Re: [PATCH 1/3] hwmon: (oxp-sensors): remove static board variable

[Guenter Roeck]

On 7/4/23 09:14, Greg Kroah-Hartman wrote:
> On Tue, Jul 04, 2023 at 07:14:54AM -0700, Guenter Roeck wrote:
> > On 7/4/23 06:44, Greg Kroah-Hartman wrote:
> > > On Tue, Jul 04, 2023 at 06:39:07AM -0700, Guenter Roeck wrote:
> > > > On 7/4/23 06:17, Greg Kroah-Hartman wrote:
> > > > > Drivers should not have a single static variable for the type of device
> > > > > they are bound to.  While this driver is really going to only have one
> > > > > device at a time in the system, remove the static variable and instead,
> > > > > look up the device type when needed.
> > > >
> > > > This is expensive. I think it would be much better to just move
> > > > the board type detection into the init code and not instantiate
> > > > the driver in the fist place if the board type is unknown.
> > >
> > > The board type detection is all over the place in the driver, it's not
> > > just for "unknown" types, so how about just saving the board type at
> > > probe time and using it then for all other places?
> >
> > I must be missing something. The current code detects the board type
> > only once, in the probe function. Otherwise the static variable is used.
> > You are replacing it with repeated calls to get_board_type().
> > The whole point of the static variable is to avoid the cost of repeated
> > calls to dmi_first_match().
>
> Ah, ok, yes, I was refering to the fact that the driver relies on the
> detection of the device type in lots of different places (and doesn't
> ever error out from the detection call.)

I am lost again. Current code:

        dmi_entry = dmi_first_match(dmi_table);
        if (!dmi_entry || boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
                return -ENODEV;



> > > > We can handle the static variable separately if it really bothers
> > > > you that much.
> > >
> > > I did this change to make patch 2/3 more "obvious" what is happening
> > > when the in_visible() callback happens, so that you don't have to worry
> > > about the saved value or not.  But this whole patch isn't really needed
> > > if you don't mind the lookup just happening in the in_visible() callback
> > > for the first time.
> >
> > That would at least be a minimal change, and just add one extra lookup
> > which is only called once (or zero, if it is used to save the board type).
>
> Ok, I'll switch it up, but really, it's just a simple table lookup loop,
> and none of the detection calls are on a "hot path" that I can
> determine.  Or am I missing something?
>
> > As I said, my solution would be to move the board type detection
> > into the init function and not instantiate the driver in the first
> > place if the probe function would bail out anyway.
>
> That's not the case today, the only way the probe function would fail
> today is if the registering of the sysfs files fail.  It does not matter
> if the board detection call passes or not.

Again,
        dmi_entry = dmi_first_match(dmi_table);
        if (!dmi_entry || boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
                return -ENODEV;
                ^^^^^^^^^^^^^^^

What am I missing ?

Guenter