On 22.12.2018 00:22, Marek Vasut wrote:
> On 12/15/2018 06:38 PM, Heiner Kallweit wrote:
>> On 15.12.2018 18:01, Andrew Lunn wrote:
>>>> +static struct tja11xx_phy_stats tja11xx_hw_stats[] = {
>>>> + { "phy_symbol_error_count", 20, 0, 0xffff },
>>>> + { "phy_overtemp_error", 21, 1, BIT(1) },
>>>> + { "phy_undervolt_error", 21, 3, BIT(3) },
>>>> + { "phy_polarity_detect", 25, 6, BIT(6) },
>>>> + { "phy_open_detect", 25, 7, BIT(7) },
>>>> + { "phy_short_detect", 25, 8, BIT(8) },
>>>
>>> Hi Marek
>>>
>>> You have a number of one bit counters here, which is pretty unusual.
>>> The names also don't really suggest they are counters.
>>>
>> Apart from few counters the values seem to be flags. I just think
>> it could be done in a little bit more readable form, e.g. instead of
>> { "phy_short_detect", 25, 8, BIT(8) } use
>> { "phy_short_detect", 25, BIT(8) } and in tja11xx_get_stats() then
>> use FIELD_GET (see linux/bitfield.h).
>
> This doesn't work with the counters, it only works with flags. The array
> contains both.
>
Maybe we have a misunderstanding here, but FIELD_GET would work also
with the counters. FIELD_GET calculates the field offset, so you don't
need to specify it. Let's say you would have an entry in your array like
this: { "phy_xxx_error_count", 20, 4, 0xfff0 }
Then you would do:
#define XXX_ERROR_CNT_MASK GENMASK(15, 4)
{ "phy_xxx_error_count", 20, XXX_ERROR_CNT_MASK }
and access the value by
val = FIELD_GET(XXX_ERROR_CNT_MASK, reg)
>> The idea of HWMON attributes sounds good to me because it allows
>> to use the flags to trigger actions in a structured way. And I
>> assume in case of e.g. "PHY undervolt" some monitoring system
>> would like to be informed (especially because we talk about
>> automotive here).
>
> I added HWMON_T_CRIT_ALARM and HWMON_I_LCRIT_ALARM for
> phy_overtemp_error and phy_undervolt_error respectively.
> Are there any other hwmon attributes I can use to replace the flags in
> the tja11xx_hw_stats array ?
>
I think that's it.
