[openib-general] [PATCH] (resend) mthca_eq improvements

[Michael S. Tsirkin]

Hello!
Quoting r. Roland Dreier (roland at topspin.com) "Re: [openib-general] [PATCH] (resend) mthca_eq improvements":
>     Michael> Hello, Roland, all!  Here are some mthca eq improvements,
>     Michael> rolled into one patch.  These are all touching the same
>     Michael> code and so are related, but if its too big to be
>     Michael> readable, let me know and I'll try to split it up.
> 
> It looks like you concatenated more than one patch
No, I re-ordered them by hand for clarity :)

>  (there are out of
> order chunks, which at least my version of patch complains about).

Mine doesnt :)

> Anyway, easy to split up...

Fine, let me know if you need help.

>     Michael> 1. general interrupt handling speedup I noticed that
>     Michael> several operations where performed multiple times, for
>     Michael> example get_eqe was called multiple times, as another
>     Michael> example, consumer index was truncated to eq size upon
>     Michael> each eqe found.
> 
>     Michael> 2. non-MSI interrupt handling simplification This is the
>     Michael> 3'd chunk from the end.
> 
> Looks great, I'll apply these two pieces.  With MSI-X I get about a 3%
> increase in throughput with netperf over IPoIB with these changes.


>     Michael> 3. Added memfree support in the eq initialisation
>     Michael> command.  This is in the last 2 chunks.
> 
> This looks wrong to me:
> 
> +	if (dev->hca_type == ARBEL_NATIVE)
> +		page = MTHCA_KAR_PAGE;
> +	else
> +		page = 0;
> 
> the test is reversed, right?

Yes it is ... the reason it works (I did test it :)) is because if you use
page 1 you dont have to supply the page in the context at all -
KAR_PAGE is priviledged, so you can do whatever you want with it.

> I'm going to hold off applying this change until I have the rest of
> mem-free EQ support ready.
> 
> Thanks,
>   Roland

Just for this last one - probably a good idea.