On Mon, Feb 5, 2024 at 10:11 AM Peng Fan <peng.fan@xxxxxxx> wrote:
> [Me]
> > This is very much to the core of the problem isn't it?
>
> Yes. Code size should be as small as possible.
This is a valid argument, I don't know how big your SCMI FW is, or
if it has to fit into some on-chip memory or so.
> And using SCMI generic pinconf, there will be multiple
> SCMI calls(not MMIO access), such as setting mux(ops->set_mux)
> needs an SCMI call, pad settings(ops->pin_config_set) needs an SCMI call.
> And maybe ops->get_function_name needs an extra SCMI call.
>
> With current i.MX design, only one SCMI call is used, which
> use less time.
I think this could be fixed in the spec, let's see what the spec author says.
In any case: pin control is pretty much never on a hot path, a few
microseconds more or less doesn't matter. It's usually just used
and probe, suspend/resume and maybe shutdown. All usecases on
the slowpath, so I think it's a premature optimization.
> > Over the years I have come to regret it a bit, I think insisting on groups and
> > functions as strings is better for abstraction. And the point of firmware is to
> > abstract things so they work the same on all systems.
>
> With current:
> pinctrl_uart1: uart1grp {
> fsl,pins = <
> IMX95_PAD_UART1_RXD__AONMIX_TOP_LPUART1_RX 0x31e
> IMX95_PAD_UART1_TXD__AONMIX_TOP_LPUART1_TX 0x31e
> >;
> };
>
> It is very easy for people to know the meaning from reading reference manual.
Yes If the defines are provided it's quite readable. And I think Freescale
people really get used to it.
But from a maintenance and standardization point of view, it's nice if
everything works the same way and looks the same.
> If using generic pinconf, the dt node will be like
> Uartgrp {
> pins = "uart1txd", "uart1rxd";
> functions = "uart1";
> bias-xx
> drive-strength =
> };
Well that looks good to me :)
It looks the same as e.g. Qualcomm or nVidia pins. So it's easy for me
to read and understand, and that's my perspective as a maintainer.
> The firmware needs add more code to export functions, pack the conf settings,
> each pins needs a function name per current i.MX HW logic.
Indeed, but I think any standard requires a bit of code to implement.
How much is "too much" code is a matter of taste and physical contraints.
(A PC UEFI BIOS isn't exactly small either.)
I understand your point of view, and I think it is pretty much
the same stance that the Freescale maintainers put forward for the DT
bindings for Freescale.
Yours,
Linus Walleij
