[RFC v5 2/5] drivers: firmware: scmi: Introduce scmi_get_max_msg_size function

Cristian Marussi cristian.marussi at arm.com
Thu Nov 2 08:04:43 PDT 2023


On Thu, Nov 02, 2023 at 01:57:24PM +0000, Oleksii Moisieiev wrote:
> Hi Cristian,
> 

Hi,

> Just found an interesting note in the PINCTRL_CONFIG_SET command 
> description:
> 
> The maximum value of this field is limited by
> the transport used. The agent needs to specify
> this field such that the entire command can be
> accommodated within the transport chosen.
> 

Yes I am aware of that.

> Furthermore, I observed the absence of a skip_configs parameter.
> 
>  From my understanding, this implies that the maximum number of 
> configurations should not exceed the msg_max_size allowed by the 
> protocol, enabling the transmission of only one message to the SCMI 
> server at a time.
>

Yes that is correct, my understanding is that the transmitter is in
charge of building a message whose payload can fit into the maximum
message size allowed by the underlying configured transport.

> Given this constraint, it seems we might not require additional helper 
> functions. We could potentially just verify against msg_max_size.
> 

Indeed for that reason the scmi_get_max_msg_size that you introduced is
just enough since it allows you to peek into the transport and get the
max_msg_size...the misunderstanding is around the fact that I was simply
meaning that you should plug it into some new helper_ops so that yo can
call it like:

  max_msg = ph->hops->get_max_msg_size(ph);

(like iterators or get_extended_name)

Because in this way you could use it also when the protocol is build as
a loadable module, thing that now it is possible only for vendor defined
protocols, but we could also easily switch all the base protocols to be
selectable via Kconfig and =m in the future (if ever)

Your helper is fine by itself it is just that it cannot be called by a
protocol defined to loaded as a module because the symbol is not
exported and, indeed, we introduced the ph->hops thing just for this
reason, i.e. to have a set of common protocol utilities that can be
called even from loadable modules protocols without the need to export
every single symbol.

The reference to iterators and extended_name was misleading
probably...my bad...or I am still missing something :D

Thanks,
Cristian



More information about the linux-arm-kernel mailing list