Skip to contentSkip to conversation

kbuild updates

PATCH In review Sam Ravnborg 2 participants 3 messages Mar 10, 2005 – Mar 11, 2005

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 kbuild updates

Hi Linus.

Please pull latest kbuild patches.
Full list below - the most important stuff:

o default value for INSTALL_PATH set to /boot
o Use -Wno-pointer-sign for gcc 4.0
o arch/i386: make install no longer check vmlinux
o Introdude KBUILD_NOCMDDEP

Except a few trivial things it has been in -mm for a while with no comments.

Sam

Please do a

bk pull bk://linux-sam.bkbits.net/kbuild

This will update the following files:

 Makefile                              |   47 +++++++++++++++++++++++-----------
 arch/i386/Makefile                    |    5 ++-
 arch/i386/kernel/Makefile             |    2 -
 drivers/net/wireless/prism54/Makefile |    2 -
 drivers/video/console/Makefile        |    4 +-
 include/linux/module.h                |   13 +--------
 include/linux/moduleparam.h           |   19 +++++++++++--
 kernel/Makefile                       |    2 -
 kernel/kallsyms.c                     |    4 +-
 scripts/Makefile.lib                  |   28 ++++++++++++--------
 scripts/Makefile.modinst              |    5 ++-
 scripts/genksyms/genksyms.h           |   16 ++++++++---
 scripts/kconfig/Makefile              |    8 +++++
 scripts/mod/modpost.c                 |    5 +--
 scripts/mod/modpost.h                 |    4 +-
 scripts/namespace.pl                  |    5 +++
 16 files changed, 109 insertions(+), 60 deletions(-)

through these ChangeSets:

<[email protected]> (05/03/10 1.2007)
kconfig: Add explicit depedencies

Without these I could not do make menuconfig when using O= This is the shipped rule that plays tricks here.

<[email protected]> (05/03/10 1.2006)
kbuild: Install external modules in a path relative to their own path

When an external module is being built in down in a directory structure keep the relative directory when installing the module.

Example:
fs/ contains a Makefile used to build both modules:
obj-y := myfs/ oldfs/
Install directories
fs/myfs/myfs.ko => Will be installed in /lib/modules/<version>/extra/fs/myfs/
fs/oldfs/oldfs.o => Will be installed in /lib/modules/<version>/extra/fs/oldfs/

<[email protected]> (05/03/10 1.2005)
video/console: fix spurious rebuild

kbuild does have troubles with assignmnets including '#'.
The '#' is seen as a comment marker and this will in the end cause kbuild to think the commandline to build promcon_tbl.c has changed. This happens because the commandline is stored in the file: .promcon_tbl.c.cmd

Although a bit complex the command to build promcon_tbl.c is unlikely to change so the workaround is to skip the check for a changed commandline. Now promcon_tbl.c is only rebuilt if the .uni file is newer than the .c file.

<[email protected]> (05/02/02 1.2004)
prismtech: Avoid recompile when changing compile dir

-I$(PWD) is superflous - and caused absolute path to be stored in build command - this casuses recompile when using symlink to kernel.
Also deleted commented out -DCONFIG_PRISM_WDS. CONFIG_PRISM_WDS are not present in any of the source files.

<[email protected]> (05/02/02 1.2003)
kbuild: Fix debugging leftover

So now check for commandline options actually works again.

<[email protected]> (05/01/31 1.2002)
[PATCH] kbuild: skip depmod if not executable

I've cross-compiled Linux on i386-netbsdelf2.0 for arm-linux for quite some time now and everything seems to be working perfectly except for one minor glitch in the build process that halts module installation (needlessly, IMHO). Specifically, if System.map exists $(DEPMOD) is run ("for convenience" as the comment says in the Makefile). However, on NetBSD I don't have $(DEPMOD) available so the command fails and make exits with non-zero exit status. Please consider the attached patch to add a check for $(DEPMOD) so that missing $(DEPMOD) won't halt the whole build process.

From: Tero Niemela <[email protected]>

<[email protected]> (05/01/31 1.2001)
[PATCH] kbuild: fix for i386 cross compile

I used to be be able to cross compile for i386 on my x86_64 machine, but recently something (gcc/binutils?) changed, and it stopped working.

Following patch makes cross compile work with:

make ARCH=i386 CFLAGS_KERNEL="-m32" AFLAGS_KERNEL="-m32" bzImage

Without the patch I'm getting the following error:

SYSCALL arch/i386/kernel/vsyscall-syms.o
/usr/lib/gcc/x86_64-pc-linux-gnu/3.4.3/../../../../x86_64-pc-linux-gnu/bin/ld:
Relocatable linking with relocations from format elf32-i386 (arch/i386/kernel/vsyscall-sysenter.o) to format elf64-x86-64 (arch/i386/kernel/vsyscall-syms.o) is not supported
collect2: ld returned 1 exit status

<[email protected]> (05/01/31 1.2000)
[PATCH] kbuild: no redundant srctree in tags file

Avoid cluttering the tags/TAGS generated file with $(srctree) in the paths if this is not needed.

This has two advantages:

- Saving about 20M on the size of the resulting tags file (which are used currently to store the absolute path of the file names rather than the relative one) when KBUILD_OUTPUT is not set.

- Keeping the tags file valid when the directory is renamed.

No change is done for who does make tags O=..., if this is wanted (I would find that incommodous and non-typical for a developer, but anyway I've not ruined functionality in that case).

<[email protected]> (05/01/31 1.1999)
[PATCH] kbuild: Use -Wno-pointer-sign for gcc 4.0

Compiling an allyesconfig kernel straight with a gcc 4.0 snapshot gives nearly 10k new warnings like:

warning: pointer targets in passing argument 5 of `cpuid' differ in signedness

Since the sheer number of these warnings was too much even for the most determined kernel janitors (I actually asked ;-) and I don't think it's a very serious issue to have these mismatches I submitted an new option to gcc to disable it. It was incorporated in gcc mainline now.

This patch makes the kernel compilation use it. There are still quite a lot of new warnings with 4.0 (mostly about uninitialized variables), but the compile log looks much nicer nnow.

<[email protected]> (05/01/30 1.1998)
The wrong version of the parmtype patch was merged, incompletely, and the part that got merged got broken on the way. Here are the fixes:

Move __MODULE_INFO to modparam.h: This macro is used in modparam.h; there are users who include this header but not module.h. The latter includes modparam.h already.

__MODULE_INFO(parmtype, name##type, #name ":" #type) does not evaluate to __MODULE_INFO(parmtype, footype, "foo:int") as was the idea, but to __MODULE_INFO(parmtype, fooint, "foo:int") when type is bound to int. In more complicated cases, we get syntax erros. Re-introduce the __MODULE_PARM_TYPE macro; this is cleaner than renaming the type parameter.

Add the parmtype definition which was dropped during the merge to to the obsolete but still heavily used MODULE_PARM macro.

<[email protected]> (05/01/30 1.1997)
kbuild: make 'make help' show all *config targets and update descriptions slightly.

"make help" doesn't show "make randconfig" nor "make config" as options and the description of oldconfig could be better (IMHO). Patch below adds the missing targets to the help and updates the description of oldconfig.

<[email protected]> (05/01/30 1.1996)
kbuild: update scripts/namespace.pl

The patch below removes some false positives I've observed.

<[email protected]> (05/01/30 1.1995)
kallsyms: kallsyms.c - make some code static

This patch makes some needlessly global code static.

<[email protected]> (05/01/30 1.1994)
kernel/configs.c: make a variable static

This patch makes a needlessly global variable static.

<[email protected]> (05/01/30 1.1993)
kbuild: Dont include absolute filenames in binaries

The kbuild utilities are compiled with absolute patch names, so paths starting with $RPM_BUILD_ROOT would end up in the binaries. To avoid this, remove all references to __FILE__ (directly and indirectly via assert()).

<[email protected]> (05/01/30 1.1992)
kbuild: Nicer printout when Module.symvers is missing

<[email protected]> (05/01/30 1.1991)
kbuild: Warn when building external modules without modversions

This adds a warning when building external modules (M= or SUBDIRS= syntax) and there is no Module.symvers in the object tree. A missing Module.symvers is a clear sign that the kernel tree itself was never compiled. The resulting modules will work, but no symbol version information will be attached to kernel symbols the module uses (because that information comes from Module.symvers), and so the module will be more unsafe.
Futhermore the external module will not record what other modules it is depended on.

The test works with CONFIG_MODVERSIONS enabled or disabled.

<[email protected]> (05/01/30 1.1990)
kbuild: add '--extra=+f' to ctags in Makefile in order to search for file names

From: John Kacur <[email protected]>

<[email protected]> (05/01/30 1.1989)
kbuild: (trivial) spelling fix in comment in Makefile

From: John Kacur <[email protected]>

<[email protected]> (05/01/29 1.1987)
kbuild: default value for INSTALL_PATH set to /boot

Most architectures uses /boot for there kernel image, so let this be reflected by the kernel.
If INSTALL_PATH shell variable is set then this will have effect. If INSTALL_PATH is set one the commanline to make like this:
make INSTALL_PATH=/nfs/boot install
then this will override both kbuild and shell variable.

If an arch prefer another default this must be set in the arch Makefile

<[email protected]> (05/01/29 1.1986)
kbuild arch/i386: make install no longer check vmlinux

make install is often executed as root or on a different mechine via NFS To avoid updating vmlinux due to directory changes or similar the install target for i386 no longer has vmlinux as a prerequisite.
Now modules_install and install are aligned in this respect.

<[email protected]> (05/01/29 1.1985)
kbuild: Introdude KBUILD_NOCMDDEP

When tossing around with different gcc compilers there is no way to tell kbuild to ignore the new name of the compiler. The new option KBUILD_NOCMDDEP tell kbuild not to check the commandline for changes.
This should be used with care because the resulting kernel may become inconsistent if one part is build with 2.96, and another part build with 3.3.4. So use only when you know what you are doing.

Syntax:
make KBUILD_NOCMDDEP=1

Original request for this feature came from hpa.

<[email protected]> (05/01/29 1.1984)
kbuild: Makefile.lib - small cleanup

Combine duplicate code in two smaller 'functions'

Signed-off-by Sam Ravnborg

<[email protected]> (05/01/10 1.1966.36.1)
kallsyms: gate page patch breaks module lookups

Quoted 3 lines

Your recent patch looks to break module kallsyms lookups.... It looks like if CONFIG_KALLSYMS_ALL is set then we never look up module addresses.

Separate lookups for kernel and modules when CONFIG_KALLSYMS_ALL=y.

#1 lore raw
Quoted 7 lines · Sam Ravnborg

...
Full list below - the most important stuff:
...
o Use -Wno-pointer-sign for gcc 4.0
...
Except a few trivial things it has been in -mm for a while with no comments.

If this is the same version as in 2.6.11-mm2 (you didn't offer a GNU patch so that I could check it), the following is still present:

http://www.ussg.iu.edu/hypermail/linux/kernel/0502.2/1507.html

Quoted 2 lines

Sam
...

cu
Adrian

Signature · 4 lines
       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed