Re: [PATCH v5 01/11] i2c: Enhance i2c_new_ancillary_device API

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Wed, Jun 14, 2023 at 08:21:38AM +0000, Biju Das wrote:
> Hi Laurent,
> 
> Thanks for the feedback.
> 
> > Subject: Re: [PATCH v5 01/11] i2c: Enhance i2c_new_ancillary_device API
> > 
> > On Tue, Jun 13, 2023 at 07:31:46PM +0000, Biju Das wrote:
> > > > Subject: RE: [PATCH v5 01/11] i2c: Enhance i2c_new_ancillary_device
> > > > API
> > > > > Subject: RE: [PATCH v5 01/11] i2c: Enhance
> > > > > i2c_new_ancillary_device API
> > > > > > Subject: Re: [PATCH v5 01/11] i2c: Enhance
> > > > > > i2c_new_ancillary_device API
> > > > > >
> > > > > > Hi everyone,
> > > > > >
> > > > > > > Perhaps we should first think through what an ancillary device
> > > > > > > really is.  My understanding is that it is used to talk to
> > > > > > > secondary addresses of a multi-address I2C slave device.
> > > > > >
> > > > > > As I mentioned somewhere before, this is not the case. Ancillary
> > > > > > devices are when one *driver* handles more than one address.
> > > > > > Everything else has been handled differently in the past (for
> > > > > > all the uses I am aware of).
> > > > > >
> > > > > > Yet, I have another idea which is so simple that I wonder if it
> > > > > > maybe has already been discussed so far?
> > > > > >
> > > > > > * have two regs in the bindings
> > > > >
> > > > > OK, it is inline with DT maintainers expectation as it is matching
> > > > > with real hw as single device node having two regs.
> > > > >
> > > > > > * use the second reg with i2c_new_client_device to instantiate the
> > > > > >   RTC sibling. 'struct i2c_board_info', which is one parameter, should
> > > > > >   have enough options to pass data, e.g it has a software_node.
> > > > >
> > > > > OK, I can see the below can be passed from PMIC to new client device.
> > > > >
> > > > > 	client->addr = info->addr;
> > > > >
> > > > > 	client->init_irq = info->irq;
> > > > >
> > > > > >
> > > > > > Should work or did I miss something here?
> > > > >
> > > > > I guess it will work. We instantiate appropriate device based On
> > > > > PMIC revision and slave address and IRQ resource passed through
> > > > > 'struct i2c_board_info'
> > > > >
> > > > > Will check this and update you.
> > > >
> > > > info.irq = irq; -->Irq fine
> > > > info.addr = addr; -->slave address fine size = strscpy(info.type,
> > > > name, sizeof(info.type)); -->instantiation based on PMIC version
> > > > fine.
> > > >
> > > > 1) How do we share clk details on instantiated device to find is it
> > > > connected to external crystal or external clock source? as we cannot
> > > > pass of_node between PMIC and "i2c_board_info" as it results in
> > > > pinctrl failure. info->platformdata and
> > > > Client->dev.platformdata to retrieve this info??
> > >
> > > Or
> > >
> > > I2C instantiation based on actual oscillator bit value, ie, two
> > > i2c_device_id's with one for setting oscillator bit and another for
> > > clearing oscillator bit
> > >
> > > PMIC driver parses the clock details. Based on firmware version and
> > > clock, It instantiates either i2c_device_id with setting oscillator
> > > bit or clearing oscillator bit.
> > 
> > I don't like that hack. I still think that two DT nodes is the best
> > option, I think you're trying hard to hack around a problem that is
> > actually not a problem.
> 
> Why do you think it is a hack? I believe rather it is actual solution
> 
> PMIC is a single device, with 2 regs, clocks, pinctrl and IRQ properties.
> So it will be represented as single node with single compatible.

The chip is a single package that contains two independent devices. This
is not different than bundling many IP cores in an SoC, we have one DT
node per IP core, not a single DT node for the SoC. The fact that we're
dealing with an external physical component here isn't relevant.

> By instating a client device, we are sharing the relevant resources to
> RTC device driver.

By instantiating a client device, you create a second struct device,
which is the kernel abstraction of a hardware device. This shows in my
opinion that we're dealing with two devices here, hence my
recommendation of using two DT nodes.

As you've noticed, with two devices and a single DT node, pinctrl
complains. You can hack around that by moving the pinctrl configuration
from the PMIC DT node to another DT node, and that's one first hack.
Then, you'll need to have two different device IDs depending on the PMIC
version to let the RTC driver set the oscillator bit correctly, and
that's a second hack.

A solution with two DT nodes models the hardware better and is cleaner.

-- 
Regards,

Laurent Pinchart



[Index of Archives]     [Linux Samsung SOC]     [Linux Wireless]     [Linux Kernel]     [ATH6KL]     [Linux Bluetooth]     [Linux Netdev]     [Kernel Newbies]     [IDE]     [Security]     [Git]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux ATA RAID]     [Samba]     [Device Mapper]

  Powered by Linux