By default, sparse assumes a 64bit machine when compiled on x86-64
and 32bit when compiled on anything else.

This can of course create all sort of problems for the other archs, like
issuing false warnings ('shift too big (32) for type unsigned long'), or
worse, failing to emit legitimate warnings.

Fix this by adding the -m32/-m64 flag, depending on CONFIG_64BIT,
to CHECKFLAGS in the main Makefile (and so for all archs).
Also, remove the now unneeded -m32/-m64 in arch specific Makefiles.
Signed-off-by: NLuc Van Oostenryck <luc.vanoostenryck@gmail.com>
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

The kernel depends on macros like __BYTE_ORDER__,
__BIG_ENDIAN__ or __LITTLE_ENDIAN__.

OTOH, sparse doesn't know about the endianness of the kernel and
by default uses the same as the machine on which sparse was built.

Ensure that sparse can predefine the macros corresponding to
how the kernel was configured by adding -m{big,little}-endian
to CHECKFLAGS in the main Makefile (and so for all archs).
Also, remove the equivalent done in arch specific Makefiles.
Signed-off-by: NLuc Van Oostenryck <luc.vanoostenryck@gmail.com>
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

Currently, $(CHECK) receives NOSTDINC_FLAGS twice:
* first directly in the main Makefile via CHECKFLAGS,
* then indirectly in scripts/Makefile.build via c_flags.

Since once is enough, leave the occurence via c_flags and
remove the one via CHECKFLAGS.
Signed-off-by: NLuc Van Oostenryck <luc.vanoostenryck@gmail.com>
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

The kernel configuration phase is now tightly coupled with the compiler
in use.  It will be nice to show the compiler information in Kconfig.

The compiler information will be displayed like this:

  $ make ARCH=arm64 CROSS_COMPILE=aarch64-linux-gnu- config
  scripts/kconfig/conf  --oldaskconfig Kconfig
  *
  * Linux/arm64 4.16.0-rc1 Kernel Configuration
  *
  *
  * Compiler: aarch64-linux-gnu-gcc (Linaro GCC 7.2-2017.11) 7.2.1 20171011
  *
  *
  * General setup
  *
  Compile also drivers which will not load (COMPILE_TEST) [N/y/?]

If you use GUI methods such as menuconfig, it will be displayed in the
top menu.

This is simply implemented by using the 'comment' statement.  So, it
will be saved into the .config file as well.

This commit has a very important meaning.  If the compiler is upgraded,
Kconfig must be re-run since different compilers have different sets
of supported options.

All referenced environments are written to include/config/auto.conf.cmd
so that any environment change triggers syncconfig, and prompt the user
to input new values if needed.

With this commit, something like follows will be added to
include/config/auto.conf.cmd

  ifneq "$(CC_VERSION_TEXT)" "aarch64-linux-gnu-gcc (Linaro GCC 7.2-2017.11) 7.2.1 20171011"
  include/config/auto.conf: FORCE
  endif
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: NKees Cook <keescook@chromium.org>

Now that 'shell' function is supported, this can be self-contained in
Kconfig.
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: NKees Cook <keescook@chromium.org>
Reviewed-by: NUlf Magnusson <ulfalizer@gmail.com>

To get access to environment variables, Kconfig needs to define a
symbol using "option env=" syntax.  It is tedious to add a symbol entry
for each environment variable given that we need to define much more
such as 'CC', 'AS', 'srctree' etc. to evaluate the compiler capability
in Kconfig.

Adding '$' for symbol references is grammatically inconsistent.
Looking at the code, the symbols prefixed with 'S' are expanded by:
 - conf_expand_value()
   This is used to expand 'arch/$ARCH/defconfig' and 'defconfig_list'
 - sym_expand_string_value()
   This is used to expand strings in 'source' and 'mainmenu'

All of them are fixed values independent of user configuration.  So,
they can be changed into the direct expansion instead of symbols.

This change makes the code much cleaner.  The bounce symbols 'SRCARCH',
'ARCH', 'SUBARCH', 'KERNELVERSION' are gone.

sym_init() hard-coding 'UNAME_RELEASE' is also gone.  'UNAME_RELEASE'
should be replaced with an environment variable.

ARCH_DEFCONFIG is a normal symbol, so it should be simply referenced
without '$' prefix.

The new syntax is addicted by Make.  The variable reference needs
parentheses, like $(FOO), but you can omit them for single-letter
variables, like $F.  Yet, in Makefiles, people tend to use the
parenthetical form for consistency / clarification.

At this moment, only the environment variable is supported, but I will
extend the concept of 'variable' later on.

The variables are expanded in the lexer so we can simplify the token
handling on the parser side.

For example, the following code works.

[Example code]

  config MY_TOOLCHAIN_LIST
          string
          default "My tools: CC=$(CC), AS=$(AS), CPP=$(CPP)"

[Result]

  $ make -s alldefconfig && tail -n 1 .config
  CONFIG_MY_TOOLCHAIN_LIST="My tools: CC=gcc, AS=as, CPP=gcc -E"
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: NKees Cook <keescook@chromium.org>

Kbuild provides a couple of ways to specify CROSS_COMPILE:

[1] Command line
[2] Environment
[3] arch/*/Makefile (only some architectures)
[4] CONFIG_CROSS_COMPILE

[4] is problematic for the compiler capability tests in Kconfig.
CONFIG_CROSS_COMPILE allows users to change the compiler prefix from
'make menuconfig', etc.  It means, the compiler options would have
to be all re-calculated everytime CONFIG_CROSS_COMPILE is changed.

To avoid complexity and performance issues, I'd like to evaluate
the shell commands statically, i.e. only parsing Kconfig files.

I guess the majority is [1] or [2].  Currently, there are only
5 defconfig files that specify CONFIG_CROSS_COMPILE.
  arch/arm/configs/lpc18xx_defconfig
  arch/hexagon/configs/comet_defconfig
  arch/nds32/configs/defconfig
  arch/openrisc/configs/or1ksim_defconfig
  arch/openrisc/configs/simple_smp_defconfig
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: NKees Cook <keescook@chromium.org>

The kbuild cache was introduced to remember the result of shell
commands, some of which are expensive to compute, such as
$(call cc-option,...).

However, this turned out not so clever as I had first expected.
Actually, it is problematic.  For example, "$(CC) -print-file-name"
is cached.  If the compiler is updated, the stale search path causes
build error, which is difficult to figure out.  Another problem
scenario is cache files could be touched while install targets are
running under the root permission.  We can patch them if desired,
but the build infrastructure is getting uglier and uglier.

Now, we are going to move compiler flag tests to the configuration
phase.  If this is completed, the result of compiler tests will be
naturally cached in the .config file.  We will not have performance
issues of incremental building since this testing only happens at
Kconfig time.

To start this work with a cleaner code base, remove the kbuild
cache first.

Revert the following commits:
Commit 9a234a2e ("kbuild: create directory for make cache only when necessary")
Commit e17c400a ("kbuild: shrink .cache.mk when it exceeds 1000 lines")
Commit 4e562071 ("kbuild: Cache a few more calls to the compiler")
Commit 3298b690 ("kbuild: Add a cache for generated variables")
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: NKees Cook <keescook@chromium.org>

Modules do not tend to cope with -ffunction-sections, even though they
do not link with -gc-sections. It may be possible for unused symbols to
be trimmed from modules, but in general that would take much more work
in architecture module linker scripts.

For now, enable these only for kernel build.
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

CONFIG_HAVE_UNDERSCORE_SYMBOL_PREFIX was selected by BLACKFIN, METAG.
They were removed by commit 4ba66a97 ("arch: remove blackfin port"),
commit bb6fb6df ("metag: Remove arch/metag/"), respectively.

No more architecture enables CONFIG_HAVE_UNDERSCORE_SYMBOL_PREFIX,
hence the last argument of scripts/depmod.sh can be removed.
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: NSam Ravnborg <sam@ravnborg.org>

Since commit e501ce95 ("x86: Force asm-goto"), aarch64 build on
distributions which enable PIE by default (e.g. openSUSE Tumbleweed) does
not detect support for asm goto correctly. The problem is that ARM specific
part of scripts/gcc-goto.sh fails with PIE even with recent gcc versions.
Moving the asm goto detection up in Makefile put it before the place where
we disable PIE. As a result, kernel is built without jump label support.

Move the lines disabling PIE before the asm goto test to make it work.

Fixes: e501ce95 ("x86: Force asm-goto")
Reported-by: NAndreas Faerber <afaerber@suse.com>
Signed-off-by: NMichal Kubecek <mkubecek@suse.cz>
Acked-by: NPeter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

Clang does not recognize this compiler option.
Reported-by: NPrasad Sodagudi <psodagud@codeaurora.org>
Signed-off-by: NNick Desaulniers <ndesaulniers@google.com>
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

The __FILE__ macro is used everywhere in the kernel to locate the file
printing the log message, such as WARN_ON(), etc.  If the kernel is
built out of tree, this can be a long absolute path, like this:

  WARNING: CPU: 1 PID: 1 at /path/to/build/directory/arch/arm64/kernel/foo.c:...

This is because Kbuild runs in the objtree instead of the srctree,
then __FILE__ is expanded to a file path prefixed with $(srctree)/.

Commit 9da0763b ("kbuild: Use relative path when building in a
subdir of the source tree") improved this to some extent; $(srctree)
becomes ".." if the objtree is a child of the srctree.

For other cases of out-of-tree build, __FILE__ is still the absolute
path.  It also means the kernel image depends on where it was built.

A brand-new option from GCC, -fmacro-prefix-map, solves this problem.
If your compiler supports it, __FILE__ is the relative path from the
srctree regardless of O= option.  This provides more readable log and
more reproducible builds.

Please note __FILE__ is always an absolute path for external modules.
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

Our convention is to distinguish file types by suffixes with a period
as a separator.

*-asn1.[ch] is a different pattern from other generated sources such
as *.lex.c, *.tab.[ch], *.dtb.S, etc.  More confusing, files with
'-asn1.[ch]' are generated files, but '_asn1.[ch]' are checked-in
files:
  net/netfilter/nf_conntrack_h323_asn1.c
  include/linux/netfilter/nf_conntrack_h323_asn1.h
  include/linux/sunrpc/gss_asn1.h

Rename generated files to *.asn1.[ch] for consistency.
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

Clean up these patterns from the top Makefile to omit 'clean-files'
in each Makefile.
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

Files suffixed by .lex.c, .tab.[ch] are generated lexers, parsers,
respectively.  Clean them up globally from the top Makefile.

Some of the final host programs those lexer/parser are linked into
are necessary for building external modules, but the intermediates
are unneeded.  They can be cleaned away by 'make clean' instead of
'make mrproper'.
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: NFrank Rowand <frowand.list@gmail.com>

The variable 'PYTHON' allows users to specify a proper executable
name in case the default 'python' does not work.  However, this does
not address the case where both Python 2.x and 3.x scripts are used
in one source tree.

PEP 394 (https://www.python.org/dev/peps/pep-0394/) provides a
convention for Python scripts portability.  Here is a quotation:

  In order to tolerate differences across platforms, all new code
  that needs to invoke the Python interpreter should not specify
  'python', but rather should specify either 'python2' or 'python3'.
  This distinction should be made in shebangs, when invoking from a
  shell script, when invoking via the system() call, or when invoking
  in any other context.
  One exception to this is scripts that are deliberately written to
  be source compatible with both Python 2.x and 3.x. Such scripts may
  continue to use python on their shebang line without affecting their
  portability.

To meet this requirement, this commit adds new variables 'PYTHON2'
and 'PYTHON3'.

arch/ia64/scripts/unwcheck.py is the only script that has ever used
$(PYTHON).  Recent commit bd5edbe6 ("ia64: convert unwcheck.py to
python3") converted it to be compatible with both Python 2.x and 3.x,
so this is the exceptional case where the use of 'python' is allowed.
So, I did not touch arch/ia64/Makefile.

tools/perf/Makefile.config sets PYTHON and PYTHON2 by itself, so it
is not affected by this commit.
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

As commit cedd55d4 ("kconfig: Remove silentoldconfig from help
and docs; fix kconfig/conf's help") mentioned, 'silentoldconfig' is a
historical misnomer.  That commit removed it from help and docs since
it is an internal interface.  If so, it should be allowed to rename
it to something more intuitive.  'syncconfig' is the one I came up
with because it updates the .config if necessary, then synchronize
include/generated/autoconf.h and include/config/* with it.

You should not manually invoke 'silentoldcofig'.  Display warning if
used in case existing scripts are doing wrong.
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: NUlf Magnusson <ulfalizer@gmail.com>

If CONFIG_TRIM_UNUSED_KSYMS is enabled and the kernel is built from
a pristine state, the vmlinux is linked twice.

[1] A user runs 'make'

[2] First build with empty autoksyms.h

[3] adjust_autoksyms.sh updates autoksyms.h and recurses 'make vmlinux'

  --------(begin sub-make)--------
  [4] Second build with new autoksyms.h

  [5] link-vmlinux.sh is invoked because vmlinux is missing
  ---------(end sub-make)---------

[6] link-vmlinux.sh is invoked again despite vmlinux is up-to-date.

The reason of [6] is probably because Make already decided to update
vmlinux at the time of [2] because vmlinux was missing when Make
built up the dependency graph.

Because if_changed is implemented based on $?, this issue can be
narrowed down to how Make handles $?.

You can test it with the following simple code:

[Test Makefile]
  A: B
          @echo newer prerequisite: $?
          cp B A

  B: C
          cp C B
          touch A

[Result]
  $ rm -f A B
  $ touch C
  $ make
  cp C B
  touch A
  newer prerequisite: B
  cp B A

Here, 'A' has been touched in the recipe of 'B'.  So, the dependency
'A: B' has already been met before the recipe of 'A' is executed.
However, Make does not notice the fact that the recipe of 'B' also
updates 'A' as a side-effect.

The situation is similar in this case; the vmlinux has actually been
updated in the vmlinux_prereq target.  Make cannot predict this, so
judges the vmlinux is old.

link-vmlinux.sh is costly, so it is better to not run it when unneeded.
Split CONFIG_TRIM_UNUSED_KSYMS recursion to a dedicated target.

The reason of commit 2441e78b ("kbuild: better abstract vmlinux
sequential prerequisites") was to cater to CONFIG_BUILD_DOCSRC, but
it was later removed by commit 18489292 ("samples: move blackfin
gptimers-example from Documentation").
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: NNicolas Pitre <nico@linaro.org>

The idea of using fixdep was inspired by Kconfig, but autoksyms
belongs to a different group.  So, I want to move those touched
files under include/config/ksym/ to include/ksym/.

The directory include/ksym/ can be removed by 'make clean' because
it is meaningless for the external module building.
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: NNicolas Pitre <nico@linaro.org>

The external module building does not need to parse this code because
KBUILD_MODULES is always set anyway.

Move this code inside the "ifeq ($(KBUILD_EXTMOD),) ... endif" block.
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: NNicolas Pitre <nico@linaro.org>

Commit d3fc425e ("kbuild: make sure autoksyms.h exists early")
moved the code that touches autoksyms.h to scripts/kconfig/Makefile
with obscure reason.

From Nicolas' comment [1], he did not seem to be sure about the root
cause.

I guess I figured it out, so here is a fix-up I think is more correct.
According to the error log in the original post [2], the build failed
in scripts/mod/devicetable-offsets.c

scripts/mod/Makefile is descended from scripts/Makefile, which is
invoked from the top-level Makefile by the 'scripts' target.

To build vmlinux and/or modules, Kbuild descend into $(vmlinux-dirs).
This depends on 'prepare' and 'scripts' as follows:

  $(vmlinux-dirs): prepare scripts

Because there is no dependency between 'prepare' and 'scripts', the
parallel building can execute them simultaneously.

'prepare' depends on 'prepare1', which touched autoksyms.h, while
'scripts' descends into script/, then scripts/mod/, which needs
<generated/autoksyms.h> if CONFIG_TRIM_UNUSED_KSYMS.  It was the
reason of the race.

I am not happy to have unrelated code in the Kconfig Makefile, so
getting it back to the top Makefile.

I removed the standalone test target because I want to use it to
create an empty autoksyms.h file.  Here is a little improvement;
unnecessary autoksyms.h is not created when CONFIG_TRIM_UNUSED_KSYMS
is disabled.

[1] https://lkml.org/lkml/2016/11/30/734
[2] https://lkml.org/lkml/2016/11/30/531Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: NNicolas Pitre <nico@linaro.org>

Just a trivial change to prepare for the next commit.
This target is still invisible from external module building.
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

Currently LDFLAGS is not cleared, so same flags are accumulated in
LDFLAGS when the top Makefile is recursively invoked.

I found unneeded rebuild for ARCH=arm64 when CONFIG_TRIM_UNUSED_KSYMS
is enabled.  If include/generated/autoksyms.h is updated, the top
Makefile is recursively invoked, then arch/arm64/Makefile adds one
more '-maarch64linux'.  Due to the command line change, modules are
rebuilt needlessly.
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: NNicolas Pitre <nico@linaro.org>

Support parallel building of clean, config, and build targets in a
single command.

For example,

  make -j<N> clean all

or

  make -j<N> mrproper defconfig all

They should be handled one by one.
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

Incremental linking is gone, so rename built-in.o to built-in.a, which
is the usual extension for archive files.

This patch does two things, first is a simple search/replace:

git grep -l 'built-in\.o' | xargs sed -i 's/built-in\.o/built-in\.a/g'

The second is to invert nesting of nested text manipulations to avoid
filtering built-in.a out from libs-y2:

-libs-y2 := $(filter-out %.a, $(patsubst %/, %/built-in.a, $(libs-y)))
+libs-y2 := $(patsubst %/, %/built-in.a, $(filter-out %.a, $(libs-y)))
Signed-off-by: NNicholas Piggin <npiggin@gmail.com>
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

In order to make sure compiler flag detection for ARM works
correctly the no-integrated-as flags need to be set before
including the arch specific Makefile.

Fixes: cfe17c9b ("kbuild: move cc-option and cc-disable-warning after incl. arch Makefile")
Signed-off-by: NStefan Agner <stefan@agner.ch>
Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>

Prasad reported that he has seen crashes in BPF subsystem with netd
on Android with arm64 in the form of (note, the taint is unrelated):

  [ 4134.721483] Unable to handle kernel paging request at virtual address 800000001
  [ 4134.820925] Mem abort info:
  [ 4134.901283]   Exception class = DABT (current EL), IL = 32 bits
  [ 4135.016736]   SET = 0, FnV = 0
  [ 4135.119820]   EA = 0, S1PTW = 0
  [ 4135.201431] Data abort info:
  [ 4135.301388]   ISV = 0, ISS = 0x00000021
  [ 4135.359599]   CM = 0, WnR = 0
  [ 4135.470873] user pgtable: 4k pages, 39-bit VAs, pgd = ffffffe39b946000
  [ 4135.499757] [0000000800000001] *pgd=0000000000000000, *pud=0000000000000000
  [ 4135.660725] Internal error: Oops: 96000021 [#1] PREEMPT SMP
  [ 4135.674610] Modules linked in:
  [ 4135.682883] CPU: 5 PID: 1260 Comm: netd Tainted: G S      W       4.14.19+ #1
  [ 4135.716188] task: ffffffe39f4aa380 task.stack: ffffff801d4e0000
  [ 4135.731599] PC is at bpf_prog_add+0x20/0x68
  [ 4135.741746] LR is at bpf_prog_inc+0x20/0x2c
  [ 4135.751788] pc : [<ffffff94ab7ad584>] lr : [<ffffff94ab7ad638>] pstate: 60400145
  [ 4135.769062] sp : ffffff801d4e3ce0
  [...]
  [ 4136.258315] Process netd (pid: 1260, stack limit = 0xffffff801d4e0000)
  [ 4136.273746] Call trace:
  [...]
  [ 4136.442494] 3ca0: ffffff94ab7ad584 0000000060400145 ffffffe3a01bf8f8 0000000000000006
  [ 4136.460936] 3cc0: 0000008000000000 ffffff94ab844204 ffffff801d4e3cf0 ffffff94ab7ad584
  [ 4136.479241] [<ffffff94ab7ad584>] bpf_prog_add+0x20/0x68
  [ 4136.491767] [<ffffff94ab7ad638>] bpf_prog_inc+0x20/0x2c
  [ 4136.504536] [<ffffff94ab7b5d08>] bpf_obj_get_user+0x204/0x22c
  [ 4136.518746] [<ffffff94ab7ade68>] SyS_bpf+0x5a8/0x1a88

Android's netd was basically pinning the uid cookie BPF map in BPF
fs (/sys/fs/bpf/traffic_cookie_uid_map) and later on retrieving it
again resulting in above panic. Issue is that the map was wrongly
identified as a prog! Above kernel was compiled with clang 4.0,
and it turns out that clang decided to merge the bpf_prog_iops and
bpf_map_iops into a single memory location, such that the two i_ops
could then not be distinguished anymore.

Reason for this miscompilation is that clang has the more aggressive
-fmerge-all-constants enabled by default. In fact, clang source code
has a comment about it in lib/AST/ExprConstant.cpp on why it is okay
to do so:

  Pointers with different bases cannot represent the same object.
  (Note that clang defaults to -fmerge-all-constants, which can
  lead to inconsistent results for comparisons involving the address
  of a constant; this generally doesn't matter in practice.)

The issue never appeared with gcc however, since gcc does not enable
-fmerge-all-constants by default and even *explicitly* states in
it's option description that using this flag results in non-conforming
behavior, quote from man gcc:

  Languages like C or C++ require each variable, including multiple
  instances of the same variable in recursive calls, to have distinct
  locations, so using this option results in non-conforming behavior.

There are also various clang bug reports open on that matter [1],
where clang developers acknowledge the non-conforming behavior,
and refer to disabling it with -fno-merge-all-constants. But even
if this gets fixed in clang today, there are already users out there
that triggered this. Thus, fix this issue by explicitly adding
-fno-merge-all-constants to the kernel's Makefile to generically
disable this optimization, since potentially other places in the
kernel could subtly break as well.

Note, there is also a flag called -fmerge-constants (not supported
by clang), which is more conservative and only applies to strings
and it's enabled in gcc's -O/-O2/-O3/-Os optimization levels. In
gcc's code, the two flags -fmerge-{all-,}constants share the same
variable internally, so when disabling it via -fno-merge-all-constants,
then we really don't merge any const data (e.g. strings), and text
size increases with gcc (14,927,214 -> 14,942,646 for vmlinux.o).

  $ gcc -fverbose-asm -O2 foo.c -S -o foo.S
    -> foo.S lists -fmerge-constants under options enabled
  $ gcc -fverbose-asm -O2 -fno-merge-all-constants foo.c -S -o foo.S
    -> foo.S doesn't list -fmerge-constants under options enabled
  $ gcc -fverbose-asm -O2 -fno-merge-all-constants -fmerge-constants foo.c -S -o foo.S
    -> foo.S lists -fmerge-constants under options enabled

Thus, as a workaround we need to set both -fno-merge-all-constants
*and* -fmerge-constants in the Makefile in order for text size to
stay as is.

  [1] https://bugs.llvm.org/show_bug.cgi?id=18538Reported-by: NPrasad Sodagudi <psodagud@codeaurora.org>
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Chenbo Feng <fengc@google.com>
Cc: Richard Smith <richard-llvm@metafoo.co.uk>
Cc: Chandler Carruth <chandlerc@gmail.com>
Cc: linux-kernel@vger.kernel.org
Tested-by: NPrasad Sodagudi <psodagud@codeaurora.org>
Acked-by: NAlexei Starovoitov <ast@kernel.org>
Signed-off-by: NAlexei Starovoitov <ast@kernel.org>

We want to start using asm-goto to guarantee the absence of dynamic
branches (and thus speculation).

A primary prerequisite for this is of course that the compiler
supports asm-goto. This effecively lifts the minimum GCC version to
build an x86 kernel to gcc-4.5.
Signed-off-by: NPeter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Cc: torvalds@linux-foundation.org
Link: https://lkml.kernel.org/r/20180319201327.GJ4043@hirez.programming.kicks-ass.net