one of the pm8001 plugs. I may have the cables for that... 
Let me check.

> 3. This series needs to be merged into or go after [0]
> 
> Patch 1/6 could be merged now.
> 
> [0] https://lore.kernel.org/linux-ide/1654770559-101375-1-git-send-email-john.garry@huawei.com/
> [1] https://lore.kernel.org/linux-ide/369448ed-f89a-c2db-1850-91450d8b5998@opensource.wdc.com/
> 
> Any comments welcome - please have a look.
> 
> Based on v6.0-rc4 and tested for QEMU AHCI and libsas.
> 
> John Garry (6):
>    scsi: core: Use SCSI_SCAN_RESCAN in  __scsi_add_device()
>    scsi: scsi_transport_sas: Allocate end device target id in the rphy
>      alloc
>    scsi: core: Add scsi_get_dev()
>    ata: libata-scsi: Add ata_scsi_setup_sdev()
>    scsi: libsas: Add sas_ata_setup_device()
>    ata: libata-scsi: Allocate sdev early in port probe
> 
>   drivers/ata/libata-eh.c           |  4 +++
>   drivers/ata/libata-scsi.c         | 45 +++++++++++++++++++++----------
>   drivers/ata/libata.h              |  1 +
>   drivers/scsi/libsas/sas_ata.c     | 20 ++++++++++++++
>   drivers/scsi/scsi_scan.c          | 28 ++++++++++++++++++-
>   drivers/scsi/scsi_transport_sas.c | 25 +++++++++++------
>   include/linux/libata.h            |  2 ++
>   include/scsi/scsi_host.h          |  3 +++
>   8 files changed, 105 insertions(+), 23 deletions(-)
> 

-- 
Damien Le Moal
Western Digital Research

Re: [PATCH RFC 0/6] libata/scsi/libsas: Allocate SCSI device earlier for ata port probe

[John Garry]

On 21/09/2022 05:04, Damien Le Moal wrote:
> On 9/20/22 19:27, John Garry wrote:
>> Currently for libata the SCSI device (sdev) associated with an ata_device
>> is allocated when the port probe has completed.
>>
>> It's useful to have the SCSI device and its associated request queue
>> available earlier for the port probe. Specifically if we have the
>> request queue available, then we can:
>> - Easily put ATA qc in SCSI cmnd priv data
>> - Send ATA internal commands on SCSI device request queue for [0]. The
>>    current solution there is to use the shost sdev request queue, which
>>    isn't great.
>> This series changes the ata port probe to alloc the sdev in the
>> ata_device revalidation, and then just do a SCSI starget scan afterwards.
>>
>> Why an RFC?
>> 1. IPR  driver needs to be fixed up - it does not use ATA EH port probe
>>     Mail [1] needs following up
> 
> Yes. If IPR could be converted to ata error_handler, a lot of code can 
> be simplified in libata too.

Hmmm... yeah, it would be good to see progress there.

> 
>> 2. SATA PMP support needs verification, but I don't have a setup
> 
> Port multiplier behind a sas HBA will be challenging to setup :)
> I can try, but I will need to open up one of my servers and hook a small 
> PMP box to one of the pm8001 plugs. I may have the cables for that... 
> Let me check.

I was more thinking of just AHCI with a port multiplier.

As for SAS controllers, I don't think it's something to be concerned 
about. For a start, I know for sure that hisi_sas HW does not support 
port multipliers, and I don't think that pm8001 does either. In 
addition, libsas does not even support it - I did see a series in the 
scsi list from years ago (to support it), but it did not progress.

Thanks,
John

Re: [PATCH RFC 0/6] libata/scsi/libsas: Allocate SCSI device earlier for ata port probe

[Damien Le Moal]

On 9/21/22 17:24, John Garry wrote:
> On 21/09/2022 05:04, Damien Le Moal wrote:
>> On 9/20/22 19:27, John Garry wrote:
>>> Currently for libata the SCSI device (sdev) associated with an 
>>> ata_device
>>> is allocated when the port probe has completed.
>>>
>>> It's useful to have the SCSI device and its associated request queue
>>> available earlier for the port probe. Specifically if we have the
>>> request queue available, then we can:
>>> - Easily put ATA qc in SCSI cmnd priv data
>>> - Send ATA internal commands on SCSI device request queue for [0]. The
>>>    current solution there is to use the shost sdev request queue, which
>>>    isn't great.
>>> This series changes the ata port probe to alloc the sdev in the
>>> ata_device revalidation, and then just do a SCSI starget scan 
>>> afterwards.
>>>
>>> Why an RFC?
>>> 1. IPR  driver needs to be fixed up - it does not use ATA EH port probe
>>>     Mail [1] needs following up
>>
>> Yes. If IPR could be converted to ata error_handler, a lot of code can 
>> be simplified in libata too.
> 
> Hmmm... yeah, it would be good to see progress there.
> 
>>
>>> 2. SATA PMP support needs verification, but I don't have a setup
>>
>> Port multiplier behind a sas HBA will be challenging to setup :)
>> I can try, but I will need to open up one of my servers and hook a 
>> small PMP box to one of the pm8001 plugs. I may have the cables for 
>> that... Let me check.
> 
> I was more thinking of just AHCI with a port multiplier.

OK. I got confused :)
Easy then, my test box is all hooked up already.
Will give this a spin.

> 
> As for SAS controllers, I don't think it's something to be concerned 
> about. For a start, I know for sure that hisi_sas HW does not support 
> port multipliers, and I don't think that pm8001 does either. In 
> addition, libsas does not even support it - I did see a series in the 
> scsi list from years ago (to support it), but it did not progress.
> 
> Thanks,
> John

-- 
Damien Le Moal
Western Digital Research

Re: [PATCH RFC 3/6] scsi: core: Add scsi_get_dev()

[John Garry]

On 20/09/2022 22:58, Damien Le Moal wrote:
>> +extern struct scsi_device *scsi_get_dev(struct device *parent, int 
>> channel, uint id, u64 lun);
> 
> You should not need extern here, and long line...
> 
>> +
>>   /*
>> +
> 
> white line change.
> 
>>    * DIF defines the exchange of protection information between
>>    * initiator and SBC block device. 

ok, I'll tidy them.

Thanks,
John

Re: [PATCH RFC 2/6] scsi: scsi_transport_sas: Allocate end device target id in the rphy alloc

[John Garry]

On 20/09/2022 23:02, Damien Le Moal wrote:
>>
>>       return &rdev->rphy;
>>   }
>>   EXPORT_SYMBOL(sas_end_device_alloc);
>> @@ -1500,6 +1505,16 @@ struct sas_rphy *sas_expander_alloc(struct 
>> sas_port *parent,
>>   }
>>   EXPORT_SYMBOL(sas_expander_alloc);
>> +static bool sas_rphy_end_device_valid_tproto(struct sas_rphy *rphy)
>> +{
>> +    struct sas_identify *identify = &rphy->identify;
>> +
>> +    if (identify->target_port_protocols &
>> +        (SAS_PROTOCOL_SSP | SAS_PROTOCOL_STP | SAS_PROTOCOL_SATA))
>> +        return true;
>> +    return false;
> 
> You could just do:
> 
> return identify->target_port_protocols &
>      (SAS_PROTOCOL_SSP | SAS_PROTOCOL_STP | SAS_PROTOCOL_SATA))

OK

> 
>> +}
>> +
>>   /**
>>    * sas_rphy_add  -  add a SAS remote PHY to the device hierarchy
>>    * @rphy:    The remote PHY to be added
>> @@ -1529,16 +1544,10 @@ int sas_rphy_add(struct sas_rphy *rphy)
>>       mutex_lock(&sas_host->lock);
>>       list_add_tail(&rphy->list, &sas_host->rphy_list);
>> -    if (identify->device_type == SAS_END_DEVICE &&
>> -        (identify->target_port_protocols &
>> -         (SAS_PROTOCOL_SSP | SAS_PROTOCOL_STP | SAS_PROTOCOL_SATA)))
>> -        rphy->scsi_target_id = sas_host->next_target_id++;
>> -    else if (identify->device_type == SAS_END_DEVICE)
>> -        rphy->scsi_target_id = -1;
>>       mutex_unlock(&sas_host->lock);
>>       if (identify->device_type == SAS_END_DEVICE &&
> 
> You could move this check inside sas_rphy_end_device_valid_tproto(),
> no ?
> 

Yeah, I suppose I could. I might require a function rename with that.

>> -        rphy->scsi_target_id != -1) {
>> +        sas_rphy_end_device_valid_tproto(rphy)) {
>>           int lun;
>>           if (identify->target_port_protocols & SAS_PROTOCOL_SSP)
>> @@ -1667,7 +1676,7 @@ static int sas_user_scan(struct Scsi_Host 
>> *shost, uint channel,
>>       mutex_lock(&sas_host->lock);
>>       list_for_each_entry(rphy, &sas_host->rphy_list, list) {
>>           if (rphy->identify.device_type != SAS_END_DEVICE ||
>> -            rphy->scsi_target_id == -1)
>> +            !sas_rphy_end_device_valid_tproto(rphy))
>>               continue; 

Thanks,
John