RE: [PATCH 3/3] scsi: ufs-sysfs: Add sysfs entries for temperature notification

From: Avri Altman
Date: Thu Sep 02 2021 - 02:52:25 EST


>
> On 9/1/21 5:37 AM, Avri Altman wrote:
> > +What:
> /sys/bus/platform/drivers/ufshcd/*/attributes/case_rough_temp
> > +Date: September 2021
> > +Contact: Avri Altman <avri.altman@xxxxxxx>
> > +Description: The device case rough temperature
> (bDeviceCaseRoughTemperature
> > + attribute). It is termed "rough" due to the inherent inaccuracy
> > + of the temperature sensor inside a semiconductor device,
> > + e.g. +- 10 degrees centigrade error range.
>
> My understanding is that the word Celsius is more common than centigrade
> so please use Celsius instead of centigrade. See also
> https://www.brannan.co.uk/celsius-centigrade-and-fahrenheit/
Done.

>
> > + allowable range is [-79..170].
> > + The temperature readings are in decimal degrees Celsius.
> > +
> > + Please note that the Tcase validity depends on the state of the
> > + wExceptionEventControl attribute: it is up to the user to
> > + verify that the applicable mask (TOO_HIGH_TEMP_EN, and / or
> > + TOO_LOW_TEMP_EN) is set for the exception handling control.
> > + This can be either done by ufs-bsg or ufs-debugfs.
>
> Instead of making the user verify whether case_rough_temp is valid,
> please modify the kernel code such that case_rough_temp only reports a
> value if that value is valid. One possible approach is to make the show
> method return an error code if case_rough_temp is not valid.
But it does.
Just wanted to document that exception control is controlled from user space,
And avoid the eyebrow raises when getting invalid temperature reading.

>Another and
> probably better approach is to define a sysfs attribute group and to
> make case_rough_temp visible only if it is valid.
>
> > diff --git a/drivers/scsi/ufs/ufs-sysfs.c b/drivers/scsi/ufs/ufs-sysfs.c
> > index 5c405ff7b6ea..a9abe33c40e4 100644
> > --- a/drivers/scsi/ufs/ufs-sysfs.c
> > +++ b/drivers/scsi/ufs/ufs-sysfs.c
> > @@ -1047,6 +1047,86 @@ static inline bool ufshcd_is_wb_attrs(enum
> attr_idn idn)
> > idn <= QUERY_ATTR_IDN_CURR_WB_BUFF_SIZE;
> > }
> >
> > +static inline bool ufshcd_is_temp_attrs(enum attr_idn idn)
> > +{
> > + return idn >= QUERY_ATTR_IDN_CASE_ROUGH_TEMP &&
> > + idn <= QUERY_ATTR_IDN_LOW_TEMP_BOUND;
> > +}
>
> Modern compilers are good at deciding when to inline a function so
> please leave out the 'inline' keyword from the above function.
Done.

>
> > +static bool ufshcd_case_temp_legal(struct ufs_hba *hba)\
>
> Please use another word than "legal" since the primary meaning of
> "legal" is "of or relating to law".
Done.

>
> > + ufshcd_rpm_get_sync(hba);
> > + ret = ufshcd_query_attr(hba, UPIU_QUERY_OPCODE_READ_ATTR,
> > + QUERY_ATTR_IDN_EE_CONTROL, 0, 0, &ee_mask);
> > + ufshcd_rpm_put_sync(hba);
>
> Are there any ufshcd_query_attr() calls that are not surrounded by
> ufshcd_rpm_{get,put}_sync()? If not, please move the
> ufshcd_rpm_{get,put}_sync() calls into ufshcd_query_attr().
Will check.

Thanks,
Avri
>
> Thanks,
>
> Bart.