linux.git/include/linux/compiler_types.h, branch v5.5.12

Merge tag 'compiler-attributes-for-linus-v5.4' of git://github.com/ojeda/linux

2019-09-21T16:47:19+00:00

Pull asm inline support from Miguel Ojeda:
 "Make use of gcc 9's "asm inline()" (Rasmus Villemoes):

  gcc 9+ (and gcc 8.3, 7.5) provides a way to override the otherwise
  crude heuristic that gcc uses to estimate the size of the code
  represented by an asm() statement. From the gcc docs

      If you use 'asm inline' instead of just 'asm', then for inlining
      purposes the size of the asm is taken as the minimum size, ignoring
      how many instructions GCC thinks it is.

  For compatibility with older compilers, we obviously want a

      #if [understands asm inline]
      #define asm_inline asm inline
      #else
      #define asm_inline asm
      #endif

  But since we #define the identifier inline to attach some attributes,
  we have to use an alternate spelling of that keyword. gcc provides
  both __inline__ and __inline, and we currently #define both to inline,
  so they all have the same semantics.

  We have to free up one of __inline__ and __inline, and the latter is
  by far the easiest.

  The two x86 changes cause smaller code gen differences than I'd
  expect, but I think we do want the asm_inline thing available sooner
  or later, so this is just to get the ball rolling"

* tag 'compiler-attributes-for-linus-v5.4' of git://github.com/ojeda/linux:
  x86: bug.h: use asm_inline in _BUG_FLAGS definitions
  x86: alternative.h: use asm_inline for all alternative variants
  compiler-types.h: add asm_inline definition
  compiler_types.h: don't #define __inline
  lib/zstd/mem.h: replace __inline by inline
  staging: rtl8723bs: replace __inline by inline

compiler-types.h: add asm_inline definition

2019-09-15T18:14:15+00:00

This adds an asm_inline macro which expands to "asm inline" [1] when
the compiler supports it. This is currently gcc 9.1+, gcc 8.3
and (once released) gcc 7.5 [2]. It expands to just "asm" for other
compilers.

Using asm inline("foo") instead of asm("foo") overrules gcc's
heuristic estimate of the size of the code represented by the asm()
statement, and makes gcc use the minimum possible size instead. That
can in turn affect gcc's inlining decisions.

I wasn't sure whether to make this a function-like macro or not - this
way, it can be combined with volatile as

  asm_inline volatile()

but perhaps we'd prefer to spell that

  asm_inline_volatile()

anyway.

The Kconfig logic is taken from an RFC patch by Masahiro Yamada [3].

[1] Technically, asm __inline, since both inline and __inline__
are macros that attach various attributes, making gcc barf if one
literally does "asm inline()". However, the third spelling __inline is
available for referring to the bare keyword.

[2] https://lore.kernel.org/lkml/20190907001411.GG9749@gate.crashing.org/

[3] https://lore.kernel.org/lkml/1544695154-15250-1-git-send-email-yamada.masahiro@socionext.com/

Signed-off-by: Rasmus Villemoes 
Signed-off-by: Miguel Ojeda

compiler_types.h: don't #define __inline

2019-09-15T18:14:15+00:00

The spellings __inline and __inline__ should be reserved for uses
where one really wants to refer to the inline keyword, regardless of
whether or not the spelling "inline" has been #defined to something
else. Due to use of __inline__ in uapi headers, we can't easily get
rid of the definition of __inline__. However, almost all users of
__inline have been converted to inline, so we can get rid of that
#define.

The exception is include/acpi/platform/acintel.h. However, that header
is only included when using the intel compiler (does anybody actually
build the kernel with that?), and the ACPI_INLINE macro is only used
in the definition of utterly trivial stub functions, where I doubt a
small change of semantics (lack of __gnu_inline) changes anything.

Signed-off-by: Rasmus Villemoes 
[Fix trivial typo in message]
Signed-off-by: Miguel Ojeda

kbuild: allow Clang to find unused static inline functions for W=1 build

2019-09-09T14:55:43+00:00

GCC and Clang have different policy for -Wunused-function; GCC does not
warn unused static inline functions at all whereas Clang does if they
are defined in source files instead of included headers although it has
been suppressed since commit abb2ea7dfd82 ("compiler, clang: suppress
warning for unused static inline functions").

We often miss to delete unused functions where 'static inline' is used
in *.c files since there is no tool to detect them. Unused code remains
until somebody notices. For example, commit 075ddd75680f ("regulator:
core: remove unused rdev_get_supply()").

Let's remove __maybe_unused from the inline macro to allow Clang to
start finding unused static inline functions. For now, we do this only
for W=1 build since it is not a good idea to sprinkle warnings for the
normal build (e.g. 35 warnings for arch/x86/configs/x86_64_defconfig).

My initial attempt was to add -Wno-unused-function for no W= build
(https://lore.kernel.org/patchwork/patch/1120594/)

Nathan Chancellor pointed out that would weaken Clang's checks since
we would no longer get -Wunused-function without W=1. It is true GCC
would catch unused static non-inline functions, but it would weaken
Clang as a standalone compiler, at least.

Hence, here is a counter implementation. The current problem is, W=...
only controls compiler flags, which are globally effective. There is
no way to address only 'static inline' functions.

This commit defines KBUILD_EXTRA_WARN[123] corresponding to W=[123].
When KBUILD_EXTRA_WARN1 is defined, __maybe_unused is omitted from
the 'inline' macro.

The new macro __inline_maybe_unused makes the code a bit uglier, so I
hope we can remove it entirely after fixing most of the warnings.

If you contribute to code clean-up, please run "make CC=clang W=1"
and check -Wunused-function warnings. You will find lots of unused
functions.

Some of them are false-positives because the call-sites are disabled
by #ifdef. I do not like to abuse the inline keyword for suppressing
unused-function warnings because it is intended to be a hint for the
compiler optimization. I prefer #ifdef around the definition, or
__maybe_unused if #ifdef would make the code too ugly.

Signed-off-by: Masahiro Yamada 
Reviewed-by: Nathan Chancellor 
Tested-by: Nathan Chancellor

bpf: Disable GCC -fgcse optimization for ___bpf_prog_run()

2019-07-18T19:01:06+00:00

On x86-64, with CONFIG_RETPOLINE=n, GCC's "global common subexpression
elimination" optimization results in ___bpf_prog_run()'s jumptable code
changing from this:

select_insn:
jmp *jumptable(, %rax, 8)
...
ALU64_ADD_X:
...
jmp *jumptable(, %rax, 8)
ALU_ADD_X:
...
jmp *jumptable(, %rax, 8)

to this:

select_insn:
mov jumptable, %r12
jmp *(%r12, %rax, 8)
...
ALU64_ADD_X:
...
jmp *(%r12, %rax, 8)
ALU_ADD_X:
...
jmp *(%r12, %rax, 8)

The jumptable address is placed in a register once, at the beginning of
the function. The function execution can then go through multiple
indirect jumps which rely on that same register value. This has a few
issues:

1) Objtool isn't smart enough to be able to track such a register value
across multiple recursive indirect jumps through the jump table.

2) With CONFIG_RETPOLINE enabled, this optimization actually results in
a small slowdown. I measured a ~4.7% slowdown in the test_bpf
"tcpdump port 22" selftest.

This slowdown is actually predicted by the GCC manual:

Note: When compiling a program using computed gotos, a GCC
extension, you may get better run-time performance if you
disable the global common subexpression elimination pass by
adding -fno-gcse to the command line.

So just disable the optimization for this function.

Fixes: e55a73251da3 ("bpf: Fix ORC unwinding in non-JIT BPF code")
Reported-by: Randy Dunlap
Signed-off-by: Josh Poimboeuf
Signed-off-by: Thomas Gleixner
Acked-by: Alexei Starovoitov
Acked-by: Peter Zijlstra (Intel)
Link: https://lkml.kernel.org/r/30c3ca29ba037afcbd860a8672eef0021addf9fe.1563413318.git.jpoimboe@redhat.com

compiler.h: add CC_USING_PATCHABLE_FUNCTION_ENTRY

2019-06-08T10:56:28+00:00

This can be used for architectures implementing dynamic
ftrace via -fpatchable-function-entry.

Signed-off-by: Sven Schnelle 
Signed-off-by: Helge Deller

compiler: allow all arches to enable CONFIG_OPTIMIZE_INLINING

2019-05-15T02:52:48+00:00

Commit 60a3cdd06394 ("x86: add optimized inlining") introduced
CONFIG_OPTIMIZE_INLINING, but it has been available only for x86.

The idea is obviously arch-agnostic.  This commit moves the config entry
from arch/x86/Kconfig.debug to lib/Kconfig.debug so that all
architectures can benefit from it.

This can make a huge difference in kernel image size especially when
CONFIG_OPTIMIZE_FOR_SIZE is enabled.

For example, I got 3.5% smaller arm64 kernel for v5.1-rc1.

  dec       file
  18983424  arch/arm64/boot/Image.before
  18321920  arch/arm64/boot/Image.after

This also slightly improves the "Kernel hacking" Kconfig menu as
e61aca5158a8 ("Merge branch 'kconfig-diet' from Dave Hansen') suggested;
this config option would be a good fit in the "compiler option" menu.

Link: http://lkml.kernel.org/r/20190423034959.13525-12-yamada.masahiro@socionext.com
Signed-off-by: Masahiro Yamada 
Acked-by: Borislav Petkov 
Cc: Arnd Bergmann 
Cc: Benjamin Herrenschmidt 
Cc: Boris Brezillon 
Cc: Brian Norris 
Cc: Christophe Leroy 
Cc: David Woodhouse 
Cc: Heiko Carstens 
Cc: "H. Peter Anvin" 
Cc: Ingo Molnar 
Cc: Marek Vasut 
Cc: Mark Rutland 
Cc: Mathieu Malaterre 
Cc: Miquel Raynal 
Cc: Paul Mackerras 
Cc: Ralf Baechle 
Cc: Richard Weinberger 
Cc: Russell King 
Cc: Stefan Agner 
Cc: Thomas Gleixner 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

include/linux/compiler_types.h: don't pollute userspace with macro definitions

2018-12-14T15:57:16+00:00

Macros 'inline' and '__gnu_inline' used to be defined in compiler-gcc.h,
which was (and is) included entirely in (__KERNEL__ && !__ASSEMBLY__).
Commit 815f0ddb346c ("include/linux/compiler*.h: make compiler-*.h mutually
exclusive") had those macros exposed to userspace, unintentionally.

Then commit a3f8a30f3f00 ("Compiler Attributes: use feature checks
instead of version checks") moved '__gnu_inline' back into
(__KERNEL__ && !__ASSEMBLY__) and 'inline' was left behind. Since 'inline'
depends on '__gnu_inline', compiling error showing "unknown type name
‘__gnu_inline’" will pop up, if userspace somehow includes
.

Other macros like __must_check, notrace, etc. are in a similar situation.
So just move all these macros back into (__KERNEL__ && !__ASSEMBLY__).

Note:
  1. This patch only affects what userspace sees.
  2. __must_check (when !CONFIG_ENABLE_MUST_CHECK) and noinline_for_stack
     were once defined in __KERNEL__ only, but we believe that they can
     be put into !__ASSEMBLY__ too.

Acked-by: Nick Desaulniers 
Signed-off-by: Xiaozhou Liu 
Signed-off-by: Miguel Ojeda

include/linux/compiler*.h: define asm_volatile_goto

2018-11-06T14:30:16+00:00

asm_volatile_goto should also be defined for other compilers that support
asm goto.

Fixes commit 815f0ddb346c ("include/linux/compiler*.h: make compiler-*.h
mutually exclusive").

Signed-off-by: Nick Desaulniers 
Signed-off-by: Miguel Ojeda

Compiler Attributes: use feature checks instead of version checks

2018-09-30T18:14:03+00:00

Instead of using version checks per-compiler to define (or not)
each attribute, use __has_attribute to test for them, following
the cleanup started with commit 815f0ddb346c
("include/linux/compiler*.h: make compiler-*.h mutually exclusive"),
which is supported on gcc >= 5, clang >= 2.9 and icc >= 17.
In the meantime, to support 4.6 <= gcc < 5, we implement
__has_attribute by hand.

All the attributes that can be unconditionally defined and directly
map to compiler attribute(s) (even if optional) have been moved
to a new file include/linux/compiler_attributes.h

In an effort to make the file as regular as possible, comments
stating the purpose of attributes have been removed. Instead,
links to the compiler docs have been added (i.e. to gcc and,
if available, to clang as well). In addition, they have been sorted.

Finally, if an attribute is optional (i.e. if it is guarded
by __has_attribute), the reason has been stated for future reference.

Tested-by: Sedat Dilek  # on top of v4.19-rc5, clang 7
Reviewed-by: Nick Desaulniers 
Reviewed-by: Luc Van Oostenryck 
Signed-off-by: Miguel Ojeda