On Thu, 19 May 2016 08:28:02 +0200, Alexander Stein wrote:
On Wednesday 18 May 2016 21:33:01, Takashi Iwai wrote:
On Wed, 18 May 2016 19:20:19 +0200,
Mark Brown wrote:
On Wed, May 18, 2016 at 06:57:50PM +0200, Takashi Iwai wrote:
Mark Brown wrote:
It's just shifting the problem around... it sounds like for your use case suppressing the messages until we finish kernel boot would deal with most of the issue in a far more general fashion.
It comes to the question whether this message must be shown verbosely as an error at all. EPROBE_DEFER is usually a mechanism for the delayed probe, and it doesn't indicate an error per se. dev_err() is, OTOH, for real errors that have to be notified to user inevitably. That's why "quiet" boot option still shows it.
We can't tell the difference between something that's delayed and will come up later and something that's just never going to work - missing or misidentified components has always been a common source of errors in ASoC device bringup.
Yes, but it still means that most of cases are false-positive to show as "error" message.
I think the same way. Of course it is an error if probing never finishes due to some missing configuration/dt entry, but that needs to be checked in dmesg anyway. dev_notice still allows that.
There's also just the fact that the noise from deferred probe is not an ASoC specific thing, we need to tackle this at a system level rather than hacking individual cases.
Such other noises are usually now shown as errors with dev_err() but with dev_*() with a lower level.
I guess the patch subject and description are misleading. This doesn't suppress the messages to be printed. In usual boots without quiet boot option, there will be no visible change with this patch, the messages are still shown. It's not shown, however, when booted with quiet boot option. That's the whole point.
dev_notice() lowers the log level to somewhat between info and warning, and it looks like a sensible choice to me.
You want me to change the subject to be more clear?
If the patch can go in, yes. The changelog text should be also clearer why it's needed and what it gives, too.
Takashi