Hi all,
I started seeing build failures such as the following in the last week or so.
make[2]: [__modpost] Error 1
followed by lots of messages such as
gcc: error: arch/x86/crypto/ablk_helper.mod.c: No such file or directory
gcc: fatal error: no input files
(at least if I run make -i)
After reverting commit ea4054a238 (modpost: handle huge numbers of modules)
everything is fine.
This happens with multiple gcc versions and target platforms (arm, x86_64,
i386, powerpc). Host kernel version is 3.8.10, system is Ubuntu 12.10.
Does anyone else see this problem ? Any idea what might be wrong ?
Thanks,
Guenter
Guenter Roeck <[email protected]> writes:
> Hi all,
>
> I started seeing build failures such as the following in the last week or so.
>
> make[2]: [__modpost] Error 1
>
> followed by lots of messages such as
>
> gcc: error: arch/x86/crypto/ablk_helper.mod.c: No such file or directory
> gcc: fatal error: no input files
>
> (at least if I run make -i)
Hmm, weird! Looks like __modules contains something it shouldn't. We
changed from:
__modules := $(sort $(shell grep -h '\.ko$$' /dev/null $(wildcard $(MODVERDIR)/*.mod)))
To:
MODLISTCMD := find $(MODVERDIR) -name '*.mod' | xargs -r grep -h '\.ko$$' | sort -u
__modules := $(shell $(MODLISTCMD))
They give the same results for me:
diff --git a/scripts/Makefile.modpost b/scripts/Makefile.modpost
index 8dcdca2..1ca358d 100644
--- a/scripts/Makefile.modpost
+++ b/scripts/Makefile.modpost
@@ -62,8 +62,14 @@ modulesymfile := $(firstword $(KBUILD_EXTMOD))/Module.symvers
# Step 1), find all modules listed in $(MODVERDIR)/
MODLISTCMD := find $(MODVERDIR) -name '*.mod' | xargs -r grep -h '\.ko$$' | sort -u
__modules := $(shell $(MODLISTCMD))
+__old_modules := $(sort $(shell grep -h '\.ko$$' /dev/null $(wildcard $(MODVERDIR)/*.mod)))
modules := $(patsubst %.o,%.ko, $(wildcard $(__modules:.ko=.o)))
+.PHONY: dump_modules
+dump_modules:
+ @echo $(__modules) > __modules.list
+ @echo $(__old_modules) > __old_modules.list
+
# Stop after building .o files if NOFINAL is set. Makes compile tests quicker
_modpost: $(if $(KBUILD_MODPOST_NOFINAL), $(modules:.ko:.o),$(modules))
@@ -84,7 +90,7 @@ quiet_cmd_modpost = MODPOST $(words $(filter-out vmlinux FORCE, $^)) modules
cmd_modpost = $(MODLISTCMD) | sed 's/\.ko$$/.o/' | $(modpost) -s -T -
PHONY += __modpost
-__modpost: $(modules:.ko=.o) FORCE
+__modpost: $(modules:.ko=.o) dump_modules FORCE
$(call cmd,modpost) $(wildcard vmlinux)
quiet_cmd_kernel-mod = MODPOST $@
> After reverting commit ea4054a238 (modpost: handle huge numbers of modules)
> everything is fine.
>
> This happens with multiple gcc versions and target platforms (arm, x86_64,
> i386, powerpc). Host kernel version is 3.8.10, system is Ubuntu 12.10.
>
> Does anyone else see this problem ? Any idea what might be wrong ?
Assuming it happens with a clean tree, please send the .config, and I'll
see if I can reproduce...
Thanks,
Rusty.
Hi Rusty,
On Thu, May 09, 2013 at 10:26:00AM +0930, Rusty Russell wrote:
> Guenter Roeck <[email protected]> writes:
> > Hi all,
> >
> > I started seeing build failures such as the following in the last week or so.
> >
> > make[2]: [__modpost] Error 1
> >
> > followed by lots of messages such as
> >
> > gcc: error: arch/x86/crypto/ablk_helper.mod.c: No such file or directory
> > gcc: fatal error: no input files
> >
> > (at least if I run make -i)
>
> Hmm, weird! Looks like __modules contains something it shouldn't. We
> changed from:
>
> __modules := $(sort $(shell grep -h '\.ko$$' /dev/null $(wildcard $(MODVERDIR)/*.mod)))
>
> To:
>
> MODLISTCMD := find $(MODVERDIR) -name '*.mod' | xargs -r grep -h '\.ko$$' | sort -u
> __modules := $(shell $(MODLISTCMD))
>
> They give the same results for me:
>
> diff --git a/scripts/Makefile.modpost b/scripts/Makefile.modpost
> index 8dcdca2..1ca358d 100644
> --- a/scripts/Makefile.modpost
> +++ b/scripts/Makefile.modpost
> @@ -62,8 +62,14 @@ modulesymfile := $(firstword $(KBUILD_EXTMOD))/Module.symvers
> # Step 1), find all modules listed in $(MODVERDIR)/
> MODLISTCMD := find $(MODVERDIR) -name '*.mod' | xargs -r grep -h '\.ko$$' | sort -u
> __modules := $(shell $(MODLISTCMD))
> +__old_modules := $(sort $(shell grep -h '\.ko$$' /dev/null $(wildcard $(MODVERDIR)/*.mod)))
> modules := $(patsubst %.o,%.ko, $(wildcard $(__modules:.ko=.o)))
>
> +.PHONY: dump_modules
> +dump_modules:
> + @echo $(__modules) > __modules.list
> + @echo $(__old_modules) > __old_modules.list
> +
> # Stop after building .o files if NOFINAL is set. Makes compile tests quicker
> _modpost: $(if $(KBUILD_MODPOST_NOFINAL), $(modules:.ko:.o),$(modules))
>
> @@ -84,7 +90,7 @@ quiet_cmd_modpost = MODPOST $(words $(filter-out vmlinux FORCE, $^)) modules
> cmd_modpost = $(MODLISTCMD) | sed 's/\.ko$$/.o/' | $(modpost) -s -T -
>
> PHONY += __modpost
> -__modpost: $(modules:.ko=.o) FORCE
> +__modpost: $(modules:.ko=.o) dump_modules FORCE
> $(call cmd,modpost) $(wildcard vmlinux)
>
> quiet_cmd_kernel-mod = MODPOST $@
>
I applied the above diff, and the output is exactly the same for me as well.
But the command still fails.
> > After reverting commit ea4054a238 (modpost: handle huge numbers of modules)
> > everything is fine.
> >
> > This happens with multiple gcc versions and target platforms (arm, x86_64,
> > i386, powerpc). Host kernel version is 3.8.10, system is Ubuntu 12.10.
> >
> > Does anyone else see this problem ? Any idea what might be wrong ?
>
> Assuming it happens with a clean tree, please send the .config, and I'll
> see if I can reproduce...
>
Attached is a failing configuration for x86_64. Note that
drivers/staging/android/logger.c fails to build with this configuration,
so you have to build it with make -i to see the problem.
Looking through my logs, it seems that the problem may be related to other
build errors. At least in my latest nightly builds there was always some other
build error when this happened.
Thanks,
Guenter
Guenter Roeck <[email protected]> writes:
> Attached is a failing configuration for x86_64. Note that
> drivers/staging/android/logger.c fails to build with this configuration,
> so you have to build it with make -i to see the problem.
>
> Looking through my logs, it seems that the problem may be related to other
> build errors. At least in my latest nightly builds there was always some other
> build error when this happened.
That would explain it. Perhaps some kind of race?
I can't see this as a real problem, though.
Cheers,
Rusty.
On Mon, May 13, 2013 at 11:52:09AM +0930, Rusty Russell wrote:
> Guenter Roeck <[email protected]> writes:
> > Attached is a failing configuration for x86_64. Note that
> > drivers/staging/android/logger.c fails to build with this configuration,
> > so you have to build it with make -i to see the problem.
> >
> > Looking through my logs, it seems that the problem may be related to other
> > build errors. At least in my latest nightly builds there was always some other
> > build error when this happened.
>
> That would explain it. Perhaps some kind of race?
>
> I can't see this as a real problem, though.
>
It reduces the value of "make -i" significantly, as it creates lots of false
compile errors. If you think that is not a problem, fine with me. I can (and do)
filter the resulting errors from my auto-build scripts by now . That may cause
the scripts it to miss some real errors, but after all I can only do so much.
And it does seeem as if no one else cares, or maybe I am the only one seeing
the problem, so I guess you are right.
Thanks,
Guenter