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

[Neil Armstrong]

On 29/06/2018 01:45, Kevin Hilman wrote:
> Hi Miquel,
> 
> Miquel Raynal <miquel.raynal@xxxxxxxxxxx> writes:
> 
>> On Wed, 27 Jun 2018 16:33:43 -0700, Kevin Hilman <khilman@xxxxxxxxxxxx>
>> wrote:
>>
>>> Hi Boris,
>>>
>>> Boris Brezillon <boris.brezillon@xxxxxxxxxxx> writes:
>>>
>>>> Hi Yixun,
>>>>
>>>> On Wed, 13 Jun 2018 16:13:14 +0000
>>>> Yixun Lan <yixun.lan@xxxxxxxxxxx> wrote:
>>>>  
>>>>> From: Liang Yang <liang.yang@xxxxxxxxxxx>
>>>>>
>>>>> 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@xxxxxxxxxxx>
>>>>> Signed-off-by: Yixun Lan <yixun.lan@xxxxxxxxxxx>
>>>>> ---
>>>>>  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.

The idea was to override the mmc node by "syscon" when the board is in the
NAND case, since the eMMC controller won't be usable in this case anyway.

Neil

> 
> 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?
> 
> Kevin
>