Michel Dänzer and a couple of other people reported inexplicable random
oopses in the scheduler, and the cause turns out to be gcc mis-compiling
the load_balance() function when debugging is enabled.  The gcc bug
apparently goes back to gcc-4.5, but slight optimization changes means
that it now showed up as a problem in 4.9.0 and 4.9.1.

The instruction scheduling problem causes gcc to schedule a spill
operation to before the stack frame has been created, which in turn can
corrupt the spilled value if an interrupt comes in.  There may be other
effects of this bug too, but that's the code generation problem seen in
Michel's case.

This is fixed in current gcc HEAD, but the workaround as suggested by
Markus Trippelsdorf is pretty simple: use -fno-var-tracking-assignments
when compiling the kernel, which disables the gcc code that causes the
problem.  This can result in slightly worse debug information for
variable accesses, but that is infinitely preferable to actual code
generation problems.

Doing this unconditionally (not just for CONFIG_DEBUG_INFO) also allows
non-debug builds to verify that the debug build would be identical: we
can do

    export GCC_COMPARE_DEBUG=1

to make gcc internally verify that the result of the build is
independent of the "-g" flag (it will make the compiler build everything
twice, toggling the debug flag, and compare the results).

Without the "-fno-var-tracking-assignments" option, the build would fail
(even with 4.8.3 that didn't show the actual stack frame bug) with a gcc
compare failure.

See also gcc bugzilla:

  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61801Reported-by: NMichel Dänzer <michel@daenzer.net>
Suggested-by: NMarkus Trippelsdorf <markus@trippelsdorf.de>
Cc: Jakub Jelinek <jakub@redhat.com>
Cc: stable@kernel.org
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

All other users of Makefile.build set $(obj) to the name of the
subdirectory to build. Do the same for the packaging targets, otherwise
the build fails if $(srctree) is a relative directory:

    $ make O=build tar-pkg
    make[1]: Entering directory `/home/mmarek/linux-2.6/build'
      CHK     include/config/kernel.release
    ../scripts/Makefile.build:44: ../../scripts/package/Makefile: No such file or directory
    make[2]: *** No rule to make target `../../scripts/package/Makefile'.  Stop.

Fixes: 9da0763b ("kbuild: Use relative path when building in a subdir of the source tree")
Signed-off-by: NMichal Marek <mmarek@suse.cz>

Commit c2e28dc9 (kbuild: Print the name of the build directory) prints
the name of the build directory for O= builds, but we should not be
doing this in make -s mode, so that commands like

  make -s O=<dir> kernelrelease

can be used by scripts. This matches the behavior of make itself, where
the -s option implies --no-print-directory.
Signed-off-by: NMichal Marek <mmarek@suse.cz>

With commit 9da0763b (kbuild: Use relative path when building in a
subdir of the source tree), the compiler messages include relative
paths. These are however relative to the build directory, not the
directory where make was started. Print the "Entering directory ..."
message once, so that IDEs/editors can find the source files.
Acked-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

The directory include/config is used only for
silentoldconfig, localmodconfig, localyesconfig.
Signed-off-by: NMasahiro Yamada <yamada.m@jp.panasonic.com>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

There are no generated files under include/linux directory.
Signed-off-by: NMasahiro Yamada <yamada.m@jp.panasonic.com>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

On architectures that setup CROSS_COMPILE in their arch/*/Makefile
(arc, blackfin, m68k, mips, parisc, score, sh, tile, unicore32, xtensa),
cc-option and cc-disable-warning may check against the wrong compiler,
causing errors like

    cc1: error: unrecognized command line option "-Wno-maybe-uninitialized"

if the host gcc supports a compiler option, while the cross compiler
doesn't support that option.

Move all logic using cc-option or cc-disable-warning below the inclusion
of the arch's Makefile to fix this.

Introduced by
  - commit e74fc973 ("Turn off
    -Wmaybe-uninitialized when building with -Os"),
  - commit 61163efa ("kbuild: LLVMLinux:
    Add Kbuild support for building kernel with Clang").

As -Wno-maybe-uninitialized requires a quite recent gcc (gcc 4.6.3 on
Ubuntu 12.04 LTS doesn't support it), this only showed up recently (gcc
4.8.2 on Ubuntu 14.04 LTS does support it).
Signed-off-by: NGeert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

Both gcc (well, actually gnu as) and clang support the "-Wa,-gdwarf-2" option
(though clang does not support "-Wa,--gdwarf-2"). Since these flags are equivalent
in meaning, this patch uses the one which is better supported across compilers.
Signed-off-by: NBehan Webster <behanw@converseincode.com>

When doing make O=<subdir>, use '..' to refer to the source tree. This
allows for more readable compiler messages, and, more importantly, it
sets the VPATH to '..', so filenames in WARN_ON() etc. will be shorter.
Acked-by: NSam Ravnborg <sam@ravnborg.org>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

When not using O=, $(srctree) refers to the same directory as
$(objtree), so we can set it to '.' as well. This makes the default
include path more compact and results in more readable messages from the
compiler. The only case where we need the absolute path is when creating
the 'source' symlink in /lib/modules.
Acked-by: NSam Ravnborg <sam@ravnborg.org>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

The main Makefile sets its working directory to the object tree and
never changes it again. Therefore, we can use '.' instead of the
absolute path. The only case where we need the absolute path is when
creating the 'build' symlink in /lib/modules.
Acked-by: NSam Ravnborg <sam@ravnborg.org>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

Signed-off-by: NMasahiro Yamada <yamada.m@jp.panasonic.com>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

Signed-off-by: NMasahiro Yamada <yamada.m@jp.panasonic.com>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

Kbuild is supposed to support mixed targets. (%config and build targets)

But "make all" did nothing if it was run with configuration targets.
For example,

  $ LANG=C make defconfig all
    HOSTCC  scripts/basic/fixdep
    HOSTCC  scripts/kconfig/conf.o
    SHIPPED scripts/kconfig/zconf.tab.c
    SHIPPED scripts/kconfig/zconf.lex.c
    SHIPPED scripts/kconfig/zconf.hash.c
    HOSTCC  scripts/kconfig/zconf.tab.o
    HOSTLD  scripts/kconfig/conf
  *** Default configuration is based on 'x86_64_defconfig'
  #
  # configuration written to .config
  #
  make: Nothing to be done for `all'.

This commits allows "make %config all" and makes sure
mixed targets are built one by one in the given order.
Signed-off-by: NMasahiro Yamada <yamada.m@jp.panasonic.com>
Cc: Michal Marek <mmarek@suse.cz>
CC: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

W=... provides extra gcc checks.

Having such code in scripts/Makefile.build results in the same flags
being added to KBUILD_CFLAGS multiple times becuase
scripts/Makefile.build is invoked every time Kbuild descends into
the subdirectories.

Since the top Makefile is already too cluttered, this commit moves
all of extra gcc check stuff to a new file scripts/Makefile.extrawarn,
which is included from the top Makefile.
Signed-off-by: NMasahiro Yamada <yamada.m@jp.panasonic.com>
CC: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

Add support to toplevel Makefile for compiling with clang, both for
HOSTCC and CC. Use cc-option to prevent gcc option from breaking clang, and
from clang options from breaking gcc.

Clang 3.4 semantics are the same as gcc semantics for unsupported flags. For
unsupported warnings clang 3.4 returns true but shows a warning and gcc shows
a warning and returns false.
Signed-off-by: NBehan Webster <behanw@converseincode.com>
Signed-off-by: NJan-Simon Möller <dl9pf@gmx.de>
Signed-off-by: NMark Charlebois <charlebm@gmail.com>
Cc: PaX Team <pageexec@freemail.hu>

objdiff is useful when doing large code cleanups.  For example, when
removing checkpatch warnings and errors from new drivers in the staging
tree.

objdiff can be used in conjunction with a git rebase to confirm that
each commit made no changes to the resulting object code.  It has the
same return values as diff(1).

This was written specifically to support adding the skein and threefish
cryto drivers to the staging tree.  I needed a programmatic way to
confirm that commits changing >90% of the lines didn't inadvertently
change the code.

Temporary files (objdump output) are stored in

  /path/to/linux/.tmp_objdiff

'make mrproper' will remove this directory.
Signed-off-by: NJason Cooper <jason@lakedaemon.net>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

Kbuild supports saving output files in a separate directory.
But the build directory must be created beforehand. For example,

  $ mkdir -p dir/to/store/output/files
  $ make O=dir/to/store/output/files defconfig

Creating a build directory automatically would be useful.
Signed-off-by: NMasahiro Yamada <yamada.m@jp.panasonic.com>
Acked-by: NSam Ravnborg <sam@ravnborg.org>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

'.*.cmd' files are cleaned-up by "make clean".
The same pattern in "make distclean" is unnecessary.
Signed-off-by: NMasahiro Yamada <yamada.m@jp.panasonic.com>
Signed-off-by: NMichal Marek <mmarek@suse.cz>

As of v3.7, the UAPI changes relocated headers around such that the
kernel version header lived in a new place.

If a person is bisecting and if you go back to pre-UAPI days,
you will create an include/linux/version.h  -- then if you checkout a
post-UAPI kernel, and even run "make distclean" it still won't delete
that old version file.  So you get a situation like this:

$ grep -R LINUX_VERSION_CODE include/
include/generated/uapi/linux/version.h:#define LINUX_VERSION_CODE 200192
include/linux/version.h:#define LINUX_VERSION_CODE 132646

The value in that second line is representative of a v2.6.38 version.
And it will be sourced/used, hence leading to strange behaviours, such
as drivers/staging content (which typically hasn't been purged of version
ifdefs) failing to build.

Since it is a subtle mode of failure, lets always clobber the old
file when doing a distclean.
Signed-off-by: NPaul Gortmaker <paul.gortmaker@windriver.com>
Acked-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NMichal Marek <mmarek@suse.cz>