mlx5(4) jumbo receive

[Konstantin Belousov]

On Wed, Apr 25, 2018 at 04:04:13PM -0400, Ryan Stone wrote:
> On Tue, Apr 24, 2018 at 4:55 AM, Konstantin Belousov
> <kostikbel at gmail.com> wrote:
> > +#ifndef MLX5E_MAX_RX_BYTES
> > +#define        MLX5E_MAX_RX_BYTES MCLBYTES
> > +#endif
> 
> Why do you use a 2KB buffer rather than a PAGE_SIZE'd buffer?
> MJUMPAGESIZE should offer significantly better performance for jumbo
> frames without increasing the risk of memory fragmentation.
Part of the answer is that the patch was not written in one go (even not
by one person), but evolved, and this is how it shaped.

Another part is that indeed, as Rick stated, I am not sure about mixing
the different sizes for mbuf allocator.  This might be more FUD than
factual-based considerations, but still.

I believe that the patch as is provides the important improvements.
If developing mlx4(4) change of the same nature, I will probably take
this into the exp stage from the beginning.  For mlx5(4), I think
that the patch should be applied as is, then I might  experiment
with PAGE_SIZE as the later step.