[PATCH 2/2] mtd: rawnand: meson: add support for Amlogic NAND flash controller

[yixun.lan@xxxxxxxxxxx (Yixun Lan)]

HI Kevin

On 06/29/18 07:45, Kevin Hilman wrote:
> Hi Miquel,
> 
> Miquel Raynal <miquel.raynal at bootlin.com> writes:
> 
>> On Wed, 27 Jun 2018 16:33:43 -0700, Kevin Hilman <khilman at baylibre.com>
>> wrote:
>>
>>> Hi Boris,
>>>
>>> Boris Brezillon <boris.brezillon at bootlin.com> writes:
>>>
>>>> Hi Yixun,
>>>>
>>>> On Wed, 13 Jun 2018 16:13:14 +0000
>>>> Yixun Lan <yixun.lan at amlogic.com> wrote:
>>>>  
>>>>> From: Liang Yang <liang.yang at amlogic.com>
>>>>>
>>>>> Add initial support for the Amlogic NAND flash controller which found
>>>>> in the Meson-GXBB/GXL/AXG SoCs.
>>>>>
>>>>> Singed-off-by: Liang Yang <liang.yang at amlogic.com>
>>>>> Signed-off-by: Yixun Lan <yixun.lan at amlogic.com>
>>>>> ---
>>>>>  drivers/mtd/nand/raw/Kconfig      |    8 +
>>>>>  drivers/mtd/nand/raw/Makefile     |    3 +
>>>>>  drivers/mtd/nand/raw/meson_nand.c | 1422 +++++++++++++++++++++++++++++
>>>>>  3 files changed, 1433 insertions(+)
>>>>>  create mode 100644 drivers/mtd/nand/raw/meson_nand.c  
>>>>
>>>> Can you run checkpatch.pl --strict and fix the coding style issues?
>>>>  
>>>>>
>>>>> diff --git a/drivers/mtd/nand/raw/Kconfig b/drivers/mtd/nand/raw/Kconfig
>>>>> index 19a2b283fbbe..b3c17a3ca8f4 100644
>>>>> --- a/drivers/mtd/nand/raw/Kconfig
>>>>> +++ b/drivers/mtd/nand/raw/Kconfig
>>>>> @@ -534,4 +534,12 @@ config MTD_NAND_MTK
>>>>>  	  Enables support for NAND controller on MTK SoCs.
>>>>>  	  This controller is found on mt27xx, mt81xx, mt65xx SoCs.
>>>>>  
>>>>> +config MTD_NAND_MESON
>>>>> +	tristate "Support for NAND flash controller on Amlogic's Meson SoCs"
>>>>> +	depends on ARCH_MESON || COMPILE_TEST
>>>>> +	select COMMON_CLK_REGMAP_MESON
>>>>> +	select MFD_SYSCON
>>>>> +	help
>>>>> +	  Enables support for NAND controller on Amlogic's Meson SoCs.
>>>>> +
>>>>>  endif # MTD_NAND
>>>>> diff --git a/drivers/mtd/nand/raw/Makefile b/drivers/mtd/nand/raw/Makefile
>>>>> index 165b7ef9e9a1..cdf6162f38c3 100644
>>>>> --- a/drivers/mtd/nand/raw/Makefile
>>>>> +++ b/drivers/mtd/nand/raw/Makefile
>>>>> @@ -1,5 +1,7 @@
>>>>>  # SPDX-License-Identifier: GPL-2.0
>>>>>  
>>>>> +ccflags-$(CONFIG_MTD_NAND_MESON) += -I$(srctree)/drivers/clk/meson  
>>>>
>>>> Please don't do that. If you need to expose common regs, put them
>>>> in include/linux/soc/meson/. I'm also not sure why you need to access
>>>> the clk regs directly. Why can't you expose the MMC/NAND clk as a clk
>>>> provider whose driver would be placed in drivers/clk and which would use
>>>> the mmc syscon. This way the same clk driver could be used for both
>>>> MMC and NAND clk indifferently, and the NAND driver would be much
>>>> simpler.  
>>>
>>> [...]
>>>
>>>>> +
>>>>> +	return 0;
>>>>> +}
>>>>> +
>>>>> +static const char * sd_emmc_ext_clk0_parent_names[MUX_CLK_NUM_PARENTS];
>>>>> +
>>>>> +static struct clk_regmap sd_emmc_c_ext_clk0_sel = {
>>>>> +	.data = &(struct clk_regmap_mux_data){
>>>>> +		.offset = SD_EMMC_CLOCK,
>>>>> +		.mask = 0x3,
>>>>> +		.shift = 6,
>>>>> +	},
>>>>> +	.hw.init = &(struct clk_init_data) {
>>>>> +		.name = "sd_emmc_c_nand_clk_mux",
>>>>> +		.ops = &clk_regmap_mux_ops,
>>>>> +		.parent_names = sd_emmc_ext_clk0_parent_names,
>>>>> +		.num_parents = ARRAY_SIZE(sd_emmc_ext_clk0_parent_names),
>>>>> +		.flags = CLK_SET_RATE_PARENT,
>>>>> +	},
>>>>> +};
>>>>> +
>>>>> +static struct clk_regmap sd_emmc_c_ext_clk0_div = {
>>>>> +	.data = &(struct clk_regmap_div_data){
>>>>> +		.offset = SD_EMMC_CLOCK,
>>>>> +		.shift = 0,
>>>>> +		.width = 6,
>>>>> +		.flags = CLK_DIVIDER_ROUND_CLOSEST | CLK_DIVIDER_ONE_BASED,
>>>>> +	},
>>>>> +	.hw.init = &(struct clk_init_data) {
>>>>> +		.name = "sd_emmc_c_nand_clk_div",
>>>>> +		.ops = &clk_regmap_divider_ops,
>>>>> +		.parent_names = (const char *[]){ "sd_emmc_c_nand_clk_mux" },
>>>>> +		.num_parents = 1,
>>>>> +		.flags = CLK_SET_RATE_PARENT,
>>>>> +	},
>>>>> +};
>>>>> +
>>>>> +static int meson_nfc_clk_init(struct meson_nfc *nfc)
>>>>> +{
>>>>> +	struct clk_regmap *mux = &sd_emmc_c_ext_clk0_sel;
>>>>> +	struct clk_regmap *div = &sd_emmc_c_ext_clk0_div;
>>>>> +	struct clk *clk;
>>>>> +	int i, ret;
>>>>> +
>>>>> +	/* request core clock */
>>>>> +	nfc->core_clk = devm_clk_get(nfc->dev, "core");
>>>>> +	if (IS_ERR(nfc->core_clk)) {
>>>>> +		dev_err(nfc->dev, "failed to get core clk\n");
>>>>> +		return PTR_ERR(nfc->core_clk);
>>>>> +	}
>>>>> +
>>>>> +	/* init SD_EMMC_CLOCK to sane defaults w/min clock rate */
>>>>> +	regmap_update_bits(nfc->reg_clk, 0,
>>>>> +			CLK_SELECT_NAND | CLK_ALWAYS_ON | CLK_DIV_MASK,
>>>>> +			CLK_SELECT_NAND | CLK_ALWAYS_ON | CLK_DIV_MASK);
>>>>> +
>>>>> +	/* get the mux parents */
>>>>> +	for (i = 0; i < MUX_CLK_NUM_PARENTS; i++) {
>>>>> +		char name[16];
>>>>> +
>>>>> +		snprintf(name, sizeof(name), "clkin%d", i);
>>>>> +		clk = devm_clk_get(nfc->dev, name);
>>>>> +		if (IS_ERR(clk)) {
>>>>> +			if (clk != ERR_PTR(-EPROBE_DEFER))
>>>>> +				dev_err(nfc->dev, "Missing clock %s\n", name);
>>>>> +			return PTR_ERR(clk);
>>>>> +		}
>>>>> +
>>>>> +		sd_emmc_ext_clk0_parent_names[i] = __clk_get_name(clk);
>>>>> +	}
>>>>> +
>>>>> +	mux->map = nfc->reg_clk;
>>>>> +	clk = devm_clk_register(nfc->dev, &mux->hw);
>>>>> +	if (WARN_ON(IS_ERR(clk)))
>>>>> +		return PTR_ERR(clk);
>>>>> +
>>>>> +	div->map = nfc->reg_clk;
>>>>> +	nfc->device_clk = devm_clk_register(nfc->dev, &div->hw);
>>>>> +	if (WARN_ON(IS_ERR(nfc->device_clk)))
>>>>> +		return PTR_ERR(nfc->device_clk);
>>>>> +
>>>>> +	ret = clk_prepare_enable(nfc->core_clk);
>>>>> +	if (ret) {
>>>>> +		dev_err(nfc->dev, "failed to enable core clk\n");
>>>>> +		return ret;
>>>>> +	}
>>>>> +
>>>>> +	ret = clk_prepare_enable(nfc->device_clk);
>>>>> +	if (ret) {
>>>>> +		dev_err(nfc->dev, "failed to enable device clk\n");
>>>>> +		clk_disable_unprepare(nfc->core_clk);
>>>>> +		return ret;
>>>>> +	}
>>>>> +
>>>>> +	return 0;
>>>>> +}  
>>>>
>>>>
>>>> As said above, I don't like having a clk driver here, especially since
>>>> the registers you're accessing are not part of the NAND controller
>>>> registers. Please try to create a driver in drivers/clk/ for that.  
>>>
>>> We went back and forth on this one on some off-list reviews.
>>>
>>> Had we known that the NAND controller was (re)using the clock registers
>>> internal to the MMC IP block from the beginning, we would have written a
>>> clock provider in drivers/clk for this, and shared it.
>>>
>>> However, when I wrote the MMC driver[1] (already upstream) along with
>>> the bindings[2], we did not fathom that the internal mux and divider
>>> would be "borrowed" by another device. :(
>>>
>>> We only recently found out that the NAND controller "borrows" one of the
>>> MMC clocks, whose registers are inside the MMC range.  Taking the clock
>>> out of the MMC driver and into its own clock-provider implies redoing
>>> the MMC driver, as well as its bindings, which we wanted to avoid
>>> (especially the binding changes.)
>>>
>>> We (I can take the blame) decided that since the MMC and NAND are
>>> mutually exclusive (they also share pins), that allowing NAND to reuse
>>> the MMC range would be a good compromise.  The DT still accurately
>>> describes the hardware, but we don't have to throw a large wrench into
>>> the DT bindings just for a newly discovered shared clock.
>>>
>>> I agree, it's not the prettiest thing, but when we cannot know the full
>>> details of the hardware when we start, sometimes we end up in a bit of a
>>> mess that requires some compromise.
>>
>> I totally understand your situation but as MMC and NAND are mutually
>> exclusive, how is this a problem to have a dedicated clock driver used
>> only by the NAND controller (as maybe a first step)? I mean, if you
>> don't change the MMC bindings, then the MMC driver will still use its
>> own 'local' clock driver, right?
> 
> Yeah, I think you're right.  That would work too.
> 
>> I don't know if you can have two nodes reserving the same address
>> range though.
> 
> You can, but it's a race who gets to claim the region.
> 
> You'd have to have the new clock-controler disabled by default, and have
> any boards that use the NAND disable the MMC and enable the clock
> controller node.
> 
> But I think that should work.
> 
> Yixun, can you give this approach a try?
Yes, It actually works here, I'll send out the clock patches later for
review.

Thanks everyone for the suggestion
> 
> Kevin
> 
> .
>