your change to prevent needless re-building of vmlinux.o

RFC PATCH In review Jan Beulich 2 participants 7 messages May 16, 2008 – May 19, 2008 Kbuild

Share & tools

Jan Beulich

author

2008-05-16 16:08 UTC #0 lore raw

Sam,

while also looking into this issue (found time only now) I realized this is already worked around in .26-rc2. However, it would seem to me that suppressing the dependency on init/built-in.o altogether is likely to cause problems in the future. Going through the version generating logic again I realized that since vmlinux.o is always built before the first of the (perhaps multiple, due to kallsyms) final linking stages, it ought to be possible to simply move the version generation into the vmlinux.o rule. And indeed, this

define rule_vmlinux-modpost

	:
	+$(call cmd,vmlinux_version)
	+$(call cmd,vmlinux-modpost)
	$(Q)$(MAKE) -f $(srctree)/scripts/Makefile.modpost $@
	$(Q)echo 'cmd_$@ := $(cmd_vmlinux-modpost)' > $(dot-target).cmd

endef

with the two other invocations of vmlinux_version removed, seems to do what is wanted.

What do you think? Am I overlooking some aspect here?

Btw., another slight anomaly I noticed is that modpost on the modules is always performed (even when nothing changed in the tree), while on vmlinux.o it is only performed when something caused a re-link. Is that intentional, or just a to-be-tolerated side effect?

Thanks, Jan

Sam Ravnborg

2008-05-17 12:52 UTC #1 lore raw

Hi Jan.

Quoted 6 lines · Jan Beulich

Sam,
while also looking into this issue (found time only now) I realized this is already worked around in .26-rc2. However, it would seem to me that suppressing the dependency on init/built-in.o altogether is likely to cause problems in the future.

What is not obvious is that the only reason we link vmlinux.o is to support the section mismatch check.
I once had a version where I used vmlinux.o as an intermediate step but that did not speed up the following link so I dropped it again. Therefore the suppressed init/built-in.o dependency will only cause us to miss a section mismatch check in very rare cases and will not cause troubles for a normal kernel link.

I have also had in mind to make the vmlinux.o link dependent on CONFIG_DEBUG_SECTION_MISMATCH so we could skip it for a kernel build where we do not do this analysis.
But the vmlinux.o link were kept on the rationale that CONFIG_DEBUG_SECTION_MISMATCH should be used to trigger an analysis for all links and in the normal situation we would do the check only on vmlinux.o and for the individual modules.

So I hope this explains why init/built-in.o is supressed today.

Quoted 12 lines

Going through the version generating
logic again I realized that since vmlinux.o is always built before the first of the (perhaps multiple, due to kallsyms) final linking stages, it ought to be possible to simply move the version generation into the vmlinux.o rule. And indeed, this
define rule_vmlinux-modpost
	:
	+$(call cmd,vmlinux_version)
	+$(call cmd,vmlinux-modpost)
	$(Q)$(MAKE) -f $(srctree)/scripts/Makefile.modpost $@
	$(Q)echo 'cmd_$@ := $(cmd_vmlinux-modpost)' > $(dot-target).cmd

Quoted 4 lines

endef
with the two other invocations of vmlinux_version removed, seems to do what is wanted.A

So what you suggest is something like this:

Makefile +2 -6

index 3140145..3dbfffc 100644
@@ -657,7 +657,6 @@ vmlinux-init := $(head-y) $(init-y)
 vmlinux-main := $(core-y) $(libs-y) $(drivers-y) $(net-y)
 vmlinux-all  := $(vmlinux-init) $(vmlinux-main)
 vmlinux-lds  := arch/$(SRCARCH)/kernel/vmlinux.lds
-export KBUILD_VMLINUX_OBJS := $(vmlinux-all)
 
 # Rule to link vmlinux - also used during CONFIG_KALLSYMS
 # May be overridden by arch/$(ARCH)/Makefile

@@ -690,8 +689,6 @@ quiet_cmd_sysmap = SYSMAP
 # First command is ':' to allow us to use + in front of the rule
 define rule_vmlinux__
 	:
-	$(if $(CONFIG_KALLSYMS),,+$(call cmd,vmlinux_version))
-
 	$(call cmd,vmlinux__)
 	$(Q)echo 'cmd_$@ := $(cmd_vmlinux__)' > $(@D)/.$(@F).cmd

@@ -747,8 +744,7 @@ endef
 # First command is ':' to allow us to use + in front of this rule
 cmd_ksym_ld = $(cmd_vmlinux__)
 define rule_ksym_ld
-	: 
-	+$(call cmd,vmlinux_version)
+	:
 	$(call cmd,vmlinux__)
 	$(Q)echo 'cmd_$@ := $(cmd_vmlinux__)' > $(@D)/.$(@F).cmd
 endef

@@ -798,6 +794,7 @@ quiet_cmd_vmlinux-modpost = LD      $@
 define rule_vmlinux-modpost
 	:
 	+$(call cmd,vmlinux-modpost)
+	+$(call cmd,vmlinux_version)
 	$(Q)$(MAKE) -f $(srctree)/scripts/Makefile.modpost $@
 	$(Q)echo 'cmd_$@ := $(cmd_vmlinux-modpost)' > $(dot-target).cmd
 endef

@@ -810,7 +807,6 @@ endif
 ifdef CONFIG_SAMPLES
 	$(Q)$(MAKE) $(build)=samples
 endif
-	$(call vmlinux-modpost)
 	$(call if_changed_rule,vmlinux__)
 	$(Q)rm -f .old_version

This is a welcome simplifications and I like it.
If we decide to keep the intermediate vmlinux.o link than we should do so.
If you agree I can try to give it a spin in -next. [There are two unrelated clean-ups in the patch - both are noops].

Everything that can simplify the final link of vmlinux is welcome.

I have several times been tempted to move it all to a shell script. What has prevented me to do so has only been:
1) that um has some subtle requirement and therefore redefine the vmlinux link
2) that last time I reworked this stuff Andrew could not boot one of his systems and I never figured out why
3) lack of time..

But if someone comes up with a patch that moves all this stuff out of the top-level Makefile I am all open for it.

Quoted 7 lines

What do you think? Am I overlooking some aspect here?
Btw., another slight anomaly I noticed is that modpost on the modules is always performed (even when nothing changed in the tree), while on vmlinux.o it is only performed when something caused a re-link. Is that intentional, or just a to-be-tolerated side effect?

It is intentional as in we do not have information available if a module has been build or not.
modpost take care not to overwrite module.mod.c files if the generated file equals the old file. But could we skip the modpost entirely it would be better. But that was not trivial last I looked at it (loong time ago).

Sam

Sam Ravnborg

2008-05-18 18:51 UTC #2 lore raw

RFC PATCH 1/1 move link of vmlinux from Makefile to a script

Quoted 11 lines

I have several times been tempted to move it all to a shell script. What has prevented me to do so has only been:
1) that um has some subtle requirement and therefore redefine the vmlinux link
2) that last time I reworked this stuff Andrew could not boot one of his systems and I never figured out why
3) lack of time..
But if someone comes up with a patch that moves all this stuff out of the top-level Makefile I am all open for it.

I gave it a quick try. This is first try and comments are welcome. It does not address the um issue - but I think that is doable with an additional enviroment variable.

The conditionals around KALLSYMS may be simplified - I just did not try to do so.

What I like with the patch below is that it is 100 times more readable than the Makefile stuff.
The drawback are that we either link or not - we cannot restart from the middle of the link. But that it not a real issue.

This patch futhermore remove a debug target to assist in kallysms issues - I have not seen it used for a long time now. And I was also considering removing the extra kallsyms pass as I dunnot how usefull it is and for an allyesconfig build it take too much time.

This patch includes your version simplifactions too allthough that is not obvious.

Sam

 Makefile                |  196 ++++-------------------------------------------
 scripts/link-vmlinux.sh |  132 +++++++++++++++++++++++++++++++
 scripts/mksysmap        |   44 -----------
 3 files changed, 148 insertions(+), 224 deletions(-)

Makefile +15 ~~-179~~

index 3140145..6589ae5 100644
@@ -628,204 +628,40 @@ libs-y		:= $(libs-y1) $(libs-y2)
 
 # Build vmlinux
 # ---------------------------------------------------------------------------
-# vmlinux is built from the objects selected by $(vmlinux-init) and
-# $(vmlinux-main). Most are built-in.o files from top-level directories
+# vmlinux is built from the objects selected by $(KBUILD_VMLINUX_INIT) and
+# $(KBUILD_VMLINUX_MAIN). Most are built-in.o files from top-level directories
 # in the kernel tree, others are specified in arch/$(ARCH)/Makefile.
-# Ordering when linking is important, and $(vmlinux-init) must be first.
+# Ordering when linking is important, and $(KBUILD_VMLINUX_INIT) must be first.
 #
 # vmlinux
 #   ^
 #   |
-#   +-< $(vmlinux-init)
+#   +-< $(KBUILD_VMLINUX_INIT)
 #   |   +--< init/version.o + more
 #   |
-#   +--< $(vmlinux-main)
-#   |    +--< driver/built-in.o mm/built-in.o + more
-#   |
-#   +-< kallsyms.o (see description in CONFIG_KALLSYMS section)
-#
-# vmlinux version (uname -v) cannot be updated during normal
-# descending-into-subdirs phase since we do not yet know if we need to
-# update vmlinux.
-# Therefore this step is delayed until just before final link of vmlinux -
-# except in the kallsyms case where it is done just before adding the
-# symbols to the kernel.
-#
-# System.map is generated to document addresses of all kernel symbols
-
-vmlinux-init := $(head-y) $(init-y)
-vmlinux-main := $(core-y) $(libs-y) $(drivers-y) $(net-y)
-vmlinux-all  := $(vmlinux-init) $(vmlinux-main)
-vmlinux-lds  := arch/$(SRCARCH)/kernel/vmlinux.lds
-export KBUILD_VMLINUX_OBJS := $(vmlinux-all)
-
-# Rule to link vmlinux - also used during CONFIG_KALLSYMS
-# May be overridden by arch/$(ARCH)/Makefile
-quiet_cmd_vmlinux__ ?= LD      $@
-      cmd_vmlinux__ ?= $(LD) $(LDFLAGS) $(LDFLAGS_vmlinux) -o $@ \
-      -T $(vmlinux-lds) $(vmlinux-init)                          \
-      --start-group $(vmlinux-main) --end-group                  \
-      $(filter-out $(vmlinux-lds) $(vmlinux-init) $(vmlinux-main) vmlinux.o FORCE ,$^)
-
-# Generate new vmlinux version
-quiet_cmd_vmlinux_version = GEN     .version
-      cmd_vmlinux_version = set -e;                     \
-	if [ ! -r .version ]; then			\
-	  rm -f .version;				\
-	  echo 1 >.version;				\
-	else						\
-	  mv .version .old_version;			\
-	  expr 0$$(cat .old_version) + 1 >.version;	\
-	fi;						\
-	$(MAKE) $(build)=init
-
-# Generate System.map
-quiet_cmd_sysmap = SYSMAP
-      cmd_sysmap = $(CONFIG_SHELL) $(srctree)/scripts/mksysmap
-
-# Link of vmlinux
-# If CONFIG_KALLSYMS is set .version is already updated
-# Generate System.map and verify that the content is consistent
-# Use + in front of the vmlinux_version rule to silent warning with make -j2
-# First command is ':' to allow us to use + in front of the rule
-define rule_vmlinux__
-	:
-	$(if $(CONFIG_KALLSYMS),,+$(call cmd,vmlinux_version))
-
-	$(call cmd,vmlinux__)
-	$(Q)echo 'cmd_$@ := $(cmd_vmlinux__)' > $(@D)/.$(@F).cmd
-
-	$(Q)$(if $($(quiet)cmd_sysmap),                                      \
-	  echo '  $($(quiet)cmd_sysmap)  System.map' &&)                     \
-	$(cmd_sysmap) $@ System.map;                                         \
-	if [ $$? -ne 0 ]; then                                               \
-		rm -f $@;                                                    \
-		/bin/false;                                                  \
-	fi;
-	$(verify_kallsyms)
-endef
-
-
-ifdef CONFIG_KALLSYMS
-# Generate section listing all symbols and add it into vmlinux $(kallsyms.o)
-# It's a three stage process:
-# o .tmp_vmlinux1 has all symbols and sections, but __kallsyms is
-#   empty
-#   Running kallsyms on that gives us .tmp_kallsyms1.o with
-#   the right size - vmlinux version (uname -v) is updated during this step
-# o .tmp_vmlinux2 now has a __kallsyms section of the right size,
-#   but due to the added section, some addresses have shifted.
-#   From here, we generate a correct .tmp_kallsyms2.o
-# o The correct .tmp_kallsyms2.o is linked into the final vmlinux.
-# o Verify that the System.map from vmlinux matches the map from
-#   .tmp_vmlinux2, just in case we did not generate kallsyms correctly.
-# o If CONFIG_KALLSYMS_EXTRA_PASS is set, do an extra pass using
-#   .tmp_vmlinux3 and .tmp_kallsyms3.o.  This is only meant as a
-#   temporary bypass to allow the kernel to be built while the
-#   maintainers work out what went wrong with kallsyms.
-
-ifdef CONFIG_KALLSYMS_EXTRA_PASS
-last_kallsyms := 3
-else
-last_kallsyms := 2
-endif
-
-kallsyms.o := .tmp_kallsyms$(last_kallsyms).o
-
-define verify_kallsyms
-	$(Q)$(if $($(quiet)cmd_sysmap),                                      \
-	  echo '  $($(quiet)cmd_sysmap)  .tmp_System.map' &&)                \
-	  $(cmd_sysmap) .tmp_vmlinux$(last_kallsyms) .tmp_System.map
-	$(Q)cmp -s System.map .tmp_System.map ||                             \
-		(echo Inconsistent kallsyms data;                            \
-		 echo Try setting CONFIG_KALLSYMS_EXTRA_PASS;                \
-		 rm .tmp_kallsyms* ; /bin/false )
-endef
-
-# Update vmlinux version before link
-# Use + in front of this rule to silent warning about make -j1
-# First command is ':' to allow us to use + in front of this rule
-cmd_ksym_ld = $(cmd_vmlinux__)
-define rule_ksym_ld
-	: 
-	+$(call cmd,vmlinux_version)
-	$(call cmd,vmlinux__)
-	$(Q)echo 'cmd_$@ := $(cmd_vmlinux__)' > $(@D)/.$(@F).cmd
-endef
-
-# Generate .S file with all kernel symbols
-quiet_cmd_kallsyms = KSYM    $@
-      cmd_kallsyms = $(NM) -n $< | $(KALLSYMS) \
-                     $(if $(CONFIG_KALLSYMS_ALL),--all-symbols) > $@
+#   +--< $(KBUILD_VMLINUX_MAIN)
+#        +--< driver/built-in.o mm/built-in.o + more
 
-.tmp_kallsyms1.o .tmp_kallsyms2.o .tmp_kallsyms3.o: %.o: %.S scripts FORCE
-	$(call if_changed_dep,as_o_S)
+export KBUILD_VMLINUX_INIT := $(head-y) $(init-y)
+export KBUILD_VMLINUX_MAIN := $(core-y) $(libs-y) $(drivers-y) $(net-y)
+export KBUILD_VMLINUX_LDS  := arch/$(SRCARCH)/kernel/vmlinux.lds
 
-.tmp_kallsyms%.S: .tmp_vmlinux% $(KALLSYMS)
-	$(call cmd,kallsyms)
-
-# .tmp_vmlinux1 must be complete except kallsyms, so update vmlinux version
-.tmp_vmlinux1: $(vmlinux-lds) $(vmlinux-all) FORCE
-	$(call if_changed_rule,ksym_ld)
-
-.tmp_vmlinux2: $(vmlinux-lds) $(vmlinux-all) .tmp_kallsyms1.o FORCE
-	$(call if_changed,vmlinux__)
-
-.tmp_vmlinux3: $(vmlinux-lds) $(vmlinux-all) .tmp_kallsyms2.o FORCE
-	$(call if_changed,vmlinux__)
-
-# Needs to visit scripts/ before $(KALLSYMS) can be used.
-$(KALLSYMS): scripts ;
-
-# Generate some data for debugging strange kallsyms problems
-debug_kallsyms: .tmp_map$(last_kallsyms)
-
-.tmp_map%: .tmp_vmlinux% FORCE
-	($(OBJDUMP) -h $< | $(AWK) '/^ +[0-9]/{print $$4 " 0 " $$2}'; $(NM) $<) | sort > $@
-
-.tmp_map3: .tmp_map2
-
-.tmp_map2: .tmp_map1
-
-endif # ifdef CONFIG_KALLSYMS
-
-# Do modpost on a prelinked vmlinux. The finally linked vmlinux has
-# relevant sections renamed as per the linker script.
-quiet_cmd_vmlinux-modpost = LD      $@
-      cmd_vmlinux-modpost = $(LD) $(LDFLAGS) -r -o $@                          \
-	 $(vmlinux-init) --start-group $(vmlinux-main) --end-group             \
-	 $(filter-out $(vmlinux-init) $(vmlinux-main) FORCE ,$^)
-define rule_vmlinux-modpost
-	:
-	+$(call cmd,vmlinux-modpost)
-	$(Q)$(MAKE) -f $(srctree)/scripts/Makefile.modpost $@
-	$(Q)echo 'cmd_$@ := $(cmd_vmlinux-modpost)' > $(dot-target).cmd
-endef
+vmlinux-deps := $(KBUILD_VMLINUX_INIT) $(KBUILD_VMLINUX_MAIN)
+vmlinux-deps += $(KBUILD_VMLINUX_LDS)
 
 # vmlinux image - including updated kernel symbols
-vmlinux: $(vmlinux-lds) $(vmlinux-init) $(vmlinux-main) vmlinux.o $(kallsyms.o) FORCE
+vmlinux: $(vmlinux-deps)
 ifdef CONFIG_HEADERS_CHECK
 	$(Q)$(MAKE) -f $(srctree)/Makefile headers_check
 endif
 ifdef CONFIG_SAMPLES
 	$(Q)$(MAKE) $(build)=samples
 endif
-	$(call vmlinux-modpost)
-	$(call if_changed_rule,vmlinux__)
-	$(Q)rm -f .old_version
-
-# build vmlinux.o first to catch section mismatch errors early
-ifdef CONFIG_KALLSYMS
-.tmp_vmlinux1: vmlinux.o
-endif
-
-modpost-init := $(filter-out init/built-in.o, $(vmlinux-init))
-vmlinux.o: $(modpost-init) $(vmlinux-main) FORCE
-	$(call if_changed_rule,vmlinux-modpost)
+	$(Q)$(CONFIG_SHELL) $(srctree)/scripts/link-vmlinux.sh
 
-# The actual objects are generated when descending, 
+# The actual objects are generated when descending,
 # make sure no implicit rule kicks in
-$(sort $(vmlinux-init) $(vmlinux-main)) $(vmlinux-lds): $(vmlinux-dirs) ;
+$(sort $(vmlinux-deps)): $(vmlinux-dirs) ;
 
 # Handle descending into subdirectories listed in $(vmlinux-dirs)
 # Preset locale variables to speed up the build process. Limit locale

scripts/link-vmlinux.sh new file +132 -0

132 lines not on this page · raw message

scripts/mksysmap deleted +0 ~~-44~~

44 lines not on this page · raw message

Jan Beulich

author

2008-05-19 07:42 UTC #3 lore raw

Re: [RFC PATCH] move link of vmlinux from Makefile to a script

Quoted 13 lines

Quoted 10 lines
Quoted 1 line
Sam Ravnborg <[email protected]> 18.05.08 20:52 >>>
I have several times been tempted to move it all to a shell script. What has prevented me to do so has only been:
1) that um has some subtle requirement and therefore redefine the vmlinux link
2) that last time I reworked this stuff Andrew could not boot one of his systems and I never figured out why
3) lack of time..
Quoted 2 lines
But if someone comes up with a patch that moves all this stuff out of the top-level Makefile I am all open for it.

Quoted 6 lines

I gave it a quick try. This is first try and comments are welcome. It does not address the um issue - but I think that is doable with an additional enviroment variable.
The conditionals around KALLSYMS may be simplified - I just did not try to do so.

What I fail to see is how the multi-stage (necessary even without CONFIG_KALLSYMS_EXTRA_PASS) kallsyms linking is now happening.

Quoted 2 lines

What I like with the patch below is that it is 100 times more readable than the Makefile stuff.

Fully agree, and hence I'd favor this for .27 over the simple version handling fix (which I'd hope to see in .26, and maybe even .25.x).

Quoted 6 lines

The drawback are that we either link or not - we cannot restart from the middle of the link. But that it not a real issue.
This patch futhermore remove a debug target to assist in kallysms issues - I have not seen it used for a long time now.

Is this what I referred to above? If so, I don't think these are just debug targets...

Quoted 3 lines

And I was also considering removing the extra kallsyms pass as I dunnot how usefull it is and for an allyesconfig build it take too much time.

I'd vote for it, but would of course think that whoever introduced it should voice his opinion.

Quoted 2 lines

This patch includes your version simplifactions too allthough that is not obvious.

Thanks, Jan

Sam Ravnborg

2008-05-19 08:19 UTC #4 lore raw

Re: [RFC PATCH] move link of vmlinux from Makefile to a script

Quoted 17 lines · Jan Beulich

Quoted 17 lines
Quoted 10 lines
Quoted 1 line
Sam Ravnborg <[email protected]> 18.05.08 20:52 >>>
I have several times been tempted to move it all to a shell script. What has prevented me to do so has only been:
1) that um has some subtle requirement and therefore redefine the vmlinux link
2) that last time I reworked this stuff Andrew could not boot one of his systems and I never figured out why
3) lack of time..
Quoted 2 lines
But if someone comes up with a patch that moves all this stuff out of the top-level Makefile I am all open for it.
I gave it a quick try. This is first try and comments are welcome. It does not address the um issue - but I think that is doable with an additional enviroment variable.

Quoted 5 lines

Quoted 2 lines
The conditionals around KALLSYMS may be simplified - I just did not try to do so.
What I fail to see is how the multi-stage (necessary even without CONFIG_KALLSYMS_EXTRA_PASS) kallsyms linking is now happening.

I will recheck my script tonight or tomorrow.
But the principle is like this:

-> Link vmlinux.o
-> modpost vmlinux.o
-> link vmlinux if not KALLSYMS

-> link .tmp_vmlinux_kall if KALLSYMS
-> link vmlinux with new kernel symbols if not KALLSYM_EXTRA_PASS

-> link .tml_vmlinux_extra_pass if KALLSYM_EXTRA_PASS -> link vmlinux with new kernel symbols
-> check that System.map are equal in the above two cases

Quoted 6 lines

Quoted 3 lines
What I like with the patch below is that it is 100 times more readable than the Makefile stuff.
Fully agree, and hence I'd favor this for .27 over the simple version handling fix (which I'd hope to see in .26, and maybe even .25.x).

Does it fix anything or is it just a clean-up? I see it as a cleanup only and thus not -rc material.

Quoted 9 lines

Quoted 6 lines
The drawback are that we either link or not - we cannot restart from the middle of the link. But that it not a real issue.
This patch futhermore remove a debug target to assist in kallysms issues - I have not seen it used for a long time now.
Is this what I referred to above? If so, I don't think these are just debug targets...

Nope - we can create maps of kernel symbols to assist when we have kernel symbols bugs (when EXTRA_PASS detect inconsistency). When I added this commit I told Keith Ownes that I planned to remove that again when we no longer saw KALLSYMS issues where we needed this to debug it.
And I have not seen such reports for a long time now.

Quoted 5 lines

Quoted 4 lines
And I was also considering removing the extra kallsyms pass as I dunnot how usefull it is and for an allyesconfig build it take too much time.

Quoted 8 lines

I'd vote for it, but would of course think that whoever introduced it should voice his opinion.
Quoted 2 lines
This patch includes your version simplifactions too allthough that is not obvious.
Thanks, Jan

I will post an updated patch later this week.

Sam

Jan Beulich

author

2008-05-19 08:35 UTC #5 lore raw

Re: [RFC PATCH] move link of vmlinux from Makefile to a script

Quoted 2 lines

Does it fix anything or is it just a clean-up? I see it as a cleanup only and thus not -rc material.

It's just an inconvenience it fixes in .25.x (unless you pushed your init.o exclusion patch to -stable already), and perhaps more a theoretical issue in .26, but anyway (and irrespective of that for consistency I'd think it should be done the same way in .25.x and .26).

Jan

Jan Beulich

author Acked-by

2008-05-19 07:35 UTC #6 lore raw

In reply to Sam Ravnborg

Quoted 1 line

So what you suggest is something like this:

Yes, exactly:

Acked-by Jan Beulich <[email protected]>

Makefile +2 -6

index 3140145..3dbfffc 100644
@@ -657,7 +657,6 @@ vmlinux-init := $(head-y) $(init-y)
 vmlinux-main := $(core-y) $(libs-y) $(drivers-y) $(net-y)
 vmlinux-all  := $(vmlinux-init) $(vmlinux-main)
 vmlinux-lds  := arch/$(SRCARCH)/kernel/vmlinux.lds
-export KBUILD_VMLINUX_OBJS := $(vmlinux-all)
 
 # Rule to link vmlinux - also used during CONFIG_KALLSYMS
 # May be overridden by arch/$(ARCH)/Makefile

@@ -690,8 +689,6 @@ quiet_cmd_sysmap = SYSMAP
 # First command is ':' to allow us to use + in front of the rule
 define rule_vmlinux__
 	:
-	$(if $(CONFIG_KALLSYMS),,+$(call cmd,vmlinux_version))
-
 	$(call cmd,vmlinux__)
 	$(Q)echo 'cmd_$@ := $(cmd_vmlinux__)' > $(@D)/.$(@F).cmd

@@ -747,8 +744,7 @@ endef
 # First command is ':' to allow us to use + in front of this rule
 cmd_ksym_ld = $(cmd_vmlinux__)
 define rule_ksym_ld
-	: 
-	+$(call cmd,vmlinux_version)
+	:
 	$(call cmd,vmlinux__)
 	$(Q)echo 'cmd_$@ := $(cmd_vmlinux__)' > $(@D)/.$(@F).cmd 
 endef

@@ -798,6 +794,7 @@ quiet_cmd_vmlinux-modpost = LD      $@
 define rule_vmlinux-modpost
 	:
 	+$(call cmd,vmlinux-modpost)
+	+$(call cmd,vmlinux_version)
 	$(Q)$(MAKE) -f $(srctree)/scripts/Makefile.modpost $@
 	$(Q)echo 'cmd_$@ := $(cmd_vmlinux-modpost)' > $(dot-target).cmd
 endef

@@ -810,7 +807,6 @@ endif
 ifdef CONFIG_SAMPLES
 	$(Q)$(MAKE) $(build)=samples
 endif
-	$(call vmlinux-modpost)
 	$(call if_changed_rule,vmlinux__)
 	$(Q)rm -f .old_version

Everything that can simplify the final link of vmlinux is welcome.

But if someone comes up with a patch that moves all this stuff out of the top-level Makefile I am all open for it.

Quoted 7 lines

What do you think? Am I overlooking some aspect here?
Btw., another slight anomaly I noticed is that modpost on the modules is always performed (even when nothing changed in the tree), while on vmlinux.o it is only performed when something caused a re-link. Is that intentional, or just a to-be-tolerated side effect?

Sam