Re: [PATCH v3 06/10] mfd: da9063: Add custom regmap for DA9063L

[Marek Vasut <marek.vasut@gmail.com>]

On 06/06/2018 11:47 AM, Steve Twiss wrote:
> Hi Marek and Geert,
> 
> On 06 June 2018 00:02 Marek Vasut wrote,
> 
>> Subject: Re: [PATCH v3 06/10] mfd: da9063: Add custom regmap for DA9063L
>>
>> On 06/05/2018 10:17 PM, Steve Twiss wrote:
>>> Hi Marek and Geert,
>>>
>>> On 04 June 2018 17:25 Marek Vasut wrote,
>>>
>>>> Subject: Re: [PATCH v3 06/10] mfd: da9063: Add custom regmap for DA9063L
>>>>
>>>> On 06/04/2018 09:39 AM, Geert Uytterhoeven wrote:
>>>>> Hi Marek, Steve,
>>>>>
>>>>> On Sat, Jun 2, 2018 at 12:11 PM, Marek Vasut <marek.vasut@gmail.com> wrote:
>>>>>> While the datasheet for DA9063L (2v1, 23-Mar-2017) lists the RTC register
>>>>>> block, the DA9063L does not have an RTC. Add custom regmap for DA9063L to
>>>>>> prevent access into that register block.
>>>
>>> Ok. I've said previously in [v3 07/10], but I'll copy again:
>>> There is now an internal Dialog request to remove the RTC references from the DA9063L datasheet.
>>> Adding that first part to the sentence in the commit log: "While the datasheet for DA9063L
>>> (2v1, 23-Mar-2017) lists the RTC register block" -- it exists in error for the register map table
>>> on page 91, but the datasheet also identifies those registers in Table 102 on page 126 as
>>> "Reserved".
>>>
>>> Pointing out the ambiguity in this version of the datasheet seems redundant in the commit log.
>>> Also Dialog do not store a history of Datasheets on their website so once this is updated (although
>>> this update is not in my hands) the datasheet will be replaced. So, it seems this comment could
>>> make the commit message just as misleading as the current datasheet.
>>>
>>> How about something simpler?
>>> "The DA9063L does not have an RTC. Add custom regmap for DA9063L to prevent access
>>> into that register block."
>>>
>>>>>>
>>>>>> Signed-off-by: Marek Vasut <marek.vasut+renesas@gmail.com>
>>>>>
>>>>> Thanks for your patch!
>>>>>
>>>>>> --- a/drivers/mfd/da9063-i2c.c
>>>>>> +++ b/drivers/mfd/da9063-i2c.c
>>>>>> @@ -254,6 +341,10 @@ static int da9063_i2c_probe(struct i2c_client *i2c,
>>>>>
>>>>> Note that the line above doesn't check da9063->type, but da9063-
>>>>> variant_code...
>>>>>
>>>>>>                 da9063_regmap_config.rd_table = &da9063_ad_readable_table;
>>>>>>                 da9063_regmap_config.wr_table = &da9063_ad_writeable_table;
>>>>>>                 da9063_regmap_config.volatile_table = &da9063_ad_volatile_table;
>>>>>> +       } else if (da9063->type == PMIC_TYPE_DA9063L) {
>>>>>
>>>>> ... so this may be slightly confusing.
>>>>
>>>> I know.
>>>>
>>>>>> +               da9063_regmap_config.rd_table = &da9063l_bb_readable_table;
>>>>>> +               da9063_regmap_config.wr_table = &da9063l_bb_writeable_table;
>>>>>> +               da9063_regmap_config.volatile_table = &da9063l_bb_volatile_table;
>>>>>>         } else {
>>>>>>                 da9063_regmap_config.rd_table = &da9063_bb_readable_table;
>>>>>>                 da9063_regmap_config.wr_table = &da9063_bb_writeable_table;
>>>>>
>>>>> However, da9063->variant_code doesn't seem to have been filled in at this
>>>>> point yet (the call to da9063_device_init() doing so is below, at the end
>>>>> of the probe function!), so commit 9cb42e2a8ed06e91 ("mfd: da9063: Add
>>>>> support for AD silicon variant") never actually handled the AD silicon variant
>>>>> correctly? Or am I missing something?
>>>
>>> Okay ... No. You're not missing anything. I had noticed that.
>>> The AD chip model is not referenced and by default only the BB chip model is used.
>>>
>>>> Ha, that is a good point.
>>>
>>> Yeah, it's a good point, but it's not an amusing point.
>>> The device tree only distinguishes a "dlg,da9063", there is no AD type in the DT schema.
>>> There is no datasheet listing AD registers supported by Dialog, only BB.
>>>
>>> But, AD registers were added back into the header file in commit 9cb42e2a8ed06e91
>>> and the RTC driver was updated to distinguish between the AD and BB according to
>>> the type of variant detected at run-time during the da9063_device_init() call.
>>>
>>> The real problem is that this leads to two competing chip detection methods for the
>>> DA9063. The function da9063_device_init() autodetects the chip variant, but
>>> autodetection cannot define the chip model. It's circular: the chip model cannot be
>>> autodetected because a chip model is needed to access the register used during
>>> autodetection.
>>>
>>> Which leads me back to what I said two paragraphs up:
>>>> The device tree only distinguishes a "dlg,da9063", there is no AD type in the DT schema.
>>>> There is no datasheet listing AD registers supported by Dialog, only BB.
>>>
>>> This is not how it is done in the DA9062 and DA9061 driver: the variant code is only
>>> used to print the information to the console during start-up and it is the DT that defines
>>> the chip model based upon "dlg,da9062" or "dlg,da9061".
>>
>> So the AD was broken since forever and noone noticed ? :)
> 
> Not quite. 
> The AD support is working, but the driver doesn't work like everybody
> expects because it uses the BB chip model. But it does work because the chip
> model for BB is valid for AD; in this case BB represents a superset of AD
> registers (and any mismatches are never accessed or mean anything in AD).
> 
>> Do you have an AD hardware and can you fix it ?
> 
> Part of my work is to support the community and I think this is fixable.
> 
> But all of this shouldn't affect your DA9063L submission should it?

I think there might be conflict between those patchsets, so let me send
out a V5 so you can play around with the AD and fix that too.

-- 
Best regards,
Marek Vasut