powerpc: Fix lwsync feature fixup vs. modules on 64-bit
Benjamin Herrenschmidt [Fri, 26 Feb 2010 07:29:17 +0000 (18:29 +1100)]
Anton's commit enabling the use of the lwsync fixup mechanism on 64-bit
breaks modules. The lwsync fixup section uses .long instead of the
FTR_ENTRY_OFFSET macro used by other fixups sections, and thus will
generate 32-bit relocations that our module loader cannot resolve.

This changes it to use the same type as other feature sections.

Note however that we might want to consider using 32-bit for all the
feature fixup offsets and add support for R_PPC_REL32 to module_64.c
instead as that would reduce the size of the kernel image. I'll leave
that as an exercise for the reader for now...

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

arch/powerpc/include/asm/feature-fixups.h
arch/powerpc/lib/feature-fixups.c

index cbd4dfa..96a7d06 100644 (file)
@@ -165,7 +165,7 @@ label##2:                                           \
        .pushsection sect,"a";                          \
        .align 2;                                       \
 label##3:                                              \
-       .long label##1b-label##3b;                      \
+       FTR_ENTRY_OFFSET label##1b-label##3b;           \
        .popsection;
 
 #endif /* __ASM_POWERPC_FEATURE_FIXUPS_H */
index 4dee652..e640175 100644 (file)
@@ -112,7 +112,8 @@ void do_feature_fixups(unsigned long value, void *fixup_start, void *fixup_end)
 
 void do_lwsync_fixups(unsigned long value, void *fixup_start, void *fixup_end)
 {
-       int *start, *end, *dest;
+       long *start, *end;
+       unsigned int *dest;
 
        if (!(value & CPU_FTR_LWSYNC))
                return ;