On Tue, 2024-08-13 at 13:10 -0400, Olivier Langlois wrote:
>
> ---
> include/linux/io_uring_types.h | 12 +++++-
> io_uring/fdinfo.c | 4 ++
> io_uring/napi.c | 76 ++++++++++++++++++++++++++++++--
> --
> io_uring/napi.h | 11 +----
> 4 files changed, 86 insertions(+), 17 deletions(-)
>
> diff --git a/include/linux/io_uring_types.h
> b/include/linux/io_uring_types.h
> index 3315005df117..c1d1b28f8cca 100644
> --- a/include/linux/io_uring_types.h
> +++ b/include/linux/io_uring_types.h
> @@ -217,6 +217,16 @@ struct io_alloc_cache {
> size_t elem_size;
> };
>
> +#ifdef CONFIG_NET_RX_BUSY_POLL
> +struct io_napi_tracking_ops {
> + void (*add_id)(struct io_kiocb *req);
> + bool (*do_busy_loop)(struct io_ring_ctx *ctx,
> + void *loop_end_arg);
> + void (*show_fdinfo)(struct io_ring_ctx *ctx,
> + struct seq_file *m);
> +};
> +#endif
> +
I have kept thinking about the critic...
add_id is either NULL or equal to dynamic_tracking_add_id and the
pointer is even tested before calling it. This pointer is easily
removed.
show_fdinfo, well, this is is so unimportant, if you don't like it, it
is very easily removable too. nobody will notice.
the only thing that would remains is do_busy_loop. Its value can either
be:
- no_tracking_do_busy_loop
- dynamic_tracking_do_busy_loop
- static_tracking_do_busy_loop
so the whole io_napi_tracking_ops could be replaced by a single
function pointer
At this point, the only thing remaining point to determine is which
between calling a function pointer of calling a 2 conditional branches
code is more efficient. and I am of the opinion that the function
pointer is better due to my C++ background but this is debatable...
