Skip to contentSkip to conversation

compiler_types: Introduce __nonstring_array

PATCH In review Kees Cook 2 participants 3 messages Mar 10, 2025 ACPI

lore
Share & tools
lore

Kees Cook

author
#0 lore raw

PATCH 1/1 compiler_types: Introduce __nonstring_array

GCC has expanded support of the "nonstring" attribute so that it can be applied to arrays of character arrays[1], which is needed to identify correct static initialization of those kinds of objects. Since this was not supported prior to GCC 15, we need to distinguish the usage of Linux's existing __nonstring macro for the attribute for non-multi-dimensional char arrays. Until GCC 15 is the minimum version, use __nonstring_array to mark arrays of non-string character arrays. (Regular non-string character arrays can continue to use __nonstring.)

This allows for changes like this:

-static const char table_sigs[][ACPI_NAMESEG_SIZE] __initconst = { +static const char table_sigs[][ACPI_NAMESEG_SIZE] __nonstring_array __initconst = {
ACPI_SIG_BERT, ACPI_SIG_BGRT, ACPI_SIG_CPEP, ACPI_SIG_ECDT,

Which will silence the coming -Wunterminated-string-initialization warnings in GCC 15:

In file included from ../include/acpi/actbl.h:371, from ../include/acpi/acpi.h:26, from ../include/linux/acpi.h:26,

                 from ../drivers/acpi/tables.c:19:
../include/acpi/actbl1.h:30:33: warning: initializer-string for array of 'char' truncates NUL terminator but destination lacks 'nonstring' attribute (5 chars into 4 available) [-Wunterminated-string-initialization]
   30 | #define ACPI_SIG_BERT           "BERT"  /* Boot Error Record Table */
      |                                 ^~~~~~                                                      ../drivers/acpi/tables.c:400:9: note: in expansion of macro 'ACPI_SIG_BERT'                           400 |         ACPI_SIG_BERT, ACPI_SIG_BGRT, ACPI_SIG_CPEP, ACPI_SIG_ECDT,
      |         ^~~~~~~~~~~~~                                                                       ../include/acpi/actbl1.h:31:33: warning: initializer-string for array of 'char' truncates NUL terminator but destination lacks 'nonstring' attribute (5 chars into 4 available) [-Wunterminated-string-initialization]
   31 | #define ACPI_SIG_BGRT           "BGRT"  /* Boot Graphics Resource Table */
      |                                 ^~~~~~

---
Cc: Miguel Ojeda <[email protected]>
Cc: Arnd Bergmann <[email protected]>
Cc: "Rafael J. Wysocki" <[email protected]>
Cc: Len Brown <[email protected]>
Cc: Marco Elver <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Przemek Kitszel <[email protected]>
Cc: Hao Luo <[email protected]>
Cc: Masahiro Yamada <[email protected]>
Cc: Christophe Leroy <[email protected]>
Cc: Johannes Weiner <[email protected]>
Cc: [email protected]

---
 include/linux/compiler_types.h | 12 ++++++++++++
 init/Kconfig                   |  3 +++
 2 files changed, 15 insertions(+)
include/linux/compiler_types.h +12 -0
index 981cc3d7e3aa..7ccea700b46d 100644
@@ -348,6 +348,18 @@ struct ftrace_likely_data {
 # define __counted_by(member)
 #endif
 
+/*
+ * Optional: only supported since gcc >= 15
+ * Optional: not supported by Clang
+ *
+ * gcc: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117178
+ */
+#ifdef CONFIG_CC_HAS_MULTIDIMENSIONAL_NONSTRING
+# define __nonstring_array		__attribute__((__nonstring__))
+#else
+# define __nonstring_array
+#endif
+
 /*
  * Apply __counted_by() when the Endianness matches to increase test coverage.
  */
init/Kconfig +3 -0
index d0d021b3fa3b..723dc69507d6 100644
@@ -129,6 +129,9 @@ config CC_HAS_COUNTED_BY
 	# https://github.com/llvm/llvm-project/pull/112636
 	depends on !(CC_IS_CLANG && CLANG_VERSION < 190103)
 
+config CC_HAS_MULTIDIMENSIONAL_NONSTRING
+	def_bool $(success,echo 'char tag[][4] __attribute__((__nonstring__)) = { };' | $(CC) $(CLANG_FLAGS) -x c - -c -o /dev/null -Werror)
+
 config RUSTC_HAS_COERCE_POINTEE
 	def_bool RUSTC_VERSION >= 108400
 
Signature · 1 line
2.34.1
author
#1 lore raw
Quoted 2 lines · Kees Cook

In file included from ../include/acpi/actbl.h:371, from ../include/acpi/acpi.h:26, from ../include/linux/acpi.h:26,

                 from ../drivers/acpi/tables.c:19:

Ugh, this whole paste went poorly. I've fixed it locally. It should be:

In file included from ../include/acpi/actbl.h:371,

                 from ../include/acpi/acpi.h:26,
                 from ../include/linux/acpi.h:26,
                 from ../drivers/acpi/tables.c:19:
../include/acpi/actbl1.h:30:33: warning: initializer-string for array of 'char' truncates NUL terminator but destination lacks 'nonstring' attribute (5 chars into 4 available) [-Wunterminated-string-initialization]
   30 | #define ACPI_SIG_BERT           "BERT"  /* Boot Error Record Table */
      |                                 ^~~~~~
../drivers/acpi/tables.c:400:9: note: in expansion of macro 'ACPI_SIG_BERT'
  400 |         ACPI_SIG_BERT, ACPI_SIG_BGRT, ACPI_SIG_CPEP, ACPI_SIG_ECDT,
      |         ^~~~~~~~~~~~~
../include/acpi/actbl1.h:31:33: warning: initializer-string for array of 'char' truncates NUL terminator but destination lacks 'nonstring' attribute (5 chars into 4 available) [-Wunterminated-string-initialization]
   31 | #define ACPI_SIG_BGRT           "BGRT"  /* Boot Graphics Resource Table */
      |                                 ^~~~~~
Signature · 1 line
Kees Cook
Acked-by on v1 1/1
#2 lore raw
In reply to Kees Cook
Quoted 9 lines · Kees Cook

GCC has expanded support of the "nonstring" attribute so that it can be applied to arrays of character arrays[1], which is needed to identify correct static initialization of those kinds of objects. Since this was not supported prior to GCC 15, we need to distinguish the usage of Linux's existing __nonstring macro for the attribute for non-multi-dimensional char arrays. Until GCC 15 is the minimum version, use __nonstring_array to mark arrays of non-string character arrays. (Regular non-string character arrays can continue to use __nonstring.)

Sounds reasonable. So this means that, when the GCC minimum goes over 15 in some years, we can both remove it and move it to compiler_attributes.h, right?

Perhaps we should mention part of this commit message (or the sentence above) in the comment to clarify (it will also help to clarify what that "only supported since GCC >= 15" means).

If you are going to use it in a series that has a use case:

Thanks!

Cheers,
Miguel