On Wed, Aug 26, 2020 at 01:14:19PM -0700, Nick Desaulniers wrote:
> During Plumbers 2020, we voted to just support the latest release of
> Clang for now. Add a compile time check for this.
>
> Older clang's may work, but we will likely drop workarounds for older
> versions.
I think this part of the commit message is a little wishy-washy. If we
are breaking the build for clang < 10.0.1, we are not saying "may work",
we are saying "won't work". Because of this, we should take the
opportunity to clean up behind us and revert/remove parts of:
87e0d4f0f37f ("kbuild: disable clang's default use of -fmerge-all-constants")
b0fe66cf0950 ("ARM: 8905/1: Emit __gnu_mcount_nc when using Clang 10.0.0 or newer")
b9249cba25a5 ("arm64: bti: Require clang >= 10.0.1 for in-kernel BTI support")
3acf4be23528 ("arm64: vdso: Fix compilation with clang older than 8")
This could be a series or a part of this commit, I do not have a
strong preference. If we are not going to clean up behind us, this
should be a warning and not an error.
> Link: https://github.com/ClangBuiltLinux/linux/issues/9
> Link: https://github.com/ClangBuiltLinux/linux/issues/941
> Suggested-by: Sedat Dilek <[email protected]>
> Signed-off-by: Nick Desaulniers <[email protected]>
Otherwise, I agree with Kees's comments. With them addressed, feel free
to add:
Acked-by: Nathan Chancellor <[email protected]>
> ---
> include/linux/compiler-clang.h | 8 ++++++++
> 1 file changed, 8 insertions(+)
>
> diff --git a/include/linux/compiler-clang.h b/include/linux/compiler-clang.h
> index cee0c728d39a..7338d3ffd240 100644
> --- a/include/linux/compiler-clang.h
> +++ b/include/linux/compiler-clang.h
> @@ -3,6 +3,14 @@
> #error "Please don't include <linux/compiler-clang.h> directly, include <linux/compiler.h> instead."
> #endif
>
> +#define CLANG_VERSION (__clang_major__ * 10000 \
> + + __clang_minor__ * 100 \
> + + __clang_patchlevel__)
> +
> +#if CLANG_VERSION < 100001
> +# error Sorry, your compiler is too old - please upgrade it.
> +#endif
> +
> /* Compiler specific definitions for Clang compiler */
>
> /* same as gcc, this was present in clang-2.6 so we can assume it works
> --
> 2.28.0.297.g1956fa8f8d-goog
>
On Wed, Aug 26, 2020 at 02:42PM -0700, Nathan Chancellor wrote:
> On Wed, Aug 26, 2020 at 01:14:19PM -0700, Nick Desaulniers wrote:
> > During Plumbers 2020, we voted to just support the latest release of
> > Clang for now. Add a compile time check for this.
> >
> > Older clang's may work, but we will likely drop workarounds for older
> > versions.
>
> I think this part of the commit message is a little wishy-washy. If we
> are breaking the build for clang < 10.0.1, we are not saying "may work",
> we are saying "won't work". Because of this, we should take the
> opportunity to clean up behind us and revert/remove parts of:
>
> 87e0d4f0f37f ("kbuild: disable clang's default use of -fmerge-all-constants")
> b0fe66cf0950 ("ARM: 8905/1: Emit __gnu_mcount_nc when using Clang 10.0.0 or newer")
> b9249cba25a5 ("arm64: bti: Require clang >= 10.0.1 for in-kernel BTI support")
> 3acf4be23528 ("arm64: vdso: Fix compilation with clang older than 8")
>
> This could be a series or a part of this commit, I do not have a
> strong preference. If we are not going to clean up behind us, this
> should be a warning and not an error.
There are also some other documentation that would go stale. We probably
have to change KASAN docs to look something like the below.
I wish we could also remove the "but detection of out-of-bounds accesses
for global variables is only supported since Clang 11", but Clang 10 is
a vast improvement so I'm not complaining. :-)
Acked-by: Marco Elver <[email protected]>
Thanks,
-- Marco
------ >8 ------
From 13d03b55c69dec813d94c1481dcb294971f164ef Mon Sep 17 00:00:00 2001
From: Marco Elver <[email protected]>
Date: Thu, 27 Aug 2020 20:56:34 +0200
Subject: [PATCH] kasan: Remove mentions of unsupported Clang versions
Since the kernel now requires at least Clang 10.0.1, remove any mention
of old Clang versions and simplify the documentation.
Signed-off-by: Marco Elver <[email protected]>
---
Documentation/dev-tools/kasan.rst | 4 ++--
lib/Kconfig.kasan | 9 ++++-----
2 files changed, 6 insertions(+), 7 deletions(-)
diff --git a/Documentation/dev-tools/kasan.rst b/Documentation/dev-tools/kasan.rst
index 38fd5681fade..4abc84b1798c 100644
--- a/Documentation/dev-tools/kasan.rst
+++ b/Documentation/dev-tools/kasan.rst
@@ -13,10 +13,10 @@ KASAN uses compile-time instrumentation to insert validity checks before every
memory access, and therefore requires a compiler version that supports that.
Generic KASAN is supported in both GCC and Clang. With GCC it requires version
-8.3.0 or later. With Clang it requires version 7.0.0 or later, but detection of
+8.3.0 or later. Any supported Clang version is compatible, but detection of
out-of-bounds accesses for global variables is only supported since Clang 11.
-Tag-based KASAN is only supported in Clang and requires version 7.0.0 or later.
+Tag-based KASAN is only supported in Clang.
Currently generic KASAN is supported for the x86_64, arm64, xtensa, s390 and
riscv architectures, and tag-based KASAN is supported only for arm64.
diff --git a/lib/Kconfig.kasan b/lib/Kconfig.kasan
index 047b53dbfd58..033a5bc67ac4 100644
--- a/lib/Kconfig.kasan
+++ b/lib/Kconfig.kasan
@@ -54,9 +54,9 @@ config KASAN_GENERIC
Enables generic KASAN mode.
This mode is supported in both GCC and Clang. With GCC it requires
- version 8.3.0 or later. With Clang it requires version 7.0.0 or
- later, but detection of out-of-bounds accesses for global variables
- is supported only since Clang 11.
+ version 8.3.0 or later. Any supported Clang version is compatible,
+ but detection of out-of-bounds accesses for global variables is
+ supported only since Clang 11.
This mode consumes about 1/8th of available memory at kernel start
and introduces an overhead of ~x1.5 for the rest of the allocations.
@@ -78,8 +78,7 @@ config KASAN_SW_TAGS
Enables software tag-based KASAN mode.
This mode requires Top Byte Ignore support by the CPU and therefore
- is only supported for arm64. This mode requires Clang version 7.0.0
- or later.
+ is only supported for arm64. This mode requires Clang.
This mode consumes about 1/16th of available memory at kernel start
and introduces an overhead of ~20% for the rest of the allocations.
--
2.28.0.297.g1956fa8f8d-goog
On Thu, Aug 27, 2020 at 9:02 PM Marco Elver <[email protected]> wrote:
>
> On Wed, Aug 26, 2020 at 02:42PM -0700, Nathan Chancellor wrote:
> > On Wed, Aug 26, 2020 at 01:14:19PM -0700, Nick Desaulniers wrote:
> > > During Plumbers 2020, we voted to just support the latest release of
> > > Clang for now. Add a compile time check for this.
> > >
> > > Older clang's may work, but we will likely drop workarounds for older
> > > versions.
> >
> > I think this part of the commit message is a little wishy-washy. If we
> > are breaking the build for clang < 10.0.1, we are not saying "may work",
> > we are saying "won't work". Because of this, we should take the
> > opportunity to clean up behind us and revert/remove parts of:
> >
> > 87e0d4f0f37f ("kbuild: disable clang's default use of -fmerge-all-constants")
> > b0fe66cf0950 ("ARM: 8905/1: Emit __gnu_mcount_nc when using Clang 10.0.0 or newer")
> > b9249cba25a5 ("arm64: bti: Require clang >= 10.0.1 for in-kernel BTI support")
> > 3acf4be23528 ("arm64: vdso: Fix compilation with clang older than 8")
> >
> > This could be a series or a part of this commit, I do not have a
> > strong preference. If we are not going to clean up behind us, this
> > should be a warning and not an error.
>
> There are also some other documentation that would go stale. We probably
> have to change KASAN docs to look something like the below.
>
> I wish we could also remove the "but detection of out-of-bounds accesses
> for global variables is only supported since Clang 11", but Clang 10 is
> a vast improvement so I'm not complaining. :-)
>
> Acked-by: Marco Elver <[email protected]>
>
> Thanks,
> -- Marco
>
> ------ >8 ------
>
> From 13d03b55c69dec813d94c1481dcb294971f164ef Mon Sep 17 00:00:00 2001
> From: Marco Elver <[email protected]>
> Date: Thu, 27 Aug 2020 20:56:34 +0200
> Subject: [PATCH] kasan: Remove mentions of unsupported Clang versions
>
> Since the kernel now requires at least Clang 10.0.1, remove any mention
> of old Clang versions and simplify the documentation.
>
> Signed-off-by: Marco Elver <[email protected]>
> ---
> Documentation/dev-tools/kasan.rst | 4 ++--
> lib/Kconfig.kasan | 9 ++++-----
> 2 files changed, 6 insertions(+), 7 deletions(-)
>
> diff --git a/Documentation/dev-tools/kasan.rst b/Documentation/dev-tools/kasan.rst
> index 38fd5681fade..4abc84b1798c 100644
> --- a/Documentation/dev-tools/kasan.rst
> +++ b/Documentation/dev-tools/kasan.rst
> @@ -13,10 +13,10 @@ KASAN uses compile-time instrumentation to insert validity checks before every
> memory access, and therefore requires a compiler version that supports that.
>
> Generic KASAN is supported in both GCC and Clang. With GCC it requires version
> -8.3.0 or later. With Clang it requires version 7.0.0 or later, but detection of
> +8.3.0 or later. Any supported Clang version is compatible, but detection of
> out-of-bounds accesses for global variables is only supported since Clang 11.
>
> -Tag-based KASAN is only supported in Clang and requires version 7.0.0 or later.
> +Tag-based KASAN is only supported in Clang.
>
> Currently generic KASAN is supported for the x86_64, arm64, xtensa, s390 and
> riscv architectures, and tag-based KASAN is supported only for arm64.
> diff --git a/lib/Kconfig.kasan b/lib/Kconfig.kasan
> index 047b53dbfd58..033a5bc67ac4 100644
> --- a/lib/Kconfig.kasan
> +++ b/lib/Kconfig.kasan
> @@ -54,9 +54,9 @@ config KASAN_GENERIC
> Enables generic KASAN mode.
>
> This mode is supported in both GCC and Clang. With GCC it requires
> - version 8.3.0 or later. With Clang it requires version 7.0.0 or
> - later, but detection of out-of-bounds accesses for global variables
> - is supported only since Clang 11.
> + version 8.3.0 or later. Any supported Clang version is compatible,
> + but detection of out-of-bounds accesses for global variables is
> + supported only since Clang 11.
>
> This mode consumes about 1/8th of available memory at kernel start
> and introduces an overhead of ~x1.5 for the rest of the allocations.
> @@ -78,8 +78,7 @@ config KASAN_SW_TAGS
> Enables software tag-based KASAN mode.
>
> This mode requires Top Byte Ignore support by the CPU and therefore
> - is only supported for arm64. This mode requires Clang version 7.0.0
> - or later.
> + is only supported for arm64. This mode requires Clang.
>
> This mode consumes about 1/16th of available memory at kernel start
> and introduces an overhead of ~20% for the rest of the allocations.
> --
> 2.28.0.297.g1956fa8f8d-goog
>
Reviewed-by: Andrey Konovalov <[email protected]>
Thanks!
On Wed, Aug 26, 2020 at 2:42 PM Nathan Chancellor
<[email protected]> wrote:
>
> On Wed, Aug 26, 2020 at 01:14:19PM -0700, Nick Desaulniers wrote:
> > During Plumbers 2020, we voted to just support the latest release of
> > Clang for now. Add a compile time check for this.
> >
> > Older clang's may work, but we will likely drop workarounds for older
> > versions.
>
> I think this part of the commit message is a little wishy-washy. If we
Yep, you're right. I'm still in denial. Let me rip that bandaid off
and send a v2, with your and Kees' suggestions.
Sorry, the docs patch already got picked up. Let's follow up with
additional patches to docs separately.
> are breaking the build for clang < 10.0.1, we are not saying "may work",
> we are saying "won't work". Because of this, we should take the
> opportunity to clean up behind us and revert/remove parts of:
>
> 87e0d4f0f37f ("kbuild: disable clang's default use of -fmerge-all-constants")
> b0fe66cf0950 ("ARM: 8905/1: Emit __gnu_mcount_nc when using Clang 10.0.0 or newer")
> b9249cba25a5 ("arm64: bti: Require clang >= 10.0.1 for in-kernel BTI support")
> 3acf4be23528 ("arm64: vdso: Fix compilation with clang older than 8")
I'd prefer to see this land in mainline first; otherwise, I'm worried
about this patch "racing" to mainline with those patches if they go
via separate trees. Thoughts?
--
Thanks,
~Nick Desaulniers
On Mon, Aug 31, 2020 at 4:32 PM Nick Desaulniers
<[email protected]> wrote:
>
> On Wed, Aug 26, 2020 at 2:42 PM Nathan Chancellor
> <[email protected]> wrote:
> >
> > On Wed, Aug 26, 2020 at 01:14:19PM -0700, Nick Desaulniers wrote:
> > > During Plumbers 2020, we voted to just support the latest release of
> > > Clang for now. Add a compile time check for this.
> > >
> > > Older clang's may work, but we will likely drop workarounds for older
> > > versions.
> >
> > I think this part of the commit message is a little wishy-washy. If we
> > are breaking the build for clang < 10.0.1, we are not saying "may work",
> > we are saying "won't work". Because of this, we should take the
> > opportunity to clean up behind us and revert/remove parts of:
> >
> > 87e0d4f0f37f ("kbuild: disable clang's default use of -fmerge-all-constants")
> > b0fe66cf0950 ("ARM: 8905/1: Emit __gnu_mcount_nc when using Clang 10.0.0 or newer")
> > b9249cba25a5 ("arm64: bti: Require clang >= 10.0.1 for in-kernel BTI support")
> > 3acf4be23528 ("arm64: vdso: Fix compilation with clang older than 8")
>
> I'd prefer to see this land in mainline first; otherwise, I'm worried
> about this patch "racing" to mainline with those patches if they go
> via separate trees. Thoughts?
Maybe I should send such a series (including Marco's recommendations)
to Mr. Morton or Yamada-san?
--
Thanks,
~Nick Desaulniers