Re: [PATCH v2] hwmon: (sbtsi) Don't read sensor more than once if it doesn't respond
From: Guenter Roeck
Date: Tue Apr 06 2021 - 17:35:15 EST
On 4/6/21 2:09 PM, Konstantin Aladyshev wrote:
> Thanks for the answer!
> Sorry for the confusion, by the "CPU is off" I meant "CPU is present,
> but currently it is in the powered off state".
> Therefore it is not possible to put these checks only in a probe
> function. And I don't know either if it is a good idea to cache
> config/min/max values.
>
> I use this driver on an OpenBMC system, which uses other software
> rather than lm-sensors utility. I guess that is why my priorities are
> shifted.
>
> By the way, I've noticed that the mutex check is absent in a
> SBTSI_REG_CONFIG read call both in the original driver version and in
> my patch, is this an error?
>
What do you mean with "mutex check" ?
Thanks,
Guenter
> Best regards,
> Konstantin Aladyshev
>
>
> On Tue, Apr 6, 2021 at 11:09 PM Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
>>
>> On 4/6/21 12:20 PM, Konstantin Aladyshev wrote:
>>> Thanks for the comment.
>>> Yes, you are correct, this patch adds an extra 'i2c_smbus_read_byte_data' call for the temp_max/temp_min reads.
>>> I guess I did that intentionally because I just wanted to keep the restructured code concise. After all I thought, 'temp_input' generally is read more often than 'temp_max/temp_min'.
>>> As I understand now, it seems like it is not acceptable. Therefore could you point me in the right direction about what I should do?
>>> Should I just stick with the original driver version and simply add two more i2c call checks for the first operations for min/max?
>>>
>>
>> Correct, it is not acceptable. A normal use case for hwmon devices is to use the "sensors"
>> command which _will_ read all attributes. i2c reads are expensive, and unnecessary read
>> operations should be avoided.
>>
>> There are several ways to solve the problem. Checking return values after each
>> read is the simple option. There are other possibilities, such as reading the limits
>> and the read order only once during probe, but I don't know enough about the
>> hardware to suggest a more sophisticated solution. For example, I don't know
>> what "CPU is off" means. Offline ? Not present ? If it means "not present",
>> or if the status is permanent, the condition should be handled in the is_visible
>> function (or the driver should not be instantiated in the first place).
>> Otherwise, the code should possibly return -ENODATA instead of -ETIMEDOUT
>> on error. But, again, I can not really suggest a better solution since
>> I don't know enough (nothing, actually) about the hardware (and the public
>> part of the SBTSI specification doesn't say anything about expected behavior
>> for offline CPUs or CPU cores).
>>
>> What I did find, though, is that the driver does not implement temperature
>> offset support, and it that doesn't support reporting alerts. I'd have assumed
>> this to be more important than optimizing error handling, but that is just
>> my personal opinion.
>>
>> Thanks,
>> Guenter
>>
>>> Best regards,
>>> Konstantin Aladyshev
>>>
>>>
>>> On Tue, Apr 6, 2021 at 9:42 PM Guenter Roeck <linux@xxxxxxxxxxxx <mailto:linux@xxxxxxxxxxxx>> wrote:
>>>
>>> On 4/6/21 11:16 AM, Konstantin Aladyshev wrote:
>>> > SBTSI sensors don't work when the CPU is off.
>>> > In this case every 'i2c_smbus_read_byte_data' function would fail
>>> > by a timeout.
>>> > Currently temp1_max/temp1_min file reads can cause two such timeouts
>>> > for every read.
>>> > Restructure code so there will be no more than one timeout for every
>>> > read operation.
>>> >
>>> > Signed-off-by: Konstantin Aladyshev <aladyshev22@xxxxxxxxx <mailto:aladyshev22@xxxxxxxxx>>
>>> > ---
>>> > Changes in v2:
>>> > - Fix typo in a commit message
>>> > - Don't swap temp_int/temp_dec checks at the end of the 'sbtsi_read' function
>>> >
>>>
>>> This doesn't explain the reason for the extra read operation for
>>> limits. Preventing a second read in error cases is not an argument
>>> for adding an extra read for non-error cases.
>>>
>>> Guenter
>>>
>>> > drivers/hwmon/sbtsi_temp.c | 55 +++++++++++++++++++-------------------
>>> > 1 file changed, 27 insertions(+), 28 deletions(-)
>>> >
>>> > diff --git a/drivers/hwmon/sbtsi_temp.c b/drivers/hwmon/sbtsi_temp.c
>>> > index e35357c48b8e..4ae48635bb31 100644
>>> > --- a/drivers/hwmon/sbtsi_temp.c
>>> > +++ b/drivers/hwmon/sbtsi_temp.c
>>> > @@ -74,48 +74,47 @@ static int sbtsi_read(struct device *dev, enum hwmon_sensor_types type,
>>> > u32 attr, int channel, long *val)
>>> > {
>>> > struct sbtsi_data *data = dev_get_drvdata(dev);
>>> > + u8 temp_int_reg, temp_dec_reg;
>>> > s32 temp_int, temp_dec;
>>> > int err;
>>> >
>>> > switch (attr) {
>>> > case hwmon_temp_input:
>>> > - /*
>>> > - * ReadOrder bit specifies the reading order of integer and
>>> > - * decimal part of CPU temp for atomic reads. If bit == 0,
>>> > - * reading integer part triggers latching of the decimal part,
>>> > - * so integer part should be read first. If bit == 1, read
>>> > - * order should be reversed.
>>> > - */
>>> > - err = i2c_smbus_read_byte_data(data->client, SBTSI_REG_CONFIG);
>>> > - if (err < 0)
>>> > - return err;
>>> > -
>>> > - mutex_lock(&data->lock);
>>> > - if (err & BIT(SBTSI_CONFIG_READ_ORDER_SHIFT)) {
>>> > - temp_dec = i2c_smbus_read_byte_data(data->client, SBTSI_REG_TEMP_DEC);
>>> > - temp_int = i2c_smbus_read_byte_data(data->client, SBTSI_REG_TEMP_INT);
>>> > - } else {
>>> > - temp_int = i2c_smbus_read_byte_data(data->client, SBTSI_REG_TEMP_INT);
>>> > - temp_dec = i2c_smbus_read_byte_data(data->client, SBTSI_REG_TEMP_DEC);
>>> > - }
>>> > - mutex_unlock(&data->lock);
>>> > + temp_int_reg = SBTSI_REG_TEMP_INT;
>>> > + temp_dec_reg = SBTSI_REG_TEMP_DEC;
>>> > break;
>>> > case hwmon_temp_max:
>>> > - mutex_lock(&data->lock);
>>> > - temp_int = i2c_smbus_read_byte_data(data->client, SBTSI_REG_TEMP_HIGH_INT);
>>> > - temp_dec = i2c_smbus_read_byte_data(data->client, SBTSI_REG_TEMP_HIGH_DEC);
>>> > - mutex_unlock(&data->lock);
>>> > + temp_int_reg = SBTSI_REG_TEMP_HIGH_INT;
>>> > + temp_dec_reg = SBTSI_REG_TEMP_HIGH_DEC;
>>> > break;
>>> > case hwmon_temp_min:
>>> > - mutex_lock(&data->lock);
>>> > - temp_int = i2c_smbus_read_byte_data(data->client, SBTSI_REG_TEMP_LOW_INT);
>>> > - temp_dec = i2c_smbus_read_byte_data(data->client, SBTSI_REG_TEMP_LOW_DEC);
>>> > - mutex_unlock(&data->lock);
>>> > + temp_int_reg = SBTSI_REG_TEMP_LOW_INT;
>>> > + temp_dec_reg = SBTSI_REG_TEMP_LOW_DEC;
>>> > break;
>>> > default:
>>> > return -EINVAL;
>>> > }
>>> >
>>> > + /*
>>> > + * ReadOrder bit specifies the reading order of integer and
>>> > + * decimal part of CPU temp for atomic reads. If bit == 0,
>>> > + * reading integer part triggers latching of the decimal part,
>>> > + * so integer part should be read first. If bit == 1, read
>>> > + * order should be reversed.
>>> > + */
>>> > + err = i2c_smbus_read_byte_data(data->client, SBTSI_REG_CONFIG);
>>> > + if (err < 0)
>>> > + return err;
>>> > +
>>> > + mutex_lock(&data->lock);
>>> > + if (err & BIT(SBTSI_CONFIG_READ_ORDER_SHIFT)) {
>>> > + temp_dec = i2c_smbus_read_byte_data(data->client, temp_dec_reg);
>>> > + temp_int = i2c_smbus_read_byte_data(data->client, temp_int_reg);
>>> > + } else {
>>> > + temp_int = i2c_smbus_read_byte_data(data->client, temp_int_reg);
>>> > + temp_dec = i2c_smbus_read_byte_data(data->client, temp_dec_reg);
>>> > + }
>>> > + mutex_unlock(&data->lock);
>>> >
>>> > if (temp_int < 0)
>>> > return temp_int;
>>> >
>>>
>>