Re: stable build: 203 builds: 4 failed, 199 passed, 5 errors, 41 warnings (v4.10.1)

[Jiri Olsa]

On Thu, Mar 16, 2017 at 12:03:38PM -0300, Arnaldo Carvalho de Melo wrote:
> Em Thu, Mar 16, 2017 at 03:39:36PM +0100, Arnd Bergmann escreveu:
> > On Thu, Mar 16, 2017 at 2:59 PM, Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
> > > On Thu, Mar 16, 2017 at 02:44:45PM +0100, Arnd Bergmann wrote:
> > >> On Thu, Mar 16, 2017 at 1:49 PM, Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
> > >> > On Thu, Mar 16, 2017 at 09:29:07AM -0300, Arnaldo Carvalho de Melo wrote:
> > >> >> Em Wed, Mar 15, 2017 at 02:15:22PM +0100, Arnd Bergmann escreveu:
> > >> >> > On Wed, Mar 15, 2017 at 8:22 AM, gregkh <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> > >> >> > >
> > >> >> > > All now queued up in the stable trees, thanks.
> > >> >> >
> > >> >> > Like 4.9.y it builds clean except for a couple of stack frame size warnings
> > >> >> > and this one that continues to puzzle me.
> > >> >> >
> > >> >> > /bin/sh: 1: /home/buildslave/workspace/kernel-builder/arch/x86/defconfig/allmodconfig+CONFIG_OF=n/label/builder/next/build-x86/tools/objtool//fixdep:
> > >> >> > Permission denied
> > >> >>
> > >> >> Jiri? Josh?
> > >> >
> > >> > hum, looks like it imight be related to this fix we did for perf:
> > >> >   abb26210a395 perf tools: Force fixdep compilation at the start of the build
> > >> >
> > >> > it's forcing fixdep to be build as first.. having it as a simple dependency
> > >> > (which AFAICS is objtool case), the make -jX occasionaly raced on high cpu
> > >> > servers, and executed unfinished binary, hence the permission fail
> > >>
> > >> It's probably another variation of this bug, but the commit you cite got merged
> > >> into 4.10-rc1, while the problem still persists in mainline (4.11-rc2+).
> > >
> > > the problem is in objtool build right? the fix was for perf build
> > 
> > Ah, got it. Yes, that must be it then. I supposed we coul duplicate what you
> > did for perf in objtool, but a cleaner way would be to generalize it for all of
> > tools/, right?

right, the thing is that objtool is standalone application like perf,
and before their builds can go the 'fixdep' needs to be there.. that's
a condition to use the tools/build framework

not sure how offensive it'd be to current Makefiles if we come with some
generalized code to do that.. I'll think about it, but I think we might
be better of the way we are now

> 
> Humm, can't we have just one fixdep?

we have.. it's just the matter who will build it first ;-)

jirka