On Tue, 2004-10-26 at 15:35, Roland Dreier wrote: > Hal> Because with just your changes, the port did not become > Hal> active. It appeared that there needed to be some > Hal> synchronization between the MAD layer completing its > Hal> initialization and the agents completing theirs. This was > Hal> the simplest way I could think of. I'm sure there are other > Hal> solutions too. > > That's strange, my tree is working fine for me. If you do > > ib_register_client(&mad_client) > ib_register_client(&agent_client) > > then it's guaranteed that the MAD layer is done initializing each > device before the agent layer starts. > > Oh well. > > Hal> Also, while your changes were fewer lines of code change, > Hal> these changes are less lines of code total. Is there an issue > Hal> with doing it this way ? > > I guess the only downside I see is that it ties ib_mad.c and > ib_agent.c together event more tightly (there is agent initialization > code in ib_mad.c, rather than keeping all the agent code in ib_agent.c)
Is it worth undoing this and investigating what was going on more to make it work in the more decoupled way ? -- Hal _______________________________________________ openib-general mailing list [EMAIL PROTECTED] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
