Re: [PATCH v7 2/7] scsi: libsas: Define NCQ Priority sysfs attributes for SATA devices

[Igor Pylypiv <ipylypiv@google.com>]

On Thu, Mar 07, 2024 at 10:51:37AM +0100, Niklas Cassel wrote:
> On Wed, Mar 06, 2024 at 11:28:42AM -0800, Igor Pylypiv wrote:
> > On Wed, Mar 06, 2024 at 11:54:52AM +0100, Niklas Cassel wrote:
> > > On Tue, Mar 05, 2024 at 05:22:21PM -0800, Igor Pylypiv wrote:
> > > > Libata sysfs attributes cannot be used for libsas managed SATA devices
> > > > because the ata_port location is different for libsas.
> > > > 
> > > > Defined sysfs attributes (visible for SATA devices only):
> > > > - /sys/block/sda/device/ncq_prio_enable
> > > > - /sys/block/sda/device/ncq_prio_supported
> > > > 
> > > > The newly defined attributes will pass the correct ata_port to libata
> > > > helper functions.
> > > > 
> > > > Reviewed-by: John Garry <john.g.garry@oracle.com>
> > > > Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
> > > > Reviewed-by: Jason Yan <yanaijie@huawei.com>
> > > > Signed-off-by: Igor Pylypiv <ipylypiv@google.com>
> > > > ---
> > > >  drivers/scsi/libsas/sas_ata.c | 94 +++++++++++++++++++++++++++++++++++
> > > >  include/scsi/sas_ata.h        |  6 +++
> > > >  2 files changed, 100 insertions(+)
> > > > 
> > > > diff --git a/drivers/scsi/libsas/sas_ata.c b/drivers/scsi/libsas/sas_ata.c
> > > > index 12e2653846e3..04b0bd9a4e01 100644
> > > > --- a/drivers/scsi/libsas/sas_ata.c
> > > > +++ b/drivers/scsi/libsas/sas_ata.c
> > > > @@ -964,3 +964,97 @@ int sas_execute_ata_cmd(struct domain_device *device, u8 *fis, int force_phy_id)
> > > >  			       force_phy_id, &tmf_task);
> > > >  }
> > > >  EXPORT_SYMBOL_GPL(sas_execute_ata_cmd);
> > > > +
> > > > +static ssize_t sas_ncq_prio_supported_show(struct device *device,
> > > > +					   struct device_attribute *attr,
> > > > +					   char *buf)
> > > > +{
> > > > +	struct scsi_device *sdev = to_scsi_device(device);
> > > > +	struct domain_device *ddev = sdev_to_domain_dev(sdev);
> > > > +	bool supported;
> > > > +	int rc;
> > > > +
> > > > +	/* This attribute shall be visible for SATA devices only */
> > > > +	if (WARN_ON_ONCE(!dev_is_sata(ddev)))
> > > > +		return -EINVAL;
> > > 
> > > Like Hannes commented, I don't believe this is needed.
> > > 
> > 
> > The intention for the check is to serve as a fail-safe in case 'is_visible()'
> > callback gets incorrectly modified and stops hiding the sysfs attributes
> > for non-SATA devices.
> > 
> > Just want to clarify should I remove the WARN_ON_ONCE and keep the fail-safe
> > check or should I get rid of the check completely and trust 'is_visible()'
> > to always hide the sysfs attributes for non-SATA devices?
> 
> I think that you can remove both the WARN_ON_ONCE and the
> if (!dev_is_sata()).
> 
> We usually don't keep code around "just in case someone modifies some
> other function sometime in the future".
> 
> 
> If someone changes is_visible() to remove the dev_is_sata() check,
> then they would introuce a bug. I don't see why anyone would change that,
> but if someone tried to remove that check from is_visible() anyway,
> I'm assuming that someone would catch it during code review.
> 

Ack, makes sense. I'll remove the checks in v8. Thanks!
> 
> > 
> > > 
> > > > +
> > > > +	rc = ata_ncq_prio_supported(ddev->sata_dev.ap, sdev, &supported);
> > > > +	if (rc)
> > > > +		return rc;
> > > > +
> > > > +	return sysfs_emit(buf, "%d\n", supported);
> > > > +}
> > > > +
> > > 
> > > While this is a bit different depending on file, the most common way is to
> > > have no blank link before the DEVICE_ATTR().
> > >
> > 
> > In "[PATCH 1/3] ata: libata-sata: Factor out NCQ Priority configuration helpers"
> > Damien asked to keep the blank link before the DEVICE_ATTR() in libata-sata.c.
> > 
> > Non-prio sysfs attributes in libata-sata.c don't have blank lines
> > before DEVICE_ATTR() so I'm more inclined to remove the lines.
> > 
> > I'm fine with either of ways, just want to get a consensus and make it 
> > consistent for both libata-sata.c and sas_ata.c.
> 
> While it is a bit different depending on file, it is slightly more common
> to no have a extra blank line before DEVICE_ATTR():
> 
> $ git grep -B 1 DEVICE_ATTR | grep "}" | wc -l
> 2167
> 
> $ git grep -B 1 DEVICE_ATTR | grep -- "c-$" | wc -l
> 1725
> 
> But I'm fine to keep it like it is, especially if Damien already had expresed
> a preference.
>

Thank you Niklas. Let's keep the extra blank line as Damien suggested.

Thanks,
Igor

> 
> Kind regards,
> Niklas
>
> >  
> > > 
> > > > +DEVICE_ATTR(ncq_prio_supported, S_IRUGO, sas_ncq_prio_supported_show, NULL);
> > > > +
> > > > +static ssize_t sas_ncq_prio_enable_show(struct device *device,
> > > > +					struct device_attribute *attr,
> > > > +					char *buf)
> > > > +{
> > > > +	struct scsi_device *sdev = to_scsi_device(device);
> > > > +	struct domain_device *ddev = sdev_to_domain_dev(sdev);
> > > > +	bool enabled;
> > > > +	int rc;
> > > > +
> > > > +	/* This attribute shall be visible for SATA devices only */
> > > > +	if (WARN_ON_ONCE(!dev_is_sata(ddev)))
> > > > +		return -EINVAL;
> > > > +
> > > > +	rc = ata_ncq_prio_enabled(ddev->sata_dev.ap, sdev, &enabled);
> > > > +	if (rc)
> > > > +		return rc;
> > > > +
> > > > +	return sysfs_emit(buf, "%d\n", enabled);
> > > > +}
> > > > +
> > > > +static ssize_t sas_ncq_prio_enable_store(struct device *device,
> > > > +					 struct device_attribute *attr,
> > > > +					 const char *buf, size_t len)
> > > > +{
> > > > +	struct scsi_device *sdev = to_scsi_device(device);
> > > > +	struct domain_device *ddev = sdev_to_domain_dev(sdev);
> > > > +	bool enable;
> > > > +	int rc;
> > > > +
> > > > +	/* This attribute shall be visible for SATA devices only */
> > > > +	if (WARN_ON_ONCE(!dev_is_sata(ddev)))
> > > > +		return -EINVAL;
> > > > +
> > > > +	rc = kstrtobool(buf, &enable);
> > > > +	if (rc)
> > > > +		return rc;
> > > > +
> > > > +	rc = ata_ncq_prio_enable(ddev->sata_dev.ap, sdev, enable);
> > > > +	if (rc)
> > > > +		return rc;
> > > > +
> > > > +	return len;
> > > > +}
> > > > +
> > > > +DEVICE_ATTR(ncq_prio_enable, S_IRUGO | S_IWUSR,
> > > > +	    sas_ncq_prio_enable_show, sas_ncq_prio_enable_store);
> > > > +
> > > > +static struct attribute *sas_ata_sdev_attrs[] = {
> > > > +	&dev_attr_ncq_prio_supported.attr,
> > > > +	&dev_attr_ncq_prio_enable.attr,
> > > > +	NULL
> > > > +};
> > > > +
> > > > +static umode_t sas_ata_attr_is_visible(struct kobject *kobj,
> > > > +				       struct attribute *attr, int i)
> > > > +{
> > > > +	struct device *dev = kobj_to_dev(kobj);
> > > > +	struct scsi_device *sdev = to_scsi_device(dev);
> > > > +	struct domain_device *ddev = sdev_to_domain_dev(sdev);
> > > > +
> > > > +	if (!dev_is_sata(ddev))
> > > > +		return 0;
> > > > +
> > > > +	return attr->mode;
> > > > +}
> > > > +
> > > > +const struct attribute_group sas_ata_sdev_attr_group = {
> > > > +	.attrs = sas_ata_sdev_attrs,
> > > > +	.is_visible = sas_ata_attr_is_visible,
> > > > +};
> > > > +EXPORT_SYMBOL_GPL(sas_ata_sdev_attr_group);
> > > > diff --git a/include/scsi/sas_ata.h b/include/scsi/sas_ata.h
> > > > index 2f8c719840a6..92e27e7bf088 100644
> > > > --- a/include/scsi/sas_ata.h
> > > > +++ b/include/scsi/sas_ata.h
> > > > @@ -39,6 +39,9 @@ int smp_ata_check_ready_type(struct ata_link *link);
> > > >  int sas_discover_sata(struct domain_device *dev);
> > > >  int sas_ata_add_dev(struct domain_device *parent, struct ex_phy *phy,
> > > >  		    struct domain_device *child, int phy_id);
> > > > +
> > > > +extern const struct attribute_group sas_ata_sdev_attr_group;
> > > > +
> > > >  #else
> > > >  
> > > >  static inline void sas_ata_disabled_notice(void)
> > > > @@ -123,6 +126,9 @@ static inline int sas_ata_add_dev(struct domain_device *parent, struct ex_phy *p
> > > >  	sas_ata_disabled_notice();
> > > >  	return -ENODEV;
> > > >  }
> > > > +
> > > > +#define sas_ata_sdev_attr_group ((struct attribute_group) {})
> > > > +
> > > >  #endif
> > > >  
> > > >  #endif /* _SAS_ATA_H_ */
> > > > -- 
> > > > 2.44.0.278.ge034bb2e1d-goog
> > > >