Re: [Scst-devel] Convincing the Initiator to look again

[Bart V. A. <bva...@ac...>]

On Mon, Sep 26, 2011 at 2:37 AM, Curtis Maloney <cu...@ti...> wrote:

> On 26/09/11 03:05, Bart Van Assche wrote:
>
>> On Sat, Sep 24, 2011 at 2:26 AM, Curtis <cu...@ti...
>> <mailto:cu...@ti...>> wrote:
>>
>>    my problem isn't specifically a SCST one, but since it's related, I'm
>>    hoping someone here has a solution.
>>
>>    I've got SRP working wonderfully with SCST 2.1, and am testing with
>>    VMWare and Win2k8 initiators.
>>
>>    The issue begins when I reboot the SAN box.  Windows comes in too
>> early,
>>    doesn't see a target, and won't rescan.
>>
>>    The sequence appears to be:
>>    1) SAN box starts IB drivers, turning line state to ACTIVE
>>    2) Win box sees line state change, and rescans for targets.
>>    3) Win box doesn't see any SRP Targets, and ignores (a)
>>    4) SAN box brings up SCST
>>    5) ???
>>
>>    (a) Since the Win box knows a STP Target _used_ to live there, it
>>    actually shows a STP Miniport Driver as broken in the devices list.
>>
>>    Now, I can cause Windows to look again by replugging the IB cable, but
>>    that's hardly a practical solution.  I've been toying with using
>>    "ibportstate", but haven't got it to work yet.
>>
>>    This morning I've been chasing down how to cause a unit attention to be
>>    issued, but if Windows doesn't even think there's a target there, it
>>    won't hear it.
>>
>>    If anyone has any ideas, I'll open to suggestions.
>>
>>   What triggers Windows to rescan after the SAN reboot ? If the trigger
>> is
>> an IB port state change: the IB port state is controlled by the IB
>> switch and hence the switch should allow to toggle the HCA port state
>> after SCST and ib_srpt have been initialized.
>>
>
> From what I've read on the WinOFED list, a port state change will do it.
>  However, currently in my test setup I don't have a switch.  We have another
> setup which does, so we'll test that today.
>

The port state is controlled by the subnet manager - that's either the IB
switch or opensmd. The InfiniBand spec states that it is possible to modify
*PortInfo:InitType *and that should allow to reinitialize a HCA port.
Unfortunately most HCAs do not (yet ?) implement this (see also the
IB_DEVICE_INIT_TYPE capability in the Linux kernel source tree). And a
user-space tool to trigger this functionality even doesn't exist yet.

Bart.