Re: [PATCH v4] scsi: libsas: Fix the failure of adding phy with zero-address to port

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi, John

On 2023/11/29 20:54, John Garry wrote:
On 28/11/2023 03:45, yangxingui wrote:

On 2023/11/28 3:28, John Garry wrote:
On 24/11/2023 02:27, yangxingui wrote:
We already do this in sas_ex_join_wide_port(), right?
No, If the addr of ex_phy matches dev->parent, sas_ex_join_wide_port() will not be called, but sas_add_parent_port() will be called  as follows:
static int sas_ex_discover_dev(struct domain_device *dev, int phy_id)
{
         struct expander_device *ex = &dev->ex_dev;
         struct ex_phy *ex_phy = &ex->ex_phy[phy_id];
         struct domain_device *child = NULL;
         int res = 0;

     <...>
         /* Parent and domain coherency */
         if (!dev->parent && sas_phy_match_port_addr(dev->port, ex_phy)) {
                 sas_add_parent_port(dev, phy_id);
                 return 0;
         }
         if (dev->parent && sas_phy_match_dev_addr(dev->parent, ex_phy)) {
                 sas_add_parent_port(dev, phy_id);
                 if (ex_phy->routing_attr == TABLE_ROUTING)
                         sas_configure_phy(dev, phy_id, dev->port->sas_addr, 1);
                 return 0;
         }
     <...>
}


I am not saying that what we do now does not have a problem - I am just trying to understand what currently happens

ok, because ex_phy->port is not set when calling sas_add_parent_port(), when deleting phy from the parent wide port, it is not removed from the phy_list of the parent wide port as follows:
static void sas_unregister_devs_sas_addr(struct domain_device *parent,
                                          int phy_id, bool last)
{
     <...>
     // Since ex_phy->port is not set, this branch will not be enter

But then how does this ever work? It is because we follow path sas_rediscover_dev() -> sas_discover_new() -> sas_ex_discover_devices() -> sas_ex_discover_dev() -> sas_add_parent_port(), and not sas_rediscover_dev() -> sas_discover_new() -> sas_ex_join_wide_port()? If so, is that because ephy->sas_attached_phy == 0 in sas_discover_new() -> sas_ex_join_wide_port() and it fails?

BTW, about something mentioned earlier - adding the phy19 with SAS_ADDR

Yes,
For phy19, when the phy is attached and added to the parent wide port, the path is:
sas_rediscover()
     ->sas_discover_new()
         ->sas_ex_discover_devices()
             ->sas_ex_discover_dev()
                 -> sas_add_parent_port().

ok, so then the change to set ex_phy->port = ex->parent_port looks ok. Maybe we can put this in a helper with the sas_port_add_phy() call, as it is duplicated in sas_ex_join_wide_port()

Do we also need to set ex_phy->phy_state (like sas_ex_join_wide_port())?

Well, okay, as follows?
+++ b/drivers/scsi/libsas/sas_expander.c
@@ -856,9 +856,7 @@ static bool sas_ex_join_wide_port(struct domain_device *parent, int phy_id)

if (!memcmp(phy->attached_sas_addr, ephy->attached_sas_addr,
                            SAS_ADDR_SIZE) && ephy->port) {
-                       sas_port_add_phy(ephy->port, phy->phy);
-                       phy->port = ephy->port;
-                       phy->phy_state = PHY_DEVICE_DISCOVERED;
+                       sas_port_add_ex_phy(ephy->port, phy);
                        return true;
                }
        }
diff --git a/drivers/scsi/libsas/sas_internal.h b/drivers/scsi/libsas/sas_internal.h
index e860d5b19880..39ffa60a9a01 100644
--- a/drivers/scsi/libsas/sas_internal.h
+++ b/drivers/scsi/libsas/sas_internal.h
@@ -189,6 +189,13 @@ static inline void sas_phy_set_target(struct asd_sas_phy *p, struct domain_devic
        }
 }

+static inline void sas_port_add_ex_phy(struct sas_port *port, struct ex_phy *ex_phy)
+{
+       sas_port_add_phy(port, ex_phy->phy);
+       ex_phy->port = port;
+       ex_phy->phy_state = PHY_DEVICE_DISCOVERED;
+}
+
static inline void sas_add_parent_port(struct domain_device *dev, int phy_id)
 {
        struct expander_device *ex = &dev->ex_dev;
@@ -201,8 +208,7 @@ static inline void sas_add_parent_port(struct domain_device *dev, int phy_id)
                BUG_ON(sas_port_add(ex->parent_port));
                sas_port_mark_backlink(ex->parent_port);
        }
-       sas_port_add_phy(ex->parent_port, ex_phy->phy);
+       sas_port_add_ex_phy(ex->parent_port, ex_phy);
 }


And the path called when it is removed from parent wide port is:
sas_rediscover()
     ->sas_unregister_devs_sas_addr() // The sas address of phy19 becomes 0. Since ex_phy->port is NULL, phy19 is not removed from the parent wide port's phy_list.

For phy0, it is connected to a new sata device.
sas_rediscover()
     ->sas_discover_new()->sas_ex_phy_discover()
                             ->sas_ex_phy_discover_helper()
                                 ->sas_set_ex_phy() // The device type is stp. Since the linkrate is 5 and less than 1.5G, sas_address is set to 0.

Then when we get the proper linkrate later, will we then rediscover and set the proper SAS address? I am just wondering if this change is really required?
Yes, but in fact it has not reached that stage yet. After setting the address to 0, it will continue to create a new port and try to add other phys with the same address as it to this new port.


BTW, Even with the change to set ex_phy->port = ex->parent_port, are we still joining the host-attached expander phy (19) to a port with SAS address == 0?
Yes, in order to avoid this situation, in the current patch, we will not force the SAS address to be set to 0 when the device type is not NULL, but will still use the address obtained after requesting the expander.


Thanks,
Xingui




[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Index of Archives]     [SCSI Target Devel]     [Linux SCSI Target Infrastructure]     [Kernel Newbies]     [IDE]     [Security]     [Git]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux ATA RAID]     [Linux IIO]     [Samba]     [Device Mapper]

  Powered by Linux