Re: [PATCH] iio: core: WARN in case sample bits do not fit storage bits

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Tue, Mar 22, 2022 at 08:43:10AM +0100, Nuno Sá wrote:
> On Mon, 2022-03-21 at 20:46 +0100, Marek Vasut wrote:
> > On 3/21/22 17:10, Andy Shevchenko wrote:
> > > On Mon, Mar 21, 2022 at 03:46:51PM +0100, Marek Vasut wrote:
> > > > On 3/21/22 11:40, Andy Shevchenko wrote:
> > > > > On Sun, Mar 20, 2022 at 07:15:42PM +0100, Marek Vasut wrote:
> > > > > > Add runtime check to verify whether storagebits are at least
> > > > > > as big
> > > > > > as shifted realbits. This should help spot broken drivers
> > > > > > which may
> > > > > > set realbits + shift above storagebits.
> > > > > 
> > > > > Thanks!
> > > > > 
> > > > > ...
> > > > > 
> > > > > > +                       /* Verify that sample bits fit into
> > > > > > storage */
> > > > > > +                       WARN_ON(channels[i].scan_type.storage
> > > > > > bits <
> > > > > > +                               channels[i].scan_type.realbit
> > > > > > s +
> > > > > > +                               channels[i].scan_type.shift);
> > > > > 
> > > > > Not sure WARN is a good level (it might be fatal on some setups
> > > > > and we won't that),
> > > > > besides the fact that we may use dev_WARN(). Perhaps dev_warn()
> > > > > would suffice?
> > > > 
> > > > I was actually thinking about BUG(), but that might crash
> > > > existing systems.
> > > > I think we want a strong indicator that something wrong is going
> > > > on which
> > > > must be fixed and the splat produced by WARN_ON() is a good
> > > > indicator of
> > > > that. It also does not crash existing systems,
> > > 
> > > It does crash _some_ of them, unfortunately.
> > 
> > Details please ?
> > 
> > WARN_ON() shouldn't cause crash outright, or do I miss something ?
> 
> Arghh, completely forgot about this... Andy is right, maybe there are
> other cases (in which case, it would be nice to share :D), but this one
> is definitely one of them:
> 
> https://elixir.bootlin.com/linux/latest/source/kernel/panic.c#L579
> 
> You can have a cmdline parameter to panic on _WARN() and some systems
> may have it.

Yes, I meant panic on warning.

And I can't imagine that this driver can be system critical to the extent
that we have to crash the system.

> That said, the "nice" stack_dump using WARN is way more explicit about
> saying that something is seriously wrong and must be fixed. dev_warn()
> is easier to ignore... But surely it is not nice to brick existing
> systems.  
> 
> Not really sure here...

-- 
With Best Regards,
Andy Shevchenko