Re: [GIT PULL 0/7] perf/core improvements and fixes

[Jiri Olsa]

On Tue, Nov 24, 2015 at 10:26:18AM +0100, Jiri Olsa wrote:
> On Tue, Nov 24, 2015 at 09:42:10AM +0100, Ingo Molnar wrote:
> 
> SNIP
> 
> > > 
> > > > 
> > > > Occasionally it also triggers a build failure:
> > > > 
> > > >   CC       plugin_kvm.o
> > > >   CC       arch/common.o
> > > >   CC       util/db-export.o
> > > >   LD       plugin_kmem-in.o
> > > > fixdep: error opening depfile: ./.plugin_kmem.o.d: No such file or directory
> > > > /home/mingo/tip/tools/build/Makefile.build:77: recipe for target 'plugin_kmem.o' 
> > > > failed
> > > > make[3]: *** [plugin_kmem.o] Error 2
> > > > Makefile:189: recipe for target 'plugin_kmem-in.o' failed
> > > > make[2]: *** [plugin_kmem-in.o] Error 2
> > > > Makefile.perf:424: recipe for target 'install-traceevent-plugins' failed
> > > > make[1]: *** [install-traceevent-plugins] Error 2
> > > > make[1]: *** Waiting for unfinished jobs....
> > > > 
> > > > that too seems to be a result of unwanted, over-eager parallelism.
> > > 
> > > hm, haven't seen that one for long time now ;-) will check
> > 
> > So maybe that's a side effect, because I do:
> > 
> >   make clean install
> > 
> > and maybe 'clean' happens in parallel with 'install'?
> 
> the fixdep error indicates that it was invoked after building object
> plugin_kmem-in.o, but it could not find its dep file (.plugin_kmem-in.o.d)
> which is built within the object build.. so seems like race with clean
> or other build of the same object
> 
> however ;-)
> 
> The 'Makefile' processing itself isn't paralel and will exec
> clean and install targets serialized:
> 
> ---
> [root@intel-wildcatpass-07 perf]# make clean install
> make -f Makefile.perf --no-print-directory -j88 O=  clean
>   CLEAN    libtraceevent
>   CLEAN    libapi
>   CLEAN    libbpf
>   CLEAN    config
>   CLEAN    core-objs
>   CLEAN    core-progs
>   CLEAN    core-gen
>   SUBDIR   Documentation
>   CLEAN    Documentation
>   CLEAN    python
>   BUILD:   Doing 'make -j88' parallel build
> make -f Makefile.perf --no-print-directory -j88 O=  install
> 
> Auto-detecting system features:
> ...                         dwarf: [ on  ]
> ...                         glibc: [ on  ]
> ---
> 
> I can't make that failure on 88 cpus server, I assume you can
> reproduce this fairly easily?
> 
> Could you please share failing build output from:
>   $ make V=1 clean install
> 
> thanks,
> jirka
> 

I think I found one race..
 - having install-traceevent-plugins depend on $(LIBTRACEEVENT),
   plugins will not be built as its prereq. and the target
   install-traceevent-plugins itself will trigger plugins build
 - but plugins build is also triggered by perf build itself
   via libtraceevent_plugins target

so those 2 might race.. but as I said, I've never reproduced ;-)

Could you please give it a try?

thanks,
jirka


---
diff --git a/tools/perf/Makefile.perf b/tools/perf/Makefile.perf
index 0d19d5447d6c..929a32ba15f5 100644
--- a/tools/perf/Makefile.perf
+++ b/tools/perf/Makefile.perf
@@ -420,7 +420,7 @@ $(LIBTRACEEVENT)-clean:
 	$(call QUIET_CLEAN, libtraceevent)
 	$(Q)$(MAKE) -C $(TRACE_EVENT_DIR) O=$(OUTPUT) clean >/dev/null
 
-install-traceevent-plugins: $(LIBTRACEEVENT)
+install-traceevent-plugins: libtraceevent_plugins
 	$(Q)$(MAKE) -C $(TRACE_EVENT_DIR) $(LIBTRACEEVENT_FLAGS) O=$(OUTPUT) install_plugins
 
 $(LIBAPI): fixdep FORCE
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/