Re: Re [patch RFC] mm/slab: introduce KZALLOC_FREE() cleanup-ed allocation macro

[Przemek Kitszel]

On 3/25/24 20:00, Dan Williams wrote:
> Przemek Kitszel wrote:
> > > From: Jiri Pirko <jiri@xxxxxxxxxx>
> > >
> > > With introduction of __free() macro using cleanup infrastructure, it
> > > will very likely become quite common to see following pattern:
> > > 	type *var __free(kfree) = kzalloc(sizeof(*var), GFP_KERNEL);
> > >
> > > To follow the CLASS() flow from cleanup.h, introduce a simple macro
> > > KZALLOC_FREE() to wrap this over and allow the same flow.
> > >
> > > Show an example usage in gpio-sim driver.
> > >
> > > Signed-off-by: Jiri Pirko <jiri@xxxxxxxxxx>
> > > ---
> > >   drivers/gpio/gpio-sim.c | 3 +--
> > >   include/linux/slab.h    | 3 +++
> > >   2 files changed, 4 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/gpio/gpio-sim.c b/drivers/gpio/gpio-sim.c
> > > index c4106e37e6db..997237b3d80c 100644
> > > --- a/drivers/gpio/gpio-sim.c
> > > +++ b/drivers/gpio/gpio-sim.c
> > > @@ -1496,8 +1496,7 @@ gpio_sim_config_make_device_group(struct config_group *group, const char *name)
> > >   {
> > >   	int id;
> > >   
> > > -	struct gpio_sim_device *dev __free(kfree) = kzalloc(sizeof(*dev),
> > > -							    GFP_KERNEL);
> > > +	KZALLOC_FREE(struct gpio_sim_device *, dev, GFP_KERNEL);
> > >   	if (!dev)
> > >   		return ERR_PTR(-ENOMEM);
> > >   
> > > diff --git a/include/linux/slab.h b/include/linux/slab.h
> > > index b5f5ee8308d0..baee6acd58d3 100644
> > > --- a/include/linux/slab.h
> > > +++ b/include/linux/slab.h
> > > @@ -711,6 +711,9 @@ static inline __alloc_size(1) void *kzalloc(size_t size, gfp_t flags)
> > >   	return kmalloc(size, flags | __GFP_ZERO);
> > >   }
> > >   
> > > +#define KZALLOC_FREE(_type, var, _gfp_t)				\
> > > +	_type var __free(kfree) = kzalloc(sizeof(*var), _gfp_t)
> > > +
> >
> > Nice, but I would rather see this wrapper in the cleanup.h file, that have all
> > of the rest of related stuff.
> >
> > On top of that, I want to propose also a wrapper that is simpler in that it
> > does not allocate but just assigns null, with that in mind `_FREE` part of your
> > proposed name does not sound right.
>
> No, do not hide assignments within macros

As most general advice I agree, but here we have a specific case:
declare variable via macro; and that, (given the macro name would be
clearer), is expected to have assignment (or default (un)init).
I would even go one step further and remove also the asterisk from the
call site (and *hide* it in the macro definition).

See _DEFINE_FLEX() as example:
(there we change on-stack instead $this_thread on-heap-autofree)
https://elixir.bootlin.com/linux/v6.9-rc1/source/include/linux/overflow.h#L401 


> http://lore.kernel.org/r/CAHk-=whYxkfLVtBW_B-PgNqhKOAThTbfoH5CxtOTkwOB6VOt6w@xxxxxxxxxxxxxx

Your thread is a more complex thing to what we have here.
And BTW, your original proposed solution is nice, and even if it hides
flow inside, it's almost obvious (the `return -EINTR` statement
is verbatim at call site). Allowing `else return -EINTR;` solution
proposed by @Linus is nicer, makes a good idiom, but is less obvious:
Imagine two developers that don't know the API (well), one writes:
`scoped_cond_guard(args);` and forgets to handle the error case,
the other by just looking at the code have no idea to append
`else handle_err();`.

> I.e. the amount of incremenal cleverness that include/linux/cleanup.h
> will tolerate is low. Any helper should look like typical C