Re: Deleting timers

[Andrew Morton]

On Fri, 26 Jun 2009 15:50:54 -0400 (EDT) Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> 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@xxxxxxxxxx>
: AuthorDate: Thu Jun 23 00:08:59 2005 -0700
: Commit:     Linus Torvalds <torvalds@xxxxxxxxxxxxxxx>
: 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@xxxxxxx>
: AuthorDate: Fri Feb 16 01:28:13 2007 -0800
: Commit:     Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxxxxxxxx>
: 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.
--
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/