Re: [PATCH 01/16] eal: add function attributes for allocation functions

[Stephen Hemminger <stephen@networkplumber.org>]

On Fri, 27 Sep 2024 18:09:22 -0400
David Marchand <david.marchand@redhat.com> wrote:

> On Fri, Sep 27, 2024 at 4:48 PM Stephen Hemminger
> <stephen@networkplumber.org> wrote:
> >
> > The allocation functions take a alignment argument that
> > can be useful to hint the compiler optimizer.
> >
> > This is supported by Gcc and Clang but only useful with
> > Gcc because Clang gives warning if alignment is 0.
> >
> > Recent versions of GCC have a malloc attribute that can
> > be used to find mismatches between allocation and free;
> > the typical problem caught is a pointer allocated with
> > rte_malloc() that is then incorrectly freed using free().  
> 
> Interesting tool.
> 
> >
> > Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
> > ---
> >  lib/eal/include/rte_common.h | 30 ++++++++++++++++++++++++++++++
> >  lib/eal/include/rte_malloc.h | 24 ++++++++++++++++--------
> >  2 files changed, 46 insertions(+), 8 deletions(-)
> >
> > diff --git a/lib/eal/include/rte_common.h b/lib/eal/include/rte_common.h
> > index eec0400dad..1b3781274d 100644
> > --- a/lib/eal/include/rte_common.h
> > +++ b/lib/eal/include/rte_common.h
> > @@ -228,6 +228,36 @@ typedef uint16_t unaligned_uint16_t;
> >  #define __rte_alloc_size(...)
> >  #endif
> >
> > +/**
> > + * Tells the compiler that the function returns a value that points to
> > + * memory aligned by a function argument.
> > + * Not enabled on clang because it warns if align argument is zero.
> > + */
> > +#if defined(RTE_CC_GCC)
> > +#define __rte_alloc_align(align_arg) \
> > +       __attribute__((alloc_align(align_arg)))
> > +#else
> > +#define __rte_alloc_align(...)
> > +#endif
> > +
> > +/**
> > + * Tells the compiler this is a function like malloc and that the pointer
> > + * returned cannot alias any other pointer (ie new memory).
> > + *
> > + * Also, with recent GCC versions also able to track that proper
> > + * dealloctor function is used for this pointer.
> > + */
> > +#if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION >= 110000)  
> 
> Even though it is probably equivalent, GCC_VERSION is set with RTE_CC_GCC.
> 
> > +#define __rte_alloc_func(free_func) \
> > +       __attribute__((malloc, malloc(free_func)))  
> 
> I read that this malloc attribute can also make use of the arg index
> to assume the pointer is freed.
> 
> Did you try this feature?

Yes, but all DPDK functions use first arg, so not really that relevant