[Users] SRP pkey support
Blake Caldwell
blakec at ornl.gov
Tue Apr 23 15:51:55 PDT 2013
Following up on a discussion at the OFA user's day last week, I
investigated whether ib_srpt (part of linux SCST) supported pkeys other
than 0xffff. It doesn't, since the pkey_index is hard coded to 0 in
srpt_init_ch_qp(). This means the target will always use pkey 0xffff.
Discussions on this being implemented is better suited to the linux-rdma
list, but the pkey verification and srp commands might be useful to
others on the list.
With this in partitions.conf:
Partition=0x8004, ipoib : 0x0002c903004bfd7b=full, 0x0002c903004ae31b=full ;
I could see that opensm was using my partitions.conf and the nodes got
the 0x8004 pkey. I ran opensm with debug options of 0x80 (not sure this
is necessary to get these particular messages).
Apr 23 20:45:23 435418 [41007940] 0x08 ->
pkey_mgr_process_physical_port: pkey 0x8004 was updated for node
0x0002c903004ae31a port 1
Apr 23 20:45:23 435422 [41007940] 0x08 ->
pkey_mgr_process_physical_port: pkey 0x8004 was updated for node
0x0002c903004bfd7a port 1
Apr 23 20:45:23 435425 [41007940] 0x08 ->
pkey_mgr_process_physical_port: pkey 0x7fff was updated for node
0x0002c903004ae31a port 1
Apr 23 20:45:23 435428 [41007940] 0x08 ->
pkey_mgr_process_physical_port: pkey 0xffff was updated for node
0x0002c903004bfd7a port 1
You can't use srp_daemon because it hard codes a pkey of 0xffff, so if
the target was to support other pkeys, you'd have to do a manual srp
login. Get the login information from ibsrpdm -c and substitute the pkey
8004 for ffff. Log into the target with the following:
echo -n
id_ext=[guid],ioc_guid=[guid],dgid=fe800000000000000002c903004ae31b,pkey=8004,service_id=[guid]
> /sys/class/infiniband_srp/srp-mlx4_1-1/add_target
Dmesg tells the story:
[5263703.357312] host12: ib_srp: new target: id_ext [guid] ioc_guid
[guid] pkey 8004 service_id [guid] dgid
fe80:0000:0000:0000:0002:c903:004a:e31b
[5263703.369485] scsi12 : SRP.T10:1AE34A0003C90200
[5263708.856427] host12: SRP abort called
[5263713.843543] host12: SRP reset_device called
[5263718.830818] host12: ib_srp: SRP reset_host called state 0 qp_err 0
[5263738.795576] host12: SRP abort called
[5263743.782699] host12: SRP reset_device called
[5263748.770978] host12: ib_srp: SRP reset_host called state 0 qp_err 0
[5263758.761393] scsi 12:0:0:0: scsi: Device offlined - not ready after
error recovery
[5263758.761400] scsi 12:0:0:0: timing out command, waited 22s
The same login command with pkey=ffff is successful, because pkey_index
0 is still 0xffff.
-Blake
More information about the Users
mailing list