[ofa-general] Re: [ewg] Re: [PATCH] IB/ipoib: avoid WARN_ON on NULL path->ah

[Eli Cohen]

On Wed, Sep 24, 2008 at 11:08:05AM -0700, Roland Dreier wrote:
> Would it be simpler to just not update path->ah if the query fails?  In
> the case of the new optimization on path flushing, path->valid will be 0
> so there's no reason to change path->ah to NULL if we're not going to
> change path->valid anyway.
> 
> In fact this seems more in keeping with the idea behind ee1e2c82
> ("IPoIB: Refresh paths instead of flushing them on SM change events"),
> because your path kills the ah in the neigh struct as soon as a path
> record query fails, when the point of the original patch was to keep
> using addresses that are probably still valid across an SM failure.
> 
> The patch I'm proposing is the following.  If this looks OK, I'll send
> it to Linus tomorrow.
> 

I am OK with this.