Re: [dpdk-dev] [PATCH 2/3] mbuf: avoid cast-align warning in pktmbuf mtod offset macro

[Olivier Matz <olivier.matz@6wind.com>]

Hi Eli,

On Tue, Oct 19, 2021 at 09:41:56AM +0300, Eli Britstein wrote:
> Hi Olivier,
> 
> On 8/1/2021 11:06 AM, Eli Britstein wrote:
> > 
> > On 7/30/2021 2:10 PM, Olivier Matz wrote:
> > > External email: Use caution opening links or attachments
> > > 
> > > 
> > > Hi Eli,
> > > 
> > > On Thu, Jul 29, 2021 at 10:13:45AM +0300, Eli Britstein wrote:
> > > > On 7/28/2021 6:28 PM, Olivier Matz wrote:
> > > > > External email: Use caution opening links or attachments
> > > > > 
> > > > > 
> > > > > On Tue, Jul 13, 2021 at 09:49:09AM +0300, Eli Britstein wrote:
> > > > > > In rte_pktmbuf_mtod_offset macro, there is a casting
> > > > > > from char * to type
> > > > > > 't', which may cause cast-align warning when using gcc flags
> > > > > > '-Werror -Wcast-align':
> > > > > > 
> > > > > > .../include/rte_mbuf_core.h:723:3: error: cast increases
> > > > > > required alignment
> > > > > >       of target type [-Werror=cast-align]
> > > > > >     723 |  ((t)((char *)(m)->buf_addr + (m)->data_off + (o)))
> > > > > >         |   ^
> > > > > > 
> > > > > > As the code assumes correct alignment, add first a (void
> > > > > > *) casting, to
> > > > > > avoid the warning.
> > > > > > 
> > > > > > Fixes: af75078fece3 ("first public release")
> > > > > > Cc: stable@dpdk.org
> > > > > > 
> > > > > > Signed-off-by: Eli Britstein <elibr@nvidia.com>
> > > > > My initial thinking was that it's the problem of the application: if
> > > > > -Werror=cast-align is used, it is up to the application to cast the
> > > > > return value of rte_pktmbuf_mtod_offset() to (void *) before
> > > > > casting it
> > > > > to the network type.
> > > > > 
> > > > > But, if I understand correctly, the problem is not about the
> > > > > application
> > > > > code itself, but about inlined code in the header files of dpdk
> > > > > (i.e. compiling an empty C file that just includes the dpdk
> > > > > headers with
> > > > > -Werror=cast-align). Is it correct? If yes I think it should be
> > > > > highlighted in the commit log.
> > > > I think yes, though in this specific patch it is not even an inline
> > > > function, but a macro.
> > > > 
> > > > However, I don't have a synthetic application example to show those
> > > > warnings, thus didn't put such in the commit msg.
> > > For this patch, I think it would be useful to have a way to reproduce
> > > the issue first, so we can check whether it is the proper place to fix
> > > the problem.
> > --- a/examples/l2fwd/Makefile
> > +++ b/examples/l2fwd/Makefile
> > @@ -22,6 +22,7 @@ static: build/$(APP)-static
> >         ln -sf $(APP)-static build/$(APP)
> > 
> >  PC_FILE := $(shell $(PKGCONF) --path libdpdk 2>/dev/null)
> > +CFLAGS += -Wcast-align=strict
> >  CFLAGS += -O3 $(shell $(PKGCONF) --cflags libdpdk)
> > 
> > gcc (Ubuntu 9.3.0-10ubuntu2) 9.3.0
> > Copyright (C) 2019 Free Software Foundation, Inc.
> > This is free software; see the source for copying conditions. There is NO
> > warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR
> > PURPOSE.
> > 
> > make -C examples/l2fwd clean static
> 
> To reproduce locally with DPDK only, no need to change any file. Only run:
> 
> CFLAGS="-Wcast-align=strict" make V=1 -C examples/l2fwd clean static
> 
> How would you like to proceed?

Sorry, I missed your previous message. I reproduced the issue, with
a slightly modified command:

  # no error, my gcc is 8.3.0-6 (debian)
  CFLAGS="-Wcast-align=strict" make V=1 -C examples/l2fwd clean static


  # bad option name with clang
  CC=clang CFLAGS="-Wcast-align=strict" make V=1 -C examples/l2fwd clean static
  ...
  warning: unknown warning option '-Wcast-align=strict'; did you mean '-Wcast-align'? [-Wunknown-warning-option]


  # problem reproduced with clang
  CC=clang CFLAGS="-Wcast-align" make V=1 -C examples/l2fwd clean static
  main.c:170:8: warning: cast from 'char *' to 'struct rte_ether_hdr *' increases required alignment from 1 to 2 [-Wcast-align]
          eth = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
                ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  /usr/local/include/rte_mbuf_core.h:830:32: note: expanded from macro 'rte_pktmbuf_mtod'
  #define rte_pktmbuf_mtod(m, t) rte_pktmbuf_mtod_offset(m, t, 0)
                                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  /usr/local/include/rte_mbuf_core.h:816:3: note: expanded from macro 'rte_pktmbuf_mtod_offset'
          ((t)((char *)(m)->buf_addr + (m)->data_off + (o)))


I confirm the patch fixes the issue.

Acked-by: Olivier Matz <olivier.matz@6wind.com>

Thanks