Re: [PATCH v2 1/2] spi: Add Renesas R-Car Gen3 RPC SPI controller driver

[Marek Vasut <marek.vasut@xxxxxxxxx>]

On 12/05/2018 02:31 PM, Geert Uytterhoeven wrote:
> Hi Marek,

Hi,

> On Wed, Dec 5, 2018 at 2:25 PM Marek Vasut <marek.vasut@xxxxxxxxx> wrote:
>> On 12/05/2018 02:15 PM, Geert Uytterhoeven wrote:
>>> On Wed, Dec 5, 2018 at 1:57 PM Marek Vasut <marek.vasut@xxxxxxxxx> wrote:
>>>> On 12/05/2018 08:44 AM, masonccyang@xxxxxxxxxxx wrote:
>>>>>> "Marek Vasut" <marek.vasut@xxxxxxxxx>
>>>>>> 2018/12/05 上午 10:04
>>>>>> On 12/03/2018 10:18 AM, Mason Yang wrote:
>>>>>>> Add a driver for Renesas R-Car Gen3 RPC SPI controller.
>>>>>>>
>>>>>>> Signed-off-by: Mason Yang <masonccyang@xxxxxxxxxxx>
>>>
>>>>>>> +static void rpc_spi_hw_init(struct rpc_spi *rpc)
>>>>>>> +{
>>>>>>> +   /*
>>>>>>> +    * NOTE: The 0x260 are undocumented bits, but they must be set.
>>>>>>> +    *   RPC_PHYCNT_STRTIM is strobe timing adjustment bit,
>>>>>>> +    *   0x0 : the delay is biggest,
>>>>>>> +    *   0x1 : the delay is 2nd biggest,
>>>>>>> +    *   0x3 or 0x6 is a recommended value.
>>>>>>> +    */
>>>>>>
>>>>>> Doesn't this vary by SoC ? I think H3 ES1.0 had different value here,
>>>>>> but I might be wrong.
>>>>>
>>>>> I check the Renesas bare-metal code, mini_monitor v4.01.
>>>>> It set 0x03 or 0x0 and I test them w/ 0x0, 0x3 and 0x6 are all OK.
>>>>
>>>> Shouldn't this somehow use the soc_device_match() then and configure it
>>>> accordingly for various chips ? Like eg. the r8a7795-cpg-mssr driver does.
>>>
>>> Please don't use soc_device_match() for per-SoC configuration, if
>>> you already have of_device_id.data.
>>
>> I mean, the value is different on H3 ES1 and ES2 iirc, that's what
>> soc_device_match() is for, right ?
> 
> Oh, it differs between revisions, too?
> Yes, in that case you need soc_device_match().
> 
>>> BTW, this drivers support r8a7795 only (for now), as per the compatible
>>> value.
>>
>> 77995
> 
> Sorry, typo on my side. So H3 is not yet supported ;-)

It's coming the minute this lands mainline, so we should make it easy to
add.

>>>>>>> +#ifdef CONFIG_RESET_CONTROLLER
>>>>>>
>>>>>> Just make the driver depend on reset controller.
>>>>>
>>>>> ?
>>>>> please refer to
>>>>> https://github.com/torvalds/linux/blob/master/drivers/clk/renesas/renesas-cpg-mssr.c
>>>>>
>>>>> line 124 ~ 126
>>>>
>>>> This seems like a stopgap measure for systems which do not have a reset
>>>> api compatible controller. Geert ?
>>>
>>> So far CONFIG_RESET_CONTROLLER is optional.
>>
>> My understanding is that for this IP, it can well be mandatory, since
>> all the chips have a reset wired to the IP internally.
> 
> That's what I was trying to find out, hence my question about the purpose.
> 
>>>>>>> +         regmap_write(rpc->regmap, RPC_SMWDR0,
>>>>>>> +                 *(u32 *)(tx_buf + pos));
>>>>>>
>>>>>> *(u32 *) cast is probably not needed , fix casts globally.
>>>>>
>>>>> It must have it!
>>>>
>>>> Why ?
>>>
>>> Else you get a compiler warning, as tx_bug is void *.
>>
>> Don't you need some get_unaligned() in that case ? txbuf+pos can well be
>> unaligned if it's void * .
> 
> True, but IIRC, arm64 can handle that, right?
> Don't know about SuperH.

Oh, that's right, there are SH systems with RPC. Right.

>>>>>>> +#ifdef CONFIG_PM_SLEEP
>>>>>>> +static int rpc_spi_suspend(struct device *dev)
>>>>>>> +{
>>>>>>> +   struct platform_device *pdev = to_platform_device(dev);
>>>>>>> +   struct spi_master *master = platform_get_drvdata(pdev);
>>>>>>> +
>>>>>>> +   return spi_master_suspend(master);
>>>>>>
>>>>>> Won't the SPI NOR lose state across suspend ? Is that a problem ?
>>>>>
>>>>> I don't think so.
>>>>> Because when the device is not in operation and CS# is high,
>>>>> it is put in stand-by mode.
>>>>
>>>> Is the power to the SPI NOR retained ?
>>>
>>> Not if PSCI system suspend turns of power to the SoC.
>>
>> And is that a problem ?
> 
> Good question!

That's what we need an answer to :)

-- 
Best regards,
Marek Vasut