Re: p = kmalloc(sizeof(*p), )

[Denis Vlasenko <vda@ilport.com.ua>]

On Sunday 18 September 2005 17:39, Al Viro wrote:
> On Sun, Sep 18, 2005 at 12:04:34PM +0100, Alan Cox wrote:
>  
> > Other good practice in many cases is a single routine which allocates
> > and initialises the structure and is used by all allocators of that
> > object. That removes duplicate initialisers, stops people forgetting to
> > update all cases, allows better debug and far more.
> 
> Indeed.  IMO, argument for sizeof(*p) is bullshit - "I've changed a pointer
> type and forgot to update the allocation and initialization, but this will
> magically save my arse" is missing "except that initialization will remain
> bogus" part.
> 
> I've seen a lot of bugs around bogus kmalloc+initialization, but I can't
> recall a single case when such bug would be prevented by using that form.
> If somebody has a different experience, please post pointers to changesets
> in question.

Do these qualify?

http://www.uwsg.iu.edu/hypermail/linux/kernel/0105.1/0579.html
o Fix wrong kmalloc sizes in ixj/emu10k1 (David Chan) 

http://www.mail-archive.com/alsa-cvslog@lists.sourceforge.net/msg02483.html
Update of /cvsroot/alsa/alsa-kernel/isa
In directory sc8-pr-cvs1:/tmp/cvs-serv4034

Modified Files:
        es18xx.c cmi8330.c 
Log Message:
Fixed wrong kmalloc


After looking at output of grep -r -C10 'malloc.*sizeof' .
(epic picture) I think that maybe Alan's typechecking kmalloc
would be useful:

On Sunday 18 September 2005 14:04, Alan Cox wrote:
> If it bugs people add a kmalloc_object macro that is
> 
> #define new_object(foo, gfp) (foo *)kmalloc(sizeof(foo), (gfp))
> 
> then you can
> 
> 	x = new_object(struct frob, GFP_KERNEL)

This will emit a warning if x is not a struct frob*,
which plain kmalloc doesn't do.
--
vda