Re: [RFC PATCH 04/10] pipe: Use head and tail pointers for the ring, not cursor and length [ver #2]

[Linus Torvalds]

This still has signs of that earlier series:

On Wed, Oct 23, 2019 at 4:17 PM David Howells <dhowells@xxxxxxxxxx> wrote:
>
>                 if (rem >= ibuf->len) {
>                         *obuf = *ibuf;
>                         ibuf->ops = NULL;
> -                       pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
> -                       pipe->nrbufs--;
> +                       tail++;
> +                       pipe_commit_read(pipe, tail);
>                 } else {
>                         if (!pipe_buf_get(pipe, ibuf))
>                                 goto out_free;

with those odd "pipe_commit_read/write()" helpers.

They make no sense, and they don't make things more legible.

It's shorter and more obvious to just write

   pipe->head = head;

than it is to write

   pipe_commit_write(pipe, head);

Even when the addition of the notifications,  it's all under the
pipe->wait.lock, so it's all just regular assignments.

Now, if at some point it starts doing fancy lockless things, at _that_
point the updates might become more complex, but that's a potential
future thing that wouldn't be relevant for a while, and isn't a reason
to make the code more obscure now.

Hmm?

             Linus