Re: [RESEND PATCH 05/11] platform/chrome: cros_ec_proto: separate cros_ec_wait_until_complete()

[Tzung-Bi Shih <tzungbi@kernel.org>]

On Wed, Jul 13, 2022 at 11:15:47AM -0700, Guenter Roeck wrote:
> On Mon, Jun 27, 2022 at 7:49 PM Tzung-Bi Shih <tzungbi@kernel.org> wrote:
> > -static int cros_ec_send_command(struct cros_ec_device *ec_dev, struct cros_ec_command *msg)
> > +static int cros_ec_wait_until_complete(struct cros_ec_device *ec_dev, uint32_t *result)
> >  {
> > -       int ret = cros_ec_xfer_command(ec_dev, msg);
> > +       struct cros_ec_command *msg;
> > +       struct ec_response_get_comms_status *status;
> > +       int ret = 0, i;
> > +
> > +       msg = kzalloc(sizeof(*msg) + sizeof(*status), GFP_KERNEL);
> > +       if (!msg)
> > +               return -ENOMEM;
> 
> AFAICS this is always 24 bytes. I would suggest to allocate it on the
> stack to reduce overhead.

Ack.

> > +               ret = cros_ec_xfer_command(ec_dev, msg);
> > +               if (ret == -EAGAIN)
> > +                       continue;
> > +               if (ret < 0)
> > +                       break;
> 
> With the command allocated on the stack, this can return immediately.

Nack, the function has no goto labels.  `return ret` follows the loop
immediately.  The `break` here doesn't make it to become too complicated.
I would prefer to keep it.

> > +
> > +               *result = msg->result;
> > +               if (msg->result != EC_RES_SUCCESS)
> > +                       break;
> 
> Again, this can return immediately if the command buffer is on the stack.

Nack.  See above.

> > -               kfree(status_msg);
> > +               if (!(status->flags & EC_COMMS_STATUS_PROCESSING))
> > +                       break;
> 
> Can return immediately.

Nack.  See above.

> > +       kfree(msg);
> > +       return ret;
> 
> What should this return on timeout ?

It returns either:
* -EAGAIN, if cros_ec_xfer_command() returned -EAGAIN
* 0, if EC_COMMS_STATUS_PROCESSING flag was on
for EC_COMMAND_RETRIES times so far.

This is a "move" refactor.  I would prefer to keep it as is and change the
behavior in later patch.