Re: [PATCH] Fix kunmap() argument in sg_miter_stop

[Ingo Molnar <mingo@elte.hu>]

* Jens Axboe <jens.axboe@oracle.com> wrote:

> On Mon, Nov 17 2008, Jens Axboe wrote:
> > On Mon, Nov 17 2008, Ingo Molnar wrote:
> > > 
> > > * Jens Axboe <jens.axboe@oracle.com> wrote:
> > > 
> > > > On Mon, Nov 17 2008, Ingo Molnar wrote:
> > > > > 
> > > > > * Jens Axboe <jens.axboe@oracle.com> wrote:
> > > > > 
> > > > > > +#define kunmap(p)			\
> > > > > > +	do {				\
> > > > > > +		struct page *__p;	\
> > > > > > +		(void) (&__p == &(p));	\
> > > > > > +		__kunmap(p);		\
> > > > > > +	} while (0)
> > > > > > +
> > > > > > +#define kunmap_atomic(a, t)		\
> > > > > > +	do {				\
> > > > > > +		void *__p;		\
> > > > > > +		(void) (&__p == &(a));	\
> > > > > > +		__kunmap_atomic(a, t);	\
> > > > > > +	} while (0)
> > > > > 
> > > > > Agreed - but please use the typecheck() primitive. (linux/typecheck.h)
> > > > 
> > > > Neat, didn't know about that, thanks.
> > > 
> > > and ack on your patch obviously. Feel free to push it via the block 
> > > tree straight away, it doesnt collide with anything pending in the x86 
> > > tree.
> > > 
> > > Acked-by: Ingo Molnar <mingo@elte.hu>
> > 
> > The kunmap() bit is easy to do as I mentioned, but the kunmap_atomic()
> > gets a bit more ugly. Lots of users can just be switched to void *
> > types, but some get ugly like:
> > 
> > static struct page **shmem_dir_map(struct page *page)
> >         return (struct page **)kmap_atomic(page, KM_USER0);
> >  }
> >  
> > -static inline void shmem_dir_unmap(struct page **dir)
> > +static inline void shmem_dir_unmap(void *dir)
> >  {
> >         kunmap_atomic(dir, KM_USER0);
> >  }
> > 
> > and others again like fs/exec.c:remove_arg_zero() would really like to
> > use a char * type since it dereferences it.
> > 
> > I think the kunmap_atomic() change it the most needed part of the patch,
> > but I don't think it can come for free (eg, we have to add variable to
> > the above function).
> 
> OK, (void *) cast works fine when I fixed it. Here's a normal build diff
> to show approximately how bad it is. I expected it to be worse, so it
> looks quite doable I think.

hm:

   32 files changed, 89 insertions(+), 62 deletions(-)

regarding churn it looks a bit like on the high side. OTOH it's 
trivial, most of the churn is spread-out in various architectures, and 
it can catch real bugs so i'm still acking it.

Could you please send the patch with a full changelog? I've Cc:-ed 
Linus and Andrew - please let us know if you think that this too much 
churn for -rc6.

	Ingo