queue-3.18/arm-8833-1-ensure-that-neon-code-always-compiles-wit.patch

   1 From ffb91ff7e5711dfd82ea69cba1744a9041acae2b Mon Sep 17 00:00:00 2001
   2 From: Nathan Chancellor <natechancellor@gmail.com>
   3 Date: Sat, 2 Feb 2019 03:34:36 +0100
   4 Subject: ARM: 8833/1: Ensure that NEON code always compiles with Clang
   5
   6 [ Upstream commit de9c0d49d85dc563549972edc5589d195cd5e859 ]
   7
   8 While building arm32 allyesconfig, I ran into the following errors:
   9
  10   arch/arm/lib/xor-neon.c:17:2: error: You should compile this file with
  11   '-mfloat-abi=softfp -mfpu=neon'
  12
  13   In file included from lib/raid6/neon1.c:27:
  14   /home/nathan/cbl/prebuilt/lib/clang/8.0.0/include/arm_neon.h:28:2:
  15   error: "NEON support not enabled"
  16
  17 Building V=1 showed NEON_FLAGS getting passed along to Clang but
  18 __ARM_NEON__ was not getting defined. Ultimately, it boils down to Clang
  19 only defining __ARM_NEON__ when targeting armv7, rather than armv6k,
  20 which is the '-march' value for allyesconfig.
  21
  22 >From lib/Basic/Targets/ARM.cpp in the Clang source:
  23
  24   // This only gets set when Neon instructions are actually available, unlike
  25   // the VFP define, hence the soft float and arch check. This is subtly
  26   // different from gcc, we follow the intent which was that it should be set
  27   // when Neon instructions are actually available.
  28   if ((FPU & NeonFPU) && !SoftFloat && ArchVersion >= 7) {
  29     Builder.defineMacro("__ARM_NEON", "1");
  30     Builder.defineMacro("__ARM_NEON__");
  31     // current AArch32 NEON implementations do not support double-precision
  32     // floating-point even when it is present in VFP.
  33     Builder.defineMacro("__ARM_NEON_FP",
  34                         "0x" + Twine::utohexstr(HW_FP & ~HW_FP_DP));
  35   }
  36
  37 Ard Biesheuvel recommended explicitly adding '-march=armv7-a' at the
  38 beginning of the NEON_FLAGS definitions so that __ARM_NEON__ always gets
  39 definined by Clang. This doesn't functionally change anything because
  40 that code will only run where NEON is supported, which is implicitly
  41 armv7.
  42
  43 Link: https://github.com/ClangBuiltLinux/linux/issues/287
  44
  45 Suggested-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
  46 Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
  47 Acked-by: Nicolas Pitre <nico@linaro.org>
  48 Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
  49 Reviewed-by: Stefan Agner <stefan@agner.ch>
  50 Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
  51 Signed-off-by: Sasha Levin <sashal@kernel.org>
  52 ---
  53  Documentation/arm/kernel_mode_neon.txt | 4 ++--
  54  arch/arm/lib/Makefile                  | 2 +-
  55  arch/arm/lib/xor-neon.c                | 2 +-
  56  lib/raid6/Makefile                     | 2 +-
  57  4 files changed, 5 insertions(+), 5 deletions(-)
  58
  59 diff --git a/Documentation/arm/kernel_mode_neon.txt b/Documentation/arm/kernel_mode_neon.txt
  60 index 525452726d31..b9e060c5b61e 100644
  61 --- a/Documentation/arm/kernel_mode_neon.txt
  62 +++ b/Documentation/arm/kernel_mode_neon.txt
  63 @@ -6,7 +6,7 @@ TL;DR summary
  64  * Use only NEON instructions, or VFP instructions that don't rely on support
  65    code
  66  * Isolate your NEON code in a separate compilation unit, and compile it with
  67 -  '-mfpu=neon -mfloat-abi=softfp'
  68 +  '-march=armv7-a -mfpu=neon -mfloat-abi=softfp'
  69  * Put kernel_neon_begin() and kernel_neon_end() calls around the calls into your
  70    NEON code
  71  * Don't sleep in your NEON code, and be aware that it will be executed with
  72 @@ -87,7 +87,7 @@ instructions appearing in unexpected places if no special care is taken.
  73  Therefore, the recommended and only supported way of using NEON/VFP in the
  74  kernel is by adhering to the following rules:
  75  * isolate the NEON code in a separate compilation unit and compile it with
  76 -  '-mfpu=neon -mfloat-abi=softfp';
  77 +  '-march=armv7-a -mfpu=neon -mfloat-abi=softfp';
  78  * issue the calls to kernel_neon_begin(), kernel_neon_end() as well as the calls
  79    into the unit containing the NEON code from a compilation unit which is *not*
  80    built with the GCC flag '-mfpu=neon' set.
  81 diff --git a/arch/arm/lib/Makefile b/arch/arm/lib/Makefile
  82 index 0573faab96ad..830b2ddcc346 100644
  83 --- a/arch/arm/lib/Makefile
  84 +++ b/arch/arm/lib/Makefile
  85 @@ -46,7 +46,7 @@ $(obj)/csumpartialcopy.o:     $(obj)/csumpartialcopygeneric.S
  86  $(obj)/csumpartialcopyuser.o:  $(obj)/csumpartialcopygeneric.S
  87
  88  ifeq ($(CONFIG_KERNEL_MODE_NEON),y)
  89 -  NEON_FLAGS                   := -mfloat-abi=softfp -mfpu=neon
  90 +  NEON_FLAGS                   := -march=armv7-a -mfloat-abi=softfp -mfpu=neon
  91    CFLAGS_xor-neon.o            += $(NEON_FLAGS)
  92    obj-$(CONFIG_XOR_BLOCKS)     += xor-neon.o
  93  endif
  94 diff --git a/arch/arm/lib/xor-neon.c b/arch/arm/lib/xor-neon.c
  95 index 2c40aeab3eaa..c691b901092f 100644
  96 --- a/arch/arm/lib/xor-neon.c
  97 +++ b/arch/arm/lib/xor-neon.c
  98 @@ -14,7 +14,7 @@
  99  MODULE_LICENSE("GPL");
 100
 101  #ifndef __ARM_NEON__
 102 -#error You should compile this file with '-mfloat-abi=softfp -mfpu=neon'
 103 +#error You should compile this file with '-march=armv7-a -mfloat-abi=softfp -mfpu=neon'
 104  #endif
 105
 106  /*
 107 diff --git a/lib/raid6/Makefile b/lib/raid6/Makefile
 108 index c7dab0645554..1c3330dc5eab 100644
 109 --- a/lib/raid6/Makefile
 110 +++ b/lib/raid6/Makefile
 111 @@ -23,7 +23,7 @@ endif
 112  ifeq ($(CONFIG_KERNEL_MODE_NEON),y)
 113  NEON_FLAGS := -ffreestanding
 114  ifeq ($(ARCH),arm)
 115 -NEON_FLAGS += -mfloat-abi=softfp -mfpu=neon
 116 +NEON_FLAGS += -march=armv7-a -mfloat-abi=softfp -mfpu=neon
 117  endif
 118  ifeq ($(ARCH),arm64)
 119  CFLAGS_REMOVE_neon1.o += -mgeneral-regs-only
 120 --
 121 2.19.1
 122