RE: [PATCH] ntb: ntb perf tool

["Allen Hubbe" <Allen.Hubbe@emc.com>]

Jon Mason <jdmason@kudzu.us>:
> On Tue, Jan 5, 2016 at 6:37 PM, Dave Jiang <dave.jiang@intel.com> wrote:

> > +static void perf_link_event(void *ctx)
> > +{
> > +       struct perf_ctx *perf = ctx;
> > +
> > +       if (ntb_link_is_up(perf->ntb, NULL, NULL) == 1)
> 
> The comparison to one seems unnecessary
> 

It api doc says it returns one, zero, or negative to indicate an error.  The Intel NTB driver does not return an error.  Would you prefer to change the api, to make the return simply a boolean, and have some other mechanism to query an error?

> > +       if (!perf->link_is_up)
> > +               cancel_delayed_work_sync(&perf->link_work);
> > +
> > +       for (i = 0; i < MAX_SPAD; i++)
> 
> Complete aside here, but we might need to add logic to the core NTB
> code to query the number of available SPADs and error out of we try to
> get more than that.

Use ntb_spad_count(), and the count does vary.  For example, there are half the number of spads in RP/TB topology than B2B.