[PATCH 1/8] Compiler attributes: GCC function alignment workarounds

Mark Rutland mark.rutland at arm.com
Mon Jan 9 05:58:21 PST 2023


>From local testing, contemporary versions of of GCC (e.g. GCC 12.2.0)
don't respect '-falign-functions=N' in all cases. This is unfortunate,
as (for non-zero values of N) CONFIG_FUNCTION_ALIGNMENT=N will set
'-falign-functions=N', but this won't take effect for all functions.
LLVM appears to respect '-falign-functions=N' in call cases.

Today, for x86 this turns out to be functionally benign, though it does
somewhat undermine the CONFIG_FUNCTION_ALIGNMENT option, and it means
that CONFIG_DEBUG_FORCE_FUNCTION_ALIGN_64B is not as robust as we'd
like.

On arm64 we'd like to use CONFIG_FUNCTION_ALIGNMENT to implement ftrace
functionality, and we'll require function alignment to be respected for
functional reasons.

As far as I can tell, GCC doesn't respect '-falign-functions=N':

* When the __weak__ attribute is used

  GCC seems to forget the alignment specified by '-falign-functions=N',
  but will respect the '__aligned__(N)' function attribute. Thus, we can
  work around this by explciitly setting the alignment for weak
  functions.

* When the __cold__ attribute is used

  GCC seems to forget the alignment specified by '-falign-functions=N',
  and also doesn't seem to respect the '__aligned__(N)' function
  attribute. The only way to work around this is to not use the __cold__
  attibute.

This patch implements workarounds for these two cases, using a function
attribute to set the alignment of __weak__ functions, and preventing the
use of the __cold__ attribute when CONFIG_FUNCTION_ALIGNMENT is
non-zero.

I've tested this by selecting CONFIG_DEBUG_FORCE_FUNCTION_ALIGN_64B=y,
building and booting a kernel, and looking for misaligned text symbols:

* arm64:

  Before:
    # grep ' [Tt] ' /proc/kallsyms | grep -iv '[048c]0 [Tt] ' | wc -l
    4939

  After:
    # grep ' [Tt] ' /proc/kallsyms | grep -iv '[048c]0 [Tt] ' | wc -l
    908

* x86_64:

  Before:
    # grep ' [Tt] ' /proc/kallsyms | grep -iv '[048c]0 [Tt] ' | wc -l
    7969

  After:
    # grep ' [Tt] ' /proc/kallsyms | grep -iv '[048c]0 [Tt] ' | wc -l
    2057

With the patch applied, the remaining unaligned text labels are a
combination of static call trampolines, non-function labels in assembly,
and ACPICA functions, which will be dealt with in subsequent patches.

Signed-off-by: Mark Rutland <mark.rutland at arm.com>
Cc: Catalin Marinas <catalin.marinas at arm.com>
Cc: Florent Revest <revest at chromium.org>
Cc: Masami Hiramatsu <mhiramat at kernel.org>
Cc: Peter Zijlstra <peterz at infradead.org>
Cc: Steven Rostedt <rostedt at goodmis.org>
Cc: Will Deacon <will at kernel.org>
Cc: Miguel Ojeda <ojeda at kernel.org>
Cc: Nick Desaulniers <ndesaulniers at google.com>
---
 include/linux/compiler_attributes.h | 23 ++++++++++++++++++++++-
 1 file changed, 22 insertions(+), 1 deletion(-)

diff --git a/include/linux/compiler_attributes.h b/include/linux/compiler_attributes.h
index 898b3458b24a0..dcb7ac67b764f 100644
--- a/include/linux/compiler_attributes.h
+++ b/include/linux/compiler_attributes.h
@@ -33,6 +33,17 @@
 #define __aligned(x)                    __attribute__((__aligned__(x)))
 #define __aligned_largest               __attribute__((__aligned__))
 
+/*
+ * Contemporary versions of GCC (e.g. 12.2.0) don't always respect
+ * '-falign-functions=N', and require alignment to be specificed via a function
+ * attribute in some cases.
+ */
+#if CONFIG_FUNCTION_ALIGNMENT > 0
+#define __function_aligned		__aligned(CONFIG_FUNCTION_ALIGNMENT)
+#else
+#define __function_aligned
+#endif
+
 /*
  * Note: do not use this directly. Instead, use __alloc_size() since it is conditionally
  * available and includes other attributes. For GCC < 9.1, __alloc_size__ gets undefined
@@ -78,8 +89,15 @@
 /*
  *   gcc: https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html#index-cold-function-attribute
  *   gcc: https://gcc.gnu.org/onlinedocs/gcc/Label-Attributes.html#index-cold-label-attribute
+ *
+ * GCC drops function alignment when the __cold__ attribute is used. Avoid the
+ * __cold__ attribute if function alignment is required.
  */
+#if !defined(CONFIG_CC_IS_GCC) || (CONFIG_FUNCTION_ALIGNMENT == 0)
 #define __cold                          __attribute__((__cold__))
+#else
+#define __cold
+#endif
 
 /*
  * Note the long name.
@@ -369,8 +387,11 @@
 /*
  *   gcc: https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html#index-weak-function-attribute
  *   gcc: https://gcc.gnu.org/onlinedocs/gcc/Common-Variable-Attributes.html#index-weak-variable-attribute
+ *
+ * GCC drops function alignment when the __weak__ attribute is used. This can
+ * be restored with function attributes.
  */
-#define __weak                          __attribute__((__weak__))
+#define __weak                          __attribute__((__weak__)) __function_aligned
 
 /*
  * Used by functions that use '__builtin_return_address'. These function
-- 
2.30.2




More information about the linux-arm-kernel mailing list