kbuild: linker script do not match C names unless LD_DEAD_CODE_DATA_ELIMINATION is configured

The .data and .bss sections were modified in the generic linker script to
pull in sections named .data.<C identifier>, which are generated by gcc with
-ffunction-sections and -fdata-sections options.

The problem with this pattern is it can also match section names that Linux
defines explicitly, e.g., .data.unlikely. This can cause Linux sections to
get moved into the wrong place.

The way to avoid this is to use ".." separators for explicit section names
(the dot character is valid in a section name but not a C identifier).
However currently there are sections which don't follow this rule, so for
now just disable the wild card by default.

Example: http://marc.info/?l=linux-arm-kernel&m=150106824024221&w=2

Cc: <[email protected]> # 4.9
Fixes: b67067f ("kbuild: allow archs to select link dead code/data elimination")
Signed-off-by: Nicholas Piggin <[email protected]>
Signed-off-by: Masahiro Yamada <[email protected]>

Author: npiggin

include/asm-generic/vmlinux.lds.h

@@ -59,6 +59,22 @@
 /* Align . to a 8 byte boundary equals to maximum function alignment. */
 #define ALIGN_FUNCTION()  . = ALIGN(8)
 
+/*
+ * LD_DEAD_CODE_DATA_ELIMINATION option enables -fdata-sections, which
+ * generates .data.identifier sections, which need to be pulled in with
+ * .data. We don't want to pull in .data..other sections, which Linux
+ * has defined. Same for text and bss.
+ */
+#ifdef CONFIG_LD_DEAD_CODE_DATA_ELIMINATION
+#define TEXT_MAIN .text .text.[0-9a-zA-Z_]*
+#define DATA_MAIN .data .data.[0-9a-zA-Z_]*
+#define BSS_MAIN .bss .bss.[0-9a-zA-Z_]*
+#else
+#define TEXT_MAIN .text
+#define DATA_MAIN .data
+#define BSS_MAIN .bss
+#endif
+
 /*
  * Align to a 32 byte boundary equal to the
  * alignment gcc 4.5 uses for a struct
@@ -198,12 +214,9 @@
 
 /*
  * .data section
- * LD_DEAD_CODE_DATA_ELIMINATION option enables -fdata-sections generates
- * .data.identifier which needs to be pulled in with .data, but don't want to
- * pull in .data..stuff which has its own requirements. Same for bss.
  */
 #define DATA_DATA							\
-	*(.data .data.[0-9a-zA-Z_]*)					\
+	*(DATA_MAIN)							\
 	*(.ref.data)							\
 	*(.data..shared_aligned) /* percpu related */			\
 	MEM_KEEP(init.data)						\
@@ -434,16 +447,17 @@
 		VMLINUX_SYMBOL(__security_initcall_end) = .;		\
 	}
 
-/* .text section. Map to function alignment to avoid address changes
+/*
+ * .text section. Map to function alignment to avoid address changes
  * during second ld run in second ld pass when generating System.map
- * LD_DEAD_CODE_DATA_ELIMINATION option enables -ffunction-sections generates
- * .text.identifier which needs to be pulled in with .text , but some
- * architectures define .text.foo which is not intended to be pulled in here.
- * Those enabling LD_DEAD_CODE_DATA_ELIMINATION must ensure they don't have
- * conflicting section names, and must pull in .text.[0-9a-zA-Z_]* */
+ *
+ * TEXT_MAIN here will match .text.fixup and .text.unlikely if dead
+ * code elimination is enabled, so these sections should be converted
+ * to use ".." first.
+ */
 #define TEXT_TEXT							\
 		ALIGN_FUNCTION();					\
-		*(.text.hot .text .text.fixup .text.unlikely)		\
+		*(.text.hot TEXT_MAIN .text.fixup .text.unlikely)	\
 		*(.ref.text)						\
 	MEM_KEEP(init.text)						\
 	MEM_KEEP(exit.text)						\
@@ -613,7 +627,7 @@
 		BSS_FIRST_SECTIONS					\
 		*(.bss..page_aligned)					\
 		*(.dynbss)						\
-		*(.bss .bss.[0-9a-zA-Z_]*)				\
+		*(BSS_MAIN)						\
 		*(COMMON)						\
 	}