Re: [PATCH v2 4/5] iio: adc: ti-adc161s626: log buffer setup failure in probe

[Jonathan Cameron]

On Thu, 25 Jun 2026 17:03:11 +0300
Andy Shevchenko <andriy.shevchenko@xxxxxxxxx> wrote:

> On Thu, Jun 25, 2026 at 12:09:52PM +0100, Jonathan Cameron wrote:
> > On Wed, 24 Jun 2026 14:26:52 +0300
> > Andy Shevchenko <andriy.shevchenko@xxxxxxxxx> wrote:  
> > > On Wed, Jun 24, 2026 at 03:43:44PM +0530, Prashant Rahul wrote:  
> > > > On Wed Jun 24, 2026 at 1:06 AM IST, Jonathan Cameron wrote:    
> 
> ...
> 
> > > > > Why this particular one but not the devm_iio_device_register() below it?
> > > > > Both pretty much only fail on driver bugs.  Or out of memory for
> > > > > which dev_err_probe() doesn't print anything anyway (on basis that
> > > > > is normally pretty noisy).    
> > > > 
> > > > My apologies, I wasnt aware that dev_err_probe simply ignores `-ENOMEM`.
> > > > Looking at drivers/base/core.c:5096, it is pretty evident that it does.
> > > > 
> > > > My goal with the patch series was to change dev_err into dev_err_probe,
> > > > as my understanding of dev_err_probe was that it's a specialization of
> > > > dev_err for probe contexts.    
> > > 
> > > Jonathan, actually the triggered buffer setup might return not -ENOMEM error
> > > code for the surprise. After all this series does add not a dead code, but
> > > a quite rare to be used one.
> > > 
> > > https://elixir.bootlin.com/linux/v7.1.1/source/drivers/iio/buffer/industrialio-triggered-buffer.c#L57  
> > 
> > Agreed and I'm not even sure why that is there (lost to the mists of time
> > / my failing memory ;)  However I think it's an obvious driver bug case
> > that should be trivial to see in basic tests so to me it's marginal on
> > whether a print on these is useful.
> > 
> > Maybe given that one is so rare we should just add a print in the core?
> > I'm not particularly keen on mass patches to print errors from the core
> > code but this one is perhaps a reasonable trade off vs series like this
> > one adding prints to drivers for something we don't think realistically
> > ever happens.  
> 
> While the series ends up not adding a dead code, the rareness of the case
> doesn't justify it, so I am still on the side of _not_ applying this series.
> 
> For the core printing, I think for devm_*() or other managed resource APIs
> it makes a lot of sense (in general). Not sure about plain calls, though.
> 
For now I've picked up the two straight conversions to dev_err_probe()
because I'm keen to (slowly) stop using dev_err() in probe functions
but lets leave the rest for now at least.  If I get some time (you never
know) I might throw together an RFC that prints appropriate messages from
these IIO registration calls so we can have a debate about whether they
belong in the core or the drivers (or not at all).

Thanks,

Jonathan