Hi Neil,
On Mon, Jun 26, 2023 at 03:20:35PM +0200, Neil Armstrong wrote:
> Hi,
>
> On 26/06/2023 15:01, Jeff LaBundy wrote:
> > Hi Neil,
> >
> > On Mon, Jun 26, 2023 at 09:02:16AM +0200, Neil Armstrong wrote:
> >
> > [...]
> >
> > > > > +static int goodix_berlin_spi_probe(struct spi_device *spi)
> > > > > +{
> > > > > + struct regmap_config *regmap_config;
> > > > > + struct regmap *regmap;
> > > > > + size_t max_size;
> > > > > + int error = 0;
> > > > > +
> > > > > + regmap_config = devm_kmemdup(&spi->dev, &goodix_berlin_spi_regmap_conf,
> > > > > + sizeof(*regmap_config), GFP_KERNEL);
> > > > > + if (!regmap_config)
> > > > > + return -ENOMEM;
> > > >
> > > > Is there any reason we cannot simply pass goodix_berlin_spi_regmap_conf to
> > > > devm_regmap_init() below? Why to duplicate and pass the copy?
> > > >
> > > > For reference, BMP280 in IIO is a similar example of a device with regmap
> > > > sitting atop a bespoke SPI protocol; it does not seem to take this extra
> > > > step.
> > >
> > > The goodix_berlin_spi_regmap_conf copy is modified after with the correct
> > > max raw read/write size, and I'm not a fan of modifying a global structure
> > > that could be use for multiple probes, I can make a copy in a stack variable
> > > if it feels simpler.
> >
> > Ah, that makes sense; in that case, the existing implementation seems fine
> > to me. No changes necessary.
> >
> > Correct me if I'm wrong, but the stack variable method wouldn't work since
> > that memory is gone after goodix_berlin_spi_probe() returns.
>
> The config is only needed for the devm_regmap_init() duration, so keeping
> the memory allocated for the whole lifetime of the device seems useless.
I revisted the regmap code, and you are indeed correct. I agree with your
suggestion.
>
> Neil
>
> >
> > Kind regards,
> > Jeff LaBundy
>
Kind regards,
Jeff LaBundy
