Skip to contentSkip to conversation

objtool build improvements

PATCH Superseded Ian Rogers 4 participants 15 messages Jan 5, 2023 – Jan 26, 2023 objtool

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Ian Rogers

author
#0 lore raw

v3 cover objtool build improvements

Install libsubcmd and then get headers from there, this avoids inadvertent dependencies on things in tools/lib. Fix V=1 support. Clean up how HOSTCC is used to override CC to avoid CFLAGS being set for say gcc, and then CC being overridden to clang. Support HOSTCFLAGS as a make option.

v3. Is a rebase that removes the merged "tools lib subcmd: Add install
target" patch. In:
https://lore.kernel.org/lkml/CAKwvOd=kgXmpfbVa1wiEvwL0tX3gu+dDTGi-HEiRXSojwCLRrg@mail.gmail.com/
Nick rightly points out that:
WARNINGS := $(EXTRA_WARNINGS) -Wno-switch-default -Wno-switch-enum -Wno-packed -Wno-nested-externs
became:
WARNINGS := -Wno-switch-default -Wno-switch-enum -Wno-packed -Wno-nested-externs
losing the EXTRA_WARNINGS which v3 now adds back in. Previous
testing had added the warnings to the end rather than the
beginning, thereby causing unexpected build issues that aren't present in v3.
v2. Include required "tools lib subcmd: Add install target" that is
already in Arnaldo's tree:
https://lore.kernel.org/lkml/[email protected]/
https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/commit/?h=perf/core&id=630ae80ea1dd253609cb50cff87f3248f901aca3
When building libsubcmd.a from objtool's Makefile, clear the
subdir to avoid it being appended onto OUTPUT and breaking the
build.

Ian Rogers (3):
objtool: Install libsubcmd in build
objtool: Properly support make V=1
objtool: Alter how HOSTCC is forced

 tools/objtool/Build    |  2 --
 tools/objtool/Makefile | 66 ++++++++++++++++++++++++++++++------------
 2 files changed, 47 insertions(+), 21 deletions(-)
Signature · 1 line
2.39.0.314.g84b9a713c41-goog
author
#1 lore raw

PATCH v3 1/3 objtool: Install libsubcmd in build

Including from tools/lib can create inadvertent dependencies. Install libsubcmd in the objtool build and then include the headers from there.

---
 tools/objtool/Build    |  2 --
 tools/objtool/Makefile | 33 +++++++++++++++++++++++++--------
 2 files changed, 25 insertions(+), 10 deletions(-)
tools/objtool/Build +0 -2
index 33f2ee5a46d3..a3cdf8af6635 100644
@@ -16,8 +16,6 @@ objtool-y += libctype.o
 objtool-y += str_error_r.o
 objtool-y += librbtree.o
 
-CFLAGS += -I$(srctree)/tools/lib
-
 $(OUTPUT)libstring.o: ../lib/string.c FORCE
 	$(call rule_mkdir)
 	$(call if_changed_dep,cc_o_c)
tools/objtool/Makefile +25 -8
index a3a9cc24e0e3..fd9b3e3113c6 100644
@@ -12,9 +12,15 @@ srctree := $(patsubst %/,%,$(dir $(CURDIR)))
 srctree := $(patsubst %/,%,$(dir $(srctree)))
 endif
 
-SUBCMD_SRCDIR		= $(srctree)/tools/lib/subcmd/
-LIBSUBCMD_OUTPUT	= $(or $(OUTPUT),$(CURDIR)/)
-LIBSUBCMD		= $(LIBSUBCMD_OUTPUT)libsubcmd.a
+LIBSUBCMD_DIR = $(srctree)/tools/lib/subcmd/
+ifneq ($(OUTPUT),)
+  LIBSUBCMD_OUTPUT = $(abspath $(OUTPUT))/libsubcmd
+else
+  LIBSUBCMD_OUTPUT = $(CURDIR)/libsubcmd
+endif
+LIBSUBCMD_DESTDIR = $(LIBSUBCMD_OUTPUT)
+LIBSUBCMD = $(LIBSUBCMD_OUTPUT)/libsubcmd.a
+CFLAGS += -I$(LIBSUBCMD_OUTPUT)/include
 
 OBJTOOL    := $(OUTPUT)objtool
 OBJTOOL_IN := $(OBJTOOL)-in.o
@@ -28,7 +34,8 @@ INCLUDES := -I$(srctree)/tools/include \
 	    -I$(srctree)/tools/arch/$(HOSTARCH)/include/uapi \
 	    -I$(srctree)/tools/arch/$(SRCARCH)/include	\
 	    -I$(srctree)/tools/objtool/include \
-	    -I$(srctree)/tools/objtool/arch/$(SRCARCH)/include
+	    -I$(srctree)/tools/objtool/arch/$(SRCARCH)/include \
+	    -I$(LIBSUBCMD_OUTPUT)/include
 WARNINGS := $(EXTRA_WARNINGS) -Wno-switch-default -Wno-switch-enum -Wno-packed -Wno-nested-externs
 CFLAGS   := -Werror $(WARNINGS) $(KBUILD_HOSTCFLAGS) -g $(INCLUDES) $(LIBELF_FLAGS)
 LDFLAGS  += $(LIBELF_LIBS) $(LIBSUBCMD) $(KBUILD_HOSTLDFLAGS)
@@ -38,6 +45,7 @@ elfshdr := $(shell echo '$(pound)include <libelf.h>' | $(CC) $(CFLAGS) -x c -E -
 CFLAGS += $(if $(elfshdr),,-DLIBELF_USE_DEPRECATED)
 
 AWK = awk
+MKDIR = mkdir
 
 BUILD_ORC := n
 
@@ -57,13 +65,22 @@ $(OBJTOOL): $(LIBSUBCMD) $(OBJTOOL_IN)
 	$(QUIET_LINK)$(CC) $(OBJTOOL_IN) $(LDFLAGS) -o $@
 
 
-$(LIBSUBCMD): fixdep FORCE
-	$(Q)$(MAKE) -C $(SUBCMD_SRCDIR) OUTPUT=$(LIBSUBCMD_OUTPUT)
+$(LIBSUBCMD_OUTPUT):
+	@$(MKDIR) -p $@
+
+$(LIBSUBCMD): fixdep FORCE $(LIBSUBCMD_OUTPUT)
+	@$(MAKE) -C $(LIBSUBCMD_DIR) O=$(LIBSUBCMD_OUTPUT) \
+		DESTDIR=$(LIBSUBCMD_DESTDIR) prefix= subdir= \
+		$@ install_headers
+
+$(LIBSUBCMD)-clean:
+	$(call QUIET_CLEAN, libsubcmd)
+	$(Q)$(RM) -r -- $(LIBSUBCMD_OUTPUT)
 
-clean:
+clean: $(LIBSUBCMD)-clean
 	$(call QUIET_CLEAN, objtool) $(RM) $(OBJTOOL)
 	$(Q)find $(OUTPUT) -name '*.o' -delete -o -name '\.*.cmd' -delete -o -name '\.*.d' -delete
-	$(Q)$(RM) $(OUTPUT)arch/x86/lib/inat-tables.c $(OUTPUT)fixdep $(LIBSUBCMD)
+	$(Q)$(RM) $(OUTPUT)arch/x86/lib/inat-tables.c $(OUTPUT)fixdep
 
 FORCE:
 
Signature · 1 line
2.39.0.314.g84b9a713c41-goog
Reviewed-by on v3 1/3
#2 lore raw

Re: [PATCH v3 1/3] objtool: Install libsubcmd in build

Quoted 12 lines · On Thu, Jan 05, 2023 at 01:01:53AM -0800 Ian Rogers wrote:

Including from tools/lib can create inadvertent dependencies. Install libsubcmd in the objtool build and then include the headers from there.

---
 tools/objtool/Build    |  2 --
 tools/objtool/Makefile | 33 +++++++++++++++++++++++++--------
 2 files changed, 25 insertions(+), 10 deletions(-)

[...]

Quoted 8 lines
tools/objtool/Makefile +0 -0
index a3a9cc24e0e3..fd9b3e3113c6 100644
@@ -12,9 +12,15 @@ srctree := $(patsubst %/,%,$(dir $(CURDIR)))
 srctree := $(patsubst %/,%,$(dir $(srctree)))
 endif
 

tools/objtool/Makefile srctree := $(patsubst %/,%,$(dir $(CURDIR))) · line 21

Quoted 10 lines
tools/objtool/Makefile +7 -3
-SUBCMD_SRCDIR		= $(srctree)/tools/lib/subcmd/
-LIBSUBCMD_OUTPUT	= $(or $(OUTPUT),$(CURDIR)/)
-LIBSUBCMD		= $(LIBSUBCMD_OUTPUT)libsubcmd.a
+LIBSUBCMD_DIR = $(srctree)/tools/lib/subcmd/
+ifneq ($(OUTPUT),)
+  LIBSUBCMD_OUTPUT = $(abspath $(OUTPUT))/libsubcmd
+else
+  LIBSUBCMD_OUTPUT = $(CURDIR)/libsubcmd
+endif
+LIBSUBCMD_DESTDIR = $(LIBSUBCMD_OUTPUT)

Hi Ian,

Is there a reason for distinguishing between $(LIBSUBCMD_DESTDIR) and $(LIBSUBCMD_OUTPUT)?

author
#3 lore raw

Re: [PATCH v3 1/3] objtool: Install libsubcmd in build

Quoted 36 lines · Nicolas Schier
Quoted 12 lines · On Thu, Jan 05, 2023 at 01:01:53AM -0800 Ian Rogers wrote:

Including from tools/lib can create inadvertent dependencies. Install libsubcmd in the objtool build and then include the headers from there.

---
 tools/objtool/Build    |  2 --
 tools/objtool/Makefile | 33 +++++++++++++++++++++++++--------
 2 files changed, 25 insertions(+), 10 deletions(-)

[...]

Quoted 8 lines
tools/objtool/Makefile +0 -0
index a3a9cc24e0e3..fd9b3e3113c6 100644
@@ -12,9 +12,15 @@ srctree := $(patsubst %/,%,$(dir $(CURDIR)))
 srctree := $(patsubst %/,%,$(dir $(srctree)))
 endif
 

tools/objtool/Makefile srctree := $(patsubst %/,%,$(dir $(CURDIR))) · line 21

Quoted 10 lines
tools/objtool/Makefile +7 -3
-SUBCMD_SRCDIR                = $(srctree)/tools/lib/subcmd/
-LIBSUBCMD_OUTPUT     = $(or $(OUTPUT),$(CURDIR)/)
-LIBSUBCMD            = $(LIBSUBCMD_OUTPUT)libsubcmd.a
+LIBSUBCMD_DIR = $(srctree)/tools/lib/subcmd/
+ifneq ($(OUTPUT),)
+  LIBSUBCMD_OUTPUT = $(abspath $(OUTPUT))/libsubcmd
+else
+  LIBSUBCMD_OUTPUT = $(CURDIR)/libsubcmd
+endif
+LIBSUBCMD_DESTDIR = $(LIBSUBCMD_OUTPUT)

Hi Ian,

Quoted 2 lines

Is there a reason for distinguishing between $(LIBSUBCMD_DESTDIR) and $(LIBSUBCMD_OUTPUT)?

I believe the naming aligns with their use when passed to the corresponding sub-make variable, so "DESTDIR=$(LIBSUBCMD_DESTDIR)". It is possible to do other this, a slightly different convention is in some BPF related code:
https://lore.kernel.org/lkml/[email protected]/

Thanks,
Ian

Quoted 1 line
maintainer
#4 lore raw

Re: [PATCH v3 1/3] objtool: Install libsubcmd in build

In reply to Ian Rogers
Quoted 7 lines · Ian Rogers

Including from tools/lib can create inadvertent dependencies. Install libsubcmd in the objtool build and then include the headers from there.

After a make, "git status" shows:

Untracked files:
(use "git add <file>..." to include in what will be committed)
tools/objtool/libsubcmd/

So tools/objtool/.gitignore needs an update.

Quoted 6 lines
tools/objtool/Makefile +0 -0
@@ -12,9 +12,15 @@ srctree := $(patsubst %/,%,$(dir $(CURDIR)))
 srctree := $(patsubst %/,%,$(dir $(srctree)))
 endif
 

tools/objtool/Makefile srctree := $(patsubst %/,%,$(dir $(CURDIR))) · line 21

Quoted 10 lines
tools/objtool/Makefile +7 -3
-SUBCMD_SRCDIR		= $(srctree)/tools/lib/subcmd/
-LIBSUBCMD_OUTPUT	= $(or $(OUTPUT),$(CURDIR)/)
-LIBSUBCMD		= $(LIBSUBCMD_OUTPUT)libsubcmd.a
+LIBSUBCMD_DIR = $(srctree)/tools/lib/subcmd/
+ifneq ($(OUTPUT),)
+  LIBSUBCMD_OUTPUT = $(abspath $(OUTPUT))/libsubcmd
+else
+  LIBSUBCMD_OUTPUT = $(CURDIR)/libsubcmd
+endif
+LIBSUBCMD_DESTDIR = $(LIBSUBCMD_OUTPUT)

Similar to Nicolas' comment, it's confusing to have two variables with the same value. Please s/LIBSUBCMD_DESTDIR/LIBSUBCMD_OUTPUT/

tools/objtool/Makefile srctree := $(patsubst %/,%,$(dir $(CURDIR))) · line 23

Quoted 2 lines
tools/objtool/Makefile +2 -0
+LIBSUBCMD = $(LIBSUBCMD_OUTPUT)/libsubcmd.a
+CFLAGS += -I$(LIBSUBCMD_OUTPUT)/include

As far as I can tell, this CFLAGS addition is both ineffective (it's overwritten later) and unnecessary (it's made redundant by the same addition to the INCLUDES variable).

Signature · 1 line
Josh
author
#5 lore raw

Re: [PATCH v3 1/3] objtool: Install libsubcmd in build

Quoted 10 lines · Josh Poimboeuf
Quoted 7 lines · Ian Rogers

Including from tools/lib can create inadvertent dependencies. Install libsubcmd in the objtool build and then include the headers from there.

Quoted 7 lines

After a make, "git status" shows:

Untracked files:
(use "git add <file>..." to include in what will be committed)
tools/objtool/libsubcmd/

So tools/objtool/.gitignore needs an update.

Done.

Quoted 6 lines
Quoted 6 lines
tools/objtool/Makefile +0 -0
@@ -12,9 +12,15 @@ srctree := $(patsubst %/,%,$(dir $(CURDIR)))
 srctree := $(patsubst %/,%,$(dir $(srctree)))
 endif
 
Quoted 13 lines

tools/objtool/Makefile srctree := $(patsubst %/,%,$(dir $(CURDIR))) · line 21

Quoted 10 lines
tools/objtool/Makefile +7 -3
-SUBCMD_SRCDIR                = $(srctree)/tools/lib/subcmd/
-LIBSUBCMD_OUTPUT     = $(or $(OUTPUT),$(CURDIR)/)
-LIBSUBCMD            = $(LIBSUBCMD_OUTPUT)libsubcmd.a
+LIBSUBCMD_DIR = $(srctree)/tools/lib/subcmd/
+ifneq ($(OUTPUT),)
+  LIBSUBCMD_OUTPUT = $(abspath $(OUTPUT))/libsubcmd
+else
+  LIBSUBCMD_OUTPUT = $(CURDIR)/libsubcmd
+endif
+LIBSUBCMD_DESTDIR = $(LIBSUBCMD_OUTPUT)

Similar to Nicolas' comment, it's confusing to have two variables with the same value. Please s/LIBSUBCMD_DESTDIR/LIBSUBCMD_OUTPUT/

Done. I feel this is more of a meta question for the standard in the current build. DESTDIR is related to installation, in this case header files, OUTPUT is a build directory. Anyway, LIBSUBCMD_DESTDIR is gone in v4.

Quoted 6 lines

tools/objtool/Makefile srctree := $(patsubst %/,%,$(dir $(CURDIR))) · line 23

Quoted 2 lines
tools/objtool/Makefile +2 -0
+LIBSUBCMD = $(LIBSUBCMD_OUTPUT)/libsubcmd.a
+CFLAGS += -I$(LIBSUBCMD_OUTPUT)/include

As far as I can tell, this CFLAGS addition is both ineffective (it's overwritten later) and unnecessary (it's made redundant by the same addition to the INCLUDES variable).

Agreed, removed in v4.

Quoted 2 lines

--
Josh

author
#6 lore raw

PATCH v3 2/3 objtool: Properly support make V=1

In reply to Ian Rogers

The Q variable was being used but never correctly set up. Add the setting up and use in place of @.

---
 tools/objtool/Makefile | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)
tools/objtool/Makefile +10 -4
index fd9b3e3113c6..61a00b7acae9 100644
@@ -47,6 +47,12 @@ CFLAGS += $(if $(elfshdr),,-DLIBELF_USE_DEPRECATED)
 AWK = awk
 MKDIR = mkdir
 
+ifeq ($(V),1)
+  Q =
+else
+  Q = @
+endif
+
 BUILD_ORC := n
 
 ifeq ($(SRCARCH),x86)
@@ -58,18 +64,18 @@ export srctree OUTPUT CFLAGS SRCARCH AWK
 include $(srctree)/tools/build/Makefile.include
 
 $(OBJTOOL_IN): fixdep FORCE
-	@$(CONFIG_SHELL) ./sync-check.sh
-	@$(MAKE) $(build)=objtool
+	$(Q)$(CONFIG_SHELL) ./sync-check.sh
+	$(Q)$(MAKE) $(build)=objtool
 
 $(OBJTOOL): $(LIBSUBCMD) $(OBJTOOL_IN)
 	$(QUIET_LINK)$(CC) $(OBJTOOL_IN) $(LDFLAGS) -o $@
 
 
 $(LIBSUBCMD_OUTPUT):
-	@$(MKDIR) -p $@
+	$(Q)$(MKDIR) -p $@
 
 $(LIBSUBCMD): fixdep FORCE $(LIBSUBCMD_OUTPUT)
-	@$(MAKE) -C $(LIBSUBCMD_DIR) O=$(LIBSUBCMD_OUTPUT) \
+	$(Q)$(MAKE) -C $(LIBSUBCMD_DIR) O=$(LIBSUBCMD_OUTPUT) \
 		DESTDIR=$(LIBSUBCMD_DESTDIR) prefix= subdir= \
 		$@ install_headers
 
Signature · 1 line
2.39.0.314.g84b9a713c41-goog
Reviewed-by on v3 2/3
#7 lore raw

Re: [PATCH v3 2/3] objtool: Properly support make V=1

Quoted 10 lines · On Thu, Jan 05, 2023 at 01:01:54AM -0800 Ian Rogers wrote:

The Q variable was being used but never correctly set up. Add the setting up and use in place of @.

---
 tools/objtool/Makefile | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)
author
#8 lore raw

PATCH v3 3/3 objtool: Alter how HOSTCC is forced

In reply to Ian Rogers

HOSTCC is always wanted when building objtool. Setting CC to HOSTCC happens after tools/scripts/Makefile.include is included, meaning flags are set assuming say CC is gcc, but then it can be later set to HOSTCC which may be clang. tools/scripts/Makefile.include is needed for host set up and common macros in objtool's Makefile. Rather than override CC to HOSTCC, just pass CC as HOSTCC to Makefile.build, the libsubcmd builds and the linkage step. This means the Makefiles don't see things like CC changing and tool flag determination, and similar, work properly. To avoid mixing CFLAGS from different compilers just the objtool CFLAGS are determined with the exception of EXTRA_WARNINGS. HOSTCFLAGS is added to these so that command line flags can add to the CFLAGS.

---
 tools/objtool/Makefile | 25 +++++++++++++++----------
 1 file changed, 15 insertions(+), 10 deletions(-)
tools/objtool/Makefile +15 -10
index 61a00b7acae9..49956f4f58b9 100644
@@ -2,16 +2,12 @@
 include ../scripts/Makefile.include
 include ../scripts/Makefile.arch
 
-# always use the host compiler
-AR	 = $(HOSTAR)
-CC	 = $(HOSTCC)
-LD	 = $(HOSTLD)
-
 ifeq ($(srctree),)
 srctree := $(patsubst %/,%,$(dir $(CURDIR)))
 srctree := $(patsubst %/,%,$(dir $(srctree)))
 endif
 
+MAKE = make -S
 LIBSUBCMD_DIR = $(srctree)/tools/lib/subcmd/
 ifneq ($(OUTPUT),)
   LIBSUBCMD_OUTPUT = $(abspath $(OUTPUT))/libsubcmd
@@ -37,12 +33,19 @@ INCLUDES := -I$(srctree)/tools/include \
 	    -I$(srctree)/tools/objtool/arch/$(SRCARCH)/include \
 	    -I$(LIBSUBCMD_OUTPUT)/include
 WARNINGS := $(EXTRA_WARNINGS) -Wno-switch-default -Wno-switch-enum -Wno-packed -Wno-nested-externs
-CFLAGS   := -Werror $(WARNINGS) $(KBUILD_HOSTCFLAGS) -g $(INCLUDES) $(LIBELF_FLAGS)
-LDFLAGS  += $(LIBELF_LIBS) $(LIBSUBCMD) $(KBUILD_HOSTLDFLAGS)
+OBJTOOL_CFLAGS := -Werror $(WARNINGS) -g $(INCLUDES) $(LIBELF_FLAGS) $(HOSTCFLAGS)
+OBJTOOL_LDFLAGS := $(LIBELF_LIBS) $(LIBSUBCMD)
 
 # Allow old libelf to be used:
 elfshdr := $(shell echo '$(pound)include <libelf.h>' | $(CC) $(CFLAGS) -x c -E - | grep elf_getshdr)
-CFLAGS += $(if $(elfshdr),,-DLIBELF_USE_DEPRECATED)
+OBJTOOL_CFLAGS += $(if $(elfshdr),,-DLIBELF_USE_DEPRECATED)
+
+# Always want host compilation.
+HOST_OVERRIDES := CC="$(HOSTCC)" EXTRA_CFLAGS="$(OBJTOOL_CFLAGS)" \
+		  LD="$(HOSTLD)" AR="$(HOSTAR)"
+BUILD_HOST_OVERRIDES := CC="$(HOSTCC)" CFLAGS="$(OBJTOOL_CFLAGS)" \
+			LD="$(HOSTLD)" LDFLAGS="$(OBJTOOL_LDFLAGS)" \
+			AR="$(HOSTAR)"
 
 AWK = awk
 MKDIR = mkdir
@@ -65,10 +68,11 @@ include $(srctree)/tools/build/Makefile.include
 
 $(OBJTOOL_IN): fixdep FORCE
 	$(Q)$(CONFIG_SHELL) ./sync-check.sh
-	$(Q)$(MAKE) $(build)=objtool
+	$(Q)$(MAKE) $(build)=objtool $(BUILD_HOST_OVERRIDES)
+
 
 $(OBJTOOL): $(LIBSUBCMD) $(OBJTOOL_IN)
-	$(QUIET_LINK)$(CC) $(OBJTOOL_IN) $(LDFLAGS) -o $@
+	$(QUIET_LINK)$(HOSTCC) $(OBJTOOL_IN) $(KBUILD_HOSTLDFLAGS) $(OBJTOOL_LDFLAGS) -o $@
 
 
 $(LIBSUBCMD_OUTPUT):
@@ -77,6 +81,7 @@ $(LIBSUBCMD_OUTPUT):
 $(LIBSUBCMD): fixdep FORCE $(LIBSUBCMD_OUTPUT)
 	$(Q)$(MAKE) -C $(LIBSUBCMD_DIR) O=$(LIBSUBCMD_OUTPUT) \
 		DESTDIR=$(LIBSUBCMD_DESTDIR) prefix= subdir= \
+		$(HOST_OVERRIDES) \
 		$@ install_headers
 
 $(LIBSUBCMD)-clean:
Signature · 1 line
2.39.0.314.g84b9a713c41-goog
Reviewed-by Tested-by on v3 3/3
#9 lore raw

Re: [PATCH v3 3/3] objtool: Alter how HOSTCC is forced

Quoted 14 lines · Ian Rogers

HOSTCC is always wanted when building objtool. Setting CC to HOSTCC happens after tools/scripts/Makefile.include is included, meaning flags are set assuming say CC is gcc, but then it can be later set to HOSTCC which may be clang. tools/scripts/Makefile.include is needed for host set up and common macros in objtool's Makefile. Rather than override CC to HOSTCC, just pass CC as HOSTCC to Makefile.build, the libsubcmd builds and the linkage step. This means the Makefiles don't see things like CC changing and tool flag determination, and similar, work properly. To avoid mixing CFLAGS from different compilers just the objtool CFLAGS are determined with the exception of EXTRA_WARNINGS. HOSTCFLAGS is added to these so that command line flags can add to the CFLAGS.

Quoted 1 line

Thanks Ian, and happy new year! Sorry for delays reviewing this; I'm just getting back online today. Assuming Peter and Josh may be too, otherwise can Peter or Josh PTAL?

(I'm happy that this allows me to now pass HOSTCFLAGS on to objtool successfully)

Quoted 74 lines
---
 tools/objtool/Makefile | 25 +++++++++++++++----------
 1 file changed, 15 insertions(+), 10 deletions(-)
tools/objtool/Makefile +15 -11
index 61a00b7acae9..49956f4f58b9 100644
@@ -2,16 +2,12 @@
 include ../scripts/Makefile.include
 include ../scripts/Makefile.arch
 
-# always use the host compiler
-AR      = $(HOSTAR)
-CC      = $(HOSTCC)
-LD      = $(HOSTLD)
-
 ifeq ($(srctree),)
 srctree := $(patsubst %/,%,$(dir $(CURDIR)))
 srctree := $(patsubst %/,%,$(dir $(srctree)))
 endif
 
+MAKE = make -S
 LIBSUBCMD_DIR = $(srctree)/tools/lib/subcmd/
 ifneq ($(OUTPUT),)
   LIBSUBCMD_OUTPUT = $(abspath $(OUTPUT))/libsubcmd
@@ -37,12 +33,19 @@ INCLUDES := -I$(srctree)/tools/include \
            -I$(srctree)/tools/objtool/arch/$(SRCARCH)/include \
            -I$(LIBSUBCMD_OUTPUT)/include
 WARNINGS := $(EXTRA_WARNINGS) -Wno-switch-default -Wno-switch-enum -Wno-packed -Wno-nested-externs
-CFLAGS   := -Werror $(WARNINGS) $(KBUILD_HOSTCFLAGS) -g $(INCLUDES) $(LIBELF_FLAGS)
-LDFLAGS  += $(LIBELF_LIBS) $(LIBSUBCMD) $(KBUILD_HOSTLDFLAGS)
+OBJTOOL_CFLAGS := -Werror $(WARNINGS) -g $(INCLUDES) $(LIBELF_FLAGS) $(HOSTCFLAGS)
+OBJTOOL_LDFLAGS := $(LIBELF_LIBS) $(LIBSUBCMD)
 
 # Allow old libelf to be used:
 elfshdr := $(shell echo '$(pound)include <libelf.h>' | $(CC) $(CFLAGS) -x c -E - | grep elf_getshdr)
-CFLAGS += $(if $(elfshdr),,-DLIBELF_USE_DEPRECATED)
+OBJTOOL_CFLAGS += $(if $(elfshdr),,-DLIBELF_USE_DEPRECATED)
+
+# Always want host compilation.
+HOST_OVERRIDES := CC="$(HOSTCC)" EXTRA_CFLAGS="$(OBJTOOL_CFLAGS)" \
+                 LD="$(HOSTLD)" AR="$(HOSTAR)"
+BUILD_HOST_OVERRIDES := CC="$(HOSTCC)" CFLAGS="$(OBJTOOL_CFLAGS)" \
+                       LD="$(HOSTLD)" LDFLAGS="$(OBJTOOL_LDFLAGS)" \
+                       AR="$(HOSTAR)"
 
 AWK = awk
 MKDIR = mkdir
@@ -65,10 +68,11 @@ include $(srctree)/tools/build/Makefile.include
 
 $(OBJTOOL_IN): fixdep FORCE
        $(Q)$(CONFIG_SHELL) ./sync-check.sh
-       $(Q)$(MAKE) $(build)=objtool
+       $(Q)$(MAKE) $(build)=objtool $(BUILD_HOST_OVERRIDES)
+
 
 $(OBJTOOL): $(LIBSUBCMD) $(OBJTOOL_IN)
-       $(QUIET_LINK)$(CC) $(OBJTOOL_IN) $(LDFLAGS) -o $@
+       $(QUIET_LINK)$(HOSTCC) $(OBJTOOL_IN) $(KBUILD_HOSTLDFLAGS) $(OBJTOOL_LDFLAGS) -o $@
 
 
 $(LIBSUBCMD_OUTPUT):
@@ -77,6 +81,7 @@ $(LIBSUBCMD_OUTPUT):
 $(LIBSUBCMD): fixdep FORCE $(LIBSUBCMD_OUTPUT)
        $(Q)$(MAKE) -C $(LIBSUBCMD_DIR) O=$(LIBSUBCMD_OUTPUT) \
                DESTDIR=$(LIBSUBCMD_DESTDIR) prefix= subdir= \
+               $(HOST_OVERRIDES) \
                $@ install_headers
 
 $(LIBSUBCMD)-clean:
--

2.39.0.314.g84b9a713c41-goog

Signature · 2 lines
Thanks,
~Nick Desaulniers
Reviewed-by on v3 3/3
#10 lore raw

Re: [PATCH v3 3/3] objtool: Alter how HOSTCC is forced

In reply to Ian Rogers
Quoted 13 lines · On Thu, Jan 05, 2023 at 01:01:55AM -0800 Ian Rogers wrote:

HOSTCC is always wanted when building objtool. Setting CC to HOSTCC happens after tools/scripts/Makefile.include is included, meaning flags are set assuming say CC is gcc, but then it can be later set to HOSTCC which may be clang. tools/scripts/Makefile.include is needed for host set up and common macros in objtool's Makefile. Rather than override CC to HOSTCC, just pass CC as HOSTCC to Makefile.build, the libsubcmd builds and the linkage step. This means the Makefiles don't see things like CC changing and tool flag determination, and similar, work properly. To avoid mixing CFLAGS from different compilers just the objtool CFLAGS are determined with the exception of EXTRA_WARNINGS. HOSTCFLAGS is added to these so that command line flags can add to the CFLAGS.

Quoted 5 lines
---
 tools/objtool/Makefile | 25 +++++++++++++++----------
 1 file changed, 15 insertions(+), 10 deletions(-)
maintainer
#11 lore raw

Re: [PATCH v3 3/3] objtool: Alter how HOSTCC is forced

In reply to Ian Rogers
Quoted 9 lines · Ian Rogers

HOSTCC is always wanted when building objtool. Setting CC to HOSTCC happens after tools/scripts/Makefile.include is included, meaning flags are set assuming say CC is gcc, but then it can be later set to HOSTCC which may be clang. tools/scripts/Makefile.include is needed for host set up and common macros in objtool's Makefile. Rather than override CC to HOSTCC, just pass CC as HOSTCC to Makefile.build, the libsubcmd builds and the linkage step. This means the Makefiles don't see things like CC changing and tool flag determination, and similar, work properly.

I'm having trouble parsing this last sentence, can you rephrase or restructure it?

Quoted 3 lines

To avoid mixing CFLAGS from different compilers just the objtool CFLAGS are determined with the exception of EXTRA_WARNINGS.

I'm not really sure what this one means either.

Quoted 2 lines

HOSTCFLAGS is added to these so that command line
flags can add to the CFLAGS.

Overall this patch description is a big wall of dense text which I found hard to decipher. Please split it up into paragraphs and make it more legible and logical.

For example, un-jumble the ordering, with the background first, then the problem(s), then the fix(es). (At least three paragraphs)

If possible, the subject should describe the end result for the user, something like

objtool: Fix HOSTCFLAGS cmdline support

... unless I'm misunderstanding the point of the patch.

HOSTCFLAGS from the cmdline *used* to work, a git bisect shows it stopped working with

96f14fe738b6 ("kbuild: Rename HOSTCFLAGS to KBUILD_HOSTCFLAGS")

... so please add a "Fixes" tag for that commit.

Quoted 1 line

+MAKE = make -S

Why?

tools/objtool/Makefile INCLUDES := -I$(srctree)/tools/include \ · line 36

Quoted 10 lines

LIBSUBCMD_DIR = $(srctree)/tools/lib/subcmd/ ifneq ($(OUTPUT),)
LIBSUBCMD_OUTPUT = $(abspath $(OUTPUT))/libsubcmd

tools/objtool/Makefile +1 -2
@@ -37,12 +33,19 @@ INCLUDES := -I$(srctree)/tools/include \
 	    -I$(srctree)/tools/objtool/arch/$(SRCARCH)/include \
 	    -I$(LIBSUBCMD_OUTPUT)/include
 WARNINGS := $(EXTRA_WARNINGS) -Wno-switch-default -Wno-switch-enum -Wno-packed -Wno-nested-externs
-CFLAGS   := -Werror $(WARNINGS) $(KBUILD_HOSTCFLAGS) -g $(INCLUDES) $(LIBELF_FLAGS)
-LDFLAGS  += $(LIBELF_LIBS) $(LIBSUBCMD) $(KBUILD_HOSTLDFLAGS)
+OBJTOOL_CFLAGS := -Werror $(WARNINGS) -g $(INCLUDES) $(LIBELF_FLAGS) $(HOSTCFLAGS)

I *think* I understand the reason for the switch from KBUILD_HOSTCFLAGS to HOSTCFLAGS -- because KBUILD_HOSTCFLAGS is defined in the top-level kernel Makefile which isn't included here so it's undefined? -- but regardless that should be called out more explicitly as a problem being fixed in the commit log.

This issue really has me scratching my head, as I could have sworn objtool was being built with -O2.

tools/objtool/Makefile INCLUDES := -I$(srctree)/tools/include \ · line 37

Quoted 1 line
tools/objtool/Makefile +1 -0
+OBJTOOL_LDFLAGS := $(LIBELF_LIBS) $(LIBSUBCMD)

Is there a reason to not include $(HOSTLDFLAGS) here?

tools/objtool/Makefile INCLUDES := -I$(srctree)/tools/include \ · line 48

Quoted 11 lines
tools/objtool/Makefile +8 -1
 # Allow old libelf to be used:
 elfshdr := $(shell echo '$(pound)include <libelf.h>' | $(CC) $(CFLAGS) -x c -E - | grep elf_getshdr)
-CFLAGS += $(if $(elfshdr),,-DLIBELF_USE_DEPRECATED)
+OBJTOOL_CFLAGS += $(if $(elfshdr),,-DLIBELF_USE_DEPRECATED)
+
+# Always want host compilation.
+HOST_OVERRIDES := CC="$(HOSTCC)" EXTRA_CFLAGS="$(OBJTOOL_CFLAGS)" \
+		  LD="$(HOSTLD)" AR="$(HOSTAR)"
+BUILD_HOST_OVERRIDES := CC="$(HOSTCC)" CFLAGS="$(OBJTOOL_CFLAGS)" \
+			LD="$(HOSTLD)" LDFLAGS="$(OBJTOOL_LDFLAGS)" \
+			AR="$(HOSTAR)"

Maybe it depends on your perspective, but I'm thinking that some of these aren't really overrides, but rather normal expected flags. And the distinction between these two variables is muddy: it's not only Build vs non-Build, but also objtool vs libsubcmd.

How about just

  HOST_OVERRIDES := CC="$(HOSTCC) "LD="$(HOSTLD)" AR="$(HOSTAR)"

And then specifying the {EXTRA_}CFLAGS and LDFLAGS below where needed.

tools/objtool/Makefile include $(srctree)/tools/build/Makefile.include · line 75

Quoted 13 lines
tools/objtool/Makefile +3 -2
 AWK = awk
 MKDIR = mkdir
@@ -65,10 +68,11 @@ include $(srctree)/tools/build/Makefile.include
 
 $(OBJTOOL_IN): fixdep FORCE
 	$(Q)$(CONFIG_SHELL) ./sync-check.sh
-	$(Q)$(MAKE) $(build)=objtool
+	$(Q)$(MAKE) $(build)=objtool $(BUILD_HOST_OVERRIDES)
+
 
 $(OBJTOOL): $(LIBSUBCMD) $(OBJTOOL_IN)
-	$(QUIET_LINK)$(CC) $(OBJTOOL_IN) $(LDFLAGS) -o $@
+	$(QUIET_LINK)$(HOSTCC) $(OBJTOOL_IN) $(KBUILD_HOSTLDFLAGS) $(OBJTOOL_LDFLAGS) -o $@

Does KBUILD_HOSTLDFLAGS even work here?

Signature · 1 line
Josh
author
#12 lore raw

Re: [PATCH v3 3/3] objtool: Alter how HOSTCC is forced

Quoted 12 lines · Josh Poimboeuf
Quoted 9 lines · Ian Rogers

HOSTCC is always wanted when building objtool. Setting CC to HOSTCC happens after tools/scripts/Makefile.include is included, meaning flags are set assuming say CC is gcc, but then it can be later set to HOSTCC which may be clang. tools/scripts/Makefile.include is needed for host set up and common macros in objtool's Makefile. Rather than override CC to HOSTCC, just pass CC as HOSTCC to Makefile.build, the libsubcmd builds and the linkage step. This means the Makefiles don't see things like CC changing and tool flag determination, and similar, work properly.

Quoted 2 lines

I'm having trouble parsing this last sentence, can you rephrase or restructure it?

It was restating what was happening, deleted.

Quoted 5 lines
Quoted 3 lines

To avoid mixing CFLAGS from different compilers just the objtool CFLAGS are determined with the exception of EXTRA_WARNINGS.

I'm not really sure what this one means either.

Moved to an inline comment.

Quoted 6 lines
Quoted 2 lines

HOSTCFLAGS is added to these so that command line
flags can add to the CFLAGS.

Overall this patch description is a big wall of dense text which I found hard to decipher. Please split it up into paragraphs and make it more legible and logical.

Yes, I've deleted most of the text now. I'd been adding to it as v1 went to v2 and so on.

Quoted 9 lines

For example, un-jumble the ordering, with the background first, then the problem(s), then the fix(es). (At least three paragraphs)

If possible, the subject should describe the end result for the user, something like

objtool: Fix HOSTCFLAGS cmdline support

... unless I'm misunderstanding the point of the patch.

The patch is trying to fix the Makefile. Setting "CC=$(HOSTCC)" was just wrong as apparent from looking at the behavior of tools/scripts/Makefile.include. Some of that persists after this change with WARNINGS, as now noted in a comment. A side effect of fixing the Makefile is to make HOSTCFLAGS work, but I suspect with the variable renames this may not be working. I'll leave that for yourself and Nick. I told Nick I'd take a look at this as I saw the wrong use of libsubcmd's headers and that was something I wanted to clean up having done similar in tools/perf, along with the whole removal of tools/lib/traceevent from Linux 6.2.

Quoted 6 lines

HOSTCFLAGS from the cmdline *used* to work, a git bisect shows it stopped working with

96f14fe738b6 ("kbuild: Rename HOSTCFLAGS to KBUILD_HOSTCFLAGS")

... so please add a "Fixes" tag for that commit.

Left off, see later.

Quoted 3 lines
Quoted 1 line

+MAKE = make -S

Why?

Cruft, removed.

Quoted 8 lines
Quoted 8 lines

LIBSUBCMD_DIR = $(srctree)/tools/lib/subcmd/ ifneq ($(OUTPUT),)
LIBSUBCMD_OUTPUT = $(abspath $(OUTPUT))/libsubcmd

tools/objtool/Makefile +0 -1
@@ -37,12 +33,19 @@ INCLUDES := -I$(srctree)/tools/include \
          -I$(srctree)/tools/objtool/arch/$(SRCARCH)/include \
          -I$(LIBSUBCMD_OUTPUT)/include
 WARNINGS := $(EXTRA_WARNINGS) -Wno-switch-default -Wno-switch-enum -Wno-packed -Wno-nested-externs
-CFLAGS   := -Werror $(WARNINGS) $(KBUILD_HOSTCFLAGS) -g $(INCLUDES) $(LIBELF_FLAGS)
Quoted 8 lines

tools/objtool/Makefile INCLUDES := -I$(srctree)/tools/include \ · line 36

Quoted 2 lines
tools/objtool/Makefile +1 -1
-LDFLAGS  += $(LIBELF_LIBS) $(LIBSUBCMD) $(KBUILD_HOSTLDFLAGS)
+OBJTOOL_CFLAGS := -Werror $(WARNINGS) -g $(INCLUDES) $(LIBELF_FLAGS) $(HOSTCFLAGS)

I *think* I understand the reason for the switch from KBUILD_HOSTCFLAGS to HOSTCFLAGS -- because KBUILD_HOSTCFLAGS is defined in the top-level kernel Makefile which isn't included here so it's undefined? -- but regardless that should be called out more explicitly as a problem being fixed in the commit log.

I was matching tools/perf, I've switched back to KBUILD_HOSTCFLAGS in v4. There's some higher logic at play with these variable names and I'm not aware of it so I'll leave it to be fixed if necessary later.

Quoted 6 lines

This issue really has me scratching my head, as I could have sworn objtool was being built with -O2.

tools/objtool/Makefile INCLUDES := -I$(srctree)/tools/include \ · line 37

Quoted 1 line
tools/objtool/Makefile +1 -0
+OBJTOOL_LDFLAGS := $(LIBELF_LIBS) $(LIBSUBCMD)

Is there a reason to not include $(HOSTLDFLAGS) here?

Done as KBUILD_HOSTLDFLAGS as I don't see HOSTLDFLAGS set anywhere.

Quoted 17 lines

tools/objtool/Makefile INCLUDES := -I$(srctree)/tools/include \ · line 48

Quoted 11 lines
tools/objtool/Makefile +8 -1
 # Allow old libelf to be used:
 elfshdr := $(shell echo '$(pound)include <libelf.h>' | $(CC) $(CFLAGS) -x c -E - | grep elf_getshdr)
-CFLAGS += $(if $(elfshdr),,-DLIBELF_USE_DEPRECATED)
+OBJTOOL_CFLAGS += $(if $(elfshdr),,-DLIBELF_USE_DEPRECATED)
+
+# Always want host compilation.
+HOST_OVERRIDES := CC="$(HOSTCC)" EXTRA_CFLAGS="$(OBJTOOL_CFLAGS)" \
+               LD="$(HOSTLD)" AR="$(HOSTAR)"
+BUILD_HOST_OVERRIDES := CC="$(HOSTCC)" CFLAGS="$(OBJTOOL_CFLAGS)" \
+                     LD="$(HOSTLD)" LDFLAGS="$(OBJTOOL_LDFLAGS)" \
+                     AR="$(HOSTAR)"

Maybe it depends on your perspective, but I'm thinking that some of these aren't really overrides, but rather normal expected flags. And the distinction between these two variables is muddy: it's not only Build vs non-Build, but also objtool vs libsubcmd.

Quoted 5 lines

How about just

  HOST_OVERRIDES := CC="$(HOSTCC) "LD="$(HOSTLD)" AR="$(HOSTAR)"

And then specifying the {EXTRA_}CFLAGS and LDFLAGS below where needed.

Done.

Quoted 10 lines
Quoted 10 lines
tools/objtool/Makefile +2 -1
 AWK = awk
 MKDIR = mkdir
@@ -65,10 +68,11 @@ include $(srctree)/tools/build/Makefile.include
 
 $(OBJTOOL_IN): fixdep FORCE
      $(Q)$(CONFIG_SHELL) ./sync-check.sh
-     $(Q)$(MAKE) $(build)=objtool
+     $(Q)$(MAKE) $(build)=objtool $(BUILD_HOST_OVERRIDES)
+
 
Quoted 5 lines

tools/objtool/Makefile include $(srctree)/tools/build/Makefile.include · line 75

Quoted 3 lines
tools/objtool/Makefile +1 -1
 $(OBJTOOL): $(LIBSUBCMD) $(OBJTOOL_IN)
-     $(QUIET_LINK)$(CC) $(OBJTOOL_IN) $(LDFLAGS) -o $@
+     $(QUIET_LINK)$(HOSTCC) $(OBJTOOL_IN) $(KBUILD_HOSTLDFLAGS) $(OBJTOOL_LDFLAGS) -o $@

Does KBUILD_HOSTLDFLAGS even work here?

Removed, albeit now to be part of OBJTOOL_LDFLAGS as your earlier comment requested.

Thanks,
Ian

Quoted 2 lines

--
Josh

author
#13 lore raw
In reply to Ian Rogers
Quoted 25 lines · Ian Rogers

Install libsubcmd and then get headers from there, this avoids inadvertent dependencies on things in tools/lib. Fix V=1 support. Clean up how HOSTCC is used to override CC to avoid CFLAGS being set for say gcc, and then CC being overridden to clang. Support HOSTCFLAGS as a make option.

v3. Is a rebase that removes the merged "tools lib subcmd: Add install
target" patch. In:
https://lore.kernel.org/lkml/CAKwvOd=kgXmpfbVa1wiEvwL0tX3gu+dDTGi-HEiRXSojwCLRrg@mail.gmail.com/
Nick rightly points out that:
WARNINGS := $(EXTRA_WARNINGS) -Wno-switch-default -Wno-switch-enum -Wno-packed -Wno-nested-externs
became:
WARNINGS := -Wno-switch-default -Wno-switch-enum -Wno-packed -Wno-nested-externs
losing the EXTRA_WARNINGS which v3 now adds back in. Previous
testing had added the warnings to the end rather than the
beginning, thereby causing unexpected build issues that aren't present in v3.
v2. Include required "tools lib subcmd: Add install target" that is
already in Arnaldo's tree:
https://lore.kernel.org/lkml/[email protected]/
https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/commit/?h=perf/core&id=630ae80ea1dd253609cb50cff87f3248f901aca3
When building libsubcmd.a from objtool's Makefile, clear the
subdir to avoid it being appended onto OUTPUT and breaking the
build.

Quoted 8 lines

Ian Rogers (3):
objtool: Install libsubcmd in build
objtool: Properly support make V=1
objtool: Alter how HOSTCC is forced

 tools/objtool/Build    |  2 --
 tools/objtool/Makefile | 66 ++++++++++++++++++++++++++++++------------
 2 files changed, 47 insertions(+), 21 deletions(-)

Ping. Relatively small set of patches, with Reviewed-by and Tested-by, would be nice to land. Thanks!

Ian

Quoted 3 lines

--
2.39.0.314.g84b9a713c41-goog

author
#14 lore raw
Quoted 32 lines · Ian Rogers
Quoted 30 lines · Ian Rogers

Install libsubcmd and then get headers from there, this avoids inadvertent dependencies on things in tools/lib. Fix V=1 support. Clean up how HOSTCC is used to override CC to avoid CFLAGS being set for say gcc, and then CC being overridden to clang. Support HOSTCFLAGS as a make option.

v3. Is a rebase that removes the merged "tools lib subcmd: Add install
target" patch. In:
https://lore.kernel.org/lkml/CAKwvOd=kgXmpfbVa1wiEvwL0tX3gu+dDTGi-HEiRXSojwCLRrg@mail.gmail.com/
Nick rightly points out that:
WARNINGS := $(EXTRA_WARNINGS) -Wno-switch-default -Wno-switch-enum -Wno-packed -Wno-nested-externs
became:
WARNINGS := -Wno-switch-default -Wno-switch-enum -Wno-packed -Wno-nested-externs
losing the EXTRA_WARNINGS which v3 now adds back in. Previous
testing had added the warnings to the end rather than the
beginning, thereby causing unexpected build issues that aren't present in v3.
v2. Include required "tools lib subcmd: Add install target" that is
already in Arnaldo's tree:
https://lore.kernel.org/lkml/[email protected]/
https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/commit/?h=perf/core&id=630ae80ea1dd253609cb50cff87f3248f901aca3
When building libsubcmd.a from objtool's Makefile, clear the
subdir to avoid it being appended onto OUTPUT and breaking the
build.

Ian Rogers (3):
objtool: Install libsubcmd in build
objtool: Properly support make V=1
objtool: Alter how HOSTCC is forced

Quoted 8 lines
Quoted 3 lines
 tools/objtool/Build    |  2 --
 tools/objtool/Makefile | 66 ++++++++++++++++++++++++++++++------------
 2 files changed, 47 insertions(+), 21 deletions(-)

Ping. Relatively small set of patches, with Reviewed-by and Tested-by, would be nice to land. Thanks!

Ian

Ping2. Small set of patches, 2x Reviewed-by and 1x Tested-by.

Thanks,
Ian

Quoted 4 lines
Quoted 4 lines

--
2.39.0.314.g84b9a713c41-goog