On Tue, Feb 26, 2019 at 01:13:17AM -0500, Steven Rostedt wrote:
> On Mon, 25 Feb 2019 21:52:46 -0800
> Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx> wrote:
>
> > On Mon, Feb 25, 2019 at 11:36:40AM -0500, Steven Rostedt wrote:
> > > On Sun, 24 Feb 2019 21:11:59 -0800
> > > ira.weiny@xxxxxxxxx wrote:
> > > >
> > > > +static void create_mad_addr_info(struct ib_mad_send_wr_private *mad_send_wr,
> > > > + struct ib_mad_qp_info *qp_info,
> > > > + u32 *dlid, u8 *sl, u16 *pkey, u32 *rqpn,
> > > > + u32 *rqkey)
> > > > +{
> > >
> > > Instead of passing in a bunch of addresses, you may be able to pass in
> > > the entry structure directly.
> > >
> > > struct trace_event_raw_ib_mad_send_template *entry
> > >
> > > (defined from DECLARE_EVENT_CLASS(name, ...
> > >
> > > struct trace_event_raw_##name .. )
> > >
> > > And just assign everything as:
> > >
> > > entry->dlid = x, ...
> >
> > raw_tracepoint requires all args to fit into u64.
> > bpf prog cannot access it otherwise.
> > Also struct by value is pretty slow on some archs.
> > Pointer to struct is always preferred.
>
> I meant passing in the pointer, not the value. It wouldn't make sense
> otherwise, because the entire point of create_mad_addr_info(), is to
> update the __entry structure. How would passing in the value do that?
>
>
> Instead of:
>
> + create_mad_addr_info(wr, qp_info,
> + &__entry->dlid, &__entry->sl,
> + &__entry->pkey, &__entry->rqpn,
> + &__entry->rqkey);
>
> You do:
>
> create_mad_addr_info(wr, qp_info, __entry);
>
> And then in that function you assign the values directly to the structure.
Is there any reason this is preferred over what I did? I'm not seeing the
benefit?
Ira
>
> -- Steve
>
>
