Message-ID: <20170131101356.bpg5hamkfgprdbpt@dell>
Date:   Tue, 31 Jan 2017 10:13:56 +0000
From:   Lee Jones <lee.jones@...aro.org>
To:     Peter Griffin <peter.griffin@...aro.org>
Cc:     gregkh@...uxfoundation.org, jslaby@...e.com,
        linux-serial@...r.kernel.org, robh+dt@...nel.org,
        devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
        linux-arm-kernel@...ts.infradead.org, kernel@...inux.com
Subject: Re: [STLinux Kernel] [PATCH 3/8] serial: st-asc: Read in all Pinctrl
 states

On Mon, 30 Jan 2017, Peter Griffin wrote:
> On Mon, 30 Jan 2017, Lee Jones wrote:
> > On Mon, 30 Jan 2017, Peter Griffin wrote:
> > > On Mon, 30 Jan 2017, Lee Jones wrote:
> > > > On Mon, 30 Jan 2017, Peter Griffin wrote:
> > > > > On Fri, 27 Jan 2017, Lee Jones wrote:
> > > > > > On Wed, 25 Jan 2017, Peter Griffin wrote:
> > > > > > > On Tue, 24 Jan 2017, Lee Jones wrote:
> > > > > > > 
> > > > > > > > There are now 2 possible separate/different Pinctrl states which can
> > > > > > > > be provided from platform data.  One which encompasses the lines
> > > > > > > > required for HW flow-control (CTS/RTS) and another which does not
> > > > > > > > specify these lines, such that they can be used via GPIO mechanisms
> > > > > > > > for manually toggling (i.e. from a request by `stty`).
> > > > > > > > 
> > > > > > > > Signed-off-by: Lee Jones <lee.jones@...aro.org>
> > > > > > > > ---
> > > > > > > >  drivers/tty/serial/st-asc.c | 28 ++++++++++++++++++++++++++++
> > > > > > > >  1 file changed, 28 insertions(+)
> > > > > > > > 
> > > > > > > > diff --git a/drivers/tty/serial/st-asc.c b/drivers/tty/serial/st-asc.c
> > > > > > > > index 397df50..03801ed 100644
> > > > > > > > --- a/drivers/tty/serial/st-asc.c
> > > > > > > > +++ b/drivers/tty/serial/st-asc.c
> > 
> > [...]
> > 
> > > > > > > > +		pinctrl_lookup_state(ascport->pinctrl, "manual-rts");
> > > > > > > > +	if (IS_ERR(ascport->states[MANUAL_RTS]))
> > > > > > > > +		ascport->states[MANUAL_RTS] = NULL;
> > > > > > > > +
> > > > > > > 
> > > > > > > The different pinctrl states looks like a neat solution to the problem.
> > > > > > > 
> > > > > > > My only concern here is that 'default' state is implying a hw-flow-control
> > > > > > > pinmux config, and manual-rts is implying what is the current upstream
> > > > > > > 'default' pinmux config.
> > > > > > > 
> > > > > > > Which maybe ok if you update all uarts, but currently only serial0
> > > > > > > is updated. So the other uarts current 'default' is actually the same as serial0
> > > > > > > 'manual-rts' grouping, which conceptually is odd.
> > > > > > > 
> > > > > > > Would it not be better to make 'manual-rts' the default state? As that aligns
> > > > > > > to what is currently already the default for the other UARTS? And then make
> > > > > > > hw-flow-control the optional state for serial0?
> > > > > > > 
> > > > > > > That also has the advantage that 'default' has the same meaning with older DT's.
> > > > > > 
> > > > > > The reason it was done is this was because none of the other UARTs
> > > > > > require 2 separate Pinctrl configurations, only this one.  Moreover,
> > > > > > if they support RTS/CTS then I believe that the lines should be
> > > > > > defined in Pinctrl.
> > > > > 
> > > > > Yes I agree with that.
> > > > > 
> > > > > > Thus, it was my plan to update all UART's default
> > > > > > Pinctrl configs to include the RTS/CTS lines.
> > > > > > 
> > > > > 
> > > > > I still don't see the point in changing the meaning of 'default' group and breaking
> > > > > ABI if you don't need to?
> > > > > 
> > > > > As far as I can tell if you swap the meaning of 'default' and 'maunal-rts'
> > > > > groups you get all the benefits of this series whilst also maintaining backwards
> > > > > compatbility with older DT's.
> > > > 
> > > > What makes you think this will break ABI?
> > > 
> > > I've not tried it, but an older DT defines one group, 'default' which contains
> > > the same pin config as your new optional 'manual-rts' group.
> > > 
> > > The driver now reads like the manual-rts pin config is optional and should be stored in
> > > ascport->states[MANUAL_RTS]. An older DT will pass that same pin config as the default
> > > group and it will be stored in ascport->states[DEFAULT].
> > > 
> > > That seems wrong to me, and if it executes OK it wouldn't be what you
> > > expect by reading the code.
> > 
> > This makes no sense at a functional level.
> > 
> > Old kernel, old DTB:
> > 
> > ASC driver doesn't understand Pinctrl, but since only the "default"
> > state is defined, that's what will be used as a matter of course.
> > RTS/CTS aren't configured, but that doesn't matter because the DTS
> > does not advertise that HW flow-control is available.  In this
> > use-case neither HW flow-control, nor manual toggling of the RTS line
> > is possible.
> > 
> > New kernel, old DTB:
> > 
> > ASC driver demands "default" and requests "manual-rts" Pinctrl states,
> > but "manual-rts" isn't available so "default" will be the only
> > utilised state.  Unlike the first example above, "default" now
> > contains the RTS and CTS lines,
> 
> No it doesn't, default just contains 'tx' & 'rx' pins, as it has always
> done until now.
> 
> Which is IMO where the condusion arises, as it is the same pin configuration
> as what you are now calling 'manual-rts' which the driver just tried and failed
> to obtain (although in reality it has actually obtained those pins but stored
> them in DEFAULT instead.
> 
> I presume this is why it didn't make sense to you above.

I guess this is what happens when you try to explain semantics last
thing, after a long day at work.  I chopped and changed the
descriptions and the ordering of these and it looks like some
peculiarities arose as a result.  Let me try again with a fresh(ish)
mind.

New kernel, old DTB:

ASC driver demands "default" and requests "manual-rts" Pinctrl states,
but "manual-rts" isn't available so "default" will be the only
utilised state.  The RTS and CTS lines will not be present, but since
the DTB is not advertising HW flow-control as a possibility, the IP
will not try to use those lines anyway.  [DEFAULT] will contain the
"default" state as proposed by the current DTB, so that is also
semantically correct.

> >but since the DTS does not advertise
> > HW flow-control as available they will be harmlessly unused.  This
> > configuration culminates in the same result as the first example
> > i.e. no HW flow-control and no manual toggling.  However, there are no
> > detremental effects to the driver's functions. 
> >
> 
> <snip>
> 
> >New kernel, new DTB:
> > 
> > ASC driver demands "default" and requests "manual-rts" Pinctrl
> > states.  If DTS advertises that HW flow-control is possible and the
> > client requests it, ASC will use the "default" state and HW
> > flow-control will commence.  If HW flow-control is not requested by
> > the client and "manual-rts" is available, then ASC will request the
> > RTS line is handled by GPIO until such times as the client requests
> > HW flow-control, at which point ASC will disable GPIO and request the
> > "default" state again.
> 
> Unless it is uart 1 or 2, in which case 'default' still only contains
> tx & rx pins, and you have the same situation as above. 

Doesn't matter.  "default" is non-descriptive.  I could understand an
argument were you to say that the "manual-rts" should not contain a
non-manual-rts state, but the "default" state should just contain
whatever the default configuration is, and in the case of UART 1 and
UART 2 the default state (until they are HW flow-control enabled --
which I plan to do as a follow-up) is not to provide HW flow-control
pins.  These semantics are unchanged since authorship of the driver.

> > It is not possible to read C-code and make assumptions that the DTB
> > will be in a particular state as you suggest.
> > No disparity ever
> > exists and the code is always clear IMHO.
> > 
> 
> Really?

Yes.

> ascport->states[DEFAULT]: may contain "tx, rx" or "tx, rx, cts & rts"

Correct.  "DEFAULT" does not mean "HW flow-control only".  It's
whatever the default is, so can correctly contain either state,
depending on what the default state of the DTB is.

> ascport->states[MANUAL_RTS]: may contain "tx, rx", or it could be stored in DEFAULT

The last part of this is reiterating your previous point, which I
just answered.  The correct description would be; "may contain *only*
"tx, rx", allowing "rts" to be manually controlled OR, may not be
populated".  In the latter case it would not be semantically incorrect
for DEFAULT to be either HW flow-control capable "tx, rx, rts, cts" or
not "tx, rx" -- whichever is the default of the supplied DTB.

> And as the series currently is you have a mixture of the two in the same kernel
> depending on what instance of the UART you are.

Again, doesn't matter, since it's the DTB that provides the default
state.  So, back when it was authored, the default state was HW
flow-control disabled.  And in a newer DTB (again, until I follow-up
with more changes), the defaults for UART 1 and UART 2 are HW
flow-control disabled.

Your issue seems to be that you've assumed since we now provide the
possibility of a "manual-rts" state, then the "default" state should
*only* be HW flow-control capable, which is not the case.  It's the
'uart-has-rtscts' property which determines this *not* whether the
second state has been provided.  It is not logical to make any
inference using solely the presence or absence of the "manual-rts"
state.

-- 
Lee Jones
Linaro STMicroelectronics Landing Team Lead
Linaro.org │ Open source software for ARM SoCs
Follow Linaro: Facebook | Twitter | Blog

Powered by blists - more mailing lists