[RFC] w1: regression, remove need for ida and use PLATFORM_DEVID_AUTO

[Petr Cvek]

Dne 3.2.2017 v 15:07 Guenter Roeck napsal(a):
> On 02/03/2017 05:28 AM, Petr Cvek wrote:
>> Dne 3.2.2017 v 04:31 Guenter Roeck napsal(a):
>>> On 02/02/2017 05:31 PM, Petr Cvek wrote:
>>>> Dne 2.2.2017 v 15:23 Andrew F. Davis napsal(a):
>>>>> On 02/01/2017 07:05 PM, Petr Cvek wrote:
>>>>>> Patch
>>>>>>     w1: remove need for ida and use PLATFORM_DEVID_AUTO
>>>>>>     098f9fb0c962eb2fdba5f9d34f4cf7a938237184
>>>>>>
>>>>>> causes a regression in the w1_ds2760 driver. Initialization creates a name "ds2760-battery.0.auto". It seems that name of that size will cause checking code in __thermal_cooling_device_register() from thermal_core.c to fail as it checks for names shorter than 20 chars:
>>>>>>
>>>>>>     if (type && strlen(type) >= THERMAL_NAME_LENGTH)
>>>>>>         return ERR_PTR(-EINVAL);
>>>>>>
>>>>>> A second problem seems to be with magician_supplicants list as "ds2760-battery.0" does not match in a test power_supply_am_i_supplied() in ds2760_battery_update_status() from drivers/power/supply/ds2760_battery.c . This causes status flag to be forever in POWER_SUPPLY_STATUS_DISCHARGING state.
>>>>>>
>>>>>> Functionality returned after I hotfix changed (drivers/w1/slaves/w1_ds2760.c)
>>>>>>
>>>>>>     pdev = platform_device_alloc("ds2760-battery", PLATFORM_DEVID_AUTO);
>>>>>>
>>>>>> to
>>>>>>
>>>>>>     pdev = platform_device_alloc("ds2760-battery", 0);
>>>>>>
>>>>>> Which solution do you advise. Shortening the name let's say just to "ds2760" (with changes to arch files, w1 files and supply files), increasing THERMAL_NAME_LENGTH to let's say 32 (I don't know if this doesn't break something other) or just reverting to the old behavior?
>>>>>>
>>>>>
>>>>> If 20 is no longer the max thermal name then THERMAL_NAME_LENGTH should
>>>>> be changed.
>>>>
>>>> Yes it is that ".auto" addition to "ds2760-battery.0" makes 21 chars long string. With higher number (".10" onwards) it is 22 chars. Load of the module then fails at test in thermal_zone_device_register():
>>>>
>>>>     https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/drivers/thermal/thermal_core.c?id=refs/tags/v4.10-rc6#n1158
>>>>
>>>>     if (type && strlen(type) >= THERMAL_NAME_LENGTH)
>>>>         return ERR_PTR(-EINVAL);
>>>>
>>>> I searched in code further and with THERMAL_NAME_LENGTH = 30 (enough space) the module load fails at another place in __hwmon_device_register():
>>>>
>>>>     https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/drivers/hwmon/hwmon.c?id=refs/tags/v4.10-rc6#n548
>>>>
>>>>     if (name && (!strlen(name) || strpbrk(name, "-* \t\n")))
>>>>         return ERR_PTR(-EINVAL);
>>>>
>>>> That's because there is "-" in "ds2760-battery.0.auto" (and that's why shortening to "ds2760.0" works). Is there some reason for disallowing "-" in a name? If not then I'm for removal (in other case we need to change ds2760-battery and others names). Increasing THERMAL_NAME_LENGTH to 22 should be just OK for ds2760 (but maybe not enough for others) I would increased it to 32 (nice number ;-) ).
>>>>
>>>
>>> '-' is not permitted in hwmon name attributes, primarily because libsensors
>>> interprets it as a delimiter in its configuration file.
>>>
>>
>> OK so this means we need to rename ds2760-battery so libsensors works.
>>
>> Hmm is an underscore in "ds2760_battery.0.auto" OK?
> 
> Theoretically yes, but there was objection to that idea because replacing the '-'
> with '_' is considered by some to be an ABI change (why the addition of '.auto'
> doesn't matter escapes me, though).
> 

I read that discussion and both ways are not optimal. Many drivers has "-" in its
name and even "*" is a valid filename. And changing to underscore is ugly too. 
Too bad libsensors parses "-" that way.

>>
>>> Anyway, the related change in thermal has been reverted, so that should
>>
>> s/thermal/hwmon ? Or I'm confused :-D (there was no change in thermal, code stopped working because of change in w1 driver (".auto" addition))
>>
> 
> 3feb479cea37 Revert "thermal: thermal_hwmon: Convert to hwmon_device_register_with_info()"

But after applying revert on my kernel copy there still will be a problem
in arch init with missing ".auto" match and driver load still fails on a shorter string array.

Petr