linker script: use separate simpler definition for PERCPU()
Impact: fix linker screwup on x86_32
Recent x86_64 zerobased patches introduced PERCPU_VADDR() to put
.data.percpu to a predefined address and re-defined PERCPU() in terms
of it. The new macro defined one extra symbol, __per_cpu_load, for
LMA of the section so that the init data could be accessed. This new
symbol introduced the following problems to x86_32.
1. If __per_cpu_load is defined outside of .data.percpu as an absolute
symbol, relocation generation for relocatable kernel fails due to
absolute relocation.
2. If __per_cpu_load is put inside .data.percpu with absolute address
assignment to work around #1, linker gets confused and under
certain configurations ends up relocating the symbol against
.data.percpu such that the load address gets added on top of
already set load address.
As x86_32 doesn't use predefined address for .data.percpu, there's no
need for it to care about the possibility of __per_cpu_load being
different from __per_cpu_start.
This patch defines PERCPU() separately so that __per_cpu_load is
defined inside .data.percpu so that everything is ordinary
linking-wise.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index 53e21f3..5406e70 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -445,10 +445,9 @@
* section in the linker script will go there too. @phdr should have
* a leading colon.
*
- * This macro defines three symbols, __per_cpu_load, __per_cpu_start
- * and __per_cpu_end. The first one is the vaddr of loaded percpu
- * init data. __per_cpu_start equals @vaddr and __per_cpu_end is the
- * end offset.
+ * Note that this macros defines __per_cpu_load as an absolute symbol.
+ * If there is no need to put the percpu section at a predetermined
+ * address, use PERCPU().
*/
#define PERCPU_VADDR(vaddr, phdr) \
VMLINUX_SYMBOL(__per_cpu_load) = .; \
@@ -470,7 +469,20 @@
* Align to @align and outputs output section for percpu area. This
* macro doesn't maniuplate @vaddr or @phdr and __per_cpu_load and
* __per_cpu_start will be identical.
+ *
+ * This macro is equivalent to ALIGN(align); PERCPU_VADDR( , ) except
+ * that __per_cpu_load is defined as a relative symbol against
+ * .data.percpu which is required for relocatable x86_32
+ * configuration.
*/
#define PERCPU(align) \
. = ALIGN(align); \
- PERCPU_VADDR( , )
+ .data.percpu : AT(ADDR(.data.percpu) - LOAD_OFFSET) { \
+ VMLINUX_SYMBOL(__per_cpu_load) = .; \
+ VMLINUX_SYMBOL(__per_cpu_start) = .; \
+ *(.data.percpu.first) \
+ *(.data.percpu.page_aligned) \
+ *(.data.percpu) \
+ *(.data.percpu.shared_aligned) \
+ VMLINUX_SYMBOL(__per_cpu_end) = .; \
+ }