CLOCK_MONOTONIC, BOOTTIME, suspend and screensaver regression

CLOCK_MONOTONIC, BOOTTIME, suspend and screensaver regression

[Pavel Machek]

[-- Attachment #1: Type: text/plain, Size: 1129 bytes --]

Hi!

In 4.17-rc1 I have weird problems with network manager... and I also
have screensaver kicking just after resume.

Searching the git logs:

       - The most interesting change is the consolidation of clock
         MONOTONIC
	          and clock BOOTTIME.

         Clock MONOTONIC behaves now exactly like clock BOOTTIME and
         does
	          not longer ignore the time spent in suspend. A new
         clock
	          MONOTONIC_ACTIVE is provived which behaves like
         clock MONOTONIC in
	          kernels before this change. This allows applications
         to
	          programmatically check for the clock MONOTONIC
         behaviour.

         As discussed in the review thread, this has the potential of
	          breaking user space and we might have to revert
	          this. Knock on wood
		           that we can avoid that exercise.


...I believe you are breaking userspace for me. How can we check that?

									Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 181 bytes --]

Re: CLOCK_MONOTONIC, BOOTTIME, suspend and screensaver regression

[Pavel Machek]

[-- Attachment #1: Type: text/plain, Size: 1389 bytes --]

On Tue 2018-04-24 14:06:59, Pavel Machek wrote:
> Hi!
> 
> In 4.17-rc1 I have weird problems with network manager... and I also
> have screensaver kicking just after resume.
> 
> Searching the git logs:
> 
>        - The most interesting change is the consolidation of clock
>          MONOTONIC
> 	          and clock BOOTTIME.
> 
>          Clock MONOTONIC behaves now exactly like clock BOOTTIME and
>          does
> 	          not longer ignore the time spent in suspend. A new
>          clock
> 	          MONOTONIC_ACTIVE is provived which behaves like
>          clock MONOTONIC in
> 	          kernels before this change. This allows applications
>          to
> 	          programmatically check for the clock MONOTONIC
>          behaviour.
> 
>          As discussed in the review thread, this has the potential of
> 	          breaking user space and we might have to revert
> 	          this. Knock on wood
> 		           that we can avoid that exercise.
> 
> 
> ...I believe you are breaking userspace for me. How can we check that?

And actually... how is it supposed to work? If monotonic behaves same
way as boottime, we'll have nasty thundering herd problem after
resume, riight?
									Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 181 bytes --]

Re: CLOCK_MONOTONIC, BOOTTIME, suspend and screensaver regression

[Rafael J. Wysocki]

On Tuesday, April 24, 2018 2:07:00 PM CEST Pavel Machek wrote:
> 
> --CE+1k2dSO48ffgeK
> Content-Type: text/plain; charset=us-ascii
> Content-Disposition: inline
> Content-Transfer-Encoding: quoted-printable
> 
> Hi!
> 
> In 4.17-rc1 I have weird problems with network manager... and I also
> have screensaver kicking just after resume.

I seem to be able to reproduce this on an Acer Aspire S5 (Ivy Bridge
generation CPU).

This machine appears to crash during resume, but I have not been able to
collect any debug output from that so far.

It has to stay suspended for several minutes in order to trigger this
behavior, but it is reproducible enough for me to try to bisect it
(which I'm going to do tomorrow if time permits).

At this point it looks like the breakage was introduced somewhere between
commits 9abf8acea297 and bb2407a7219760, but I need to verify that the
latter really works.

Thanks,
Rafael