two nptl alpha fixes

[Richard Henderson <rth at twiddle dot net>]

NPTL testing showed up some bugs in atomic.h and clone.S.

Committed.


r~


        * sysdeps/alpha/bits/atomic.h (__arch_compare_and_exchange_bool_*_int):
        Invert the sense of the return value.
        (__arch_exchange_16_int): Fix paste-o.
        (__arch_exchange_{32,64}_int): Fix think-o.

        * sysdeps/unix/sysv/linux/alpha/clone.S: Load child_tid properly.

Index: sysdeps/alpha/bits/atomic.h
===================================================================
RCS file: /cvs/glibc/libc/sysdeps/alpha/bits/atomic.h,v
retrieving revision 1.1
diff -c -p -d -u -r1.1 atomic.h
--- sysdeps/alpha/bits/atomic.h	24 Jun 2003 17:03:59 -0000	1.1
+++ sysdeps/alpha/bits/atomic.h	30 Jun 2003 23:10:57 -0000
@@ -151,27 +151,27 @@ typedef uintmax_t uatomic_max_t;
 	: "memory");							\
 })
 
-/* For all "bool" routines, we return true if exchange succesful.  */
+/* For all "bool" routines, we return FALSE if exchange succesful.  */
 
 #define __arch_compare_and_exchange_bool_8_int(mem, new, old, mb1, mb2)	\
 ({ unsigned long __prev; int __cmp;					\
    __arch_compare_and_exchange_xxx_8_int(mem, new, old, mb1, mb2);	\
-   __cmp; })
+   !__cmp; })
 
 #define __arch_compare_and_exchange_bool_16_int(mem, new, old, mb1, mb2) \
 ({ unsigned long __prev; int __cmp;					\
    __arch_compare_and_exchange_xxx_16_int(mem, new, old, mb1, mb2);	\
-   __cmp; })
+   !__cmp; })
 
 #define __arch_compare_and_exchange_bool_32_int(mem, new, old, mb1, mb2) \
 ({ unsigned long __prev; int __cmp;					\
    __arch_compare_and_exchange_xxx_32_int(mem, new, old, mb1, mb2);	\
-   __cmp; })
+   !__cmp; })
 
 #define __arch_compare_and_exchange_bool_64_int(mem, new, old, mb1, mb2) \
 ({ unsigned long __prev; int __cmp;					\
    __arch_compare_and_exchange_xxx_64_int(mem, new, old, mb1, mb2);	\
-   __cmp; })
+   !__cmp; })
 
 /* For all "val" routines, return the old value whether exchange
    successful or not.  */
@@ -247,7 +247,7 @@ typedef uintmax_t uatomic_max_t;
   unsigned long __ret, __tmp, __addr64, __sval;				\
   __asm__ __volatile__ (						\
 		mb1							\
-	"	andnot	%[__addr8],7,%[__addr64]\n"			\
+	"	andnot	%[__addr16],7,%[__addr64]\n"			\
 	"	inswl	%[__value],%[__addr16],%[__sval]\n"		\
 	"1:	ldq_l	%[__tmp],0(%[__addr64])\n"			\
 	"	extwl	%[__tmp],%[__addr16],%[__ret]\n"		\
@@ -271,13 +271,14 @@ typedef uintmax_t uatomic_max_t;
   __asm__ __volatile__ (						\
 		mb1							\
 	"1:	ldl_l	%[__ret],%[__mem]\n"				\
-	"	mov	%[__ret],%[__tmp]\n"				\
+	"	mov	%[__val],%[__tmp]\n"				\
 	"	stl_c	%[__tmp],%[__mem]\n"				\
 	"	beq	%[__tmp],1b\n"					\
 		mb2							\
 	: [__ret] "=&r" (__ret),					\
 	  [__tmp] "=&r" (__tmp)						\
-	: [__mem] "m" (*(mem))						\
+	: [__mem] "m" (*(mem)),						\
+	  [__val] "Ir" (value)						\
 	: "memory");							\
   __ret; })
 
@@ -287,13 +288,14 @@ typedef uintmax_t uatomic_max_t;
   __asm__ __volatile__ (						\
 		mb1							\
 	"1:	ldq_l	%[__ret],%[__mem]\n"				\
-	"	mov	%[__ret],%[__tmp]\n"				\
+	"	mov	%[__val],%[__tmp]\n"				\
 	"	stq_c	%[__tmp],%[__mem]\n"				\
 	"	beq	%[__tmp],1b\n"					\
 		mb2							\
 	: [__ret] "=&r" (__ret),					\
 	  [__tmp] "=&r" (__tmp)						\
-	: [__mem] "m" (*(mem))						\
+	: [__mem] "m" (*(mem)),						\
+	  [__val] "Ir" (value)						\
 	: "memory");							\
   __ret; })
 
Index: sysdeps/unix/sysv/linux/alpha/clone.S
===================================================================
RCS file: /cvs/glibc/libc/sysdeps/unix/sysv/linux/alpha/clone.S,v
retrieving revision 1.14
diff -c -p -d -u -r1.14 clone.S
--- sysdeps/unix/sysv/linux/alpha/clone.S	24 Jun 2003 16:38:45 -0000	1.14
+++ sysdeps/unix/sysv/linux/alpha/clone.S	30 Jun 2003 23:11:03 -0000
@@ -24,9 +24,13 @@
 #define _ERRNO_H	1
 #include <bits/errno.h>
 
-/* int clone(int (*fn)(void *arg), void *child_stack, int flags, void *arg,
-		       pid_t *tid, void *tls);
- */
+/* int clone(int (*fn)(void *arg), void *child_stack, int flags,
+	     void *arg, pid_t *ptid, void *tls, pid_t *ctid);
+
+   Note that everything past ARG is technically optional, based
+   on FLAGS, and that CTID is arg 7, and thus is on the stack.
+   However, since a load from top-of-stack better be legal always,
+   we don't bother checking FLAGS.  */
 
         .text
 ENTRY(__clone)
@@ -51,11 +55,13 @@ ENTRY(__clone)
 	stq	a0,0(a1)
 	stq	a3,8(a1)
 
-	/* Shift the flags, tid and tls arguments into place; the
+	/* The syscall is of the form clone(flags, usp, ptid, ctid, tls).
+	   Shift the flags, ptid, ctid, tls arguments into place; the
 	   child_stack argument is already correct.  */
 	mov	a2,a0
 	mov	a4,a2
-	mov	a5,a3
+	ldq	a3,0(sp)
+	mov	a5,a4
 
 	/* Do the system call.  */
 	ldiq	v0,__NR_clone