Re: random.c changes for sparse irq_desc are crap

[Ingo Molnar]

* Matt Mackall <mpm@xxxxxxxxxxx> wrote:

> On Wed, 2008-12-31 at 16:14 -0800, Yinghai Lu wrote:
> > On Wed, Dec 31, 2008 at 3:40 PM, Matt Mackall <mpm@xxxxxxxxxxx> wrote:
> > > On Wed, 2008-12-31 at 15:07 -0800, Yinghai Lu wrote:
> > >> Matt Mackall wrote:
> > >> > I just noticed you merged a change that pointlessly converts two
> > >> > random.c functions into ugly random.h inlines without going through the
> > >> > maintainer.
> > >> >
> > >> > I also don't like the look of the newly-introduced sparse variants of
> > >> > these functions. Failure to find an irq descriptor in
> > >> > get_timer_rand_state is a BUG_ON should-never-happen sort of condition,
> > >> > not something to silently ignore. Letting the code try to dereference
> > >> > NULL is preferred here: we'll actually be able to find and fix the
> > >> > broken driver that's throwing around meaningless irq vectors.
> > >> >
> > >> > Throwing away the timer_state pointer in the set_timer_rand_state
> > >> > function is similarly bogus in addition to being a memory leak.
> > >> >
> > >> > Please fix this up.
> > >> >
> > >>
> > >> want something like this?
> > >
> > > Not quite.
> > >
> > > First, please turn these back into normal functions in random.c.
> > > Inlining functions is generally discouraged these days unless you have a
> > > good reason and numbers to back it up.
> > 
> > Ingo wanted to hide that #ifdef to .h
> 
> Ahh. I think that makes sense in some situations, but I'd prefer not to 
> do that here. Some headers are actually meant to be more readable than 
> the corresponding .c file. Also, we still end up with an ifdef in the .c 
> file so we're not actually winning.

Yinghai - i think Matt is right and there are a number of things we could 
do to improve cleanliness here. (Matt, sorry about this - we'll quickly 
fix it.)

The most important thing to observe is that the irq_desc->timer_rand_state 
cleanup you did should be made unconditional, and should not depend on 
SPARSE_IRQ. That will eliminate most of the #ifdefs, and it makes the 
random.c and random.h impact rather straightforward.

Another cleanliness detail is that these types of checks:

#ifndef CONFIG_SPARSE_IRQ
        if (irq >= nr_irqs)
                return;
#endif

can go away completely. They were needed from the sparseirq data model 
that had a dynamic array size - but that is not so anymore, so we can 
remove these complications.

Agreed?

	Ingo
--
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/