Re: [PATCH 1/5] lib: introduce call_once()

[Akinobu Mita]

2008/3/11, Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>:
> On Tue, 11 Mar 2008 15:10:52 +1100 Nick Piggin <nickpiggin@xxxxxxxxxxxx> wrote:
>
>  > On Tuesday 11 March 2008 14:48, Andrew Morton wrote:
>  > > On Mon, 10 Mar 2008 23:57:05 +0900 Akinobu Mita <akinobu.mita@xxxxxxxxx>
>  > wrote:
>  > > > call_once() is an utility function which has similar functionality of
>  > > > pthread_once().
>  > > >
>  > > > +/*
>  > > > + * call_once - call the initialization function only once
>  > > > + *
>  > > > + * @once_control: guarantee that the init_routine will be called only
>  > > > once + * @init_routine: initialization function
>  > > > + *
>  > > > + * The first call to call_once(), with a given once_control, shall call
>  > > > the + * init_routine with no arguments and return the value init_routine
>  > > > returned. + * If the init_routine returns zero which indicates the
>  > > > initialization + * succeeded, subsequent calls of call_once() with the
>  > > > same once_control shall + * not call the init_routine and return zero.
>  > > > + */
>  > > > +
>  > > > +static inline int call_once(struct once_control *once_control,
>  > > > +                     int (*init_rouine)(void))
>  > > > +{
>  > > > + return likely(once_control->done) ? 0
>  > > > +                 : call_once_slow(once_control, init_rouine);
>  > > > +}
>  > >
>  > > I don't believe that this shold be described in terms of an "init_routine".
>  > > This mechanism can be used for things other than initialisation routines.
>  > >
>  > > It is spelled "routine", not "rouine".
>  > >
>  > >
>  > > Would it not be simpler and more general to do:
>  > >
>  > > #define ONCE()                                              \
>  > >     ({                                              \
>  > >             static long flag;                       \
>  > >                                                     \
>  > >             return !test_and_set_bit(0, flag);      \
>  > >     })
>  > >
>  > > and then callers can do
>  > >
>  > >     if (ONCE())
>  > >             do_something();
>  > >
>  > > ?

This cannot be used directly in the conversions of patch 2 - 5.
Because the callers of call_once() in patch 2 - 5 need to wait for the
complition of "init_routine". Some blocking mechanism is needed.

>  > Isn't this usually going to be buggy if you have concurrent access
>  > here? I'd prefer to keep synchronisation details in the caller and
>  > not have this call_once at all.
>
>
> Well, I'm a bit dubious about the calue of all of this (althoug I didn't
>  review the callers).
>

I'll try to find another conversions to back up the necessity...
--
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/