[Top] [All Lists]

Re: [PATCH] use __always_inline for __xchg

To: Imre Kaloz <>
Subject: Re: [PATCH] use __always_inline for __xchg
From: Ralf Baechle <>
Date: Tue, 24 Nov 2009 13:26:01 +0000
In-reply-to: <op.sldiliqr2s3iss@richese>
Original-recipient: rfc822;
References: <op.sldiliqr2s3iss@richese>
User-agent: Mutt/1.5.19 (2009-01-05)
On Sat, Jan 29, 2005 at 05:54:44PM +0100, Imre Kaloz wrote:

> The following patch changes inline for __xchg to __always_inline in  
> include/asm-mips/system.h, following the changes in git. Without this  
> change linking the kernel fails.
> Signed-off-by: Imre Kaloz <>
> Signed-off-by: Felix Fietkau <>

Imre just pointed me at this antique patch again saying it was still needed.
Back then I ditched it because I couldn't see what the problem it was
trying to fix was - but after a little brainstorming it got obvious.

Now I don't like __always_inline which forces the optimizer to do things
which may not really be what we want, for example when building with -Os
so I found a different fix using BUILD_BUG_ON().  Patch below.



From: Ralf Baechle <>

MIPS: Fix build error if __xchg() is not getting inlined.

If __xchg() is not getting inlined the outline version of the function
will have a reference to __xchg_called_with_bad_pointer() which does not
exist remaining.  Fixed by using BUILD_BUG_ON() xchg() to check for
allowable operand sizes.

Signed-off-by: Ralf Baechle <>

diff --git a/arch/mips/include/asm/system.h b/arch/mips/include/asm/system.h
index fcf5f98..587a48a 100644
--- a/arch/mips/include/asm/system.h
+++ b/arch/mips/include/asm/system.h
@@ -193,10 +193,6 @@ extern __u64 
__xchg_u64_unsupported_on_32bit_kernels(volatile __u64 * m, __u64 v
 #define __xchg_u64 __xchg_u64_unsupported_on_32bit_kernels
-/* This function doesn't exist, so you'll get a linker error
-   if something tries to do an invalid xchg().  */
-extern void __xchg_called_with_bad_pointer(void);
 static inline unsigned long __xchg(unsigned long x, volatile void * ptr, int 
        switch (size) {
@@ -205,11 +201,17 @@ static inline unsigned long __xchg(unsigned long x, 
volatile void * ptr, int siz
        case 8:
                return __xchg_u64(ptr, x);
-       __xchg_called_with_bad_pointer();
        return x;
-#define xchg(ptr, x) ((__typeof__(*(ptr)))__xchg((unsigned long)(x), (ptr), 
+#define xchg(ptr, x)                                                   \
+({                                                                     \
+       BUILD_BUG_ON(sizeof(*(ptr)) & ~0xc);                            \
+                                                                       \
+       ((__typeof__(*(ptr)))                                           \
+               __xchg((unsigned long)(x), (ptr), sizeof(*(ptr))));     \
 extern void set_handler(unsigned long offset, void *addr, unsigned long len);
 extern void set_uncached_handler(unsigned long offset, void *addr, unsigned 
long len);

<Prev in Thread] Current Thread [Next in Thread>
  • Re: [PATCH] use __always_inline for __xchg, Ralf Baechle <=