[ofa-general] Re: [PATCH for-2.6.22] IB/ipoib: fix SRQ WR leak

[Roland Dreier]

 > Yes, it seems that we shouldn't keep a QP in error state
 > for extended periods of time, since that moves hardware
 > to the slow path.

Ugh, so I can do a local DoS by just creating a QP and moving it to
error and then going to sleep for a long time?  What hardware is
susceptible to this?

 > It seems that the right approach might be to create
 > a loopback QP in RTS, and perform post sends there.

How about using the send queue of the QP we're trying to flush?
I'll try to code this up tomorrow if no one beats me to the fix.

 - R.