building Linux for ARMv7 with llvm breaks sparse

building Linux for ARMv7 with llvm breaks sparse

[Marc Kleine-Budde]

[-- Attachment #1: Type: text/plain, Size: 2713 bytes --]

Hello,

building a 32bit ARM kernel with llvm (LLVM=1 LLVM_IAS=1) and C=1 breaks
current sparse:

| sparse -D__linux__ -Dlinux -D__STDC__ -Dunix -D__unix__ -Wbitwise
| -Wno-return-void -Wno-unknown-attribute -D__ARMEL__ -D__arm__
| --arch=arm -mlittle-endian -m32 -Wp,-MMD,scripts/mod/.empty.o.d
| -nostdinc -I/path/to/linux/arch/arm/include
| -I./arch/arm/include/generated
| -I/path/to/linux/include -I./include
| -I/path/to/linux/arch/arm/include/uapi
| -I./arch/arm/include/generated/uapi
| -I/path/to/linux/include/uapi
| -I./include/generated/uapi -include
| /path/to/linux/include/linux/compiler-version.h
| -include /path/to/linux/include/linux/kconfig.h
| -include
| /path/to/linux/include/linux/compiler_types.h
| -D__KERNEL__ -mlittle-endian -Qunused-arguments
| -fmacro-prefix-map=/path/to/linux/=
| -DKBUILD_EXTRA_WARN1 -Wall -Wundef -Werror=strict-prototypes
| -Wno-trigraphs -fno-strict-aliasing -fno-common -fshort-wchar -fno-PIE
| -Werror=implicit-function-declaration -Werror=implicit-int
| -Werror=return-type -Wno-format-security -std=gnu11
| --target=arm-linux-gnueabi -fintegrated-as
| -Werror=unknown-warning-option -Werror=ignored-optimization-argument
| -fno-dwarf2-cfi-asm -mabi=aapcs-linux -mfpu=vfp -funwind-tables
| -meabi gnu -marm -Wa,-W -D__LINUX_ARM_ARCH__=6 -march=armv6k
  ^^^^^^^^^^
| -mtune=arm1136j-s -msoft-float -Uarm -fno-delete-null-pointer-checks
| -Wno-frame-address -Wno-address-of-packed-member -O2
| -Wframe-larger-than=1024 -fstack-protector-strong -Werror -Wno-gnu
| -mno-global-merge -Wno-unused-but-set-variable
| -Wno-unused-const-variable -fomit-frame-pointer
| -fno-stack-clash-protection -Wdeclaration-after-statement -Wvla
| -Wno-pointer-sign -Wcast-function-type -fno-strict-overflow
| -fno-stack-check -Werror=date-time -Werror=incompatible-pointer-types
| -Wextra -Wunused -Wno-unused-parameter -Wmissing-declarations
| -Wmissing-format-attribute -Wmissing-prototypes -Wold-style-definition
| -Wmissing-include-dirs -Wunused-but-set-variable
| -Wunused-const-variable -Wno-missing-field-initializers
| -Wno-sign-compare -Wno-type-limits -Wno-shift-negative-value -g
| -gdwarf-4 -I /path/to/linux/scripts/mod -I
| ./scripts/mod -DKBUILD_MODFILE='"scripts/mod/empty"'
| -DKBUILD_BASENAME='"empty"' -DKBUILD_MODNAME='"empty"'
| -D__KBUILD_MODNAME=kmod_empty
| /path/to/linux/scripts/mod/empty.c
| 
| No such file: gnu

regards,
Marc

-- 
Pengutronix e.K.                 | Marc Kleine-Budde           |
Embedded Linux                   | https://www.pengutronix.de  |
Vertretung West/Dortmund         | Phone: +49-231-2826-924     |
Amtsgericht Hildesheim, HRA 2686 | Fax:   +49-5121-206917-5555 |

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: building Linux for ARMv7 with llvm breaks sparse

[Linus Torvalds]

On Thu, Mar 31, 2022 at 4:01 AM Marc Kleine-Budde <mkl@pengutronix.de> wrote:
> | -meabi gnu
>   ^^^^^^^^^^

That's certainly technically valid syntax, but it's unusual, and it
breaks the normal sparse "ignore stuff we don't know about" thing.

Does it all work if we just make the kernel use the normal

  -meabi=gnu

syntax instead?

And by that "all work" I mean not just sparse, but the actual real
clang/arm build.

            Linus

Re: building Linux for ARMv7 with llvm breaks sparse

[Linus Torvalds]

On Thu, Mar 31, 2022 at 12:17 PM Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> Does it all work if we just make the kernel use the normal
>
>   -meabi=gnu
>
> syntax instead?

IOW, just something like this

    --- a/arch/arm/Makefile
    +++ b/arch/arm/Makefile
    @@ -110,7 +110,7 @@ CFLAGS_ABI      +=-funwind-tables
     endif

     ifeq ($(CONFIG_CC_IS_CLANG),y)
    -CFLAGS_ABI += -meabi gnu
    +CFLAGS_ABI += -meabi=gnu
     endif

     ifeq ($(CONFIG_CURRENT_POINTER_IN_TPIDRURO),y)

that's basically what we do for all the other '-mxyz=abc' cases, so
it's really that use of -meabi that stands out as being the odd man
out.

        Linus

Re: building Linux for ARMv7 with llvm breaks sparse

[Marc Kleine-Budde]

[-- Attachment #1: Type: text/plain, Size: 1266 bytes --]

On 31.03.2022 12:22:08, Linus Torvalds wrote:
> On Thu, Mar 31, 2022 at 12:17 PM Linus Torvalds
> <torvalds@linux-foundation.org> wrote:
> >
> > Does it all work if we just make the kernel use the normal
> >
> >   -meabi=gnu
> >
> > syntax instead?
> 
> IOW, just something like this
> 
>     --- a/arch/arm/Makefile
>     +++ b/arch/arm/Makefile
>     @@ -110,7 +110,7 @@ CFLAGS_ABI      +=-funwind-tables
>      endif
> 
>      ifeq ($(CONFIG_CC_IS_CLANG),y)
>     -CFLAGS_ABI += -meabi gnu
>     +CFLAGS_ABI += -meabi=gnu
>      endif
> 
>      ifeq ($(CONFIG_CURRENT_POINTER_IN_TPIDRURO),y)
> 
> that's basically what we do for all the other '-mxyz=abc' cases, so
> it's really that use of -meabi that stands out as being the odd man
> out.

No :/

| clang: error: unknown argument '-meabi=gnu'; did you mean '-mabi=gnu'?

The documentation says no "=":

| https://clang.llvm.org/docs/ClangCommandLineReference.html#cmdoption-clang-meabi

regards,
Marc

-- 
Pengutronix e.K.                 | Marc Kleine-Budde           |
Embedded Linux                   | https://www.pengutronix.de  |
Vertretung West/Dortmund         | Phone: +49-231-2826-924     |
Amtsgericht Hildesheim, HRA 2686 | Fax:   +49-5121-206917-5555 |

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: building Linux for ARMv7 with llvm breaks sparse

[Linus Torvalds]

On Thu, Mar 31, 2022 at 12:31 PM Marc Kleine-Budde <mkl@pengutronix.de> wrote:
>
> No :/
>
> | clang: error: unknown argument '-meabi=gnu'; did you mean '-mabi=gnu'?
>
> The documentation says no "=":
>
> | https://clang.llvm.org/docs/ClangCommandLineReference.html#cmdoption-clang-meabi

Wow. What a horrible inconsistency. Every single other '-mxyz arg'
thing seems to use '=', so it's really -meabi that stands out as being
broken.

Oh well. Sparse does handle these kinds of architecture things in
other places, but I'm not personally really very motivated to fix it
due to just being pissed off at clang idiocy.

               Linus

[PATCH] handle clang's option "-meabi gnu"

[Luc Van Oostenryck]

From: Luc Van Oostenryck <luc.vanoostenryck@gmail.com>

Clang has an option "-meabi <arg>" which is used by the kernel for ARMv7.
This kind of option, taking a argument without a separating '=', can't
be ignored like most other options and must this be special-cased.

So, add the special case for this option and consume the argument if it's
one of the valid one.

Link: https://lore.kernel.org/r/20220331110118.vr4miyyytqlssjoi@pengutronix.de
Reported-by: Marc Kleine-Budde <mkl@pengutronix.de>
Signed-off-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com>
---
 options.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/options.c b/options.c
index 6704fc8d2c8d..0224c290d322 100644
--- a/options.c
+++ b/options.c
@@ -685,6 +685,19 @@ static const struct flag mflags[] = {
 
 static char **handle_switch_m(char *arg, char **next)
 {
+	if (!strcmp(arg, "meabi") && next[1] && next[1][0] != '-') {
+		// clang has such an option with syntax: -meabi <arg>
+		// It's used by the kernel for armv7.
+		// GCC has the same opion but with no argument.
+		// Parse it here to consume the possible argument.
+		static const char *valid[] = { "gnu", "4", "5", "default", NULL };
+		int i;
+		for (i = 0; valid[i]; i++) {
+			if (!strcmp(next[1], valid[i]))
+				return ++next;
+		}
+	}
+
 	if (!strcmp(arg, "multiarch-dir")) {
 		return handle_multiarch_dir(arg, next);
 	} else {
-- 
2.36.1

Re: [PATCH] handle clang's option "-meabi gnu"

[Marc Kleine-Budde]

[-- Attachment #1: Type: text/plain, Size: 990 bytes --]

On 22.05.2022 11:08:24, Luc Van Oostenryck wrote:
> From: Luc Van Oostenryck <luc.vanoostenryck@gmail.com>
> 
> Clang has an option "-meabi <arg>" which is used by the kernel for ARMv7.
> This kind of option, taking a argument without a separating '=', can't
> be ignored like most other options and must this be special-cased.
> 
> So, add the special case for this option and consume the argument if it's
> one of the valid one.
> 
> Link: https://lore.kernel.org/r/20220331110118.vr4miyyytqlssjoi@pengutronix.de
> Reported-by: Marc Kleine-Budde <mkl@pengutronix.de>
> Signed-off-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com>

Tested-by: Marc Kleine-Budde <mkl@pengutronix.de>

regards,
Marc

-- 
Pengutronix e.K.                 | Marc Kleine-Budde           |
Embedded Linux                   | https://www.pengutronix.de  |
Vertretung West/Dortmund         | Phone: +49-231-2826-924     |
Amtsgericht Hildesheim, HRA 2686 | Fax:   +49-5121-206917-5555 |

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]