mbox series

[RFC,0/4] i2c: core: add generic GPIO bus recovery

Message ID 20200619141904.910889-1-codrin.ciubotariu@microchip.com
Headers show
Series i2c: core: add generic GPIO bus recovery | expand

Message

Codrin Ciubotariu June 19, 2020, 2:19 p.m. UTC 
GPIO recovery has been added already for some I2C bus drivers, such as
imx, pxa and at91. These drivers use similar bindings and have more or
less the same code for recovery. For this reason, we aim to move the
GPIO bus recovery implementation to the I2C core so that other drivers
can benefit from it, with small modifications.
This implementation initializes the pinctrl states and the SDA/SCL
GPIOs based on common bindings. The I2C bus drivers can still use
different bindings or other particular recovery steps if needed.
The ugly part with this patch series is the handle of PROBE_DEFER
which could be returned by devm_gpiod_get(). This changes things a
little for i2c_register_adapter() and for this reason this step is
implemented in a sperate patch.
The at91 Microchip driver is the first to use this implementation,
with an AI to move the rest of the drivers in the following steps.

Codrin Ciubotariu (4):
  dt-binding: i2c: add generic properties for GPIO bus recovery
  i2c: core: add generic I2C GPIO recovery
  i2c: core: treat EPROBE_DEFER when acquiring SCL/SDA GPIOs
  i2c: at91: Move to generic GPIO bus recovery

 Documentation/devicetree/bindings/i2c/i2c.txt |  10 ++
 drivers/i2c/busses/i2c-at91-master.c          |  69 +--------
 drivers/i2c/i2c-core-base.c                   | 139 +++++++++++++++++-
 include/linux/i2c.h                           |  11 ++
 4 files changed, 158 insertions(+), 71 deletions(-)

Comments

Wolfram Sang July 5, 2020, 9:19 p.m. UTC | #1 
> +- pinctrl

> +	add extra pinctrl to configure SCL/SDA pins to GPIO function for bus

> +	recovery, call it "gpio" or "recovery" state


I think we should stick with "gpio" only. That is what at91 and imx have
in their bindings. pxa uses "recovery" as a pinctrl state name but I
can't find any further use or documentation of that. PXA is not fully
converted to the best of my knowledge, so maybe it is no problem for PXA
to switch to "gpio", too? We should ask Russell King (cced).

Russell, do you object naming the pinctrl state for bus recovery in
the pxa i2c driver from "recovery" to "gpio"?
Codrin Ciubotariu July 27, 2020, 10:44 a.m. UTC | #2 
On 24.07.2020 23:52, Russell King - ARM Linux admin wrote:
> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe

> 

> On Fri, Jul 24, 2020 at 09:39:13PM +0200, Wolfram Sang wrote:

>> On Sun, Jul 05, 2020 at 11:19:18PM +0200, Wolfram Sang wrote:

>>>

>>>> +- pinctrl

>>>> + add extra pinctrl to configure SCL/SDA pins to GPIO function for bus

>>>> + recovery, call it "gpio" or "recovery" state

>>>

>>> I think we should stick with "gpio" only. That is what at91 and imx have

>>> in their bindings. pxa uses "recovery" as a pinctrl state name but I

>>> can't find any further use or documentation of that. PXA is not fully

>>> converted to the best of my knowledge, so maybe it is no problem for PXA

>>> to switch to "gpio", too? We should ask Russell King (cced).

> 

> Fully converted to what?  The generic handling where the i2c core layer

> handles everything to do with recovery, including the switch between

> modes?

> 

> i2c-pxa _intentionally_ carefully handles the switch between i2c mode and

> GPIO mode, and I don't see a generic driver doing that to avoid causing

> any additional glitches on the bus.  Given the use case that this recovery

> is targetted at, avoiding glitches is very important to keep.


Why is it not possbile to handle glitches in a generic way? I guess it 
depends on the pinctl, but we could treat a worst-case scenario to 
assure the switch between states is done properly.

> 

>>> Russell, do you object naming the pinctrl state for bus recovery in

>>> the pxa i2c driver from "recovery" to "gpio"?

>>

>> No response, so far. I suggest now to support the "recovery" naming but

>> mark it as deprecated. Opinions?

> 

> I don't have a preference on the exact naming.

> 

> --

> RMK's Patch system: https://www.armlinux.org.uk/developer/patches/

> FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!

>
Russell King (Oracle) July 27, 2020, 10:50 a.m. UTC | #3 
On Mon, Jul 27, 2020 at 10:44:57AM +0000, Codrin.Ciubotariu@microchip.com wrote:
> On 24.07.2020 23:52, Russell King - ARM Linux admin wrote:

> > EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe

> > 

> > On Fri, Jul 24, 2020 at 09:39:13PM +0200, Wolfram Sang wrote:

> >> On Sun, Jul 05, 2020 at 11:19:18PM +0200, Wolfram Sang wrote:

> >>>

> >>>> +- pinctrl

> >>>> + add extra pinctrl to configure SCL/SDA pins to GPIO function for bus

> >>>> + recovery, call it "gpio" or "recovery" state

> >>>

> >>> I think we should stick with "gpio" only. That is what at91 and imx have

> >>> in their bindings. pxa uses "recovery" as a pinctrl state name but I

> >>> can't find any further use or documentation of that. PXA is not fully

> >>> converted to the best of my knowledge, so maybe it is no problem for PXA

> >>> to switch to "gpio", too? We should ask Russell King (cced).

> > 

> > Fully converted to what?  The generic handling where the i2c core layer

> > handles everything to do with recovery, including the switch between

> > modes?

> > 

> > i2c-pxa _intentionally_ carefully handles the switch between i2c mode and

> > GPIO mode, and I don't see a generic driver doing that to avoid causing

> > any additional glitches on the bus.  Given the use case that this recovery

> > is targetted at, avoiding glitches is very important to keep.

> 

> Why is it not possbile to handle glitches in a generic way? I guess it 

> depends on the pinctl, but we could treat a worst-case scenario to 

> assure the switch between states is done properly.


Please look at how i2c-pxa switches between the two, and decide whether
the generic implementation can do the same.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!
Russell King (Oracle) Aug. 3, 2020, 2:16 p.m. UTC | #4 
On Thu, Jul 30, 2020 at 09:00:36AM +0000, Codrin.Ciubotariu@microchip.com wrote:
> On 27.07.2020 13:50, Russell King - ARM Linux admin wrote:

> > EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe

> > 

> > On Mon, Jul 27, 2020 at 10:44:57AM +0000, Codrin.Ciubotariu@microchip.com wrote:

> >> On 24.07.2020 23:52, Russell King - ARM Linux admin wrote:

> >>> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe

> >>>

> >>> On Fri, Jul 24, 2020 at 09:39:13PM +0200, Wolfram Sang wrote:

> >>>> On Sun, Jul 05, 2020 at 11:19:18PM +0200, Wolfram Sang wrote:

> >>>>>

> >>>>>> +- pinctrl

> >>>>>> + add extra pinctrl to configure SCL/SDA pins to GPIO function for bus

> >>>>>> + recovery, call it "gpio" or "recovery" state

> >>>>>

> >>>>> I think we should stick with "gpio" only. That is what at91 and imx have

> >>>>> in their bindings. pxa uses "recovery" as a pinctrl state name but I

> >>>>> can't find any further use or documentation of that. PXA is not fully

> >>>>> converted to the best of my knowledge, so maybe it is no problem for PXA

> >>>>> to switch to "gpio", too? We should ask Russell King (cced).

> >>>

> >>> Fully converted to what?  The generic handling where the i2c core layer

> >>> handles everything to do with recovery, including the switch between

> >>> modes?

> >>>

> >>> i2c-pxa _intentionally_ carefully handles the switch between i2c mode and

> >>> GPIO mode, and I don't see a generic driver doing that to avoid causing

> >>> any additional glitches on the bus.  Given the use case that this recovery

> >>> is targetted at, avoiding glitches is very important to keep.

> >>

> >> Why is it not possbile to handle glitches in a generic way? I guess it

> >> depends on the pinctl, but we could treat a worst-case scenario to

> >> assure the switch between states is done properly.

> > 

> > Please look at how i2c-pxa switches between the two, and decide whether

> > the generic implementation can do the same.

> 

> The handling of glitches from initialization looks generic to me. I see 

> that there are specific clear/reset routines that are in the 

> (un)prepare_recovery() callbacks, but these callbacks are not replaced 

> by the generic i2c recovery and will still be used if given by the 

> driver. The only thing the generic recovery does is to switch the pinmux 

> state. We can discuss whether we want to change the pinmux state first 

> or call the (un)preapre_recovery().


Right, the key point i2c-pxa does is that on prepare:
- read the current state of the SCL and SDA lines and set the GPIO to
  reflect those values.
- then switch the pinmux state.

That must be preserved, otherwise if SCL is being held low by the I2C
master, and we switch to GPIO mode, SCL will be released.  So the
driver needs to be involved before the pinmux state is changed.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!