Re: [PATCH] drm/vc4: hdmi: Fix defined but not used warning

On Thu, Sep 23, 2021 at 09:54:06AM -0700, Linus Torvalds wrote:

On Thu, Sep 23, 2021 at 8:57 AM Maxime Ripard [off-list ref] wrote:

quoted

Since we select CONFIG_PM anyway, vc4_hdmi_runtime_suspend() and
vc4_hdmi_runtime_suspend() will always be used and we can thus always
assign them in struct dev_pm_ops without using the SET_RUNTIME_PM_OPS
macro.

This cannot be true.

If CONFIG_PM is always enabled, then the patch is a no-op, and the
warning you quote cannot happen:

   warning: 'vc4_hdmi_runtime_suspend' defined but not used [-Wunused-function]

So this patch is very obviously broken, the message is misleading, and
the claims in your commit message cannot _possibly_ be true.

I guess it could have been worded a bit better.

DRM_VC4 allows compilation through COMPILE_TEST and selects PM. Some
platforms don't define PM at all.

In the latter case, SET_RUNTIME_PM_OPS will be a nop, the functions
won't be used, and we'll get this warning.

Maxime, this kind of "respond to bug reports with random contents"
most not continue.

I'm not super familiar with how to deal with those kind of situations,
but it does address the warning on those platforms without affecting the
current operations of the driver. I don't see how it qualifies as
random.

You need to actually look at what the reporter is reporting, and think
about the code. Because the above fix is broken, broken, broken.

Like I said, this was a genuine attempt at fixing things. It's clear now
that you don't feel the same way and would prefer some other solution.
That's why we have review in the first place I guess? I fail to see what
that kind of personal comments brings to the discussion though.

The way people fix this is by either making the function definitions
be conditional on their uses - so that the compiler removes them
entirely - or mark them as __maybe_unused. Then a smart _linker_ can
actually remove the code if people use the smarter linker options.

The initial point of selecting CONFIG_PM was to get rid of the #ifdef,
and for all practical purposes the code will always be used when the
driver will run so __maybe_unused didn't look like a proper solution
either.

Maxime