Re: [EXT] Re: [net-next, 1/2] enetc: Configure the Time-Aware Scheduler via tc-taprio offload

[Simon Horman <simon.horman@netronome.com>]

On Tue, Nov 12, 2019 at 11:19:43AM +0000, Po Liu wrote:

...

> > > +/* class 5, command 0 */
> > > +struct tgs_gcl_conf {
> > > +     u8      atc;    /* init gate value */
> > > +     u8      res[7];
> > > +     union {
> > > +             struct {
> > > +                     u8      res1[4];
> > > +                     __le16  acl_len;
> > 
> > Given that u* types are used in this structure I think le16 would be more
> > appropriate than __le16.
>  
> Here keep the same code style of this .h file. I think it is better to have another patch to fix them all. Do you agree?
> 
> > 

> > > +                     u8      res2[2];
> > > +             };
> > > +             struct {
> > > +                     u32 cctl;
> > > +                     u32 ccth;
> > > +             };
> > 
> > I'm a little surprised to see host endian values in a structure that appears to be
> > written to hardware. Is this intentional?
> 
> Will remove.

If the HW defines these fields then I think its fine to leave them,
though with the correct byte-order.

I was more asking if it is intentional that the value for these
fields, when sent to the HW, is always zero in the context of this
patch-set. Likewise elsewhere.

...

> > > +
> > > +     gcl_data->ct = cpu_to_le32(admin_conf->cycle_time);
> > > +     gcl_data->cte = cpu_to_le32(admin_conf->cycle_time_extension);
> > > +
> > > +     for (i = 0; i < gcl_len; i++) {
> > > +             struct tc_taprio_sched_entry *temp_entry;
> > > +             struct gce *temp_gce = gce + i;
> > > +
> > > +             temp_entry = &admin_conf->entries[i];
> > > +
> > > +             temp_gce->gate = cpu_to_le32(temp_entry->gate_mask);
> > 
> >         Gate is a u8 followed by 3 reserved bytes.
> >         Perhaps there needs to be some bounds checking on
> >         the value stored there given that the source is 32bits wide.
> > 
> >         Also, its not clear to me that the above logic, which I assume
> >         takes the last significant byte of a 32bit value, works on
> >         big endian systems as the 32bit value is always little endian.
> 
> temp_entry->gate_mask is 32bit for wide possible input. Here change to hardware set 8bit wide.
> Can it just be like:
> 	temp_gce->gate = (u8) temp_entry->gate_mask;

I think that would be better.
Perhaps its best to also mask out the unwanted bits.

...