On Thu, Feb 22, 2018 at 11:40:41AM +0100, Arnd Bergmann wrote:
> gcc-8 warns that we copy TRAMPOLINE_32BIT_CODE_SIZE bytes from the pointer
> to the function into actual trampoline, when that pointer is only 8 bytes:
> 
> In file included from arch/x86/boot/compressed/pgtable_64.c:3:
> arch/x86/boot/compressed/pgtable_64.c: In function 'paging_prepare':
> arch/x86/boot/compressed/../string.h:18:23: error: '__builtin_memcpy' reading 
> 96 bytes from a region of size 8 [-Werror=stringop-overflow=]
>  #define memcpy(d,s,l) __builtin_memcpy(d,s,l)
>                        ^~~~~~~~~~~~~~~~~~~~~~~
> arch/x86/boot/compressed/pgtable_64.c:62:2: note: in expansion of macro 
> 'memcpy'
>   memcpy(trampoline + TRAMPOLINE_32BIT_CODE_OFFSET / sizeof(unsigned long),
>   ^~~~~~
> 
> I assume what was intended here is to copy the trampoline itself rather than
> the pointer to the trampoline.
> 
> Cc: Martin Sebor <mse...@gmail.com>
> Fixes: b91993a87aff ("x86/boot/compressed/64: Prepare trampoline memory")
> Signed-off-by: Arnd Bergmann <a...@arndb.de>
> ---
> This is a fairly new warning in gcc, it's possible that the code is
> correct and the warning message got it wrong, please double-check this.
> 
> Was the 32-bit trampoline code in the 64-bit decompressor tested successfully?

Yes, it was. Actually it tested on every boot of any 64-bit machine.

The fix is wrong and it breaks boot. The bug is in trampoline_32bit_src
declaration. Could you check if this helps with the warning:

diff --git a/arch/x86/boot/compressed/pgtable.h 
b/arch/x86/boot/compressed/pgtable.h
index 6e0db2260147..5e0b1c4abef4 100644
--- a/arch/x86/boot/compressed/pgtable.h
+++ b/arch/x86/boot/compressed/pgtable.h
@@ -12,7 +12,7 @@

 #ifndef __ASSEMBLER__

-extern void (*trampoline_32bit_src)(void *return_ptr);
+extern void trampoline_32bit_src(void *return_ptr);

 #endif /* __ASSEMBLER__ */
 #endif /* BOOT_COMPRESSED_PAGETABLE_H */
-- 
 Kirill A. Shutemov

Reply via email to