Re: linux-next: build failure after merge of the perf tree

[Arnaldo Carvalho de Melo]

Em Wed, Sep 07, 2022 at 09:52:48AM +0530, Anshuman Khandual escreveu:
> 
> 
> On 9/7/22 08:30, Stephen Rothwell wrote:
> > Hi all,
> > 
> > On Wed, 7 Sep 2022 08:01:34 +0530 Anshuman Khandual <anshuman.khandual@xxxxxxx> wrote:
> >>
> >> On 9/7/22 00:35, Stephen Rothwell wrote:
> >>> Hi all,
> >>>
> >>> On Wed, 31 Aug 2022 08:34:52 +1000 Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx> wrote:  
> >>>>
> >>>> After merging the perf tree, today's linux-next build (native perf)
> >>>> failed like this:
> >>>>
> >>>> In file included from /usr/include/stdio.h:866,
> >>>>                  from /home/sfr/next/next/tools/perf/util/branch.h:9,
> >>>>                  from util/branch.c:2:
> >>>> In function 'fprintf',
> >>>>     inlined from 'branch_type_stat_display' at util/branch.c:152:4:
> >>>> /usr/include/powerpc64le-linux-gnu/bits/stdio2.h:105:10: error: '%8s' directive argument is null [-Werror=format-overflow=]
> >>>>   105 |   return __fprintf_chk (__stream, __USE_FORTIFY_LEVEL - 1, __fmt,
> >>>>       |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> >>>>   106 |                         __va_arg_pack ());
> >>>>       |                         ~~~~~~~~~~~~~~~~~
> >>>> cc1: all warnings being treated as errors
> >>>>
> >>>> Presumably caused by commit
> >>>>
> >>>>   9781e500dcb8 ("perf branch: Extend branch type classification")
> >>>>
> >>>> "native" here is PowerPC64 LE.
> >>>> $ gcc --version
> >>>> gcc (Debian 11.2.0-10) 11.2.0
> >>>>
> >>>> I have used the perf tree from next-20220830 for today.  
> >>>
> >>> I am still seeing this build failure.  
> >>
> >> But did you apply the fix patch from Arnaldo that removes -Wno-format-overflow
> >> compiler option for the file util/branch.c ?
> >>
> >> https://lore.kernel.org/all/YxJBh3wvAGol+Ekq@xxxxxxxxxx/
> > 
> > No, I expected a fix to be in the perf tree ...
> > 
> > Also note that the following fixes the problem for me:
> > 
> > diff --git a/tools/perf/util/branch.c b/tools/perf/util/branch.c
> > index d40776c44b06..b7b898f2872e 100644
> > --- a/tools/perf/util/branch.c
> > +++ b/tools/perf/util/branch.c
> > @@ -88,7 +88,8 @@ const char *branch_type_name(int type)
> >  		"ERET",
> >  		"IRQ",
> >  		"SERROR",
> > -		"NO_TX"
> > +		"NO_TX",
> > +		""
> >  	};
> >  
> >  	if (type >= 0 && type < PERF_BR_MAX)
> 
> This looks right, makes sense.
> 
> > 
> > 
> > PERF_BR_MAX has been increased by one (when PERF_BR_EXTEND_ABI was
> > added), but a new string has not been added to the array ...
> 
> Right, even though new branch_new_names[] array gets queried on when type
> value is PERF_BR_EXTEND_ABI, branch_names[] should still contain an empty
> string "" just to match the now incremented PERF_BR_MAX which extends the
> array size as well. I guess the compiler detects this mismatch here and
> just complains about it.
> 
> Hello Arnaldo,
> 
> As adding empty string to the array solves the build problem, I guess we
> should fold this fix instead, rather than trying to drop the compiler
> option itself, as discussed earlier.

I'll do it, and even force push perf/core so that this problem doesn't
get into the bisection history upstream.

- Arnaldo
 
> The above fix should be folded into the following commit
> 
> 9781e500dcb8 ("perf branch: Extend branch type classification")
> 
> in the following perf tree - branch.
> 
> git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git (perf/core)
> 
> - Anshuman

-- 

- Arnaldo