Skip to contentSkip to conversation

kbuild: improve cc-option to clean up all temporary files

PATCH In review Masahiro Yamada 3 participants 8 messages Jun 14, 2020 – May 4, 2021

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 kbuild: improve cc-option to clean up all temporary files

When cc-option and friends evaluate compiler flags, the temporary file $$TMP is created as an output object, and automatically cleaned up. The actual file path of $$TMP is .<pid>.tmp, here <pid> is the process ID of $(shell ....) invoked from cc-option. (Please note $$$$ is the escape sequence of $$).

Such garbage files are cleaned up in most cases, but some compiler flags create additional output files.

For example, -gsplit-dwarf creates a .dwo file.

When CONFIG_DEBUG_INFO_SPLIT=y, you will see a bunch of .<pid>.dwo files left in the top of build directories. You may not notice them unless you do 'ls -a', but the garbage files will increase every time you run 'make'.

This commit changes the temporary object path to .tmp_<pid>/tmp, and removes .tmp_<pid> directory when exiting. The additional files .tmp_<pid>/tmp.dwo, will be cleaned away altogether.

When the compiler creates separate build artifacts, their file paths are usually determined based on the base name of the object. Another example is -ftest-coverage, which outputs the coverage data into <base-name-of-object>.gcno

---

 scripts/Kbuild.include | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)
scripts/Kbuild.include +7 -6
index 0c3dc983439b..5dfd224599b7 100644
@@ -86,20 +86,21 @@ cc-cross-prefix = $(firstword $(foreach c, $(1), \
 			$(if $(shell command -v -- $(c)gcc 2>/dev/null), $(c))))
 
 # output directory for tests below
-TMPOUT := $(if $(KBUILD_EXTMOD),$(firstword $(KBUILD_EXTMOD))/)
+TMPOUT = $(if $(KBUILD_EXTMOD),$(firstword $(KBUILD_EXTMOD))/).tmp_$$$$
 
 # try-run
 # Usage: option = $(call try-run, $(CC)...-o "$$TMP",option-ok,otherwise)
-# Exit code chooses option. "$$TMP" serves as a temporary file and is
+# Exit code chooses option. "$$TMP" serves as a temporary directory and is
 # automatically cleaned up.
 try-run = $(shell set -e;		\
-	TMP="$(TMPOUT).$$$$.tmp";	\
-	TMPO="$(TMPOUT).$$$$.o";	\
+	TMP=$(TMPOUT)/tmp;		\
+	TMPO=$(TMPOUT)/tmp.o;		\
+	mkdir -p $(TMPOUT);		\
+	trap "rm -rf $(TMPOUT)" EXIT;	\
 	if ($(1)) >/dev/null 2>&1;	\
 	then echo "$(2)";		\
 	else echo "$(3)";		\
-	fi;				\
-	rm -f "$$TMP" "$$TMPO")
+	fi)
 
 # as-option
 # Usage: cflags-y += $(call as-option,-Wa$(comma)-isa=foo,)
Signature · 1 line
2.25.1
#1 lore raw

PATCH 2/2 kconfig: unify cc-option and as-option

cc-option and as-option are almost the same; both pass the flag to $(CC). The main difference is the cc-option stops before the assemble stage (-S option) whereas as-option stops after it (-c option).

I chose -S because it is slightly faster, but $(cc-option,-gz=zlib) returns a wrong result (https://lkml.org/lkml/2020/6/9/1529).
It has been fixed by a separate patch, but using -c is more robust.

However, you cannot simply replace -S with -c because the following code would break:

    depends on $(cc-option,-gsplit-dwarf)

The combination of -c and -gsplit-dwarf does not accept /dev/null as output.

  $ cat /dev/null | gcc -gsplit-dwarf -S -x c - -o /dev/null
  $ echo $?
  0
  $ cat /dev/null | gcc -gsplit-dwarf -c -x c - -o /dev/null
  objcopy: Warning: '/dev/null' is not an ordinary file
  $ echo $?
  1
  $ cat /dev/null | gcc -gsplit-dwarf -c -x c - -o tmp.o
  $ echo $?
  0

There is another flag that creates an separate file based on the object file path:

  $ cat /dev/null | gcc -ftest-coverage -c -x c - -o /dev/null
  <stdin>:1: error: cannot open /dev/null.gcno

So, we cannot use /dev/null to sink the output.

Align the cc-option implementation with scripts/Kbuild.include.

With -c option used in cc-option, as-option is unneeded.

---

 arch/arm64/Kconfig      | 2 +-
 lib/Kconfig.debug       | 1 -
 scripts/Kconfig.include | 8 +-------
 3 files changed, 2 insertions(+), 9 deletions(-)
arch/arm64/Kconfig +1 -1
index 31380da53689..6eb18f45258e 100644
@@ -1564,7 +1564,7 @@ config CC_HAS_SIGN_RETURN_ADDRESS
 	def_bool $(cc-option,-msign-return-address=all)
 
 config AS_HAS_PAC
-	def_bool $(as-option,-Wa$(comma)-march=armv8.3-a)
+	def_bool $(cc-option,-Wa$(comma)-march=armv8.3-a)
 
 config AS_HAS_CFI_NEGATE_RA_STATE
 	def_bool $(as-instr,.cfi_startproc\n.cfi_negate_ra_state\n.cfi_endproc\n)
lib/Kconfig.debug +0 -1
index 96999d4d2dda..9ad9210d70a1 100644
@@ -229,7 +229,6 @@ config DEBUG_INFO_COMPRESSED
 	bool "Compressed debugging information"
 	depends on DEBUG_INFO
 	depends on $(cc-option,-gz=zlib)
-	depends on $(as-option,-gz=zlib)
 	depends on $(ld-option,--compress-debug-sections=zlib)
 	help
 	  Compress the debug information using zlib.  Requires GCC 5.0+ or Clang
scripts/Kconfig.include +1 -7
index c264da2b9b30..a5fe72c504ff 100644
@@ -25,18 +25,12 @@ failure = $(if-success,$(1),n,y)
 
 # $(cc-option,<flag>)
 # Return y if the compiler supports <flag>, n otherwise
-cc-option = $(success,$(CC) -Werror $(CLANG_FLAGS) $(1) -S -x c /dev/null -o /dev/null)
+cc-option = $(success,mkdir .tmp_$$$$; trap "rm -rf .tmp_$$$$" EXIT; $(CC) -Werror $(CLANG_FLAGS) $(1) -c -x c /dev/null -o .tmp_$$$$/tmp.o)
 
 # $(ld-option,<flag>)
 # Return y if the linker supports <flag>, n otherwise
 ld-option = $(success,$(LD) -v $(1))
 
-# $(as-option,<flag>)
-# /dev/zero is used as output instead of /dev/null as some assembler cribs when
-# both input and output are same. Also both of them have same write behaviour so
-# can be easily substituted.
-as-option = $(success, $(CC) $(CLANG_FLAGS) $(1) -c -x assembler /dev/null -o /dev/zero)
-
 # $(as-instr,<instr>)
 # Return y if the assembler supports <instr>, n otherwise
 as-instr = $(success,printf "%b\n" "$(1)" | $(CC) $(CLANG_FLAGS) -c -x assembler -o /dev/null -)
Signature · 7 lines
2.25.1


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
maintainer Acked-by on v1 2/2
#2 lore raw

Re: [PATCH 2/2] kconfig: unify cc-option and as-option

Quoted 54 lines · Masahiro Yamada

cc-option and as-option are almost the same; both pass the flag to $(CC). The main difference is the cc-option stops before the assemble stage (-S option) whereas as-option stops after it (-c option).

I chose -S because it is slightly faster, but $(cc-option,-gz=zlib) returns a wrong result (https://lkml.org/lkml/2020/6/9/1529).
It has been fixed by a separate patch, but using -c is more robust.

However, you cannot simply replace -S with -c because the following code would break:

    depends on $(cc-option,-gsplit-dwarf)

The combination of -c and -gsplit-dwarf does not accept /dev/null as output.

  $ cat /dev/null | gcc -gsplit-dwarf -S -x c - -o /dev/null
  $ echo $?
  0
  $ cat /dev/null | gcc -gsplit-dwarf -c -x c - -o /dev/null
  objcopy: Warning: '/dev/null' is not an ordinary file
  $ echo $?
  1
  $ cat /dev/null | gcc -gsplit-dwarf -c -x c - -o tmp.o
  $ echo $?
  0

There is another flag that creates an separate file based on the object file path:

  $ cat /dev/null | gcc -ftest-coverage -c -x c - -o /dev/null
  <stdin>:1: error: cannot open /dev/null.gcno

So, we cannot use /dev/null to sink the output.

Align the cc-option implementation with scripts/Kbuild.include.

With -c option used in cc-option, as-option is unneeded.

---

 arch/arm64/Kconfig      | 2 +-
 lib/Kconfig.debug       | 1 -
 scripts/Kconfig.include | 8 +-------
 3 files changed, 2 insertions(+), 9 deletions(-)
arch/arm64/Kconfig +0 -0
index 31380da53689..6eb18f45258e 100644

arch/arm64/Kconfig config CC_HAS_SIGN_RETURN_ADDRESS · line 1567

Quoted 5 lines
arch/arm64/Kconfig +1 -1
@@ -1564,7 +1564,7 @@ config CC_HAS_SIGN_RETURN_ADDRESS
 	def_bool $(cc-option,-msign-return-address=all)
 
 config AS_HAS_PAC
-	def_bool $(as-option,-Wa$(comma)-march=armv8.3-a)
+	def_bool $(cc-option,-Wa$(comma)-march=armv8.3-a)

For this arm64 part:

Will

#3 lore raw

Re: [PATCH 2/2] kconfig: unify cc-option and as-option

In reply to Masahiro Yamada

Hi,

So... I realized it's been a while:

Quoted 4 lines · Masahiro Yamada

cc-option and as-option are almost the same; both pass the flag to $(CC). The main difference is the cc-option stops before the assemble stage (-S option) whereas as-option stops after it (-c option).

But, I had noticed for a while now that M= build for an out-of-tree driver were causing some trouble. Not really completely "out-of-tree" but rather backported (https://backports.wiki.kernel.org/).

And then I finally narrowed it down to this commit, specifically this:

scripts/Kconfig.include failure = $(if-success,$(1),n,y) · line 28

Quoted 3 lines

 # Return y if the compiler supports <flag>, n otherwise

scripts/Kconfig.include +1 -1
-cc-option = $(success,$(CC) -Werror $(CLANG_FLAGS) $(1) -S -x c /dev/null -o /dev/null)
+cc-option = $(success,mkdir .tmp_$$$$; trap "rm -rf .tmp_$$$$" EXIT; $(CC) -Werror $(CLANG_FLAGS) $(1) -c -x c /dev/null -o .tmp_$$$$/tmp.o)

What happens is that we're doing

make -C /path/to/kernel M=/path/to/driver

But /path/to/kernel may be the installed distro kernel headers, and thus not be writable to the user doing the driver compile. Obviously, the user may need to 'sudo' anyway to install the result, but if just test- compiling, or even as better practice to not run everything as root, this ".tmp_$$" dir cannot be created.

IOW, this broke compiler option detection when KBUILD_EXTMOD=/M= is used. It seems this is still supported (documented in kbuild docs), so I'm kind of hoping it could be fixed? But OTOH, I really don't know how, perhaps just using "mktemp -d" here instead of the hardcoded temp dir?

Thanks,
johannes

#4 lore raw

Re: [PATCH 2/2] kconfig: unify cc-option and as-option

Quoted 31 lines · Johannes Berg

Hi,

So... I realized it's been a while:

Quoted 4 lines · Masahiro Yamada

cc-option and as-option are almost the same; both pass the flag to $(CC). The main difference is the cc-option stops before the assemble stage (-S option) whereas as-option stops after it (-c option).

But, I had noticed for a while now that M= build for an out-of-tree driver were causing some trouble. Not really completely "out-of-tree" but rather backported (https://backports.wiki.kernel.org/).

And then I finally narrowed it down to this commit, specifically this:

scripts/Kconfig.include failure = $(if-success,$(1),n,y) · line 28

Quoted 3 lines

# Return y if the compiler supports <flag>, n otherwise

scripts/Kconfig.include +1 -1
-cc-option = $(success,$(CC) -Werror $(CLANG_FLAGS) $(1) -S -x c /dev/null -o /dev/null)
+cc-option = $(success,mkdir .tmp_$$$$; trap "rm -rf .tmp_$$$$" EXIT; $(CC) -Werror $(CLANG_FLAGS) $(1) -c -x c /dev/null -o .tmp_$$$$/tmp.o)

What happens is that we're doing

make -C /path/to/kernel M=/path/to/driver

But /path/to/kernel may be the installed distro kernel headers, and thus not be writable to the user doing the driver compile. Obviously, the user may need to 'sudo' anyway to install the result, but if just test- compiling, or even as better practice to not run everything as root, this ".tmp_$$" dir cannot be created.

Quoted 8 lines

IOW, this broke compiler option detection when KBUILD_EXTMOD=/M= is used. It seems this is still supported (documented in kbuild docs), so I'm kind of hoping it could be fixed? But OTOH, I really don't know how, perhaps just using "mktemp -d" here instead of the hardcoded temp dir?

Thanks,
johannes

- This commit touches scripts/Kconfig.include.
- External module builds (M= builds) never invoke Kconfig

Putting these two together, your claim is really odd. If external module builds invoke Kconfig, your kernel is already broken.

If you claim this is an issue,
please describe how to reproduce it in *upstream* kernel. I do not know (or care about) your backport kernel.

--
Best Regards
Masahiro Yamada

#5 lore raw

Re: [PATCH 2/2] kconfig: unify cc-option and as-option

Quoted 5 lines · Masahiro Yamada

 - This commit touches scripts/Kconfig.include.
 - External module builds (M= builds) never invoke Kconfig

Putting these two together, your claim is really odd.

Hmm.

Quoted 2 lines

If external module builds invoke Kconfig,
your kernel is already broken.

Well, it's not about the kernel, that's just the normal upstream (or perhaps distribution) kernel.

Anyway, you're right, it's much simpler. The problem isn't invoking M= or something like that, that happens much later and sent me on the completely wrong track.

The problem is simply doing

make kernelversion

to determine the version of a tree that's not writable to the user, e.g.

	make -C /lib/modules/$(uname -r)/build/ kernelversion

Which basically also means that it's harmless, since the version is of course not affected by cc-option.

johannes

#6 lore raw

Re: [PATCH 2/2] kconfig: unify cc-option and as-option

Quoted 22 lines · Johannes Berg
Quoted 5 lines · Masahiro Yamada

- This commit touches scripts/Kconfig.include.
- External module builds (M= builds) never invoke Kconfig

Putting these two together, your claim is really odd.

Hmm.

Quoted 2 lines

If external module builds invoke Kconfig,
your kernel is already broken.

Well, it's not about the kernel, that's just the normal upstream (or perhaps distribution) kernel.

Anyway, you're right, it's much simpler. The problem isn't invoking M= or something like that, that happens much later and sent me on the completely wrong track.

The problem is simply doing

Quoted 8 lines

make kernelversion

to determine the version of a tree that's not writable to the user, e.g.

        make -C /lib/modules/$(uname -r)/build/ kernelversion

Which basically also means that it's harmless, since the version is of course not affected by cc-option.

It was fixed.

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=805b2e1d427aab4bb27fa7c51ebb9db7547551b1

If you want to make it work without that commit,

   make -C /lib/modules/$(uname -r)/build/ kernelversion M=/tmp

will work.

Pass a writable directory to M=.
M= build never touches the kernel source tree.

Quoted 2 lines

johannes

Signature · 2 lines
Best Regards
Masahiro Yamada
#7 lore raw

Re: [PATCH 2/2] kconfig: unify cc-option and as-option

Quoted 5 lines · Masahiro Yamada

It was fixed.

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=805b2e1d427aab4bb27fa7c51ebb9db7547551b1

Ah, ok, great.

Quoted 7 lines

If you want to make it work without that commit,

   make -C /lib/modules/$(uname -r)/build/ kernelversion M=/tmp

will work.

I guess that works. I was just going to ignore the errors now that I realized what it was :)

Thanks!

johannes