Re: [PATCH] cpufreq: mediatek: Use dev_err_probe in every error path in probe

[AngeloGioacchino Del Regno]

Il 02/07/24 10:43, Viresh Kumar ha scritto:
> On 02-07-24, 10:26, AngeloGioacchino Del Regno wrote:
> > Il 02/07/24 07:57, Viresh Kumar ha scritto:
> > > On 28-06-24, 15:48, Nícolas F. R. A. Prado wrote:
> > > > diff --git a/drivers/cpufreq/mediatek-cpufreq.c b/drivers/cpufreq/mediatek-cpufreq.c
> > > > @@ -629,11 +630,9 @@ static int mtk_cpufreq_probe(struct platform_device *pdev)
> > > >    	int cpu, ret;
> > > >    	data = dev_get_platdata(&pdev->dev);
> > > > -	if (!data) {
> > > > -		dev_err(&pdev->dev,
> > > > -			"failed to get mtk cpufreq platform data\n");
> > > > -		return -ENODEV;
> > > > -	}
> > > > +	if (!data)
> > > > +		return dev_err_probe(&pdev->dev, -ENODEV,
> > >
> > > What's the point of calling dev_err_probe() when we know for sure that
> > > the error isn't EPROBE_DEFER ?
> >
> > Logging consistency, that's all; the alternative would be to rewrite the dev_err()
> > messages to be consistent with what dev_err_probe() says, so that'd be
> > dev_err("error %pe: (message)");
>
> That would be better I guess. There is no point adding inefficient
> code.
>
> > > > +				     "failed to get mtk cpufreq platform data\n");
> > > >    	for_each_possible_cpu(cpu) {
> > > >    		info = mtk_cpu_dvfs_info_lookup(cpu);
> > > > @@ -643,24 +642,21 @@ static int mtk_cpufreq_probe(struct platform_device *pdev)
> > > >    		info = devm_kzalloc(&pdev->dev, sizeof(*info), GFP_KERNEL);
> > > >    		if (!info) {
> > > >    			ret = -ENOMEM;
> > > > +			dev_err_probe(&pdev->dev, ret, "Failed to allocate dvfs_info\n");
> >
> > By the way, forgot to point that out in my former review: to make it shorter,
> > instead of "ret = -ENOMEM; dev_err_probe()" you can write it as...
> >
> > ret = dev_err_probe(&pdev->dev, -ENOMEM, ".... message");
>
> `ret` will be  be used I guess with the `goto` statement to return
> error and so the change was like this.

Yes it will, but `ret = dev_err_probe(dev, -ENOMEM, "...")` is still kind of a
shortcut, as that will effectively assign -ENOMEM to ret, so that the error is
still returned like before ;-)

Cheers