On Tue, Oct 09, 2018 at 11:31:42AM -0700, Brian Norris wrote:
> On Tue, Oct 09, 2018 at 09:22:07AM +0200, Geert Uytterhoeven wrote:
> > Please note these aliases become cumbersome once you start considering
> > (dynamic) DT overlays. That's why I made them optional in the sh-sci
> > serial driver, cfr. commit 7678f4c20fa7670f ("serial: sh-sci: Add support
> > for dynamic instances").
>
> Note that as I understand it, the entire point of documenting this sort
> of thing is to help solidify the interface between a DT aware boot
> program (e.g., bootloader) and a device tree which is provided
> separately, to avoid memorizing node/path hierarchy. It doesn't need to
> (and doesn't, as I read it) enforce an OS's device naming policy.
I'm all for documenting this primarily to prevent folks from just adding
whatever they wish in /aliases. Some platforms seem to want to have
aliases for everything.
> > Relevant parts of the commit description are:
> >
> > On DT platforms, the sh-sci driver requires the presence of "serialN"
> > aliases in DT, from which instance IDs are derived. If a DT alias is
> > missing, the drivers fails to probe the corresponding serial port.
> >
> > This becomes cumbersome when considering DT overlays, as currently
> > there is no upstream support for dynamically updating the /aliases node
> > in DT.
>
> That part is not a DT spec problem :)
>
> > Furthermore, even in the presence of such support, hardcoded
> > instance IDs in independent overlays are prone to conflicts.
> >
> > Hence add support for dynamic instance IDs, to be used in the absence of
> > a DT alias. This makes serial ports behave similar to I2C and SPI
> > buses, which already support dynamic instances.
>
> This seems to be a much different sort of problem. People always love
> having predictable IDs given by the OS (myself included), but that's
> just plain hard to do and impossible in some cases. I don't think that's
> what this document is about though.
>
> IOW, this document seems pretty consistent with the above: it doesn't
> require the usage of aliases (and it seems silly to have a driver
> *require* an alias) -- it just documents how one should name such an
> alias if you expect multiple independent software components to
> understand it.
>
> > To clarify my point: R-Car M2-W has 4 different types of serial ports, for a
> > total of 18 ports, and the two ports on a board labeled 0 and 1 may not
> > correspond to the physical first two ports (what's "first" in a collection of
> > 4 different types?).
> >
> > Aliases may be fine for referring to the main serial console (labeled
> > port 0 on the device, too), and the primary Ethernet interface (so U-Boot
> > knows where to add the "local-mac-address" property), but beyond that,
> > I think they should be avoided.
This basically matches my opinion on aliases.
I'd decouple it from board labels a bit. Sometimes the numbering may
match, but others not. What if a board serial port is labeled "DBG" for
example? I think 'label' is the right way to handle human identifible
ports (and then we should have something like /dev/serial/by-label/...).
> That's fair enough. Just because the solution isn't an all-purpose tool
> doesn't mean it shouldn't be documented. The general concept is already
> in ePAPR, but it's just not very specific about property names.
Agreed. I guess the question is what to do on used, but not recommended
aliases. I would put SPI and I2C into that category BTW.
Rob
