Re: [PATCH] input: raydium_ts_i2c: Do not split tx transactions

[Dmitry Torokhov]

On Mon, Nov 30, 2020 at 10:54:46PM -0800, Furquan Shaikh wrote:
> Hello Dmitry,
> 
> On Mon, Nov 30, 2020 at 10:28 PM Dmitry Torokhov
> <dmitry.torokhov@xxxxxxxxx> wrote:
> >
> > Hi Furquan,
> >
> > On Mon, Nov 30, 2020 at 10:00:50PM -0800, Furquan Shaikh wrote:
> > > Raydium device does not like splitting of tx transactions into
> > > multiple messages - one for the register address and one for the
> > > actual data. This results in incorrect behavior on the device side.
> > >
> > > This change updates raydium_i2c_read and raydium_i2c_write to create
> > > i2c_msg arrays separately and passes those arrays into
> > > raydium_i2c_xfer which decides based on the address whether the bank
> > > switch command should be sent. The bank switch header is still added
> > > by raydium_i2c_read and raydium_i2c_write to ensure that all these
> > > operations are performed as part of a single I2C transfer. It
> > > guarantees that no other transactions are initiated to any other
> > > device on the same bus after the bank switch command is sent.
> >
> > i2c_transfer locks the bus [segment] for the entire time, so this
> > explanation on why the change is needed does not make sense.
> 
> The actual problem is with raydium_i2c_write chopping off the write
> data into 2 messages -- one for register address and other for actual
> data. Raydium devices do not like that. Hence, this change to ensure
> that the register address and actual data are packaged into a single
> message. The latter part of the above comment attempts to explain why
> the bank switch message is added to xfer[] array in raydium_i2c_read
> and raydium_i2c_write instead of sending a separate message in
> raydium_i2c_xfer i.e. to ensure that the read/write xfer and bank
> switch are sent to i2c_transfer as a single array of messages so that
> they can be handled as an atomic operation from the perspective of
> communication with this device on the bus.

OK, I see.

> 
> >
> > Also, does it help if you mark the data message as I2C_M_NOSTART in case
> > of writes?
> 
> That is a great suggestion. I think this would be helpful in this
> scenario. Let me follow-up on this to see if it helps with the current
> problem.
> 
> >
> > I also wonder if we should convert the driver to regmap, which should
> > help with handling the bank switch as well as figuring out if it can do
> > "gather write" or fall back to allocating an additional send buffer.
> 
> I will start with the above suggestion and fallback to this if that
> doesn't work.

So my understanding is that not all I2C adapters support I2C_M_NOSTART
so that is why regmap is nice as it hides it all away and figures things
on its own.

So simple solution of I2C_M_NOSTART might be a quick fix for Chrome OS
kernel, but we'd either need to always use more expensive 2nd buffer as
is in your patch, or regmap.

Thanks.

-- 
Dmitry