x86: fix switch_to() clobbers

Liu Pingfan noticed that switch_to() clobbers more registers than its
asm constraints specify.

We get away with this due to luck mostly - schedule()
by its nature only has 'local' state which gets reloaded
automatically. Fix it nevertheless, we could hit this anytime.

it turns out that with the extra constraints gcc manages to make
schedule() even more compact:

   text	   data	    bss	    dec	    hex	filename
  28626	    684	   2640	  31950	   7cce	sched.o.before
  28613	    684	   2640	  31937	   7cc1	sched.o.after
Reported-by: Liu Pingfan <kernelfans@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

include/asm-x86/system.h

@@ -29,7 +29,14 @@ struct task_struct *__switch_to(struct task_struct *prev,
  */
 #define switch_to(prev, next, last)					\
 do {									\
-	unsigned long esi, edi;						\
+	/*								\
+	 * Context-switching clobbers all registers, so we clobber	\
+	 * them explicitly, via unused output variables.		\
+	 * (EAX and EBP is not listed because EBP is saved/restored	\
+	 * explicitly for wchan access and EAX is the return value of	\
+	 * __switch_to())						\
+	 */								\
+	unsigned long ebx, ecx, edx, esi, edi;				\
 									\
 	asm volatile(							\
 		"pushfl			\n\t"	/* save    flags */	\
@@ -49,6 +56,7 @@ do {									\
 			    "=a" (last),				\
 									\
 		  /* clobbered output registers: */			\
+		  "=b" (ebx), "=c" (ecx), "=d" (edx),			\
 		  "=S" (esi), "=D" (edi)				\
 									\
 		  /* input parameters: */				\