On Wed, Oct 20, 2021 at 10:45:43AM -0700, Nathan Chancellor wrote:

On Wed, Oct 20, 2021 at 10:35:53AM -0700, Kees Cook wrote:

quoted

GCC plugins should only exist when some compiler feature needs to be
proven but does not exist in either GCC nor Clang. For example, if a
desired feature is already in Clang, it should be added to GCC upstream.
Document this explicitly.

Additionally, mark the plugins with matching upstream GCC features as
removable past their respective GCC versions.

Cc: Masahiro Yamada <masahiroy@kernel.org>
Cc: Michal Marek <redacted>
Cc: Nick Desaulniers <redacted>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: James Morris <jmorris@namei.org>
Cc: "Serge E. Hallyn" <serge@hallyn.com>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: linux-hardening@vger.kernel.org
Cc: linux-kbuild@vger.kernel.org
Cc: linux-doc@vger.kernel.org
Cc: linux-security-module@vger.kernel.org
Cc: llvm@lists.linux.dev
Signed-off-by: Kees Cook <redacted>

Seems reasonable to me.

Reviewed-by: Nathan Chancellor <nathan@kernel.org>

Thanks!

One comment below.

quoted

---
 Documentation/kbuild/gcc-plugins.rst | 26 ++++++++++++++++++++++++++
 scripts/gcc-plugins/Kconfig          |  4 ++--
 security/Kconfig.hardening           |  9 ++++++---
 3 files changed, 34 insertions(+), 5 deletions(-)

diff --git a/Documentation/kbuild/gcc-plugins.rst b/Documentation/kbuild/gcc-plugins.rst
index 3349966f213d..4b28c7a4032f 100644
--- a/Documentation/kbuild/gcc-plugins.rst
+++ b/Documentation/kbuild/gcc-plugins.rst

@@ -32,6 +32,32 @@ This infrastructure was ported from grsecurity [6]_ and PaX [7]_.
 .. [7] https://pax.grsecurity.net/
 
 
+Purpose
+=======
+
+GCC plugins are designed to provide a place to experiment with potential
+compiler features that are neither in GCC nor Clang upstream. Once
+their utility is proven, the goal is to upstream the feature into GCC
+(and Clang), and then to finally remove them from the kernel once the
+feature is available in all supported versions of GCC.
+
+Specifically, new plugins should implement only features that have no
+upstream compiler support (in either GCC or Clang).
+
+When a feature exists in Clang but not GCC, effort should be made to
+bring the feature to upstream GCC (rather than just as a kernel-specific
+GCC plugin), so the entire ecosystem can benefit from it.
+
+Similarly, even if a feature provided by a GCC plugin does *not* exist
+in Clang, but the feature is proven to be useful, effort should be spent
+to upstream the feature to GCC (and Clang).
+
+After a feature is available in upstream GCC, the plugin will be made
+unbuildable for the corresponding GCC version (and later). Once all
+kernel-supported versions of GCC provide the feature, the plugin will
+be removed from the kernel.
+
+
 Files
 =====
 

diff --git a/scripts/gcc-plugins/Kconfig b/scripts/gcc-plugins/Kconfig
index ab9eb4cbe33a..3f5d3580ec06 100644
--- a/scripts/gcc-plugins/Kconfig
+++ b/scripts/gcc-plugins/Kconfig

@@ -37,6 +37,8 @@ config GCC_PLUGIN_CYC_COMPLEXITY
 
 config GCC_PLUGIN_SANCOV
 	bool
+	# Plugin can be removed once the kernel only supports GCC 6.1.0+
+	depends on !CC_HAS_SANCOV_TRACE_PC

This symbol is not user selectable and the one place that does select it
only does so when !CC_HAS_SANCOV_TRACE_PC so this seems pointless to me.

Keep the comment, ditch the depends?

I had a similar thought, and in the end, I decided I wanted to always
enforce the GCC feature check through a depends, with a comment about
the expected version. I want to make sure we don't use plugins if an
upstream feature is already available. It happens that SANCOV was
effectively the first to do this, but it did so on the other side and I
wanted it repeated here so it was "self contained".

-Kees

-- 
Kees Cook