Re: [PATCH V10 13/15] rust: cpufreq: Extend abstractions for driver registration

[Danilo Krummrich]

On Wed, Apr 16, 2025 at 03:47:26PM +0530, Viresh Kumar wrote:
> On 16-04-25, 11:04, Danilo Krummrich wrote:
> > You need to justify why drv is a valid pointer to be passed to
> > cpufreq_register_driver(), i.e. something like
> > 
> > 	// SAFETY:
> > 	// - `drv` comes from Self::VTABLE and hence is a valid pointer to a `struct cpufreq_driver`,
> > 	// - `cpufreq_register_driver()` never attempts to modify the data `drv` points to
> 
> The cpufreq core can try to change the data pointed by `drv`. For now
> it updates the `cpufreq_driver->boost_enabled` flag.

VTABLE is const and hence ends up in the read-only section of the binary.

I assumed that struct cpufreq_driver (like most driver structures) is const.
Actually, I think it should be.

Anyways, that doesn't help for now. Unfortunately, I think you actually need to
dynamically allocate it. There's no need to revert everything though. You can
just allocate a new KBox from VTABLE, i.e.

	let vtable = KBox::new(Self::VTABLE, GFP_KERNEL)?;

This makes it easy for you to remove the dynamic allocation once (or if) cpufreq
is ever reworked to allow a static const struct cpufreq_driver.