Re: [PATCH v5 1/1] hwmon: Add driver for Astera Labs PT5161L retimer

[Guenter Roeck]

On 2/5/24 23:33, Krzysztof Kozlowski wrote:
> On 05/02/2024 17:15, Guenter Roeck wrote:
> > On Mon, Feb 05, 2024 at 04:26:08PM +0100, Krzysztof Kozlowski wrote:
> > > On 05/02/2024 16:20, Cosmo Chou wrote:
> > > > This driver implements support for temperature monitoring of Astera Labs
> > > > PT5161L series PCIe retimer chips.
> > > >
> > > > This driver implementation originates from the CSDK available at
> > > > Link: https://github.com/facebook/openbmc/tree/helium/common/recipes-lib/retimer-v2.14
> > > > The communication protocol utilized is based on the I2C/SMBus standard.
> > > >
> > > > Signed-off-by: Cosmo Chou <chou.cosmo@xxxxxxxxx>
> > > > ---
> > [ ... ]
> >
> > > > +
> > > > +static int __init pt5161l_init(void)
> > > > +{
> > > > +	pt5161l_debugfs_dir = debugfs_create_dir("pt5161l", NULL);
> > >
> > > Drivers don't need initcalls. For sure any debugfs should not be handled
> > > here but in probe.
> >
> > Lots of hwmon drivers have init functions, for basic chip detection of
> > Super-I/O chips (example: drivers/hwmon/nct6775-platform.c) and to create
> > a parent debugfs subdirectory for the driver. The probe function then adds
> > subdirecties per chip instantiation. Example for pmbus, in
> > drivers/hwmon/pmbus/pmbus_core.c:
>
> Core bus components are a bit different...
>
> > static int __init pmbus_core_init(void)
> > {
> >          pmbus_debugfs_dir = debugfs_create_dir("pmbus", NULL);
> >          if (IS_ERR(pmbus_debugfs_dir))
> >                  pmbus_debugfs_dir = NULL;
> >
> >          return 0;
> > }
> >
> > static void __exit pmbus_core_exit(void)
> > {
> >          debugfs_remove_recursive(pmbus_debugfs_dir);
> > }
> >
> > Are you saying this is all wrong ? What alternative would you suggest ?
>
> Just create parent directory in probe and only keep remove in __exit.
> But you are right that might not be much better approach.

That would just add more code and increase complexity since it would require
a mutex to protect against parallel probes. On the plus side, it would only
create the driver root directory if necessary, that is, if there is an
actual device. I am not sure if that would be worth the increased complexity.

That makes me wonder, though, if an API function such as
	struct dentry *hwmon_debugfs_get_root(char *name);
would make sense. That would move the complexity to the hwmon core and
unify its handling. I'll think about that. For the time being, I'll
accept the code as-is.

Thanks,
Guenter