On 2022/9/27 16:34, John Garry wrote:
On 27/09/2022 04:26, Jason Yan wrote:
+++ b/drivers/scsi/pm8001/pm8001_sas.c
@@ -645,22 +645,16 @@ static int pm8001_dev_found_notify(struct
domain_device *dev)
pm8001_device->dcompletion = &completion;
if (parent_dev && dev_is_expander(parent_dev->dev_type)) {
int phy_id;
- struct ex_phy *phy;
- for (phy_id = 0; phy_id < parent_dev->ex_dev.num_phys;
- phy_id++) {
- phy = &parent_dev->ex_dev.ex_phy[phy_id];
- if (SAS_ADDR(phy->attached_sas_addr)
- == SAS_ADDR(dev->sas_addr)) {
- pm8001_device->attached_phy = phy_id;
- break;
- }
- }
- if (phy_id == parent_dev->ex_dev.num_phys) {
+
+ phy_id = sas_find_attached_phy(&parent_dev->ex_dev, dev);
+ if (phy_id == -ENODEV) {
pm8001_dbg(pm8001_ha, FAIL,
"Error: no attached dev:%016llx at ex:%016llx.\n",
SAS_ADDR(dev->sas_addr),
SAS_ADDR(parent_dev->sas_addr));
res = -1;
I think that you can just pass the linux error code (-ENODEV) back here.
And for hisi_sas we change to -EINVAL for this code. I don't think it's
required, so I think that we can pass -ENODEV back there also. Using
-EINVAL seems to come from when the code was originally added in
abda97c2fe874 and from a quick glance libsas does not seem to have
special processing for -EINVAL.
Yes, libsas does not have any special processing for the return value,
so there is no difference in what value we return here. But I agree with
you that return -ENODEV is better here because we can keep it consistent
with the return value of sas_find_attached_phy().
Will update.
Thanks,
Jason
Thanks,
John
+ } else {
+ pm8001_device->attached_phy = phy_id;
}
} else {
if (dev->dev_type == SAS_SATA
.
