Date:   Wed, 29 May 2019 06:17:06 -0700
From:   Guenter Roeck <linux@...ck-us.net>
To:     "Adamski, Krzysztof (Nokia - PL/Wroclaw)" 
        <krzysztof.adamski@...ia.com>
Cc:     Jean Delvare <jdelvare@...e.com>,
        "linux-hwmon@...r.kernel.org" <linux-hwmon@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "Sverdlin, Alexander (Nokia - DE/Ulm)" <alexander.sverdlin@...ia.com>
Subject: Re: [PATCH] adm1275: support PMBUS_VIRT_*_SAMPLES

On 5/29/19 5:53 AM, Adamski, Krzysztof (Nokia - PL/Wroclaw) wrote:
> On Wed, May 29, 2019 at 05:17:47AM -0700, Guenter Roeck wrote:
>> On 5/29/19 12:11 AM, Adamski, Krzysztof (Nokia - PL/Wroclaw) wrote:
>>> On Tue, May 28, 2019 at 12:46:52PM -0700, Guenter Roeck wrote:
>>>> On Fri, May 24, 2019 at 12:49:13PM +0000, Adamski, Krzysztof (Nokia - PL/Wroclaw) wrote:
>>>>> The device supports setting the number of samples for averaging the
>>>>> measurements. There are two separate settings - PWR_AVG for averaging
>>>>> PIN and VI_AVG for averaging VIN/VAUX/IOUT, both being part of
>>>>> PMON_CONFIG register. The values are stored as exponent of base 2 of the
>>>>> actual number of samples that will be taken.
>>>>>
>>>>> Signed-off-by: Krzysztof Adamski <krzysztof.adamski@...ia.com>
>>>>> ---
>>>>>   drivers/hwmon/pmbus/adm1275.c | 68 ++++++++++++++++++++++++++++++++++-
>>>>>   1 file changed, 67 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/hwmon/pmbus/adm1275.c b/drivers/hwmon/pmbus/adm1275.c
>>>>> index f569372c9204..4efe1a9df563 100644
>>>>> --- a/drivers/hwmon/pmbus/adm1275.c
>>>>> +++ b/drivers/hwmon/pmbus/adm1275.c
>>>>> @@ -23,6 +23,8 @@
>>>>>   #include <linux/slab.h>
>>>>>   #include <linux/i2c.h>
>>>>>   #include <linux/bitops.h>
>>>>> +#include <linux/bitfield.h>
>>>>> +#include <linux/log2.h>
>>>>>   #include "pmbus.h"
>>>>>
>>>>>   enum chips { adm1075, adm1272, adm1275, adm1276, adm1278, adm1293, adm1294 };
>>>>> @@ -78,6 +80,10 @@ enum chips { adm1075, adm1272, adm1275, adm1276, adm1278, adm1293, adm1294 };
>>>>>   #define ADM1075_VAUX_OV_WARN		BIT(7)
>>>>>   #define ADM1075_VAUX_UV_WARN		BIT(6)
>>>>>
>>>>> +#define ADM1275_PWR_AVG_MASK		GENMASK(13, 11)
>>>>> +#define ADM1275_VI_AVG_MASK		GENMASK(10, 8)
>>>>> +#define ADM1275_SAMPLES_AVG_MAX	128
>>>>> +
>>>>>   struct adm1275_data {
>>>>>   	int id;
>>>>>   	bool have_oc_fault;
>>>>> @@ -90,6 +96,7 @@ struct adm1275_data {
>>>>>   	bool have_pin_max;
>>>>>   	bool have_temp_max;
>>>>>   	struct pmbus_driver_info info;
>>>>> +	struct mutex lock;
>>>>>   };
>>>>>
>>>>>   #define to_adm1275_data(x)  container_of(x, struct adm1275_data, info)
>>>>> @@ -164,6 +171,38 @@ static const struct coefficients adm1293_coefficients[] = {
>>>>>   	[18] = { 7658, 0, -3 },		/* power, 21V, irange200 */
>>>>>   };
>>>>>
>>>>> +static inline int adm1275_read_pmon_config(struct i2c_client *client, u64 mask)
>>>>
>>>> Why is the mask passed through as u64 ?
>>>
>>> Good point. I used u64 as this is the type used by bitfield machinery
>>> under the hood but I agree it doesn't make sense and is even confusing
>>> to have this in the function prototype as we are using this to mask 16
>>> bit word anyways. I will fix that in v2. I am gonna have to cast the ret
>>> to u16 when passing to FIELD_GET() to make sure the __BF_FIELD_CHECK is
>>> not complaining (since it is signed right now), though.
>>>
>>
>> Not sure I understand what you are talking about. FIELD_GET() uses typeof().
>> FIELD_GET() is used by other callers even with u8 and without any typecasts.
>> Why would it be a problem here ?
> 
> So I basically agree with you but just wanted to note why there will be
> additional cast needed in my code. The:
>     return FIELD_GET(mask, ret);
> will be changed to:
>     return FIELD_GET(mask, (u16)ret);
> 
> And the reason for that is that the __BF_FIELD_CHECK does this check at
> compile time (and breaks if this is true)
>     (_mask) > (typeof(_reg))~0ull
> 
> In my case typeof(_reg) is int, so (typeof(_reg))~0ull = -1 which is
> signed. _mask is unsigned. Depending on the type promotion, this might
> or might not be true depending on the size of _mask. When _mask was u64,
> it always worked. For _mask being u16, it will fail. For u32, this will
> fail depending on if we are compiling for 32 or 64 bit architecture.
> 
> All this might be obvious to you but it wasn't to me, thus this note.
> 

The problem here is that ret is an int, and integers are not well suited
for mask operations. The typecast thus makes sense and is ok.

>>>>
>>>>> +{
>>>>> +	int ret;
>>>>> +
>>>>> +	ret = i2c_smbus_read_word_data(client, ADM1275_PMON_CONFIG);
>>>>> +	if (ret < 0)
>>>>> +		return ret;
>>>>> +
>>>>> +	return FIELD_GET(mask, ret);
>>>>> +}
>>>>> +
>>>>> +static inline int adm1275_write_pmon_config(struct i2c_client *client, u64 mask,
>>>>> +					    u16 word)
>>>>> +{
>>>>> +	const struct pmbus_driver_info *info = pmbus_get_driver_info(client);
>>>>> +	struct adm1275_data *data = to_adm1275_data(info);
>>>>> +	int ret;
>>>>> +
>>>>> +	mutex_lock(&data->lock);
>>>>
>>>> Why is another lock on top of the lock provided by the pmbus core required ?
>>>>
>>>
>>> Good point, I was considering if I should instead add mutex_lock on
>>> update_lock in the pmbus_set_samples() function inside of pmbus_core.c
>>> instead (as this function is missing it) but figured that not all
>>> devices will need that (lm25066 didn't) so it might be a waste in most
>>> cases. But this may be cleaner approach indeed.
>>>
>>> Is this what you mean or there is some other lock I missed?
>>>
>> pmbus_set_samples() should set the pmbus lock. That was missed when
>> the function was added.
> 
> And by pmbus lock you mean the update_lock from the pmbus_data
> structure? I didn't see any other lock but wanted to double check my
> understanding.
> 

Yes, that is the lock intended to protect write operations.

Guenter

>>>>> +	ret = i2c_smbus_read_word_data(client, ADM1275_PMON_CONFIG);
>>>>> +	if (ret < 0) {
>>>>> +		mutex_unlock(&data->lock);
>>>>> +		return ret;
>>>>> +	}
>>>>> +
>>>>> +	word = FIELD_PREP(mask, word) | (ret & ~mask);
>>>>> +	ret = i2c_smbus_write_word_data(client, ADM1275_PMON_CONFIG, word);
>>>>> +	mutex_unlock(&data->lock);
>>>>> +
>>>>> +	return ret;
>>>>> +}
>>>>> +
>>>>>   static int adm1275_read_word_data(struct i2c_client *client, int page, int reg)
>>>>>   {
>>>>>   	const struct pmbus_driver_info *info = pmbus_get_driver_info(client);
>>>>> @@ -242,6 +281,19 @@ static int adm1275_read_word_data(struct i2c_client *client, int page, int reg)
>>>>>   		if (!data->have_temp_max)
>>>>>   			return -ENXIO;
>>>>>   		break;
>>>>> +	case PMBUS_VIRT_POWER_SAMPLES:
>>>>> +		ret = adm1275_read_pmon_config(client, ADM1275_PWR_AVG_MASK);
>>>>> +		if (ret < 0)
>>>>> +			break;
>>>>> +		ret = 1 << ret;
>>>>
>>>> 		ret = BIT(ret);
>>>>
>>>
>>> I intentionally used the "raw" left shift to make it more obvious this
>>> is pow2 arithmetic operation and an direct inverse to the ilog2() used
>>> on write counterpart. This is also consistent with what I used in
>>> lm25066.c driver not long time ago.
>>>
>>> I don't have strong preference but this is my reasoning. So do you still
>>> think it is better to use BIT() macro instead?
>>>
>>
>> I don't think that is a good rationale, but I'll let it go.
>>
> 
> If you don't think so, I'll change it in v2. As I said, I don't have a
> strong opinion on that.
> 
> Krzysztof
> 

Powered by blists - more mailing lists