Deleting timers

Deleting timers

[Alan Stern]

Thomas:

The major difference -- in fact, almost the only difference -- between
del_timer() and try_to_del_timer_sync() is that try_to_del_timer_sync
returns a special code (-1) if the timer couldn't be deleted because it
is currently running, whereas del_timer doesn't check this.

Furthermore, the "_sync" in the name suggests that 
try_to_del_timer_sync will wait until a running timer has finished, 
which it clearly does not do.

Despite these facts, the kerneldoc for try_to_del_timer_sync states 
that it must not be called in interrupt context.  Why not?  Isn't that
advice simply wrong?

With this in mind, would there be any objection if I renamed it to 
try_to_del_timer(), removed the comment forbidding it to be used in 
interrupt context, and made it available even on non-SMP builds?

Alan Stern

P.S.: The only other difference is that del_timer calls
timer_stats_timer_clear_start_info.  Why doesn't try_to_del_timer_sync
do the same thing?

Re: Deleting timers

[Andrew Morton]

On Fri, 26 Jun 2009 15:50:54 -0400 (EDT) Alan Stern <stern@rowland.harvard.edu> wrote:

> Thomas:

I'm not Thomas, but I play one on TV.

> The major difference -- in fact, almost the only difference -- between
> del_timer() and try_to_del_timer_sync() is that try_to_del_timer_sync
> returns a special code (-1) if the timer couldn't be deleted because it
> is currently running, whereas del_timer doesn't check this.

And del_timer() is heaps faster against a not-pending timer.  I have a
vague memory that there are some callsites which do this quite a lot.

And try_to_del_timer_sync() forgot to do timer_stats_timer_clear_start_info().

> Furthermore, the "_sync" in the name suggests that 
> try_to_del_timer_sync will wait until a running timer has finished, 
> which it clearly does not do.

yup.

> Despite these facts, the kerneldoc for try_to_del_timer_sync states 
> that it must not be called in interrupt context.  Why not?  Isn't that
> advice simply wrong?

: commit fd450b7318b75343fd76b3d95416853e34e72c95
: Author:     Oleg Nesterov <oleg@tv-sign.ru>
: AuthorDate: Thu Jun 23 00:08:59 2005 -0700
: Commit:     Linus Torvalds <torvalds@ppc970.osdl.org>
: CommitDate: Thu Jun 23 09:45:16 2005 -0700
: 
:     [PATCH] timers: introduce try_to_del_timer_sync()
:     
:     This patch splits del_timer_sync() into 2 functions.  The new one,
:     try_to_del_timer_sync(), returns -1 when it hits executing timer.
:     
:     It can be used in interrupt context, or when the caller hold locks which
:     can prevent completion of the timer's handler.
:     
:     NOTE.  Currently it can't be used in interrupt context in UP case, because
:     ->running_timer is used only with CONFIG_SMP.
:     
:     Should the need arise, it is possible to kill #ifdef CONFIG_SMP in
:     set_running_timer(), it is cheap.
: 

The changelog is somewhat vodka-fogged, but there is a bit of a problem
there.

> With this in mind, would there be any objection if I renamed it to 
> try_to_del_timer(), removed the comment forbidding it to be used in 
> interrupt context, and made it available even on non-SMP builds?

Sounds sane to me, if the set_running_timer() change is also made.

> Alan Stern
> 
> P.S.: The only other difference is that del_timer calls
> timer_stats_timer_clear_start_info.  Why doesn't try_to_del_timer_sync
> do the same thing?

This could be a day-one bug in

: commit 82f67cd9fca8c8762c15ba7ed0d5747588c1e221
: Author:     Ingo Molnar <mingo@elte.hu>
: AuthorDate: Fri Feb 16 01:28:13 2007 -0800
: Commit:     Linus Torvalds <torvalds@woody.linux-foundation.org>
: CommitDate: Fri Feb 16 08:13:59 2007 -0800
: 
:     [PATCH] Add debugging feature /proc/timer_stat

timer-stats omits accumulation for del_timer_sync() also.

Re: Deleting timers

[Alan Stern]

On Wed, 1 Jul 2009, Andrew Morton wrote:

> On Fri, 26 Jun 2009 15:50:54 -0400 (EDT) Alan Stern <stern@rowland.harvard.edu> wrote:
> 
> > Thomas:
> 
> I'm not Thomas, but I play one on TV.
> 
> > The major difference -- in fact, almost the only difference -- between
> > del_timer() and try_to_del_timer_sync() is that try_to_del_timer_sync
> > returns a special code (-1) if the timer couldn't be deleted because it
> > is currently running, whereas del_timer doesn't check this.
> 
> And del_timer() is heaps faster against a not-pending timer.  I have a
> vague memory that there are some callsites which do this quite a lot.
> 
> And try_to_del_timer_sync() forgot to do timer_stats_timer_clear_start_info().
> 
> > Furthermore, the "_sync" in the name suggests that 
> > try_to_del_timer_sync will wait until a running timer has finished, 
> > which it clearly does not do.
> 
> yup.
> 
> > Despite these facts, the kerneldoc for try_to_del_timer_sync states 
> > that it must not be called in interrupt context.  Why not?  Isn't that
> > advice simply wrong?
> 
> : commit fd450b7318b75343fd76b3d95416853e34e72c95
> : Author:     Oleg Nesterov <oleg@tv-sign.ru>
> : AuthorDate: Thu Jun 23 00:08:59 2005 -0700
> : Commit:     Linus Torvalds <torvalds@ppc970.osdl.org>
> : CommitDate: Thu Jun 23 09:45:16 2005 -0700
> : 
> :     [PATCH] timers: introduce try_to_del_timer_sync()
> :     
> :     This patch splits del_timer_sync() into 2 functions.  The new one,
> :     try_to_del_timer_sync(), returns -1 when it hits executing timer.
> :     
> :     It can be used in interrupt context, or when the caller hold locks which
> :     can prevent completion of the timer's handler.
> :     
> :     NOTE.  Currently it can't be used in interrupt context in UP case, because
> :     ->running_timer is used only with CONFIG_SMP.
> :     
> :     Should the need arise, it is possible to kill #ifdef CONFIG_SMP in
> :     set_running_timer(), it is cheap.
> : 
> 
> The changelog is somewhat vodka-fogged, but there is a bit of a problem
> there.

Okay, thanks.  That makes sense.

> > With this in mind, would there be any objection if I renamed it to 
> > try_to_del_timer(), removed the comment forbidding it to be used in 
> > interrupt context, and made it available even on non-SMP builds?
> 
> Sounds sane to me, if the set_running_timer() change is also made.

It turns out I probably don't need the enhanced functionality after 
all.  So never mind for now...

Alan Stern

Re: Deleting timers

[Oleg Nesterov]

On 07/02, Alan Stern wrote:
> On Wed, 1 Jul 2009, Andrew Morton wrote:
>
> > On Fri, 26 Jun 2009 15:50:54 -0400 (EDT) Alan Stern <stern@rowland.harvard.edu> wrote:
> >
> > > Thomas:
> >
> > I'm not Thomas, but I play one on TV.
> >
> > > The major difference -- in fact, almost the only difference -- between
> > > del_timer() and try_to_del_timer_sync() is that try_to_del_timer_sync
> > > returns a special code (-1) if the timer couldn't be deleted because it
> > > is currently running, whereas del_timer doesn't check this.
> >
> > And del_timer() is heaps faster against a not-pending timer.  I have a
> > vague memory that there are some callsites which do this quite a lot.
> >
> > And try_to_del_timer_sync() forgot to do timer_stats_timer_clear_start_info().
> >
> > > Furthermore, the "_sync" in the name suggests that
> > > try_to_del_timer_sync will wait until a running timer has finished,
> > > which it clearly does not do.
> >
> > yup.

Yes, try_to_del_timer_sync() never waits exactly because it fails if the
timer is running.

> > > Despite these facts, the kerneldoc for try_to_del_timer_sync states
> > > that it must not be called in interrupt context.  Why not?  Isn't that
> > > advice simply wrong?
> >
> > : commit fd450b7318b75343fd76b3d95416853e34e72c95
> > : Author:     Oleg Nesterov <oleg@tv-sign.ru>
> > : AuthorDate: Thu Jun 23 00:08:59 2005 -0700
> > : Commit:     Linus Torvalds <torvalds@ppc970.osdl.org>
> > : CommitDate: Thu Jun 23 09:45:16 2005 -0700
> > :
> > :     [PATCH] timers: introduce try_to_del_timer_sync()
> > :
> > :     This patch splits del_timer_sync() into 2 functions.  The new one,
> > :     try_to_del_timer_sync(), returns -1 when it hits executing timer.
> > :
> > :     It can be used in interrupt context, or when the caller hold locks which
> > :     can prevent completion of the timer's handler.
> > :
> > :     NOTE.  Currently it can't be used in interrupt context in UP case, because
> > :     ->running_timer is used only with CONFIG_SMP.
> > :
> > :     Should the need arise, it is possible to kill #ifdef CONFIG_SMP in
> > :     set_running_timer(), it is cheap.
> > :
> >
> > The changelog is somewhat vodka-fogged, but there is a bit of a problem
> > there.

Yeah. try_to_del_timer_sync() should not be used in interrupt context
because in UP case it is equal to del_timer(), this is not what we want.

But with CONFIG_SMP it can work from any context.

> Okay, thanks.  That makes sense.
>
> > > With this in mind, would there be any objection if I renamed it to
> > > try_to_del_timer(),

Not sure I understand why try_to_del_timer is better...

try_to_del_timer_sync() means: try to del_timer_sync(), that is why
"_sync" ;)

But I don't really care.

> removed the comment forbidding it to be used in
> > > interrupt context, and made it available even on non-SMP builds?
> >
> > Sounds sane to me, if the set_running_timer() change is also made.

Yes, set_running_timer() should be changed, and

	# define try_to_del_timer_sync(t)       del_timer(t)

in timer.h should be killed. I think this makes sense.

Oleg.