Makefile: rework u-boot-initial-env target
v5 cover Makefile: rework u-boot-initial-env target
From: Max Krummenacher <[email protected]>
With CONFIG_LTO enabled the current way of extracting the configured environment no longer works, i.e. the object file content changes due to LTO.
Build a host tool which prints the configured environment instead of using objcopy and friends to achive the same.
The code and Makefile changes were mostly stolen from tools/env/ i.e. the target userspace tools to access the environment.
Changes in v5:
- don't build the printinitialenv tool unconditionally but build it only as part of the u-boot-initial-env target.
This no longer fails the 'make tools-only_defconfig tools-only' use-case which is reported by Tom Rini.
Adding the $(env_h) dependencies to the tools target might give circular dependencies issues with some future tool.
- add Acked-by: Pali Rohár <[email protected]>
Changes in v4:
- add '(objtree)/' when calling the tool. Suggested by Pali Rohár.
- renamed patch, as more than just the Makefile has changes
Changes in v3:
- moved the tool from scripts/ to tools/. Suggested by Tom Rini
- changed the dependencies to '$(env_h)' and 'tools'. Suggested by Tom Rini and Pali Rohár.
- removed the sed rule which replaces \x00 with \x0A as this is already done by the tool. Suggested by Pali Rohár.
Changes in v2:
- reworked to build a host tool which prints the configured environment as proposed by Pali Rohár
https://lore.kernel.org/u-boot/[email protected]/
- renamed patch, v1 used "Makefile: fix u-boot-initial-env target if lto is enabled"
Max Krummenacher (1):
u-boot-initial-env: rework make target
Makefile | 10 ++++++---- tools/.gitignore | 1 + tools/Makefile | 4 ++++ tools/printinitialenv.c | 44 +++++++++++++++++++++++++++++++++++++++++ 4 files changed, 55 insertions(+), 4 deletions(-) create mode 100644 tools/printinitialenv.c
Signature · 1 line
2.35.3
PATCH v5 1/1 u-boot-initial-env: rework make target
From: Max Krummenacher <[email protected]>
With LTO enabled the U-Boot initial environment is no longer stored in an easy accessible section in env/common.o. I.e. the section name changes from build to build, its content maybe compressed and it is annotated with additional data.
Drop trying to read the initial env with elf tools from the compiler specific object file in favour of adding and using a host tool with the only functionality of printing the initial env to stdout.
See also:
https://lore.kernel.org/all/[email protected]/
---
Changes in v5:
- don't build the printinitialenv tool unconditionally but build it only as part of the u-boot-initial-env target.
This no longer fails the 'make tools-only_defconfig tools-only' use-case which is reported by Tom Rini.
Adding the $(env_h) dependencies to the tools target might give circular dependencies issues with some future tool.
- add Acked-by: Pali Rohár <[email protected]>
Changes in v4:
- add '(objtree)/' when calling the tool. Suggested by Pali Rohár.
- renamed patch, as more than just the Makefile has changes
Changes in v3:
- moved the tool from scripts/ to tools/. Suggested by Tom Rini
- changed the dependencies to '$(env_h)' and 'tools'. Suggested by Tom Rini and Pali Rohár.
- removed the sed rule which replaces \x00 with \x0A as this is already done by the tool. Suggested by Pali Rohár.
Changes in v2:
- reworked to build a host tool which prints the configured environment as proposed by Pali Rohár
https://lore.kernel.org/u-boot/[email protected]/
- renamed patch, v1 used "Makefile: fix u-boot-initial-env target if lto is enabled"
Makefile | 10 ++++++---- tools/.gitignore | 1 + tools/Makefile | 4 ++++ tools/printinitialenv.c | 44 +++++++++++++++++++++++++++++++++++++++++ 4 files changed, 55 insertions(+), 4 deletions(-) create mode 100644 tools/printinitialenv.c
Makefile +6 -4
@@ -2439,11 +2439,13 @@ endif $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.modpost quiet_cmd_genenv = GENENV $@ -cmd_genenv = $(OBJCOPY) --dump-section .rodata.default_environment=$@ env/common.o; \ - sed --in-place -e 's/\x00/\x0A/g' $@; sed --in-place -e '/^\s*$$/d' $@; \ - sort --field-separator== -k1,1 --stable $@ -o $@ +cmd_genenv = \ + $(objtree)/tools/printinitialenv | \ + sed -e '/^\s*$$/d' | \ + sort --field-separator== -k1,1 --stable -o $@ -u-boot-initial-env: u-boot.bin +u-boot-initial-env: $(env_h) FORCE + $(Q)$(MAKE) $(build)=tools $(objtree)/tools/printinitialenv $(call if_changed,genenv) # Consistency checks
tools/.gitignore +1 -0
@@ -28,6 +28,7 @@ /mxsboot /ncb /prelink-riscv +/printinitialenv /proftool /relocate-rela /spl_size_limit
tools/Makefile +4 -0
@@ -275,6 +275,10 @@ clean-dirs := lib common always := $(hostprogs-y) +# Host tool to dump the currently configured default environment, +# build it on demand, i.e. not add it to 'always'. +hostprogs-y += printinitialenv + # Generated LCD/video logo LOGO_H = $(objtree)/include/bmp_logo.h LOGO_DATA_H = $(objtree)/include/bmp_logo_data.h
tools/printinitialenv.c new file +44 -0
44 lines not on this page · raw message
Signature · 1 line
2.35.3
Re: [PATCH v5 1/1] u-boot-initial-env: rework make target
Hi Max,
Quoted 46 lines · Max Krummenacher
From: Max Krummenacher <[email protected]>
With LTO enabled the U-Boot initial environment is no longer stored in an easy accessible section in
env/common.o. I.e. the section name changes from build to build, its content maybe compressed and it is annotated with additional data.Drop trying to read the initial env with elf tools from the compiler specific object file in favour of adding and using a host tool with the only functionality of printing the initial env to stdout.
See also:
https://lore.kernel.org/all/[email protected]/Signed-off-by Max Krummenacher <[email protected]>Acked-by Pali Rohár <[email protected]>---
Changes in v5:
- don't build the printinitialenv tool unconditionally but build it only as part of the u-boot-initial-env target.
This no longer fails the 'make tools-only_defconfig tools-only' use-case which is reported by Tom Rini.
Adding the $(env_h) dependencies to the tools target might give circular dependencies issues with some future tool.
- add Acked-by: Pali Rohár <[email protected]>Changes in v4:
- add '(objtree)/' when calling the tool. Suggested by Pali Rohár.
- renamed patch, as more than just theMakefilehas changesChanges in v3:
- moved the tool fromscripts/totools/. Suggested by Tom Rini
- changed the dependencies to '$(env_h)' and 'tools'. Suggested by Tom Rini and Pali Rohár.
- removed the sed rule which replaces \x00 with \x0A as this is already done by the tool. Suggested by Pali Rohár.Changes in v2:
- reworked to build a host tool which prints the configured environment as proposed by Pali Rohár
https://lore.kernel.org/u-boot/[email protected]/
- renamed patch, v1 used "Makefile: fix u-boot-initial-env target if lto is enabled"
Quoted 6 lines
Makefile | 10 ++++++---- tools/.gitignore | 1 + tools/Makefile | 4 ++++ tools/printinitialenv.c | 44 +++++++++++++++++++++++++++++++++++++++++ 4 files changed, 55 insertions(+), 4 deletions(-) create mode 100644 tools/printinitialenv.c
It would be nice to have a test for this. It could go in test_env.py and you can use test_event.py as an example.
- Simon
Quoted 103 lines
Makefile +6
-4@@ -2439,11 +2439,13 @@ endif $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.modpost quiet_cmd_genenv = GENENV $@ -cmd_genenv = $(OBJCOPY) --dump-section .rodata.default_environment=$@ env/common.o; \ - sed --in-place -e 's/\x00/\x0A/g' $@; sed --in-place -e '/^\s*$$/d' $@; \ - sort --field-separator== -k1,1 --stable $@ -o $@ +cmd_genenv = \ + $(objtree)/tools/printinitialenv | \ + sed -e '/^\s*$$/d' | \ + sort --field-separator== -k1,1 --stable -o $@ -u-boot-initial-env: u-boot.bin +u-boot-initial-env: $(env_h) FORCE + $(Q)$(MAKE) $(build)=tools $(objtree)/tools/printinitialenv $(call if_changed,genenv) # Consistency checkstools/.gitignore +1
-0@@ -28,6 +28,7 @@ /mxsboot /ncb /prelink-riscv +/printinitialenv /proftool /relocate-rela /spl_size_limittools/Makefile +4
-0@@ -275,6 +275,10 @@ clean-dirs := lib common always := $(hostprogs-y) +# Host tool to dump the currently configured default environment, +# build it on demand, i.e. not add it to 'always'. +hostprogs-y += printinitialenv + # Generated LCD/video logo LOGO_H = $(objtree)/include/bmp_logo.h LOGO_DATA_H = $(objtree)/include/bmp_logo_data.htools/printinitialenv.c new file +44
-1@@ -0,0 +1,44 @@ +// SPDX-License-Identifier: GPL-2.0+ +/* +* (C) Copyright 2022 +* Max Krummenacher, Toradex +* +* Snippets taken from tools/env/fw_env.c +* +* This prints the list of default environment variables as currently +* configured. +* +*/ + +#include <stdio.h> + +/* Pull in the current config to define the default environment */ +#include <linux/kconfig.h> + +#ifndef __ASSEMBLY__ +#define __ASSEMBLY__/* get only #defines from config.h */ +#include <config.h> +#undef __ASSEMBLY__ +#else +#include <config.h> +#endif + +#define DEFAULT_ENV_INSTANCE_STATIC +#include <generated/environment.h> +#include <env_default.h> + +int main (void ) +{ +char *env, *nxt; + +for (env = default_environment; *env; env = nxt +1 ) { +for (nxt = env; *nxt; ++nxt) { +if (nxt >= &default_environment[sizeof (default_environment)]) { +fprintf (stderr,"## Error: environment not terminated\n" ); +return -1 ; + } + } +printf ("%s\n" , env); + } +return 0 ; +} --2.35.3
Re: [PATCH v5 1/1] u-boot-initial-env: rework make target
In reply to Max KrummenacherQuoted 11 lines · Max Krummenacher
From: Max Krummenacher <[email protected]>
With LTO enabled the U-Boot initial environment is no longer stored in an easy accessible section in
env/common.o. I.e. the section name changes from build to build, its content maybe compressed and it is annotated with additional data.Drop trying to read the initial env with elf tools from the compiler specific object file in favour of adding and using a host tool with the only functionality of printing the initial env to stdout.
Quoted 6 lines
See also:
https://lore.kernel.org/all/[email protected]/Signed-off-by Max Krummenacher <[email protected]>Acked-by Pali Rohár <[email protected]>Reviewed-by Simon Glass <[email protected]>
Applied to u-boot/next, thanks!
Signature · 1 line
Tom
Re: [PATCH v5 1/1] u-boot-initial-env: rework make target
Hi Tom
Quoted 14 lines · Tom Rini
Quoted 11 lines · Max Krummenacher
From: Max Krummenacher <[email protected]>
With LTO enabled the U-Boot initial environment is no longer stored in an easy accessible section in
env/common.o. I.e. the section name changes from build to build, its content maybe compressed and it is annotated with additional data.Drop trying to read the initial env with elf tools from the compiler specific object file in favour of adding and using a host tool with the only functionality of printing the initial env to stdout.
Quoted 8 lines
Quoted 6 lines
See also:
https://lore.kernel.org/all/[email protected]/Signed-off-by Max Krummenacher <[email protected]>Acked-by Pali Rohár <[email protected]>Reviewed-by Simon Glass <[email protected]>Applied to u-boot/next, thanks!
The commit not only fixes the use case on arm64 with LTO enabled, it also fixes sandbox for x86-64. For me on Fedora with a gcc (GCC) 11.3.1 20220421 both make sandbox_defconfig; make u-boot-initial-env andmake sandbox_defconfig; <unselect LTO>; make u-boot-initial-env fail with
``` GENENV u-boot-initial-env objcopy: env/common.o: can't dump section '.rodata.default_environment' - it does not exist: file format not recognized sed: can't read u-boot-initial-env: No such file or directory ```
Wouldn't that merit applying the commit to master, i.e. include it in v2023.01?
Regards
Max
P.S.:
The test in [1] catches this failure, so [1] should go in after this commit to have CI tests not failing.
[1] https://lore.kernel.org/all/[email protected]/
Quoted 3 lines
--
Tom
Re: [PATCH v5 1/1] u-boot-initial-env: rework make target
Quoted 38 lines · Max Krummenacher
Hi Tom
Quoted 14 lines · Tom Rini
Quoted 11 lines · Max Krummenacher
From: Max Krummenacher <[email protected]>
With LTO enabled the U-Boot initial environment is no longer stored in an easy accessible section in
env/common.o. I.e. the section name changes from build to build, its content maybe compressed and it is annotated with additional data.Drop trying to read the initial env with elf tools from the compiler specific object file in favour of adding and using a host tool with the only functionality of printing the initial env to stdout.
Quoted 8 lines
Quoted 6 lines
See also:
https://lore.kernel.org/all/[email protected]/Signed-off-by Max Krummenacher <[email protected]>Acked-by Pali Rohár <[email protected]>Reviewed-by Simon Glass <[email protected]>Applied to u-boot/next, thanks!
The commit not only fixes the use case on arm64 with LTO enabled, it also fixes sandbox for x86-64. For me on Fedora with a
gcc (GCC) 11.3.1 20220421bothmake sandbox_defconfig; make u-boot-initial-envandmake sandbox_defconfig; <unselect LTO>; make u-boot-initial-envfail with``` GENENV u-boot-initial-env objcopy: env/common.o: can't dump section '.rodata.default_environment' - it does not exist: file format not recognized sed: can't read u-boot-initial-env: No such file or directory ```
Quoted 1 line
Wouldn't that merit applying the commit to master, i.e. include it in v2023.01?
You can just disable LTO, and it's been an issue for a while now. It's also easy enough to cherry-pick if there's cases out there that can't wait. Thanks again for addressing the problem!
Signature · 1 line
Tom