]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/i386/i386_lock.s
xnu-792.13.8.tar.gz
[apple/xnu.git] / osfmk / i386 / i386_lock.s
index ed2710ccca30e80c1016643144175a943bdccd03..68598c3cbf5965f4d02ad24528c60900a753e71a 100644 (file)
@@ -1,26 +1,31 @@
 /*
  * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
  *
- * @APPLE_LICENSE_HEADER_START@
+ * @APPLE_LICENSE_OSREFERENCE_HEADER_START@
  * 
- * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
- * 
- * This file contains Original Code and/or Modifications of Original Code
- * as defined in and that are subject to the Apple Public Source License
- * Version 2.0 (the 'License'). You may not use this file except in
- * compliance with the License. Please obtain a copy of the License at
- * http://www.opensource.apple.com/apsl/ and read it before using this
+ * This file contains Original Code and/or Modifications of Original Code 
+ * as defined in and that are subject to the Apple Public Source License 
+ * Version 2.0 (the 'License'). You may not use this file except in 
+ * compliance with the License.  The rights granted to you under the 
+ * License may not be used to create, or enable the creation or 
+ * redistribution of, unlawful or unlicensed copies of an Apple operating 
+ * system, or to circumvent, violate, or enable the circumvention or 
+ * violation of, any terms of an Apple operating system software license 
+ * agreement.
+ *
+ * Please obtain a copy of the License at 
+ * http://www.opensource.apple.com/apsl/ and read it before using this 
  * file.
- * 
- * The Original Code and all software distributed under the License are
- * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
- * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
- * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
- * Please see the License for the specific language governing rights and
+ *
+ * The Original Code and all software distributed under the License are 
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER 
+ * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, 
+ * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, 
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. 
+ * Please see the License for the specific language governing rights and 
  * limitations under the License.
- * 
- * @APPLE_LICENSE_HEADER_END@
+ *
+ * @APPLE_LICENSE_OSREFERENCE_HEADER_END@
  */
 /*
  * @OSF_COPYRIGHT@
  * the terms and conditions for use and redistribution.
  */
 
-#include <cpus.h>
 #include <mach_rt.h>
 #include <platforms.h>
 #include <mach_ldebug.h>
 #include <i386/asm.h>
-#include <kern/etap_options.h>
 
 #include "assym.s"
 
+#define        PAUSE           rep; nop
+
 /*
  *     When performance isn't the only concern, it's
  *     nice to build stack frames...
  */
-#define        BUILD_STACK_FRAMES   ((MACH_LDEBUG || ETAP_LOCK_TRACE) && MACH_KDB)
+#define        BUILD_STACK_FRAMES   (GPROF || \
+                               ((MACH_LDEBUG || ETAP_LOCK_TRACE) && MACH_KDB))
 
 #if    BUILD_STACK_FRAMES
 
-#define        L_PC            4(%ebp)
-#define        L_ARG0          8(%ebp)
-#define        L_ARG1          12(%ebp)
+/* STack-frame-relative: */
+#define        L_PC            B_PC
+#define        L_ARG0          B_ARG0
+#define        L_ARG1          B_ARG1
+
+#define LEAF_ENTRY(name)       \
+       Entry(name);            \
+       FRAME;                  \
+       MCOUNT
 
-#define SWT_HI          -4(%ebp)
-#define SWT_LO          -8(%ebp)
-#define MISSED          -12(%ebp)
+#define LEAF_ENTRY2(n1,n2)     \
+       Entry(n1);              \
+       Entry(n2);              \
+       FRAME;                  \
+       MCOUNT
 
-#else   /* BUILD_STACK_FRAMES */
+#define LEAF_RET               \
+       EMARF;                  \
+       ret
+
+#else  /* BUILD_STACK_FRAMES */
+
+/* Stack-pointer-relative: */
+#define        L_PC            S_PC
+#define        L_ARG0          S_ARG0
+#define        L_ARG1          S_ARG1
+
+#define LEAF_ENTRY(name)       \
+       Entry(name)
+
+#define LEAF_ENTRY2(n1,n2)     \
+       Entry(n1);              \
+       Entry(n2)
+
+#define LEAF_RET               \
+       ret
 
-#undef FRAME
-#undef EMARF
-#define        FRAME
-#define        EMARF
-#define        L_PC            (%esp)
-#define        L_ARG0          4(%esp)
-#define        L_ARG1          8(%esp)
+#endif /* BUILD_STACK_FRAMES */
 
-#endif   /* BUILD_STACK_FRAMES */
 
+/* Non-leaf routines always have a stack frame: */
 
-#define        M_ILK                   (%edx)
-#define        M_LOCKED                1(%edx)
-#define        M_WAITERS               2(%edx)
-#define        M_PROMOTED_PRI  4(%edx)
+#define NONLEAF_ENTRY(name)    \
+       Entry(name);            \
+       FRAME;                  \
+       MCOUNT
+
+#define NONLEAF_ENTRY2(n1,n2)  \
+       Entry(n1);              \
+       Entry(n2);              \
+       FRAME;                  \
+       MCOUNT
+
+#define NONLEAF_RET            \
+       EMARF;                  \
+       ret
+
+
+#define        M_ILK           (%edx)
+#define        M_LOCKED        MUTEX_LOCKED(%edx)
+#define        M_WAITERS       MUTEX_WAITERS(%edx)
+#define        M_PROMOTED_PRI  MUTEX_PROMOTED_PRI(%edx)
+#define M_ITAG         MUTEX_ITAG(%edx)
+#define M_PTR          MUTEX_PTR(%edx)
 #if    MACH_LDEBUG
-#define        M_TYPE                  6(%edx)
-#define        M_PC                    10(%edx)
-#define        M_THREAD                14(%edx)
+#define        M_TYPE          MUTEX_TYPE(%edx)
+#define        M_PC            MUTEX_PC(%edx)
+#define        M_THREAD        MUTEX_THREAD(%edx)
 #endif /* MACH_LDEBUG */
 
-#include <i386/AT386/mp/mp.h>
-#if    (NCPUS > 1)
+#include <i386/mp.h>
 #define        CX(addr,reg)    addr(,reg,4)
-#else
-#define        CPU_NUMBER(reg)
-#define        CX(addr,reg)    addr
-#endif /* (NCPUS > 1) */
 
 #if    MACH_LDEBUG
 /*
  *  Routines for general lock debugging.
  */
-#define        S_TYPE          4(%edx)
-#define        S_PC            8(%edx)
-#define        S_THREAD        12(%edx)
-#define        S_DURATIONH     16(%edx)
-#define        S_DURATIONL     20(%edx)
 
 /* 
  * Checks for expected lock types and calls "panic" on
        .text                                           ;       \
 1:
 
-#define        CHECK_SIMPLE_LOCK_TYPE()                                \
-       cmpl    $ SIMPLE_LOCK_TAG,S_TYPE                ;       \
-       je      1f                                      ;       \
-       pushl   $2f                                     ;       \
-       call    EXT(panic)                              ;       \
-       hlt                                             ;       \
-       .data                                           ;       \
-2:     String  "not a simple lock!"                    ;       \
-       .text                                           ;       \
-1:
-
 /*
  * If one or more simplelocks are currently held by a thread,
  * an attempt to acquire a mutex will cause this check to fail
  * (since a mutex lock may context switch, holding a simplelock
  * is not a good thing).
  */
-#if    0 /*MACH_RT - 11/12/99 - lion@apple.com disable check for now*/
+#if    MACH_RT
 #define CHECK_PREEMPTION_LEVEL()                               \
-       movl    $ CPD_PREEMPTION_LEVEL,%eax             ;       \
-       cmpl    $0,%gs:(%eax)                           ;       \
+       cmpl    $0,%gs:CPU_PREEMPTION_LEVEL             ;       \
        je      1f                                      ;       \
        pushl   $2f                                     ;       \
        call    EXT(panic)                              ;       \
 #endif /* MACH_RT */
 
 #define        CHECK_NO_SIMPLELOCKS()                                  \
-       movl    $ CPD_SIMPLE_LOCK_COUNT,%eax            ;       \
-       cmpl    $0,%gs:(%eax)                           ;       \
+       cmpl    $0,%gs:CPU_SIMPLE_LOCK_COUNT            ;       \
        je      1f                                      ;       \
        pushl   $2f                                     ;       \
        call    EXT(panic)                              ;       \
  * Verifies return to the correct thread in "unlock" situations.
  */
 #define        CHECK_THREAD(thd)                                       \
-       movl    $ CPD_ACTIVE_THREAD,%eax                        ;       \
-       movl    %gs:(%eax),%ecx                         ;       \
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx              ;       \
        testl   %ecx,%ecx                               ;       \
        je      1f                                      ;       \
        cmpl    %ecx,thd                                ;       \
 1:
 
 #define        CHECK_MYLOCK(thd)                                       \
-       movl    $ CPD_ACTIVE_THREAD,%eax                        ;       \
-       movl    %gs:(%eax),%ecx                         ;       \
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx              ;       \
        testl   %ecx,%ecx                               ;       \
        je      1f                                      ;       \
        cmpl    %ecx,thd                                ;       \
  *
  *     Initialize a hardware lock.
  */
-ENTRY(hw_lock_init)
-       FRAME
+LEAF_ENTRY(hw_lock_init)
        movl    L_ARG0,%edx             /* fetch lock pointer */
-       xorl    %eax,%eax
-       movb    %al,0(%edx)             /* clear the lock */
-       EMARF
-       ret
+       movl    $0,0(%edx)              /* clear the lock */
+       LEAF_RET
 
 /*
  *     void hw_lock_lock(hw_lock_t)
- *     unsigned int hw_lock_to(hw_lock_t, unsigned int)
  *
  *     Acquire lock, spinning until it becomes available.
- *     XXX:  For now, we don't actually implement the timeout.
  *     MACH_RT:  also return with preemption disabled.
  */
-ENTRY2(hw_lock_lock,hw_lock_to)
-       FRAME
+LEAF_ENTRY(hw_lock_lock)
        movl    L_ARG0,%edx             /* fetch lock pointer */
 
-1:     DISABLE_PREEMPTION(%eax)
-       movb    $1,%cl
-       xchgb   0(%edx),%cl             /* try to acquire the HW lock */
-       testb   %cl,%cl                 /* success? */
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
+       DISABLE_PREEMPTION
+1:
+       movl    0(%edx), %eax
+       testl   %eax,%eax               /* lock locked? */
+       jne     3f                      /* branch if so */
+       lock; cmpxchgl  %ecx,0(%edx)    /* try to acquire the HW lock */
        jne     3f
        movl    $1,%eax                 /* In case this was a timeout call */
-       EMARF                           /* if yes, then nothing left to do */
-       ret
-
-3:     ENABLE_PREEMPTION(%eax)         /* no reason we can't be preemptable now */
+       LEAF_RET                        /* if yes, then nothing left to do */
+3:
+       PAUSE                           /* pause for hyper-threading */
+       jmp     1b                      /* try again */
 
-       movb    $1,%cl
-2:     testb   %cl,0(%edx)             /* spin checking lock value in cache */
-       jne     2b                      /* non-zero means locked, keep spinning */
-       jmp     1b                      /* zero means unlocked, try to grab it */
+/*
+ *     unsigned int hw_lock_to(hw_lock_t, unsigned int)
+ *
+ *     Acquire lock, spinning until it becomes available or timeout.
+ *     MACH_RT:  also return with preemption disabled.
+ */
+LEAF_ENTRY(hw_lock_to)
+1:
+       movl    L_ARG0,%edx             /* fetch lock pointer */
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
+       /*
+        * Attempt to grab the lock immediately
+        * - fastpath without timeout nonsense.
+        */
+       DISABLE_PREEMPTION
+       movl    0(%edx), %eax
+       testl   %eax,%eax               /* lock locked? */
+       jne     2f                      /* branch if so */
+       lock; cmpxchgl  %ecx,0(%edx)    /* try to acquire the HW lock */
+       jne     2f                      /* branch on failure */
+       movl    $1,%eax
+       LEAF_RET
+
+2:
+#define        INNER_LOOP_COUNT        1000
+       /*
+        * Failed to get the lock so set the timeout
+        * and then spin re-checking the lock but pausing
+        * every so many (INNER_LOOP_COUNT) spins to check for timeout.
+        */
+       movl    L_ARG1,%ecx             /* fetch timeout */
+       push    %edi
+       push    %ebx
+       mov     %edx,%edi
+
+       rdtsc                           /* read cyclecount into %edx:%eax */
+       addl    %ecx,%eax               /* fetch and timeout */
+       adcl    $0,%edx                 /* add carry */
+       mov     %edx,%ecx
+       mov     %eax,%ebx               /* %ecx:%ebx is the timeout expiry */
+4:
+       /*
+        * The inner-loop spin to look for the lock being freed.
+        */
+       mov     $(INNER_LOOP_COUNT),%edx
+5:
+       PAUSE                           /* pause for hyper-threading */
+       movl    0(%edi),%eax            /* spin checking lock value in cache */
+       testl   %eax,%eax
+       je      6f                      /* zero => unlocked, try to grab it */
+       decl    %edx                    /* decrement inner loop count */
+       jnz     5b                      /* time to check for timeout? */
+
+       /*
+        * Here after spinning INNER_LOOP_COUNT times, check for timeout
+        */
+       rdtsc                           /* cyclecount into %edx:%eax */
+       cmpl    %ecx,%edx               /* compare high-order 32-bits */
+       jb      4b                      /* continue spinning if less, or */
+       cmpl    %ebx,%eax               /* compare low-order 32-bits */ 
+       jb      4b                      /* continue if less, else bail */
+       xor     %eax,%eax               /* with 0 return value */
+       pop     %ebx
+       pop     %edi
+       LEAF_RET
+
+6:
+       /*
+        * Here to try to grab the lock that now appears to be free
+        * after contention.
+        */
+       movl    %gs:CPU_ACTIVE_THREAD,%edx
+       lock; cmpxchgl  %edx,0(%edi)    /* try to acquire the HW lock */
+       jne     4b                      /* no - spin again */
+       movl    $1,%eax                 /* yes */
+       pop     %ebx
+       pop     %edi
+       LEAF_RET
 
 /*
  *     void hw_lock_unlock(hw_lock_t)
@@ -261,504 +352,575 @@ ENTRY2(hw_lock_lock,hw_lock_to)
  *     Unconditionally release lock.
  *     MACH_RT:  release preemption level.
  */
-ENTRY(hw_lock_unlock)
-       FRAME
+LEAF_ENTRY(hw_lock_unlock)
        movl    L_ARG0,%edx             /* fetch lock pointer */
-       xorl    %eax,%eax
-       xchgb   0(%edx),%al             /* clear the lock... a mov instruction */
-                                       /* ...might be cheaper and less paranoid */
-       ENABLE_PREEMPTION(%eax)
-       EMARF
-       ret
+       movl    $0,0(%edx)              /* clear the lock */
+       ENABLE_PREEMPTION
+       LEAF_RET
 
 /*
  *     unsigned int hw_lock_try(hw_lock_t)
  *     MACH_RT:  returns with preemption disabled on success.
  */
-ENTRY(hw_lock_try)
-       FRAME
+LEAF_ENTRY(hw_lock_try)
        movl    L_ARG0,%edx             /* fetch lock pointer */
 
-       DISABLE_PREEMPTION(%eax)
-       movb    $1,%cl
-       xchgb   0(%edx),%cl             /* try to acquire the HW lock */
-       testb   %cl,%cl                 /* success? */
-       jne     1f                      /* if yes, let the caller know */
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
+       DISABLE_PREEMPTION
+       movl    0(%edx),%eax
+       testl   %eax,%eax
+       jne     1f
+       lock; cmpxchgl  %ecx,0(%edx)    /* try to acquire the HW lock */
+       jne     1f
 
        movl    $1,%eax                 /* success */
-       EMARF
-       ret
+       LEAF_RET
 
-1:     ENABLE_PREEMPTION(%eax)         /* failure:  release preemption... */
+1:
+       ENABLE_PREEMPTION               /* failure:  release preemption... */
        xorl    %eax,%eax               /* ...and return failure */
-       EMARF
-       ret     
+       LEAF_RET
 
 /*
  *     unsigned int hw_lock_held(hw_lock_t)
  *     MACH_RT:  doesn't change preemption state.
  *     N.B.  Racy, of course.
  */
-ENTRY(hw_lock_held)
-       FRAME
+LEAF_ENTRY(hw_lock_held)
        movl    L_ARG0,%edx             /* fetch lock pointer */
 
-       movb    $1,%cl
-       testb   %cl,0(%edx)             /* check lock value */
-       jne     1f                      /* non-zero means locked */
-       xorl    %eax,%eax               /* tell caller:  lock wasn't locked */
-       EMARF
-       ret     
+       movl    0(%edx),%eax            /* check lock value */
+       testl   %eax,%eax
+       movl    $1,%ecx
+       cmovne  %ecx,%eax               /* 0 => unlocked, 1 => locked */
+       LEAF_RET
 
-1:     movl    $1,%eax                 /* tell caller:  lock was locked */
-       EMARF
-       ret
-       
+LEAF_ENTRY(mutex_init)
+       movl    L_ARG0,%edx             /* fetch lock pointer */
+       xorl    %eax,%eax
+       movl    %eax,M_ILK              /* clear interlock */
+       movl    %eax,M_LOCKED           /* clear locked flag */
+       movw    %ax,M_WAITERS           /* init waiter count */
+       movw    %ax,M_PROMOTED_PRI
 
+#if    MACH_LDEBUG
+       movl    $ MUTEX_TAG,M_TYPE      /* set lock type */
+       movl    %eax,M_PC               /* init caller pc */
+       movl    %eax,M_THREAD           /* and owning thread */
+#endif
 
-#if    0
+       LEAF_RET
 
+NONLEAF_ENTRY2(mutex_lock,_mutex_lock)
 
-ENTRY(_usimple_lock_init)
-       FRAME
-       movl    L_ARG0,%edx             /* fetch lock pointer */
-       xorl    %eax,%eax
-       movb    %al,USL_INTERLOCK(%edx) /* unlock the HW lock */
-       EMARF
-       ret
+       movl    B_ARG0,%edx             /* fetch lock pointer */
 
-ENTRY(_simple_lock)
-       FRAME
-       movl    L_ARG0,%edx             /* fetch lock pointer */
+       CHECK_MUTEX_TYPE()
+       CHECK_NO_SIMPLELOCKS()
+       CHECK_PREEMPTION_LEVEL()
 
-       CHECK_SIMPLE_LOCK_TYPE()
+       pushf                           /* save interrupt state */
+       cli                             /* disable interrupts */
+Lml_retry:
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
 
-       DISABLE_PREEMPTION(%eax)
+Lml_get_hw:
+       movl    M_ILK,%eax              /* read interlock */
+       testl   %eax,%eax               /* unlocked? */
+       jne     Lml_ilk_fail            /* no - take the slow path */
 
-sl_get_hw:
-       movb    $1,%cl
-       xchgb   USL_INTERLOCK(%edx),%cl /* try to acquire the HW lock */
-       testb   %cl,%cl                 /* did we succeed? */
+       lock; cmpxchgl  %ecx,M_ILK      /* atomic compare and exchange */
+       jne     Lml_get_hw              /* branch on failure to retry */
 
-#if    MACH_LDEBUG
-       je      5f
-       CHECK_MYLOCK(S_THREAD)
-       jmp     sl_get_hw
-5:
-#else  /* MACH_LDEBUG */
-       jne     sl_get_hw               /* no, try again */
-#endif /* MACH_LDEBUG */
+       movl    M_LOCKED,%ecx           /* get lock owner */
+       testl   %ecx,%ecx               /* is the mutex locked? */
+       jne     Lml_fail                /* yes, we lose */
+Lml_acquire:
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
+       movl    %ecx,M_LOCKED
 
 #if    MACH_LDEBUG
-       movl    L_PC,%ecx
-       movl    %ecx,S_PC
-       movl    $ CPD_ACTIVE_THREAD,%eax
-       movl    %gs:(%eax),%ecx
-       movl    %ecx,S_THREAD
-       incl    CX(EXT(simple_lock_count),%eax)
-#if 0
-       METER_SIMPLE_LOCK_LOCK(%edx)
+       movl    %ecx,M_THREAD
+       movl    B_PC,%ecx
+       movl    %ecx,M_PC
 #endif
-#if    NCPUS == 1
-       pushf
+
+       cmpw    $0,M_WAITERS            /* are there any waiters? */
+       jne     Lml_waiters             /* yes, more work to do */
+Lml_return:
+       xorl    %eax,%eax
+       movl    %eax,M_ILK
+
+       popf                            /* restore interrupt state */
+
+       NONLEAF_RET
+
+Lml_waiters:
+       pushl   %edx                    /* save mutex address */
        pushl   %edx
-       cli
-       call    EXT(lock_stack_push)
-       popl    %edx
-       popfl
-#endif /* NCPUS == 1 */
-#endif /* MACH_LDEBUG */
+       call    EXT(lck_mtx_lock_acquire)
+       addl    $4,%esp
+       popl    %edx                    /* restore mutex address */
+       jmp     Lml_return
+
+Lml_ilk_fail:
+       /*
+        * Slow path: call out to do the spinning.
+        */
+       pushl   %edx                    /* lock address */
+       call    EXT(lck_mtx_interlock_spin)
+       popl    %edx                    /* lock pointer */
+       jmp     Lml_retry               /* try again */
+
+Lml_fail:
+       /*
+        n Check if the owner is on another processor and therefore
+        * we should try to spin before blocking.
+        */
+       testl   $(OnProc),ACT_SPF(%ecx)
+       jz      Lml_block
+
+       /*
+        * Here if owner is on another processor:
+        *  - release the interlock
+        *  - spin on the holder until release or timeout
+        *  - in either case re-acquire the interlock
+        *  - if released, acquire it
+        *  - otherwise drop thru to block.
+        */
+       xorl    %eax,%eax
+       movl    %eax,M_ILK              /* zero interlock */
+       popf
+       pushf                           /* restore interrupt state */
 
-       EMARF
-       ret
+       push    %edx                    /* lock address */
+       call    EXT(lck_mtx_lock_spin)  /* call out to do spinning */
+       addl    $4,%esp
+       movl    B_ARG0,%edx             /* refetch mutex address */
 
-ENTRY(_simple_lock_try)
-       FRAME
-       movl    L_ARG0,%edx             /* fetch lock pointer */
+       /* Re-acquire interlock */
+       cli                             /* disable interrupts */
+Lml_reget_retry:
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
 
-       CHECK_SIMPLE_LOCK_TYPE()
+Lml_reget_hw:
+       movl    M_ILK,%eax              /* read interlock */
+       testl   %eax,%eax               /* unlocked? */
+       jne     Lml_ilk_refail          /* no - slow path */
 
-       DISABLE_PREEMPTION(%eax)
+       lock; cmpxchgl  %ecx,M_ILK      /* atomic compare and exchange */
+       jne     Lml_reget_hw            /* branch on failure to retry */
 
-       movb    $1,%cl
-       xchgb   USL_INTERLOCK(%edx),%cl /* try to acquire the HW lock */
-       testb   %cl,%cl                 /* did we succeed? */
-       jne     1f                      /* no, return failure */
+       movl    M_LOCKED,%ecx           /* get lock owner */
+       testl   %ecx,%ecx               /* is the mutex free? */
+       je      Lml_acquire             /* yes, acquire */
+       
+Lml_block:
+       CHECK_MYLOCK(M_THREAD)
+       pushl   M_LOCKED
+       pushl   %edx                    /* push mutex address */
+       call    EXT(lck_mtx_lock_wait)  /* wait for the lock */
+       addl    $8,%esp
+       movl    B_ARG0,%edx             /* refetch mutex address */
+       cli                             /* ensure interrupts disabled */
+       jmp     Lml_retry               /* and try again */
 
-#if    MACH_LDEBUG
-       movl    L_PC,%ecx
-       movl    %ecx,S_PC
-       movl    $ CPD_ACTIVE_THREAD,%eax
-       movl    %gs:(%eax),%ecx
-       movl    %ecx,S_THREAD
-       incl    CX(EXT(simple_lock_count),%eax)
-#if 0
-       METER_SIMPLE_LOCK_LOCK(%edx)
-#endif
-#if    NCPUS == 1
-       pushf
-       pushl   %edx
-       cli
-       call    EXT(lock_stack_push)
-       popl    %edx
-       popfl
-#endif /* NCPUS == 1 */
-#endif /* MACH_LDEBUG */
+Lml_ilk_refail:
+       /*
+        * Slow path: call out to do the spinning.
+        */
+       pushl   %edx                    /* lock address */
+       call    EXT(lck_mtx_interlock_spin)
+       popl    %edx                    /* lock pointer */
+       jmp     Lml_reget_retry         /* try again */
 
-       movl    $1,%eax                 /* return success */
+NONLEAF_ENTRY2(mutex_try,_mutex_try)   
 
-       EMARF
-       ret
+       movl    B_ARG0,%edx             /* fetch lock pointer */
 
-1:
-       ENABLE_PREEMPTION(%eax)
+       CHECK_MUTEX_TYPE()
+       CHECK_NO_SIMPLELOCKS()
 
-       xorl    %eax,%eax               /* and return failure */
+       pushf                           /* save interrupt state */
+       cli                             /* disable interrupts */
+Lmt_retry:
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
 
-       EMARF
-       ret
+Lmt_get_hw:
+       movl    M_ILK,%eax              /* read interlock */
+       testl   %eax,%eax               /* unlocked? */
+       jne     Lmt_ilk_fail            /* no - slow path */
 
-ENTRY(_simple_unlock)
-       FRAME
-       movl    L_ARG0,%edx             /* fetch lock pointer */
+       lock; cmpxchgl  %ecx,M_ILK      /* atomic compare and exchange */
+       jne     Lmt_get_hw              /* branch on failure to retry */
 
-       CHECK_SIMPLE_LOCK_TYPE()
-       CHECK_THREAD(S_THREAD)
+       movl    M_LOCKED,%ecx           /* get lock owner */
+       testl   %ecx,%ecx               /* is the mutex locked? */
+       jne     Lmt_fail                /* yes, we lose */
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
+       movl    %ecx,M_LOCKED
 
 #if    MACH_LDEBUG
-       xorl    %eax,%eax
-       movl    %eax,S_THREAD           /* disown thread */
-       MP_DISABLE_PREEMPTION(%eax)
-       CPU_NUMBER(%eax)
-       decl    CX(EXT(simple_lock_count),%eax)
-       MP_ENABLE_PREEMPTION(%eax)
-#if 0
-       METER_SIMPLE_LOCK_UNLOCK(%edx)
+       movl    %ecx,M_THREAD
+       movl    B_PC,%ecx
+       movl    %ecx,M_PC
 #endif
-#if    NCPUS == 1
-       pushf
-       pushl   %edx
-       cli
-       call    EXT(lock_stack_pop)
-       popl    %edx
-       popfl
-#endif /* NCPUS == 1 */
-#endif /* MACH_LDEBUG */
-
-       xorb    %cl,%cl
-       xchgb   USL_INTERLOCK(%edx),%cl /* unlock the HW lock */
 
-       ENABLE_PREEMPTION(%eax)
-
-       EMARF
-       ret
+       cmpl    $0,M_WAITERS            /* are there any waiters? */
+       jne     Lmt_waiters             /* yes, more work to do */
+Lmt_return:
+       xorl    %eax,%eax
+       movl    %eax,M_ILK
+       popf                            /* restore interrupt state */
 
-#endif /* 0 */
+       movl    $1,%eax
 
+       NONLEAF_RET
 
-ENTRY(mutex_init)
-       FRAME
-       movl    L_ARG0,%edx             /* fetch lock pointer */
+Lmt_waiters:
+       pushl   %edx                    /* save mutex address */
+       pushl   %edx
+       call    EXT(lck_mtx_lock_acquire)
+       addl    $4,%esp
+       popl    %edx                    /* restore mutex address */
+       jmp     Lmt_return
+
+Lmt_ilk_fail:
+       /*
+        * Slow path: call out to do the spinning.
+        */
+       pushl   %edx                    /* lock address */
+       call    EXT(lck_mtx_interlock_spin)
+       popl    %edx                    /* lock pointer */
+       jmp     Lmt_retry               /* try again */
+
+Lmt_fail:
        xorl    %eax,%eax
-       movb    %al,M_ILK               /* clear interlock */
-       movb    %al,M_LOCKED            /* clear locked flag */
-       movw    %ax,M_WAITERS           /* init waiter count */
-       movw    %ax,M_PROMOTED_PRI
+       movl    %eax,M_ILK
 
-#if    MACH_LDEBUG
-       movl    $ MUTEX_TAG,M_TYPE      /* set lock type */
-       movl    %eax,M_PC               /* init caller pc */
-       movl    %eax,M_THREAD           /* and owning thread */
-#endif
-#if    ETAP_LOCK_TRACE
-       movl    L_ARG1,%ecx             /* fetch event type */
-       pushl   %ecx                    /* push event type */
-       pushl   %edx                    /* push mutex address */
-       call    EXT(etap_mutex_init)    /* init ETAP data */
-       addl    $8,%esp
-#endif /* ETAP_LOCK_TRACE */
-
-       EMARF
-       ret
+       popf                            /* restore interrupt state */
 
-ENTRY2(mutex_lock,_mutex_lock)
-       FRAME
+       xorl    %eax,%eax
 
-#if    ETAP_LOCK_TRACE
-       subl    $12,%esp                /* make room for locals */
-       movl    $0,SWT_HI               /* set wait time to zero (HI) */
-       movl    $0,SWT_LO               /* set wait time to zero (LO) */
-       movl    $0,MISSED               /* clear local miss marker */
-#endif /* ETAP_LOCK_TRACE */
+       NONLEAF_RET
 
-       movl    L_ARG0,%edx             /* fetch lock pointer */
+NONLEAF_ENTRY(mutex_unlock)
+       movl    B_ARG0,%edx             /* fetch lock pointer */
 
        CHECK_MUTEX_TYPE()
-       CHECK_NO_SIMPLELOCKS()
-       CHECK_PREEMPTION_LEVEL()
+       CHECK_THREAD(M_THREAD)
 
-ml_retry:
-       DISABLE_PREEMPTION(%eax)
+       pushf                           /* save interrupt state */
+       cli                             /* disable interrupts */
+Lmu_retry:
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
 
-ml_get_hw:
-       movb    $1,%cl
-       xchgb   %cl,M_ILK
-       testb   %cl,%cl                 /* did we succeed? */
-       jne     ml_get_hw               /* no, try again */
+Lmu_get_hw:
+       movl    M_ILK,%eax              /* read interlock */
+       testl   %eax,%eax               /* unlocked? */
+       jne     Lmu_ilk_fail            /* no - slow path */
 
-       movb    $1,%cl
-       xchgb   %cl,M_LOCKED            /* try to set locked flag */
-       testb   %cl,%cl                 /* is the mutex locked? */
-       jne     ml_fail                 /* yes, we lose */
+       lock; cmpxchgl  %ecx,M_ILK      /* atomic compare and exchange */
+       jne     Lmu_get_hw              /* branch on failure to retry */
 
-       pushl   %edx
-       call    EXT(mutex_lock_acquire)
-       addl    $4,%esp
-       movl    L_ARG0,%edx
+       cmpw    $0,M_WAITERS            /* are there any waiters? */
+       jne     Lmu_wakeup              /* yes, more work to do */
+
+Lmu_doit:
 
 #if    MACH_LDEBUG
-       movl    L_PC,%ecx
-       movl    %ecx,M_PC
-       movl    $ CPD_ACTIVE_THREAD,%eax
-       movl    %gs:(%eax),%ecx
-       movl    %ecx,M_THREAD
-       testl   %ecx,%ecx
-       je      3f
-       incl    TH_MUTEX_COUNT(%ecx)
-3:
+       movl    $0,M_THREAD             /* disown thread */
 #endif
 
-       xorb    %cl,%cl
-       xchgb   %cl,M_ILK
+       xorl    %ecx,%ecx
+       movl    %ecx,M_LOCKED           /* unlock the mutex */
 
-       ENABLE_PREEMPTION(%eax)
+       movl    %ecx,M_ILK
 
-#if    ETAP_LOCK_TRACE
-       movl    L_PC,%eax               /* fetch pc */
-       pushl   SWT_LO                  /* push wait time (low) */
-       pushl   SWT_HI                  /* push wait time (high) */
-       pushl   %eax                    /* push pc */
-       pushl   %edx                    /* push mutex address */
-       call    EXT(etap_mutex_hold)    /* collect hold timestamp */
-       addl    $16+12,%esp             /* clean up stack, adjusting for locals */
-#endif /* ETAP_LOCK_TRACE */
+       popf                            /* restore interrupt state */
 
-       EMARF
-       ret
+       NONLEAF_RET
 
-ml_fail:
-#if    ETAP_LOCK_TRACE
-       cmp     $0,MISSED               /* did we already take a wait timestamp? */
-       jne     ml_block                /* yup. carry-on */
-       pushl   %edx                    /* push mutex address */
-       call    EXT(etap_mutex_miss)    /* get wait timestamp */
-       movl    %eax,SWT_HI             /* set wait time (high word) */
-       movl    %edx,SWT_LO             /* set wait time (low word) */
-       popl    %edx                    /* clean up stack */
-       movl    $1,MISSED               /* mark wait timestamp as taken */
-#endif /* ETAP_LOCK_TRACE */
-
-ml_block:
-       CHECK_MYLOCK(M_THREAD)
-       xorl    %eax,%eax
-       pushl   %eax                    /* no promotion here yet */
+Lmu_ilk_fail:
+       /*
+        * Slow path: call out to do the spinning.
+        */
+       pushl   %edx                    /* lock address */
+       call    EXT(lck_mtx_interlock_spin)
+       popl    %edx                    /* lock pointer */
+       jmp     Lmu_retry               /* try again */
+
+Lmu_wakeup:
+       pushl   M_LOCKED
        pushl   %edx                    /* push mutex address */
-       call    EXT(mutex_lock_wait)    /* wait for the lock */
+       call    EXT(lck_mtx_unlock_wakeup)/* yes, wake a thread */
        addl    $8,%esp
-       movl    L_ARG0,%edx             /* refetch lock pointer */
-       jmp     ml_retry                /* and try again */
+       movl    B_ARG0,%edx             /* restore lock pointer */
+       jmp     Lmu_doit
 
-ENTRY2(mutex_try,_mutex_try)   
-       FRAME
-
-#if    ETAP_LOCK_TRACE
-       subl    $8,%esp                 /* make room for locals */
-       movl    $0,SWT_HI               /* set wait time to zero (HI) */
-       movl    $0,SWT_LO               /* set wait time to zero (LO) */
-#endif /* ETAP_LOCK_TRACE */
+/*
+ * lck_mtx_lock()
+ * lck_mtx_try_lock()
+ * lck_mutex_unlock()
+ *
+ * These are variants of mutex_lock(), mutex_try() and mutex_unlock() without
+ * DEBUG checks (which require fields not present in lck_mtx_t's).
+ */
+NONLEAF_ENTRY(lck_mtx_lock)
 
-       movl    L_ARG0,%edx             /* fetch lock pointer */
+       movl    B_ARG0,%edx             /* fetch lock pointer */
+       cmpl    $(MUTEX_IND),M_ITAG     /* is this indirect? */
+       cmove   M_PTR,%edx              /* yes - take indirection */
 
-       CHECK_MUTEX_TYPE()
        CHECK_NO_SIMPLELOCKS()
+       CHECK_PREEMPTION_LEVEL()
 
-       DISABLE_PREEMPTION(%eax)
+       pushf                           /* save interrupt state */
+       cli                             /* disable interrupts */
+Llml_retry:
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
+
+Llml_get_hw:
+       movl    M_ILK,%eax              /* read interlock */
+       testl   %eax,%eax               /* unlocked? */
+       jne     Llml_ilk_fail           /* no - slow path */
+
+       lock; cmpxchgl  %ecx,M_ILK      /* atomic compare and exchange */
+       jne     Llml_get_hw             /* branch on failure to retry */
+
+       movl    M_LOCKED,%ecx           /* get lock owner */
+       testl   %ecx,%ecx               /* is the mutex locked? */
+       jne     Llml_fail               /* yes, we lose */
+Llml_acquire:
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
+       movl    %ecx,M_LOCKED
+
+       cmpl    $0,M_WAITERS            /* are there any waiters? */
+       jne     Llml_waiters            /* yes, more work to do */
+Llml_return:
+       xorl    %eax,%eax
+       movl    %eax,M_ILK
 
-mt_get_hw:
-       movb    $1,%cl
-       xchgb   %cl,M_ILK
-       testb   %cl,%cl
-       jne             mt_get_hw
+       popf                            /* restore interrupt state */
 
-       movb    $1,%cl
-       xchgb   %cl,M_LOCKED
-       testb   %cl,%cl
-       jne             mt_fail
+       NONLEAF_RET
 
+Llml_waiters:
+       pushl   %edx                    /* save mutex address */
        pushl   %edx
-       call    EXT(mutex_lock_acquire)
+       call    EXT(lck_mtx_lock_acquire)
        addl    $4,%esp
-       movl    L_ARG0,%edx
+       popl    %edx                    /* restore mutex address */
+       jmp     Llml_return
+
+Llml_ilk_fail:
+       /*
+        * Slow path: call out to do the spinning.
+        */
+       pushl   %edx                    /* lock address */
+       call    EXT(lck_mtx_interlock_spin)
+       popl    %edx                    /* lock pointer */
+       jmp     Llml_retry              /* try again */
+
+Llml_fail:
+       /*
+        * Check if the owner is on another processor and therefore
+        * we should try to spin before blocking.
+        */
+       testl   $(OnProc),ACT_SPF(%ecx)
+       jz      Llml_block
+
+       /*
+        * Here if owner is on another processor:
+        *  - release the interlock
+        *  - spin on the holder until release or timeout
+        *  - in either case re-acquire the interlock
+        *  - if released, acquire it
+        *  - otherwise drop thru to block.
+        */
+       xorl    %eax,%eax
+       movl    %eax,M_ILK              /* zero interlock */
+       popf
+       pushf                           /* restore interrupt state */
 
-#if    MACH_LDEBUG
-       movl    L_PC,%ecx
-       movl    %ecx,M_PC
-       movl    $ CPD_ACTIVE_THREAD,%ecx
-       movl    %gs:(%ecx),%ecx
-       movl    %ecx,M_THREAD
-       testl   %ecx,%ecx
-       je      1f
-       incl    TH_MUTEX_COUNT(%ecx)
-1:
-#endif
+       pushl   %edx                    /* save mutex address */
+       pushl   %edx
+       call    EXT(lck_mtx_lock_spin)
+       addl    $4,%esp
+       popl    %edx                    /* restore mutex address */
 
-       xorb    %cl,%cl
-       xchgb   %cl,M_ILK
+       /* Re-acquire interlock */
+       cli                             /* disable interrupts */
+Llml_reget_retry:
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
 
-       ENABLE_PREEMPTION(%eax)
+Llml_reget_hw:
+       movl    M_ILK,%eax              /* read interlock */
+       testl   %eax,%eax               /* unlocked? */
+       jne     Llml_ilk_refail         /* no - slow path */
 
-#if    ETAP_LOCK_TRACE
-       movl    L_PC,%eax               /* fetch pc */
-       pushl   SWT_LO                  /* push wait time (low) */
-       pushl   SWT_HI                  /* push wait time (high) */
-       pushl   %eax                    /* push pc */
+       lock; cmpxchgl  %ecx,M_ILK      /* atomic compare and exchange */
+       jne     Llml_reget_hw           /* branch on failure to retry */
+
+       movl    M_LOCKED,%ecx           /* get lock owner */
+       testl   %ecx,%ecx               /* is the mutex free? */
+       je      Llml_acquire            /* yes, acquire */
+       
+Llml_block:
+       CHECK_MYLOCK(M_THREAD)
+       pushl   %edx                    /* save mutex address */
+       pushl   M_LOCKED
        pushl   %edx                    /* push mutex address */
-       call    EXT(etap_mutex_hold)    /* get start hold timestamp */
-       addl    $16,%esp                /* clean up stack, adjusting for locals */
-#endif /* ETAP_LOCK_TRACE */
+       call    EXT(lck_mtx_lock_wait)  /* wait for the lock */
+       addl    $8,%esp
+       popl    %edx                    /* restore mutex address */
+       cli                             /* ensure interrupts disabled */
+       jmp     Llml_retry              /* and try again */
 
-       movl    $1,%eax
+Llml_ilk_refail:
+       /*
+        * Slow path: call out to do the spinning.
+        */
+       pushl   %edx                    /* lock address */
+       call    EXT(lck_mtx_interlock_spin)
+       popl    %edx                    /* lock pointer */
+       jmp     Llml_reget_retry        /* try again */
 
-#if    MACH_LDEBUG || ETAP_LOCK_TRACE
-#if    ETAP_LOCK_TRACE
-       addl    $8,%esp                 /* pop stack claimed on entry */
-#endif
-#endif
+NONLEAF_ENTRY(lck_mtx_try_lock)
 
-       EMARF
-       ret
+       movl    B_ARG0,%edx             /* fetch lock pointer */
+       cmpl    $(MUTEX_IND),M_ITAG     /* is this indirect? */
+       cmove   M_PTR,%edx              /* yes - take indirection */
 
-mt_fail:
-#if    MACH_LDEBUG
-       movl    L_PC,%ecx
-       movl    %ecx,M_PC
-       movl    $ CPD_ACTIVE_THREAD,%ecx
-       movl    %gs:(%ecx),%ecx
-       movl    %ecx,M_THREAD
-       testl   %ecx,%ecx
-       je      1f
-       incl    TH_MUTEX_COUNT(%ecx)
-1:
-#endif
+       CHECK_NO_SIMPLELOCKS()
+       CHECK_PREEMPTION_LEVEL()
 
-       xorb    %cl,%cl
-       xchgb   %cl,M_ILK
+       pushf                           /* save interrupt state */
+       cli                             /* disable interrupts */
+Llmt_retry:
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
 
-       ENABLE_PREEMPTION(%eax)
+Llmt_get_hw:
+       movl    M_ILK,%eax              /* read interlock */
+       testl   %eax,%eax               /* unlocked? */
+       jne     Llmt_ilk_fail           /* no - slow path */
 
-#if    ETAP_LOCK_TRACE
-       movl    L_PC,%eax               /* fetch pc */
-       pushl   SWT_LO                  /* push wait time (low) */
-       pushl   SWT_HI                  /* push wait time (high) */
-       pushl   %eax                    /* push pc */
-       pushl   %edx                    /* push mutex address */
-       call    EXT(etap_mutex_hold)    /* get start hold timestamp */
-       addl    $16,%esp                /* clean up stack, adjusting for locals */
-#endif /* ETAP_LOCK_TRACE */
+       lock; cmpxchgl  %ecx,M_ILK      /* atomic compare and exchange */
+       jne     Llmt_get_hw             /* branch on failure to retry */
 
-       xorl    %eax,%eax
+       movl    M_LOCKED,%ecx           /* get lock owner */
+       testl   %ecx,%ecx               /* is the mutex locked? */
+       jne     Llmt_fail               /* yes, we lose */
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
+       movl    %ecx,M_LOCKED
 
-#if    MACH_LDEBUG || ETAP_LOCK_TRACE
-#if    ETAP_LOCK_TRACE
-       addl    $8,%esp                 /* pop stack claimed on entry */
-#endif
-#endif
+       cmpl    $0,M_WAITERS            /* are there any waiters? */
+       jne     Llmt_waiters            /* yes, more work to do */
+Llmt_return:
+       xorl    %eax,%eax
+       movl    %eax,M_ILK
 
-       EMARF
-       ret
+       popf                            /* restore interrupt state */
 
-ENTRY(mutex_unlock)
-       FRAME
-       movl    L_ARG0,%edx             /* fetch lock pointer */
+       movl    $1,%eax                 /* return success */
+       NONLEAF_RET
 
-#if    ETAP_LOCK_TRACE
-       pushl   %edx                    /* push mutex address */
-       call    EXT(etap_mutex_unlock)  /* collect ETAP data */
+Llmt_waiters:
+       pushl   %edx                    /* save mutex address */
+       pushl   %edx
+       call    EXT(lck_mtx_lock_acquire)
+       addl    $4,%esp
        popl    %edx                    /* restore mutex address */
-#endif /* ETAP_LOCK_TRACE */
+       jmp     Llmt_return
+
+Llmt_ilk_fail:
+       /*
+        * Slow path: call out to do the spinning.
+        */
+       pushl   %edx                    /* lock address */
+       call    EXT(lck_mtx_interlock_spin)
+       popl    %edx                    /* lock pointer */
+       jmp     Llmt_retry              /* try again */
+
+Llmt_fail:
+       xorl    %eax,%eax
+       movl    %eax,M_ILK
 
-       CHECK_MUTEX_TYPE()
-       CHECK_THREAD(M_THREAD)
+       popf                            /* restore interrupt state */
+
+       xorl    %eax,%eax               /* return failure */
+       NONLEAF_RET
+
+NONLEAF_ENTRY(lck_mtx_unlock)
 
-       DISABLE_PREEMPTION(%eax)
+       movl    B_ARG0,%edx             /* fetch lock pointer */
+       cmpl    $(MUTEX_IND),M_ITAG     /* is this indirect? */
+       cmove   M_PTR,%edx              /* yes - take indirection */
 
-mu_get_hw:
-       movb    $1,%cl
-       xchgb   %cl,M_ILK
-       testb   %cl,%cl                 /* did we succeed? */
-       jne     mu_get_hw               /* no, try again */
+       pushf                           /* save interrupt state */
+       cli                             /* disable interrupts */
+Llmu_retry:
+       movl    %gs:CPU_ACTIVE_THREAD,%ecx
+
+Llmu_get_hw:
+       movl    M_ILK,%eax              /* read interlock */
+       testl   %eax,%eax               /* unlocked? */
+       jne     Llmu_ilk_fail           /* no - slow path */
+
+       lock; cmpxchgl  %ecx,M_ILK      /* atomic compare and exchange */
+       jne     Llmu_get_hw             /* branch on failure to retry */
 
        cmpw    $0,M_WAITERS            /* are there any waiters? */
-       jne     mu_wakeup               /* yes, more work to do */
+       jne     Llmu_wakeup             /* yes, more work to do */
 
-mu_doit:
-#if    MACH_LDEBUG
-       xorl    %eax,%eax
-       movl    %eax,M_THREAD           /* disown thread */
-       movl    $ CPD_ACTIVE_THREAD,%eax
-       movl    %gs:(%eax),%ecx
-       testl   %ecx,%ecx
-       je      0f
-       decl    TH_MUTEX_COUNT(%ecx)
-0:
-#endif
+Llmu_doit:
+       xorl    %ecx,%ecx
+       movl    %ecx,M_LOCKED           /* unlock the mutex */
 
-       xorb    %cl,%cl
-       xchgb   %cl,M_LOCKED            /* unlock the mutex */
+       movl    %ecx,M_ILK
 
-       xorb    %cl,%cl
-       xchgb   %cl,M_ILK
+       popf                            /* restore interrupt state */
 
-       ENABLE_PREEMPTION(%eax)
+       NONLEAF_RET
 
-       EMARF
-       ret
+Llmu_ilk_fail:
+       /*
+        * Slow path: call out to do the spinning.
+        */
+       pushl   %edx                    /* lock address */
+       call    EXT(lck_mtx_interlock_spin)
+       popl    %edx                    /* lock pointer */
+       jmp     Llmu_retry              /* try again */
 
-mu_wakeup:
-       xorl    %eax,%eax
-       pushl   %eax                    /* no promotion here yet */
+Llmu_wakeup:
+       pushl   %edx                    /* save mutex address */
+       pushl   M_LOCKED
        pushl   %edx                    /* push mutex address */
-       call    EXT(mutex_unlock_wakeup)/* yes, wake a thread */
+       call    EXT(lck_mtx_unlock_wakeup)/* yes, wake a thread */
        addl    $8,%esp
-       movl    L_ARG0,%edx             /* refetch lock pointer */
-       jmp     mu_doit
-
-ENTRY(interlock_unlock)
-       FRAME
-       movl    L_ARG0,%edx
-
-       xorb    %cl,%cl
-       xchgb   %cl,M_ILK
+       popl    %edx                    /* restore mutex pointer */
+       jmp     Llmu_doit
 
-       ENABLE_PREEMPTION(%eax)
+LEAF_ENTRY(lck_mtx_ilk_unlock)
+       movl    L_ARG0,%edx             /* no indirection here */
 
-       EMARF
-       ret
+       xorl    %eax,%eax
+       movl    %eax,M_ILK
 
+       LEAF_RET
        
-ENTRY(_disable_preemption)
+LEAF_ENTRY(_disable_preemption)
 #if    MACH_RT
-       _DISABLE_PREEMPTION(%eax)
+       _DISABLE_PREEMPTION
 #endif /* MACH_RT */
-       ret
+       LEAF_RET
 
-ENTRY(_enable_preemption)
+LEAF_ENTRY(_enable_preemption)
 #if    MACH_RT
 #if    MACH_ASSERT
-       movl    $ CPD_PREEMPTION_LEVEL,%eax
-       cmpl    $0,%gs:(%eax)
+       cmpl    $0,%gs:CPU_PREEMPTION_LEVEL
        jg      1f
-       pushl   %gs:(%eax)
+       pushl   %gs:CPU_PREEMPTION_LEVEL
        pushl   $2f
        call    EXT(panic)
        hlt
@@ -767,15 +929,14 @@ ENTRY(_enable_preemption)
        .text
 1:
 #endif /* MACH_ASSERT */
-       _ENABLE_PREEMPTION(%eax)
+       _ENABLE_PREEMPTION
 #endif /* MACH_RT */
-       ret
+       LEAF_RET
 
-ENTRY(_enable_preemption_no_check)
+LEAF_ENTRY(_enable_preemption_no_check)
 #if    MACH_RT
 #if    MACH_ASSERT
-       movl    $ CPD_PREEMPTION_LEVEL,%eax
-       cmpl    $0,%gs:(%eax)
+       cmpl    $0,%gs:CPU_PREEMPTION_LEVEL
        jg      1f
        pushl   $2f
        call    EXT(panic)
@@ -785,24 +946,23 @@ ENTRY(_enable_preemption_no_check)
        .text
 1:
 #endif /* MACH_ASSERT */
-       _ENABLE_PREEMPTION_NO_CHECK(%eax)
+       _ENABLE_PREEMPTION_NO_CHECK
 #endif /* MACH_RT */
-       ret
+       LEAF_RET
        
        
-ENTRY(_mp_disable_preemption)
-#if    MACH_RT && NCPUS > 1
-       _DISABLE_PREEMPTION(%eax)
-#endif /* MACH_RT && NCPUS > 1*/
-       ret
+LEAF_ENTRY(_mp_disable_preemption)
+#if    MACH_RT
+       _DISABLE_PREEMPTION
+#endif /* MACH_RT */
+       LEAF_RET
 
-ENTRY(_mp_enable_preemption)
-#if    MACH_RT && NCPUS > 1
+LEAF_ENTRY(_mp_enable_preemption)
+#if    MACH_RT
 #if    MACH_ASSERT
-       movl    $ CPD_PREEMPTION_LEVEL,%eax
-       cmpl    $0,%gs:(%eax)
+       cmpl    $0,%gs:CPU_PREEMPTION_LEVEL
        jg      1f
-       pushl   %gs:(%eax)
+       pushl   %gs:CPU_PREEMPTION_LEVEL
        pushl   $2f
        call    EXT(panic)
        hlt
@@ -811,15 +971,14 @@ ENTRY(_mp_enable_preemption)
        .text
 1:
 #endif /* MACH_ASSERT */
-       _ENABLE_PREEMPTION(%eax)
-#endif /* MACH_RT && NCPUS > 1 */
-       ret
+       _ENABLE_PREEMPTION
+#endif /* MACH_RT */
+       LEAF_RET
 
-ENTRY(_mp_enable_preemption_no_check)
-#if    MACH_RT && NCPUS > 1
+LEAF_ENTRY(_mp_enable_preemption_no_check)
+#if    MACH_RT
 #if    MACH_ASSERT
-       movl    $ CPD_PREEMPTION_LEVEL,%eax
-       cmpl    $0,%gs:(%eax)
+       cmpl    $0,%gs:CPU_PREEMPTION_LEVEL
        jg      1f
        pushl   $2f
        call    EXT(panic)
@@ -829,48 +988,48 @@ ENTRY(_mp_enable_preemption_no_check)
        .text
 1:
 #endif /* MACH_ASSERT */
-       _ENABLE_PREEMPTION_NO_CHECK(%eax)
-#endif /* MACH_RT && NCPUS > 1 */
-       ret
+       _ENABLE_PREEMPTION_NO_CHECK
+#endif /* MACH_RT */
+       LEAF_RET
        
        
-ENTRY(i_bit_set)
-       movl    S_ARG0,%edx
-       movl    S_ARG1,%eax
+LEAF_ENTRY(i_bit_set)
+       movl    L_ARG0,%edx
+       movl    L_ARG1,%eax
        lock
-       bts     %dl,(%eax)
-       ret
+       bts     %edx,(%eax)
+       LEAF_RET
 
-ENTRY(i_bit_clear)
-       movl    S_ARG0,%edx
-       movl    S_ARG1,%eax
+LEAF_ENTRY(i_bit_clear)
+       movl    L_ARG0,%edx
+       movl    L_ARG1,%eax
        lock
-       btr     %dl,(%eax)
-       ret
+       btr     %edx,(%eax)
+       LEAF_RET
 
-ENTRY(bit_lock)
-       movl    S_ARG0,%ecx
-       movl    S_ARG1,%eax
+LEAF_ENTRY(bit_lock)
+       movl    L_ARG0,%ecx
+       movl    L_ARG1,%eax
 1:
        lock
        bts     %ecx,(%eax)
        jb      1b
-       ret
+       LEAF_RET
 
-ENTRY(bit_lock_try)
-       movl    S_ARG0,%ecx
-       movl    S_ARG1,%eax
+LEAF_ENTRY(bit_lock_try)
+       movl    L_ARG0,%ecx
+       movl    L_ARG1,%eax
        lock
        bts     %ecx,(%eax)
        jb      bit_lock_failed
-       ret                     /* %eax better not be null ! */
+       LEAF_RET                /* %eax better not be null ! */
 bit_lock_failed:
        xorl    %eax,%eax
-       ret
+       LEAF_RET
 
-ENTRY(bit_unlock)
-       movl    S_ARG0,%ecx
-       movl    S_ARG1,%eax
+LEAF_ENTRY(bit_unlock)
+       movl    L_ARG0,%ecx
+       movl    L_ARG1,%eax
        lock
        btr     %ecx,(%eax)
-       ret
+       LEAF_RET