[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20230206225018.50d62d62@kernel.org>
Date: Mon, 6 Feb 2023 22:50:18 -0800
From: Jakub Kicinski <kuba@...nel.org>
To: Jeroen de Borst <jeroendb@...gle.com>
Cc: netdev@...r.kernel.org, davem@...emloft.net,
Leon Romanovsky <leonro@...dia.com>
Subject: Re: [PATCH net-next v4] gve: Introduce a way to disable queue
formats
On Sat, 4 Feb 2023 11:29:40 -0800 Jeroen de Borst wrote:
> The device is capable of simultaneously supporting multiple
> queue formats. These queue formats are:
>
> - GQI-QPL: A queue format with in-order completions and a
> bounce-buffer (Queue Page List)
> - GQI-RDA: A queue format with in-order completions and no
> bounce-buffer (Raw DMA Access)
> - DQO-RDA: A queue format with out-of-order completions and
> no bounce buffer
Thanks but..
> With this change the driver can deliberately pick a queue format.
Driver can already do whatever it wants. Now the _user_ can pick
the format. But the user still has no understanding of what the
practical impact of picking one queue format over another will be.
Do you have a reason to believe that the description above (and in
docs) will be sufficient for user to make a decision?
I tried to search the web but got no hits to any GCP docs either.
Differently put what is you motivation to give this control to the user?
> +
> +struct bpf_prog;
> static int gve_verify_driver_compatibility(struct gve_priv *priv)
Adding the forward declaration for bpf_prog looks like a stray change.
Powered by blists - more mailing lists