[Openib-windows] OpenSM on OpenIB complib

[Yossi Leybovich]

Fab

While moving OpenSM to use IBAL complib we encounter the few issues:

1. cl_timer.c - 
we tried to call cl_timer_start from within the timer call back,
and the thread enter to deadlock .
We notice that cl_timer_star call to cl_timer_stop which call to
DeleteTimerQueueTimer with INVALI_HANDLE which cause it to wait to all the
callbacks to finish. 
Msdn:"...If this parameter is INVALID_HANDLE_VALUE, the function waits for
the timer callback function to complete before returning.
"
So if we call it within the timer callback we cause deadlock.

Why not to call DeleteTimerQueueTimer  with NULL parameter
Msdn:"If this parameter is NULL, the function marks the timer for deletion
and returns immediately. "

The kernel implementation of cl_timer_start does not call cl_timer_stop why
?

2. local traps:
>From the PRM its state that the HCA generate local trap with source lid 0
and the AL  should forward this trap to the SM .
We notice if activate 2 SM and one is the Master the second SM after
changing the SM bit in the capability mask get trap with source lid 0
While the AL should forward this trap to the Master SM (portinfo.smlid)

Did you ever check this flow? 
Do you know if the AL forward trap with slid 0 to the SM lid?


Thank
Yossi 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ofw/attachments/20050713/9ae5fb14/attachment.html>