Re: [PATCH 3/3] i2c:support new register set for ast2600

[Andy Shevchenko]

On Fri, Jun 18, 2021 at 11:53:41AM +0800, Jamin Lin wrote:
> The 06/17/2021 10:33, Andy Shevchenko wrote:
> > On Thu, Jun 17, 2021 at 05:43:40PM +0800, Jamin Lin wrote:

...

> > > + *
> > > + *  This program is free software; you can redistribute it and/or modify
> > > + *  it under the terms of the GNU General Public License version 2 as
> > > + *  published by the Free Software Foundation.
> > 
> > SPDX covers this.
> >
> Will change as following what do you think?
> // SPDX-License-Identifier: GPL-2.0-or-later
> /*
>  *  Aspeed I2C Interrupt Controller.
>  * Copyright (C) ASPEED Technology Inc.
>  * Ryan Chen <ryan_chen@xxxxxxxxxxxxxx>
>  */

Yes, something like this.

...

> > > +static const struct aspeed_i2c_base_clk i2c_base_clk[BASE_CLK_COUNT] = {
> > > +	/* name	target_freq */
> > > +	{  "base_clk0",	1000000 },	/* 1M */
> > > +	{  "base_clk1",	4000000 },	/* 4M */
> > > +	{  "base_clk2",	10000000 },	/* 10M */
> > > +	{  "base_clk3",	40000000 },	/* 40M */
> > > +};
> > 
> > Why it's not provided as the clock provider(s)?

> According to the design of ASPEED AST2600 SOC, the I2C bus is connected to PHB bus.
> The clock driver provides PHB clock and its default frequency is 100MHZ.
> AST2600 support the bus frequency of I2C from 0.1kbps to 5Mbps if PHB clock is 50MHZ.
> To meet the different bus frequency of I2C, we use this programmer to calculate the divider to 
> change the base clock.
> For example, 
> It calculates divider to change base_clock 1 to 1M to support I2C bus frequency 100KHZ
> It calculates divider to change base_clock 2 to 4M to support I2C bus frequency 400KHZ 

My question is, why don't you provide a clock provider (under drivers/clk) for
this platform and use it instead?

...

> > > +	struct clk_hw_onecell_data *onecell;
> > > +	struct clk_hw *hw;
> > > +	int err;
> > > +	u32 clk_divider = 0;
> > > +	int i, j;
> > > +	unsigned long base_freq;
> > 
> > Use reversed xmas tree order for all these blocks.
> > 
> > The rule of thumb, btw, that any comment applies to all similar places by
> > default (independently on which line it was given against).
> >
> Do you mena change as following?
> struct clk_hw_onecell_data *onecell;
> unsigned long base_freq;
> u32 clk_divider = 0;
> struct clk_hw *hw;
> int err;
> int i;
> int j;

Yes.

...

> > > +		for (i = 0; i < 0xff; i++) {
> > > +			/*
> > > +			 * i maps to div:
> > > +			 * 0x00: div 1
> > > +			 * 0x01: div 1.5
> > > +			 * 0x02: div 2
> > > +			 * 0x03: div 2.5
> > > +			 * 0x04: div 3
> > > +			 * ...
> > > +			 * 0xFE: div 128
> > > +			 * 0xFF: div 128.5
> > > +			 */
> > > +			base_freq = base_clk * 2 / (2 + i);
> > > +			if (base_freq <= i2c_base_clk[j].base_freq)
> > > +				break;
> > > +		}
> > 
> > oAre yuo sure you can't eliminate the entire for-loop? Think about it a bit,
> > please.
> >
> What do you think if we use "lookup table" instaed of above for loop?

Besides that it should be a part of clock provider, no, you may use a formula
(bit operations and so on).

...

> > > +	i2c_ic->rst = devm_reset_control_get_exclusive(&pdev->dev, NULL);

> > > +	if (IS_ERR(i2c_ic->rst)) {
> > 
> > > +		dev_dbg(&pdev->dev,
> > > +			"missing or invalid reset controller device tree entry");
> > 
> > Make it optional.
> Can I change to use "dev_err"?

What I meant here is to make the reset optional (there is even specific API for
that) and return an error in that case.

> > > +	} else {
> > > +		/* SCU I2C Reset */
> > > +		reset_control_assert(i2c_ic->rst);
> > > +		udelay(3);
> > > +		reset_control_deassert(i2c_ic->rst);
> > > +	}

...

> > > +static struct ast_i2c_timing_table aspeed_old_i2c_timing_table[] = {
> > 
> > What the ... is this for?!
> AST2600 support old/new register set for I2C controller.
> This lookup table is used to find the divisor for the specific I2C bus
> frequency for AST2600 I2C controller with old register set.
> For example
> If I2C bus frequency is 100KHZ and PHB clock is 100MHZ, it will find this table
> because 100000000/1024 < 100000
> "{ 1024, 0x00000300 | (0x5) | (0xf << 20) | (0xf << 16) | (0xf << 12) }"

Can't you derive it by formula(s)?

> > > +	/* Divisor : Base Clock : tCKHighMin : tCK High : tCK Low  */
> > > +	/* Divisor :	  [3:0] : [23: 20]   :   [19:16]:   [15:12] */

> > > +};

-- 
With Best Regards,
Andy Shevchenko