Message ID | 1390914658-18198-1-git-send-email-mjonker@synopsys.com |
---|---|
State | Accepted |
Headers | show |
Hi Mischa, On Tue, Jan 28, 2014 at 2:10 PM, Mischa Jonker <Mischa.Jonker@synopsys.com> wrote: > Signed-off-by: Mischa Jonker <mjonker@synopsys.com> > --- > package/perf/perf.mk | 4 ++++ > 1 file changed, 4 insertions(+) > > diff --git a/package/perf/perf.mk b/package/perf/perf.mk > index ffe18bc..5628ba5 100644 > --- a/package/perf/perf.mk > +++ b/package/perf/perf.mk > @@ -22,6 +22,10 @@ PERF_MAKE_FLAGS = \ > WERROR=0 \ > ASCIIDOC= > > +ifeq ($(BR2_arc),y) > + PERF_MAKE_FLAGS += NO_BACKTRACE=1 > +endif > + > ifeq ($(BR2_PACKAGE_ELFUTILS),y) > PERF_DEPENDENCIES += elfutils > else Could you give more details as to why this is needed? I see no autobuild failures for perf on ARC, so what is the problem here? If this patch is still needed, I think the commit message should be expanded for that. Thanks, Thomas
Anton, Same for this mail, thanks. On Fri, Jul 18, 2014 at 10:16 PM, Thomas De Schampheleire <patrickdepinguin@gmail.com> wrote: > Hi Mischa, > > On Tue, Jan 28, 2014 at 2:10 PM, Mischa Jonker > <Mischa.Jonker@synopsys.com> wrote: >> Signed-off-by: Mischa Jonker <mjonker@synopsys.com> >> --- >> package/perf/perf.mk | 4 ++++ >> 1 file changed, 4 insertions(+) >> >> diff --git a/package/perf/perf.mk b/package/perf/perf.mk >> index ffe18bc..5628ba5 100644 >> --- a/package/perf/perf.mk >> +++ b/package/perf/perf.mk >> @@ -22,6 +22,10 @@ PERF_MAKE_FLAGS = \ >> WERROR=0 \ >> ASCIIDOC= >> >> +ifeq ($(BR2_arc),y) >> + PERF_MAKE_FLAGS += NO_BACKTRACE=1 >> +endif >> + >> ifeq ($(BR2_PACKAGE_ELFUTILS),y) >> PERF_DEPENDENCIES += elfutils >> else > > > Could you give more details as to why this is needed? > I see no autobuild failures for perf on ARC, so what is the problem here? > If this patch is still needed, I think the commit message should be > expanded for that. > > Thanks, > Thomas
Dear Thomas De Schampheleire, On Fri, 18 Jul 2014 22:16:44 +0200, Thomas De Schampheleire wrote: > Could you give more details as to why this is needed? > I see no autobuild failures for perf on ARC, so what is the problem here? Since perf has 'depends on BR2_LINUX_KERNEL', it is never built by the autobuilders. Thomas
Hi Thomas, Call to backtrace() function fails for ARC, because for some reason unwinder from libgcc returns early. Thus usage of backtrace() should be disabled in perf explicitly, because build system only checks for existence of this call and since it is there it tries to use. However at runtime the output will not be the one expected: it will contain only several functions from the top of stack, instead of the complete backtrace. > -----Original Message----- > From: Thomas De Schampheleire [mailto:patrickdepinguin@gmail.com] > Sent: 19 July 2014 00:24 > To: Anton Kolesov > Cc: buildroot@busybox.net > Subject: Re: [Buildroot] [PATCH 2/2] perf: Don't build backtrace for ARC > > Anton, > > Same for this mail, thanks. > > On Fri, Jul 18, 2014 at 10:16 PM, Thomas De Schampheleire > <patrickdepinguin@gmail.com> wrote: > > Hi Mischa, > > > > On Tue, Jan 28, 2014 at 2:10 PM, Mischa Jonker > > <Mischa.Jonker@synopsys.com> wrote: > >> Signed-off-by: Mischa Jonker <mjonker@synopsys.com> > >> --- > >> package/perf/perf.mk | 4 ++++ > >> 1 file changed, 4 insertions(+) > >> > >> diff --git a/package/perf/perf.mk b/package/perf/perf.mk > >> index ffe18bc..5628ba5 100644 > >> --- a/package/perf/perf.mk > >> +++ b/package/perf/perf.mk > >> @@ -22,6 +22,10 @@ PERF_MAKE_FLAGS = \ > >> WERROR=0 \ > >> ASCIIDOC= > >> > >> +ifeq ($(BR2_arc),y) > >> + PERF_MAKE_FLAGS += NO_BACKTRACE=1 > >> +endif > >> + > >> ifeq ($(BR2_PACKAGE_ELFUTILS),y) > >> PERF_DEPENDENCIES += elfutils > >> else > > > > > > Could you give more details as to why this is needed? > > I see no autobuild failures for perf on ARC, so what is the problem here? > > If this patch is still needed, I think the commit message should be > > expanded for that. > > > > Thanks, > > Thomas Anton
Dear Anton Kolesov, Please fix your e-mail client to wrap lines at a reasonable length. Thanks! On Mon, 21 Jul 2014 09:39:05 +0000, Anton Kolesov wrote: > Call to backtrace() function fails for ARC, because for some reason > unwinder from libgcc returns early. Thus usage of backtrace() should > be disabled in perf explicitly, because build system only checks for > existence of this call and since it is there it tries to use. However > at runtime the output will not be the one expected: it will contain > only several functions from the top of stack, instead of the complete > backtrace. Thanks for this explanation. I've committed the patch after adding this explanation right above the code that disables backtrace for perf on ARC. Thanks, Thomas
diff --git a/package/perf/perf.mk b/package/perf/perf.mk index ffe18bc..5628ba5 100644 --- a/package/perf/perf.mk +++ b/package/perf/perf.mk @@ -22,6 +22,10 @@ PERF_MAKE_FLAGS = \ WERROR=0 \ ASCIIDOC= +ifeq ($(BR2_arc),y) + PERF_MAKE_FLAGS += NO_BACKTRACE=1 +endif + ifeq ($(BR2_PACKAGE_ELFUTILS),y) PERF_DEPENDENCIES += elfutils else
Signed-off-by: Mischa Jonker <mjonker@synopsys.com> --- package/perf/perf.mk | 4 ++++ 1 file changed, 4 insertions(+)