]>
Commit | Line | Data |
---|---|---|
9a904337 SL |
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 |