Re: [PATCH 07/11] firmware: arm_scmi: Add support for asynchronous commands and delayed response

From: Jim Quinlan
Date: Fri Jul 19 2019 - 16:20:50 EST


On Fri, Jul 19, 2019 at 7:03 AM Sudeep Holla <sudeep.holla@xxxxxxx> wrote:
>
> On Thu, Jul 18, 2019 at 05:38:06PM -0400, Jim Quinlan wrote:
> > Hi Sudeep,
> >
> > Just a comment in general. The asynchronous commands you are
> > implementing are not really asynchronous to the caller.
>
> Yes, but as per specification, the idea is to release the channel for
> other communication.
>
> > Yes it is is "async" at the low level, but there is no way to use
> > scmi_do_xfer() or scmi_do_xfer_with_response() and have the calling
> > thread be able to continue on in parallel with the command being
> > processed by the platform. This will limit the types of applications
> > that can use SCMI (perhaps this is intentional).
>
> Yes indeed, it's intentional and async is applicable for channel usage.
>
> > I was hoping that true async would be possible, and that the caller
> > could also register a callback function to be invoked when the command
> > was completed. Is this something that may be added in the future?
>
> This is how notifications are designed and must work. I would suggest
> to use notifications instead. Do you see any issues with that approach ?
>
> > It does overlap with notifications, because with those messages you
> > will need some kind of callback or handler thread.
> >
>
> Ah you do mention about overlap. I am replying as I read, sorry for that.
> Anyways, let me know if we can just use notifications. Also the current
> sync users(mainly clocks and sensors), may need even change in Linux
> infrastructure if we need to make it work in notification style.
>
> It would be good to know the use case you are referring.
Hi Sudeep,

Well, I'm just curious how you would implement notifications. Would
you have a per-protorcol callback? The Spec says that multiple agents
can receive them; in our usage we have only one agent and it is Linux.

We have one use case where that this patchset will do wonderfully. We
have another use case where we would like to go crazy on the
asynchrony of the messages (caller's perspective, that is).
This usage, which I don't think I can talk about, would like to use
notifications and a per-protocol callback function.
>
> > BTW, if scmi_do_xfer_with_response() returns --ETIMEDOUT the caller
> > has no way of knowing whether it was the command ack timeout or the
> > command execution timeout.
> >
> Yes, I did think about this but I left it as is thinking it may not be
> important. Do you think that makes a difference ? I am fine to change
> if there are users that needs to handle the difference.
I can't think of a case where it would matter. Just thought I'd mention it.

Cheers,
Jim
>
> --
> Regards,
> Sudeep