Re: [PATCH] [2.6.22] pasemi: hardware rng driver

[olof@lixom.net (Olof Johansson)]

On Thu, Apr 26, 2007 at 01:38:31AM +0200, Arnd Bergmann wrote:
> On Wednesday 25 April 2007, Olof Johansson wrote:
> 
> > +static void __iomem *rng_regs;
> > +
> > +static int pasemi_rng_data_present(struct hwrng *rng)
> > +{
> > +	return (in_le32(rng_regs + SDCRNG_CTL_REG)
> > +		& SDCRNG_CTL_FVLD_M) ? 1 : 0;
> > +}
> 
> It would be nicer to get rid of the global rng_regs variable by sticking
> it into rng->priv.

Good point.

> > +static int __devinit rng_probe(struct of_device *ofdev,
> > +				     const struct of_device_id *match)
> > +{
> > +	struct device_node *rng_np;
> > +	struct resource res;
> > +	int err = 0;
> > +
> > +	rng_np = of_find_compatible_node(NULL, "rng", "1682m-rng");
> > +	if (!rng_np)
> > +		return -ENODEV;
> 
> I would guess that the call to of_find_compatible_node is entirely bogus
> here, because the device is already passed in as ofdev in to the probe
> function.

Yup. Leftover from pre-of_platform

> > +int rng_init(void)
> > +{
> > +       return of_register_platform_driver(&rng_driver);
> > +}
> > +
> > +void rng_exit(void)
> > +{
> > +       of_unregister_platform_driver(&rng_driver);
> > +}
> > +
> > +device_initcall(rng_init);
> 
> rng_init and rng_exit should be static
> rng_init should be __init
> rng_exit should be __exit
> Since the driver is tristate in Kconfig, it would be more conventional to
> use module_init() instead of device_initcall().
> rng_exit needs to be marked as module_exit() to allow unloading the driver.

Yup, will fix.


Thanks.


-Olof