Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1433124 > unrolled thread

Re: [PATCH] PM / suspend: show workqueues busy name in suspend flow

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH] PM / suspend: show workqueues busy name in suspend flow Tejun Heo <tj@kernel.org> - 2016-06-28 19:00 +0200
    Re: [PATCH] PM / suspend: show workqueues busy name in suspend flow Roger Lu <roger.lu@mediatek.com> - 2016-06-29 06:00 +0200
      Re: [PATCH] PM / suspend: show workqueues busy name in suspend flow Tejun Heo <tj@kernel.org> - 2016-06-29 15:30 +0200

#1433124 — Re: [PATCH] PM / suspend: show workqueues busy name in suspend flow

Hello, Roger.

On Mon, Jun 27, 2016 at 05:44:05PM +0800, Roger Lu wrote:
> show_workqueue_state() is a better choice to me. However, only freezable
> workqueue is able to affect suspend flow. So, is there other mailing
> list discussing about showing freezable workqueue state only?? Maybe we
> can use that API in this case. Thanks very much.
> 
> freezable workqueue means workqueue is created with flag WQ_FREEZABLE.

It's for debugging anyway and workqueue dumps usually are pretty
short.  I don't think it's fine to use the same function.  We can add
flags in the printouts but I'm not even sure that'd be necessary.

Thanks.

-- 
tejun

[toc] | [next] | [standalone]

#1433430

Dear Tejun,

On Tue, 2016-06-28 at 12:56 -0400, Tejun Heo wrote:
> Hello, Roger.
> 
> On Mon, Jun 27, 2016 at 05:44:05PM +0800, Roger Lu wrote:
> > show_workqueue_state() is a better choice to me. However, only freezable
> > workqueue is able to affect suspend flow. So, is there other mailing
> > list discussing about showing freezable workqueue state only?? Maybe we
> > can use that API in this case. Thanks very much.
> > 
> > freezable workqueue means workqueue is created with flag WQ_FREEZABLE.
> 
> It's for debugging anyway and workqueue dumps usually are pretty
> short.  I don't think it's fine to use the same function.  We can add
> flags in the printouts but I'm not even sure that'd be necessary.
> 
> Thanks.
> 

Please allow me to elaborate my previous concern about printing
freezable workqueue info only in this case.

The benefit of it is that debugger can quickly understand which
freezable workqueues block suspend flow and assign this issue to
corresponding owner instead of extracting freezable workqueue info from
show_workqueue_state() first and, then, assigning the issue.

Adding a flag to printout the info we need is great. Perhaps we can do
that. Thanks for the advice.

Sincerely,
Roger Lu.

[toc] | [prev] | [next] | [standalone]

#1433730

Hello, Roger.

On Wed, Jun 29, 2016 at 11:54:11AM +0800, Roger Lu wrote:
> Please allow me to elaborate my previous concern about printing
> freezable workqueue info only in this case.
> 
> The benefit of it is that debugger can quickly understand which
> freezable workqueues block suspend flow and assign this issue to
> corresponding owner instead of extracting freezable workqueue info from
> show_workqueue_state() first and, then, assigning the issue.

I don't think it matters.  At that point, workqueues are generally
pretty idle anyway and it shouldn't be difficult to tell which work
items are the offending ones.  Besides, freezable and unfreezable
workqueues share the same backend pools, so it isn't easily separable
either.

Thanks.

-- 
tejun

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web