Re: [PATCH 1/2] drivers/regulator/core.c: Don't print error on EPROBE_DEFER

[Mark Brown <broonie@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 2648 bytes --]

On Tue, Dec 16, 2014 at 11:27:03AM +0100, Mike Looijmans wrote:
> On 12/09/2014 07:48 PM, Mark Brown wrote:
> >On Tue, Dec 09, 2014 at 07:12:30PM +0100, Mike Looijmans wrote:

> >>Can we instead at least reduce it to WARN or INFO level then?

> >You appear to have deleted my reply here...  one problem with your
> >suggestion is that it means we have to special case all error handling
> >on probe for deferral which isn't wonderful.

> special casing deferral may not be "wonderful", but it is what currently
> happens in many places, and it's just "the best we can do for now". A
> similar patch for MMC got approval:
> https://lkml.org/lkml/2014/10/27/477

That's just one call site, were all the other places that might see a
probe deferral updated too?

That also looks like a case of the core passing through other errors
rather than a case of primary error reporting - I guess there's a
reasonable case for the core not logging at all here since the drivers
ought to be doing it.

> >>I have to explain over and over again that there's no problem when that
> >>message comes along ten times in a row. And it causes people to overlook the
> >>messages that really are errors.

> >Can we do something with the log message that triggers on probe
> >deferral?  There tends to be a learning curve with probe deferral but
> >the fact that it's generally extremely noisy tends to be useful - I
> >usually point people at that (not just in the context at regulators) and
> >tell them not to worry unless debugging.

> Using "dev_err" is not really "tell them not to worry unless debugging", I
> think that is what "dev_dbg" was meant to do.

Well, then the core probe deferral stuff ought to be less chatty then...

> The only real solution I could come up with here is to replace "return
> -EPROBE_DEFER" with something that stores the current stack, registers the
> resource it requires and jumps back to where the driver probe originated.
> Once the resource is available, the stored stack is resumed and then the
> probe code path can continue as if nothing bad happened. This would also
> deliver excellent diagnostic data in case the resource remains absent. I've
> built something like this in Python which has a "yield" statement one can
> use for this purpose. It's a bit tougher to do in C I guess. So until then,
> we're stuck with sprinkling "if (ret == -EPROBE_DEFER)" code snippets all
> over the place.

There was a proposal the other day for a restrack framework (name might
change in future) which drivers would call in their probe and get a
callback when everything appears.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 473 bytes --]