[ofa-general] Re: [PATCH] opensm: notify master SM with trap 144 (not finished)

[Hal Rosenstock]

Sasha,

On Wed, Oct 29, 2008 at 9:52 AM, Sasha Khapyorsky <sashak at voltaire.com> wrote:
> Hi Hal,
>
> On 16:27 Tue 28 Oct     , Hal Rosenstock wrote:
>> On Sat, Oct 25, 2008 at 4:01 PM, Sasha Khapyorsky <sashak at voltaire.com> wrote:
>> >
>> > When entering standby state (after discovery) notify master SM about us.
>> > In case when SMA doesn't support trap sending (specifically trap 144 on
>> > PortInfo:CapabilityMask change - isSM bit, example is current ConnectX
>> > firmware - 2.5.0) this is only way to notify the current master SM that
>> > another SM is running.
>>
>> So is the trap sent unconditionally (since there's no way of knowing
>> whether the SMA supports this or not) ? Is the only downside the extra
>> Trap/TrapRepress when the SMA does support this ?
>
> It is not unconditional. There is such code at beginning of
> osm_send_trap144():
>
>        /* don't bother with sending trap when SMA supports this */
>        if (!local &&
>            pi->capability_mask&(IB_PORT_CAP_HAS_TRAP|IB_PORT_CAP_HAS_CAP_NTC))
>                return 0;

Oh, I see: those bits are not on in PortInfo:CapabilityMask in the C-X SMA.

Should that just be checked against HAS_CAP_NTC as there might be
other traps supported ?

-- Hal

>> Seems to me that the right fix is to the Connect-X SMA.
>
> Agree. But it is not there yet.
>
>> Also, what happens once the Connect-X SMA is fixed ? Does this code persist ?
>
> Then osm_send_trap144(..., 0) will do nothing following
> PortInfo:CapabilityMask. And actually if the problem will become
> obsolete we can remove this call safely.
>
> Sasha
>