On Fri, 28 Jun 2024 at 11:39, Krzysztof Kozlowski
<krzysztof.kozlowski@xxxxxxxxxx> wrote:
>
> On 28/06/2024 10:34, Dmitry Baryshkov wrote:
> > On Thu, Jun 27, 2024 at 05:23:44PM GMT, Krzysztof Kozlowski wrote:
> >> The driver has static 'struct regmap_config', which is then customized
> >> depending on device version. This works fine, because there should not
> >> be two devices in a system simultaneously and even less likely that such
> >> two devices would have different versions, thus different regmap config.
> >> However code is cleaner and more obvious when static data in the driver
> >> is also const - it serves as a template.
> >>
> >> Mark the 'struct regmap_config' as const and duplicate it in the probe()
> >> with devm_kmemdup to allow customizing per detected device variant.
> >>
> >> Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxx>
> >> ---
> >> sound/soc/codecs/lpass-rx-macro.c | 17 +++++++++++++----
> >> 1 file changed, 13 insertions(+), 4 deletions(-)
> >
> > Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxx>
> >
> >>
> >> diff --git a/sound/soc/codecs/lpass-rx-macro.c b/sound/soc/codecs/lpass-rx-macro.c
> >> index 59fe76b13cdb..3d8149665439 100644
> >> --- a/sound/soc/codecs/lpass-rx-macro.c
> >> +++ b/sound/soc/codecs/lpass-rx-macro.c
> >> @@ -1662,7 +1662,7 @@ static bool rx_is_readable_register(struct device *dev, unsigned int reg)
> >> return rx_is_rw_register(dev, reg);
> >> }
> >>
> >> -static struct regmap_config rx_regmap_config = {
> >> +static const struct regmap_config rx_regmap_config = {
> >> .name = "rx_macro",
> >> .reg_bits = 16,
> >> .val_bits = 32, /* 8 but with 32 bit read/write */
> >> @@ -3765,6 +3765,7 @@ static const struct snd_soc_component_driver rx_macro_component_drv = {
> >> static int rx_macro_probe(struct platform_device *pdev)
> >> {
> >> struct reg_default *reg_defaults;
> >> + struct regmap_config *reg_config;
> >> struct device *dev = &pdev->dev;
> >> kernel_ulong_t flags;
> >> struct rx_macro *rx;
> >> @@ -3851,14 +3852,22 @@ static int rx_macro_probe(struct platform_device *pdev)
> >> goto err;
> >> }
> >>
> >> - rx_regmap_config.reg_defaults = reg_defaults;
> >> - rx_regmap_config.num_reg_defaults = def_count;
> >> + reg_config = devm_kmemdup(dev, &rx_regmap_config, sizeof(*reg_config),
> >> + GFP_KERNEL);
> >> + if (!reg_config) {
> >> + ret = -ENOMEM;
> >> + goto err;
> >> + }
> >>
> >> - rx->regmap = devm_regmap_init_mmio(dev, base, &rx_regmap_config);
> >> + reg_config->reg_defaults = reg_defaults;
> >> + reg_config->num_reg_defaults = def_count;
> >> +
> >> + rx->regmap = devm_regmap_init_mmio(dev, base, reg_config);
> >> if (IS_ERR(rx->regmap)) {
> >> ret = PTR_ERR(rx->regmap);
> >> goto err;
> >> }
> >> + devm_kfree(dev, reg_config);
> >> devm_kfree(dev, reg_defaults);
> >
> > Seeing devm_kfree in the non-error path makes me feel strange. Maybe
> > it's one of the rare occasions when I can say that __free is suitable
> > here.
>
> These would have a bit different meaning in such case. The __free would
> not clean it in this spot, but on exit from the scope. I wanted to
> kfree() here, because the config (and reg_defaults) are not used by past
> regmap_init. I mentioned it briefly in previous patch msg.
>
> To me this code is readable and obvious - past this point nothing uses
> that allocation. However maybe instead of devm(), the code would be
> easier to read if non-devm-malloc + __free()?
Yes, that's what I was thinking about. But it's definitely an optional
topic. Your code is correct.
--
With best wishes
Dmitry
