[PATCH v4 1/4] compiler_types: Introduce the Clang __preserve_most function attribute

[Marco Elver]

[1]: "On X86-64 and AArch64 targets, this attribute changes the calling
convention of a function. The preserve_most calling convention attempts
to make the code in the caller as unintrusive as possible. This
convention behaves identically to the C calling convention on how
arguments and return values are passed, but it uses a different set of
caller/callee-saved registers. This alleviates the burden of saving and
recovering a large register set before and after the call in the caller.
If the arguments are passed in callee-saved registers, then they will be
preserved by the callee across the call. This doesn't apply for values
returned in callee-saved registers.

 * On X86-64 the callee preserves all general purpose registers, except
   for R11. R11 can be used as a scratch register. Floating-point
   registers (XMMs/YMMs) are not preserved and need to be saved by the
   caller.

 * On AArch64 the callee preserve all general purpose registers, except
   x0-X8 and X16-X18."

[1] https://clang.llvm.org/docs/AttributeReference.html#preserve-most

Introduce the attribute to compiler_types.h as __preserve_most.

Use of this attribute results in better code generation for calls to
very rarely called functions, such as error-reporting functions, or
rarely executed slow paths.

Beware that the attribute conflicts with instrumentation calls inserted
on function entry which do not use __preserve_most themselves. Notably,
function tracing which assumes the normal C calling convention for the
given architecture.  Where the attribute is supported, __preserve_most
will imply notrace. It is recommended to restrict use of the attribute
to functions that should or already disable tracing.

Note: The additional preprocessor check against architecture should not
be necessary if __has_attribute() only returns true where supported;
also see https://github.com/ClangBuiltLinux/linux/issues/1908. But until
__has_attribute() does the right thing, we also guard by known-supported
architectures to avoid build warnings on other architectures.

The attribute may be supported by a future GCC version (see
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110899).

Signed-off-by: Marco Elver <elver at google.com>
Reviewed-by: Miguel Ojeda <ojeda at kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers at google.com>
Acked-by: Steven Rostedt (Google) <rostedt at goodmis.org>
Acked-by: Mark Rutland <mark.rutland at arm.com>
---
v4:
* Guard attribute based on known-supported architectures to avoid
  compiler warnings about the attribute being ignored.

v3:
* Quote more from LLVM documentation about which registers are
  callee/caller with preserve_most.
* Code comment to restrict use where tracing is meant to be disabled.

v2:
* Imply notrace, to avoid any conflicts with tracing which is inserted
  on function entry. See added comments.
---
 include/linux/compiler_types.h | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)

diff --git a/include/linux/compiler_types.h b/include/linux/compiler_types.h
index 547ea1ff806e..c523c6683789 100644
--- a/include/linux/compiler_types.h
+++ b/include/linux/compiler_types.h
@@ -106,6 +106,34 @@ static inline void __chk_io_ptr(const volatile void __iomem *ptr) { }
 #define __cold
 #endif
 
+/*
+ * On x86-64 and arm64 targets, __preserve_most changes the calling convention
+ * of a function to make the code in the caller as unintrusive as possible. This
+ * convention behaves identically to the C calling convention on how arguments
+ * and return values are passed, but uses a different set of caller- and callee-
+ * saved registers.
+ *
+ * The purpose is to alleviates the burden of saving and recovering a large
+ * register set before and after the call in the caller.  This is beneficial for
+ * rarely taken slow paths, such as error-reporting functions that may be called
+ * from hot paths.
+ *
+ * Note: This may conflict with instrumentation inserted on function entry which
+ * does not use __preserve_most or equivalent convention (if in assembly). Since
+ * function tracing assumes the normal C calling convention, where the attribute
+ * is supported, __preserve_most implies notrace.  It is recommended to restrict
+ * use of the attribute to functions that should or already disable tracing.
+ *
+ * Optional: not supported by gcc.
+ *
+ * clang: https://clang.llvm.org/docs/AttributeReference.html#preserve-most
+ */
+#if __has_attribute(__preserve_most__) && (defined(CONFIG_X86_64) || defined(CONFIG_ARM64))
+# define __preserve_most notrace __attribute__((__preserve_most__))
+#else
+# define __preserve_most
+#endif
+
 /* Builtins */
 
 /*
-- 
2.41.0.694.ge786442a9b-goog