Re: [PATCH 07/16] freezer: don't distinguish nosig tasks on thaw

[Rafael J. Wysocki]

On Saturday, August 20, 2011, Tejun Heo wrote:
> Hello, Rafael.
> 
> On Fri, Aug 19, 2011 at 11:14:52PM +0200, Rafael J. Wysocki wrote:
> > On Friday, August 19, 2011, Tejun Heo wrote:
> > > There's no point in thawing nosig tasks before others.  There's no
> > > ordering requirement between the two groups on thaw, which the staged
> > > thawing can't guarantee anyway.  Simplify thaw_processes() by removing
> > > the distinction and collapsing thaw_tasks() into thaw_processes().
> > > This will help further updates to freezer.
> > 
> > I'm not sure if I like this patch.
> > 
> > Right now there are no ordering requirements between the two groups
> > of processes, but if we decide to freeze filesystems on suspend,
> > we'll need to thaw them between nosig and sig I suppose.
> 
> Hmmm... I'm not really following.  How does staged wake up affect
> thawing filesystems?  Staged freezing makes sense as a crude way to
> define dependency during freezing - ie. userland and freezable tasks
> can't have dependency in their own groups but the former can depend on
> the latter on the way to refrigerator.
> 
> However, during thawing, it doesn't make any difference regardless of
> what was frozen when and how they depend on each other.  They might as
> well have cyclic dependency and waking them in any order wouldn't make
> any difference.  The task which dependes on another task to do
> something would simply block until that task wakes up and resolves the
> dependency; moreover, performing staged wakeups doesn't really
> guarantee execution order.  It's different from staged freezing in
> that way - staged thawing doesn't have the synchronization phase
> between the two stages.  Tasks which were woken up earlier can easily
> start executing after tasks which were woken up later.
> 
> The only guaranteed effect of staged wakeups is that tasks in the
> earlier group would have had its ->state set to TASK_RUNNING before
> the tasks of the second group.  This again is a moot point because
> 
> * __refrigerator() restores task->state afterwards overwriting the
>   TASK_RUNNING once the task starts executing (in unknown order).
>   This is fundamentally broken and should be fixed so that task is
>   left in TASK_RUNNING when leaving the refrigerator.
> 
> * However, if you leave it at TASK_RUNNING, it doesn't make any
>   difference w.r.t. synchronization.  The only way task->state can
>   participate in synchronization is through wake_up() - ie. through
>   other tasks setting its state to TASK_RUNNING, so if the
>   refrigerator leaves tast state at TASK_RUNNING on return, it can't
>   hinder any synchronization.
> 
> So, AFAICS, no matter which way it's looked at, it just doesn't make
> any difference.

OK

Rafael
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/