Re: [PATCH v4 2/4] serial: mxs-auart: use mctrl_gpio helpers for handling modem signals

[Marek Vasut <marex@denx.de>]

On Friday, November 07, 2014 at 02:23:23 PM, Janusz Użycki wrote:
[...]

> >> Hardware RTS/CTS lines can be occupied by RX/TX of other AUART port
> >> in order to obtain as much uarts as possible using i.mx283.
> >> Therefore gpios can be used for "hardware" flow control.
> > 
> > Your logic is outright flawed here, the first sentence doesn't implicate
> > the second sentence. In fact, those two are completely unrelated.
> 
> I didn't write MUST but CAN. There is a choice. Is flexibility of the
> driver disadvantage?

If the flexibility brings in known problems, then yes, it is a problem. Not 
because of the flexibility, but because it brings in bugs.

> >>>>     If we change them to gpio.  Could the DMA still
> >>>> 
> >>>> works fine?
> >>>> did you test the DMA with this patch?
> >>>> 
> >>>> Add Marek for this patch too.
> >>> 
> >>> I didn't look too deep into the patch, so here's just my experience:
> >>> 
> >>> 1) The AUART block signals and GPIO block signals are not sychronised
> >>> using the
> >>> 
> >>>      same clock. Therefore, the latency between toggling of the AUART
> >>>      lines and the GPIO-driven pins will not be deterministic and will
> >>>      vary. There might be a way to approximate that, but that's
> >>>      definitelly not a reliable solution.
> >>>      
> >>>      This is very bad for example if you drive RS485 DIR line with the
> >>>      RTS pin as a GPIO ; the RTS pin will toggle at non-deterministic
> >>>      time compared to the end of UART transmission. This will trigger
> >>>      bit-loss on the RS485 line and you just don't want that.
> >>> 
> >>> 2) Speaking of RS485, there's [1] and [2]. which I believe apply to any
> >>> combo
> >>> 
> >>>      of UART+GPIO toggling.
> >>> 
> >>> So I hate to bring the bad news , but UART+GPIO combo toggling is
> >>> really a bad bad idea.
> >> 
> >> Unfortunately if hardware is limited there is no choice and UART+GPIO is
> >> necessary.
> > 
> > You will run into timing problems (see above).
> 
> A lot of 8250-compatible devices has no hardware flow control and in
> most cases
> they works and it is enough even for 115200 speed if CTS is handled by irq.
> So it depends on your needs.

I presume that in such a case , the 8250 still handles the CTS line, not some 
external GPIO block, yes ?

> > What you're proposing here is a workaround for broken hardware, which was
> > proven to be a bad idea and NAK'd already multiple times in the past
> > (please see the links I posted in my last email).
> 
> It is not broken  hardware - rather limited to lower speeds but still
> very useful solution.

What exact "lower speed" are you talking about here and why ?

> >> Your experience confirms the discussion [3] with Russell King. DMA
> >> should be disabled and
> >> the patch disables DMA support in mxs_auart_init_gpios() if RTS or CTS
> >> line is set as gpio.
> > 
> > DMA has nothing to do with those problems here. DMA can be safely ignored
> > for the purpose of the discussion altogether.
> 
> When gpios are used for RTS/CTS DMA is not used. However DMA is related
> due to the driver
> and "fsl,uart-has-rtscts". If you look into code of the driver you
> should agree.

This makes me believe that the DMA introduces too many timing fluctuations,
so it's really not possible for you to keep toggling the GPIOs such that the
bus would work. Is that the case ?

[...]

> >> Now the question: "fsl,uart-has-rtscts" name seems to be misleading now,
> >> do you agree? It rather should include "dma" word in the name. Any
> >> suggestion?
> >> 
> >> [3] http://thread.gmane.org/gmane.linux.serial/16069/focus=16077
> > 
> > The best suggestion I can give you is to fix your hardware early, before
> > you run into nasty deep s.....tuff. These workarounds do not work and
> > they will bit you later on, when it's hard to fix the hardware anymore.
> 
> The speed is limited but why don't you accept SW-HW mixed solutions?

Did you read up on the RS485 timing problems and why that solution was never
accepted for any driver ? I believe the threads explained that quite clearly.

> Exactly the same method is accepted for 8250.

Can you point out this code please ?

> It is good to have choice than not. We can comment that speed is limited
> for gpio-based hardware flow control. So please, rethink again.
[...]
--
To unsubscribe from this list: send the line "unsubscribe linux-serial" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html