X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/5d5c5d0d5b79ade9a973d55186ffda2638ba2b6e..db6096698656d32db7df630594bd9617ee54f828:/osfmk/i386/i386_lock.s?ds=inline diff --git a/osfmk/i386/i386_lock.s b/osfmk/i386/i386_lock.s index 68598c3cb..6b34073c6 100644 --- a/osfmk/i386/i386_lock.s +++ b/osfmk/i386/i386_lock.s @@ -1,31 +1,29 @@ /* - * Copyright (c) 2000 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2009 Apple Inc. All rights reserved. * - * @APPLE_LICENSE_OSREFERENCE_HEADER_START@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * - * 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 + * 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 * limitations under the License. - * - * @APPLE_LICENSE_OSREFERENCE_HEADER_END@ + * + * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* * @OSF_COPYRIGHT@ @@ -41,21 +39,26 @@ #include #include #include - +#include +#include +#include +#include + #include "assym.s" #define PAUSE rep; nop +#include + /* * When performance isn't the only concern, it's * nice to build stack frames... */ -#define BUILD_STACK_FRAMES (GPROF || \ - ((MACH_LDEBUG || ETAP_LOCK_TRACE) && MACH_KDB)) +#define BUILD_STACK_FRAMES (GPROF) #if BUILD_STACK_FRAMES -/* STack-frame-relative: */ +/* Stack-frame-relative: */ #define L_PC B_PC #define L_ARG0 B_ARG0 #define L_ARG1 B_ARG1 @@ -113,20 +116,33 @@ 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 MUTEX_TYPE(%edx) -#define M_PC MUTEX_PC(%edx) -#define M_THREAD MUTEX_THREAD(%edx) -#endif /* MACH_LDEBUG */ +/* For x86_64, the varargs ABI requires that %al indicate + * how many SSE register contain arguments. In our case, 0 */ +#if __i386__ +#define ALIGN_STACK() subl $8, %esp; andl $0xFFFFFFF0, %esp ; +#define LOAD_STRING_ARG0(label) movl $##label, (%esp) ; +#define LOAD_ARG1(x) mov x, 4(%esp) ; +#define LOAD_PTR_ARG1(x) mov x, 4(%esp) ; +#define CALL_PANIC() call EXT(panic) ; +#else +#define ALIGN_STACK() and $0xFFFFFFFFFFFFFFF0, %rsp ; +#define LOAD_STRING_ARG0(label) leaq label(%rip), %rdi ; +#define LOAD_ARG1(x) mov x, %esi ; +#define LOAD_PTR_ARG1(x) mov x, %rsi ; +#define CALL_PANIC() xorb %al,%al ; call EXT(panic) ; +#endif -#include -#define CX(addr,reg) addr(,reg,4) +#define CHECK_UNLOCK(current, owner) \ + cmp current, owner ; \ + je 1f ; \ + ALIGN_STACK() ; \ + LOAD_STRING_ARG0(2f) ; \ + CALL_PANIC() ; \ + hlt ; \ + .data ; \ +2: String "Mutex unlock attempted from non-owner thread"; \ + .text ; \ +1: #if MACH_LDEBUG /* @@ -141,8 +157,9 @@ #define CHECK_MUTEX_TYPE() \ cmpl $ MUTEX_TAG,M_TYPE ; \ je 1f ; \ - pushl $2f ; \ - call EXT(panic) ; \ + ALIGN_STACK() ; \ + LOAD_STRING_ARG0(2f) ; \ + CALL_PANIC() ; \ hlt ; \ .data ; \ 2: String "not a mutex!" ; \ @@ -157,82 +174,178 @@ */ #if MACH_RT #define CHECK_PREEMPTION_LEVEL() \ + cmpl $0,%gs:CPU_HIBERNATE ; \ + jne 1f ; \ cmpl $0,%gs:CPU_PREEMPTION_LEVEL ; \ je 1f ; \ - pushl $2f ; \ - call EXT(panic) ; \ + ALIGN_STACK() ; \ + movl %gs:CPU_PREEMPTION_LEVEL, %eax ; \ + LOAD_ARG1(%eax) ; \ + LOAD_STRING_ARG0(2f) ; \ + CALL_PANIC() ; \ hlt ; \ .data ; \ -2: String "preemption_level != 0!" ; \ +2: String "preemption_level(%d) != 0!" ; \ .text ; \ 1: #else /* MACH_RT */ #define CHECK_PREEMPTION_LEVEL() #endif /* MACH_RT */ -#define CHECK_NO_SIMPLELOCKS() \ - cmpl $0,%gs:CPU_SIMPLE_LOCK_COUNT ; \ - je 1f ; \ - pushl $2f ; \ - call EXT(panic) ; \ - hlt ; \ - .data ; \ -2: String "simple_locks_held!" ; \ - .text ; \ -1: - -/* - * Verifies return to the correct thread in "unlock" situations. - */ -#define CHECK_THREAD(thd) \ - movl %gs:CPU_ACTIVE_THREAD,%ecx ; \ - testl %ecx,%ecx ; \ - je 1f ; \ - cmpl %ecx,thd ; \ - je 1f ; \ - pushl $2f ; \ - call EXT(panic) ; \ - hlt ; \ - .data ; \ -2: String "wrong thread!" ; \ - .text ; \ -1: - -#define CHECK_MYLOCK(thd) \ - movl %gs:CPU_ACTIVE_THREAD,%ecx ; \ - testl %ecx,%ecx ; \ - je 1f ; \ - cmpl %ecx,thd ; \ +#define CHECK_MYLOCK(current, owner) \ + cmp current, owner ; \ jne 1f ; \ - pushl $2f ; \ - call EXT(panic) ; \ + ALIGN_STACK() ; \ + LOAD_STRING_ARG0(2f) ; \ + CALL_PANIC() ; \ hlt ; \ .data ; \ -2: String "mylock attempt!" ; \ +2: String "Attempt to recursively lock a non-recursive lock"; \ .text ; \ 1: -#define METER_SIMPLE_LOCK_LOCK(reg) \ - pushl reg ; \ - call EXT(meter_simple_lock) ; \ - popl reg - -#define METER_SIMPLE_LOCK_UNLOCK(reg) \ - pushl reg ; \ - call EXT(meter_simple_unlock) ; \ - popl reg - #else /* MACH_LDEBUG */ #define CHECK_MUTEX_TYPE() -#define CHECK_SIMPLE_LOCK_TYPE -#define CHECK_THREAD(thd) #define CHECK_PREEMPTION_LEVEL() -#define CHECK_NO_SIMPLELOCKS() #define CHECK_MYLOCK(thd) -#define METER_SIMPLE_LOCK_LOCK(reg) -#define METER_SIMPLE_LOCK_UNLOCK(reg) #endif /* MACH_LDEBUG */ +#define PREEMPTION_DISABLE \ + incl %gs:CPU_PREEMPTION_LEVEL + +#define PREEMPTION_LEVEL_DEBUG 1 +#if PREEMPTION_LEVEL_DEBUG +#define PREEMPTION_ENABLE \ + decl %gs:CPU_PREEMPTION_LEVEL ; \ + js 17f ; \ + jnz 19f ; \ + testl $AST_URGENT,%gs:CPU_PENDING_AST ; \ + jz 19f ; \ + PUSHF ; \ + testl $EFL_IF, S_PC ; \ + jz 18f ; \ + POPF ; \ + int $(T_PREEMPT) ; \ + jmp 19f ; \ +17: \ + call _preemption_underflow_panic ; \ +18: \ + POPF ; \ +19: +#else +#define PREEMPTION_ENABLE \ + decl %gs:CPU_PREEMPTION_LEVEL ; \ + jnz 19f ; \ + testl $AST_URGENT,%gs:CPU_PENDING_AST ; \ + jz 19f ; \ + PUSHF ; \ + testl $EFL_IF, S_PC ; \ + jz 18f ; \ + POPF ; \ + int $(T_PREEMPT) ; \ + jmp 19f ; \ +18: \ + POPF ; \ +19: +#endif + + +#if CONFIG_DTRACE + + .globl _lockstat_probe + .globl _lockstat_probemap + +/* + * LOCKSTAT_LABEL creates a dtrace symbol which contains + * a pointer into the lock code function body. At that + * point is a "ret" instruction that can be patched into + * a "nop" + */ + +#if defined(__i386__) + +#define LOCKSTAT_LABEL(lab) \ + .data ;\ + .globl lab ;\ + lab: ;\ + .long 9f ;\ + .text ;\ + 9: + +#define LOCKSTAT_RECORD(id, lck) \ + push %ebp ; \ + mov %esp,%ebp ; \ + sub $0x38,%esp /* size of dtrace_probe args */ ; \ + movl _lockstat_probemap + (id * 4),%eax ; \ + test %eax,%eax ; \ + je 9f ; \ + movl $0,36(%esp) ; \ + movl $0,40(%esp) ; \ + movl $0,28(%esp) ; \ + movl $0,32(%esp) ; \ + movl $0,20(%esp) ; \ + movl $0,24(%esp) ; \ + movl $0,12(%esp) ; \ + movl $0,16(%esp) ; \ + movl lck,4(%esp) /* copy lock pointer to arg 1 */ ; \ + movl $0,8(%esp) ; \ + movl %eax,(%esp) ; \ + call *_lockstat_probe ; \ +9: leave + /* ret - left to subsequent code, e.g. return values */ + +#elif defined(__x86_64__) +#define LOCKSTAT_LABEL(lab) \ + .data ;\ + .globl lab ;\ + lab: ;\ + .quad 9f ;\ + .text ;\ + 9: + +#define LOCKSTAT_RECORD(id, lck) \ + push %rbp ; \ + mov %rsp,%rbp ; \ + movl _lockstat_probemap + (id * 4)(%rip),%eax ; \ + test %eax,%eax ; \ + je 9f ; \ + mov lck, %rsi ; \ + mov %rax, %rdi ; \ + mov $0, %rdx ; \ + mov $0, %rcx ; \ + mov $0, %r8 ; \ + mov $0, %r9 ; \ + call *_lockstat_probe(%rip) ; \ +9: leave + /* ret - left to subsequent code, e.g. return values */ +#else +#error Unsupported architecture +#endif +#endif /* CONFIG_DTRACE */ + +/* + * For most routines, the hw_lock_t pointer is loaded into a + * register initially, and then either a byte or register-sized + * word is loaded/stored to the pointer + */ + +#if defined(__i386__) +#define HW_LOCK_REGISTER %edx +#define LOAD_HW_LOCK_REGISTER mov L_ARG0, HW_LOCK_REGISTER +#define HW_LOCK_THREAD_REGISTER %ecx +#define LOAD_HW_LOCK_THREAD_REGISTER mov %gs:CPU_ACTIVE_THREAD, HW_LOCK_THREAD_REGISTER +#define HW_LOCK_MOV_WORD movl +#define HW_LOCK_EXAM_REGISTER %eax +#elif defined(__x86_64__) +#define HW_LOCK_REGISTER %rdi +#define LOAD_HW_LOCK_REGISTER +#define HW_LOCK_THREAD_REGISTER %rcx +#define LOAD_HW_LOCK_THREAD_REGISTER mov %gs:CPU_ACTIVE_THREAD, HW_LOCK_THREAD_REGISTER +#define HW_LOCK_MOV_WORD movq +#define HW_LOCK_EXAM_REGISTER %rax +#else +#error Unsupported architecture +#endif /* * void hw_lock_init(hw_lock_t) @@ -240,8 +353,19 @@ * Initialize a hardware lock. */ LEAF_ENTRY(hw_lock_init) - movl L_ARG0,%edx /* fetch lock pointer */ - movl $0,0(%edx) /* clear the lock */ + LOAD_HW_LOCK_REGISTER /* fetch lock pointer */ + HW_LOCK_MOV_WORD $0, (HW_LOCK_REGISTER) /* clear the lock */ + LEAF_RET + + +/* + * void hw_lock_byte_init(volatile uint8_t *) + * + * Initialize a hardware byte lock. + */ +LEAF_ENTRY(hw_lock_byte_init) + LOAD_HW_LOCK_REGISTER /* fetch lock pointer */ + movb $0, (HW_LOCK_REGISTER) /* clear the lock */ LEAF_RET /* @@ -251,15 +375,15 @@ LEAF_ENTRY(hw_lock_init) * MACH_RT: also return with preemption disabled. */ LEAF_ENTRY(hw_lock_lock) - movl L_ARG0,%edx /* fetch lock pointer */ - - movl %gs:CPU_ACTIVE_THREAD,%ecx - DISABLE_PREEMPTION + LOAD_HW_LOCK_REGISTER /* fetch lock pointer */ + LOAD_HW_LOCK_THREAD_REGISTER /* get thread pointer */ + + PREEMPTION_DISABLE 1: - movl 0(%edx), %eax - testl %eax,%eax /* lock locked? */ + mov (HW_LOCK_REGISTER), HW_LOCK_EXAM_REGISTER + test HW_LOCK_EXAM_REGISTER,HW_LOCK_EXAM_REGISTER /* lock locked? */ jne 3f /* branch if so */ - lock; cmpxchgl %ecx,0(%edx) /* try to acquire the HW lock */ + lock; cmpxchg HW_LOCK_THREAD_REGISTER,(HW_LOCK_REGISTER) /* try to acquire the HW lock */ jne 3f movl $1,%eax /* In case this was a timeout call */ LEAF_RET /* if yes, then nothing left to do */ @@ -267,6 +391,28 @@ LEAF_ENTRY(hw_lock_lock) PAUSE /* pause for hyper-threading */ jmp 1b /* try again */ +/* + * void hw_lock_byte_lock(uint8_t *lock_byte) + * + * Acquire byte sized lock operand, spinning until it becomes available. + * MACH_RT: also return with preemption disabled. + */ + +LEAF_ENTRY(hw_lock_byte_lock) + LOAD_HW_LOCK_REGISTER /* Load lock pointer */ + PREEMPTION_DISABLE + movl $1, %ecx /* Set lock value */ +1: + movb (HW_LOCK_REGISTER), %al /* Load byte at address */ + testb %al,%al /* lock locked? */ + jne 3f /* branch if so */ + lock; cmpxchg %cl,(HW_LOCK_REGISTER) /* attempt atomic compare exchange */ + jne 3f + LEAF_RET /* if yes, then nothing left to do */ +3: + PAUSE /* pause for hyper-threading */ + jmp 1b /* try again */ + /* * unsigned int hw_lock_to(hw_lock_t, unsigned int) * @@ -275,17 +421,19 @@ LEAF_ENTRY(hw_lock_lock) */ LEAF_ENTRY(hw_lock_to) 1: - movl L_ARG0,%edx /* fetch lock pointer */ - movl %gs:CPU_ACTIVE_THREAD,%ecx + LOAD_HW_LOCK_REGISTER /* fetch lock pointer */ + LOAD_HW_LOCK_THREAD_REGISTER + /* * Attempt to grab the lock immediately * - fastpath without timeout nonsense. */ - DISABLE_PREEMPTION - movl 0(%edx), %eax - testl %eax,%eax /* lock locked? */ + PREEMPTION_DISABLE + + mov (HW_LOCK_REGISTER), HW_LOCK_EXAM_REGISTER + test HW_LOCK_EXAM_REGISTER,HW_LOCK_EXAM_REGISTER /* lock locked? */ jne 2f /* branch if so */ - lock; cmpxchgl %ecx,0(%edx) /* try to acquire the HW lock */ + lock; cmpxchg HW_LOCK_THREAD_REGISTER,(HW_LOCK_REGISTER) /* try to acquire the HW lock */ jne 2f /* branch on failure */ movl $1,%eax LEAF_RET @@ -297,40 +445,74 @@ LEAF_ENTRY(hw_lock_to) * and then spin re-checking the lock but pausing * every so many (INNER_LOOP_COUNT) spins to check for timeout. */ +#if __i386__ movl L_ARG1,%ecx /* fetch timeout */ push %edi push %ebx mov %edx,%edi + lfence 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 */ + mov %edi, %edx /* load lock back into %edx */ +#else + push %r9 + lfence + rdtsc /* read cyclecount into %edx:%eax */ + shlq $32, %rdx + orq %rdx, %rax /* load 64-bit quantity into %rax */ + addq %rax, %rsi /* %rsi is the timeout expiry */ +#endif + 4: /* * The inner-loop spin to look for the lock being freed. */ - mov $(INNER_LOOP_COUNT),%edx +#if __i386__ + mov $(INNER_LOOP_COUNT),%edi +#else + mov $(INNER_LOOP_COUNT),%r9 +#endif 5: PAUSE /* pause for hyper-threading */ - movl 0(%edi),%eax /* spin checking lock value in cache */ - testl %eax,%eax + mov (HW_LOCK_REGISTER),HW_LOCK_EXAM_REGISTER /* spin checking lock value in cache */ + test HW_LOCK_EXAM_REGISTER,HW_LOCK_EXAM_REGISTER je 6f /* zero => unlocked, try to grab it */ - decl %edx /* decrement inner loop count */ +#if __i386__ + decl %edi /* decrement inner loop count */ +#else + decq %r9 /* decrement inner loop count */ +#endif jnz 5b /* time to check for timeout? */ - + /* * Here after spinning INNER_LOOP_COUNT times, check for timeout */ +#if __i386__ + mov %edx,%edi /* Save %edx */ + lfence rdtsc /* cyclecount into %edx:%eax */ - cmpl %ecx,%edx /* compare high-order 32-bits */ + xchg %edx,%edi /* cyclecount into %edi:%eax */ + cmpl %ecx,%edi /* 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 +#else + lfence + rdtsc /* cyclecount into %edx:%eax */ + shlq $32, %rdx + orq %rdx, %rax /* load 64-bit quantity into %rax */ + cmpq %rsi, %rax /* compare to timeout */ + jb 4b /* continue spinning if less, or */ + xor %rax,%rax /* with 0 return value */ + pop %r9 +#endif LEAF_RET 6: @@ -338,12 +520,16 @@ LEAF_ENTRY(hw_lock_to) * 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 */ + LOAD_HW_LOCK_THREAD_REGISTER + lock; cmpxchg HW_LOCK_THREAD_REGISTER,(HW_LOCK_REGISTER) /* try to acquire the HW lock */ jne 4b /* no - spin again */ movl $1,%eax /* yes */ +#if __i386__ pop %ebx pop %edi +#else + pop %r9 +#endif LEAF_RET /* @@ -353,9 +539,22 @@ LEAF_ENTRY(hw_lock_to) * MACH_RT: release preemption level. */ LEAF_ENTRY(hw_lock_unlock) - movl L_ARG0,%edx /* fetch lock pointer */ - movl $0,0(%edx) /* clear the lock */ - ENABLE_PREEMPTION + LOAD_HW_LOCK_REGISTER /* fetch lock pointer */ + HW_LOCK_MOV_WORD $0, (HW_LOCK_REGISTER) /* clear the lock */ + PREEMPTION_ENABLE + LEAF_RET + +/* + * void hw_lock_byte_unlock(uint8_t *lock_byte) + * + * Unconditionally release byte sized lock operand. + * MACH_RT: release preemption level. + */ + +LEAF_ENTRY(hw_lock_byte_unlock) + LOAD_HW_LOCK_REGISTER /* Load lock pointer */ + movb $0, (HW_LOCK_REGISTER) /* Clear the lock byte */ + PREEMPTION_ENABLE LEAF_RET /* @@ -363,21 +562,21 @@ LEAF_ENTRY(hw_lock_unlock) * MACH_RT: returns with preemption disabled on success. */ LEAF_ENTRY(hw_lock_try) - movl L_ARG0,%edx /* fetch lock pointer */ + LOAD_HW_LOCK_REGISTER /* fetch lock pointer */ + LOAD_HW_LOCK_THREAD_REGISTER + PREEMPTION_DISABLE - movl %gs:CPU_ACTIVE_THREAD,%ecx - DISABLE_PREEMPTION - movl 0(%edx),%eax - testl %eax,%eax + mov (HW_LOCK_REGISTER),HW_LOCK_EXAM_REGISTER + test HW_LOCK_EXAM_REGISTER,HW_LOCK_EXAM_REGISTER jne 1f - lock; cmpxchgl %ecx,0(%edx) /* try to acquire the HW lock */ + lock; cmpxchg HW_LOCK_THREAD_REGISTER,(HW_LOCK_REGISTER) /* try to acquire the HW lock */ jne 1f - + movl $1,%eax /* success */ LEAF_RET 1: - ENABLE_PREEMPTION /* failure: release preemption... */ + PREEMPTION_ENABLE /* failure: release preemption... */ xorl %eax,%eax /* ...and return failure */ LEAF_RET @@ -387,531 +586,1430 @@ LEAF_ENTRY(hw_lock_try) * N.B. Racy, of course. */ LEAF_ENTRY(hw_lock_held) - movl L_ARG0,%edx /* fetch lock pointer */ - - movl 0(%edx),%eax /* check lock value */ - testl %eax,%eax + LOAD_HW_LOCK_REGISTER /* fetch lock pointer */ + mov (HW_LOCK_REGISTER),HW_LOCK_EXAM_REGISTER /* check lock value */ + test HW_LOCK_EXAM_REGISTER,HW_LOCK_EXAM_REGISTER movl $1,%ecx cmovne %ecx,%eax /* 0 => unlocked, 1 => locked */ LEAF_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 - - LEAF_RET - -NONLEAF_ENTRY2(mutex_lock,_mutex_lock) - - movl B_ARG0,%edx /* fetch lock pointer */ - - CHECK_MUTEX_TYPE() - CHECK_NO_SIMPLELOCKS() - CHECK_PREEMPTION_LEVEL() - - pushf /* save interrupt state */ - cli /* disable interrupts */ -Lml_retry: - movl %gs:CPU_ACTIVE_THREAD,%ecx -Lml_get_hw: - movl M_ILK,%eax /* read interlock */ - testl %eax,%eax /* unlocked? */ - jne Lml_ilk_fail /* no - take the slow path */ +/* + * Reader-writer lock fastpaths. These currently exist for the + * shared lock acquire, the exclusive lock acquire, the shared to + * exclusive upgrade and the release paths (where they reduce overhead + * considerably) -- these are by far the most frequently used routines + * + * The following should reflect the layout of the bitfield embedded within + * the lck_rw_t structure (see i386/locks.h). + */ +#define LCK_RW_INTERLOCK (0x1 << 16) - lock; cmpxchgl %ecx,M_ILK /* atomic compare and exchange */ - jne Lml_get_hw /* branch on failure to retry */ +#define LCK_RW_PRIV_EXCL (0x1 << 24) +#define LCK_RW_WANT_UPGRADE (0x2 << 24) +#define LCK_RW_WANT_WRITE (0x4 << 24) +#define LCK_R_WAITING (0x8 << 24) +#define LCK_W_WAITING (0x10 << 24) - 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 +#define LCK_RW_SHARED_MASK (0xffff) -#if MACH_LDEBUG - movl %ecx,M_THREAD - movl B_PC,%ecx - movl %ecx,M_PC +/* + * For most routines, the lck_rw_t pointer is loaded into a + * register initially, and the flags bitfield loaded into another + * register and examined + */ + +#if defined(__i386__) +#define LCK_RW_REGISTER %edx +#define LOAD_LCK_RW_REGISTER mov S_ARG0, LCK_RW_REGISTER +#define LCK_RW_FLAGS_REGISTER %eax +#define LOAD_LCK_RW_FLAGS_REGISTER mov (LCK_RW_REGISTER), LCK_RW_FLAGS_REGISTER +#elif defined(__x86_64__) +#define LCK_RW_REGISTER %rdi +#define LOAD_LCK_RW_REGISTER +#define LCK_RW_FLAGS_REGISTER %eax +#define LOAD_LCK_RW_FLAGS_REGISTER mov (LCK_RW_REGISTER), LCK_RW_FLAGS_REGISTER +#else +#error Unsupported architecture #endif + +#define RW_LOCK_SHARED_MASK (LCK_RW_INTERLOCK | LCK_RW_WANT_UPGRADE | LCK_RW_WANT_WRITE) +/* + * void lck_rw_lock_shared(lck_rw_t *) + * + */ +Entry(lck_rw_lock_shared) + LOAD_LCK_RW_REGISTER +1: + LOAD_LCK_RW_FLAGS_REGISTER /* Load state bitfield and interlock */ + testl $(RW_LOCK_SHARED_MASK), %eax /* Eligible for fastpath? */ + jne 3f - 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 - call EXT(lck_mtx_lock_acquire) - addl $4,%esp - popl %edx /* restore mutex address */ - jmp Lml_return + movl %eax, %ecx /* original value in %eax for cmpxchgl */ + incl %ecx /* Increment reader refcount */ + lock + cmpxchgl %ecx, (LCK_RW_REGISTER) /* Attempt atomic exchange */ + jne 2f -Lml_ilk_fail: +#if CONFIG_DTRACE /* - * Slow path: call out to do the spinning. + * Dtrace lockstat event: LS_LCK_RW_LOCK_SHARED_ACQUIRE + * Implemented by swapping between return and no-op instructions. + * See bsd/dev/dtrace/lockstat.c. */ - pushl %edx /* lock address */ - call EXT(lck_mtx_interlock_spin) - popl %edx /* lock pointer */ - jmp Lml_retry /* try again */ - -Lml_fail: + LOCKSTAT_LABEL(_lck_rw_lock_shared_lockstat_patch_point) + ret /* - 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 + Fall thru when patched, counting on lock pointer in LCK_RW_REGISTER + */ + LOCKSTAT_RECORD(LS_LCK_RW_LOCK_SHARED_ACQUIRE, LCK_RW_REGISTER) +#endif + ret +2: + PAUSE + jmp 1b +3: + jmp EXT(lck_rw_lock_shared_gen) + - /* - * 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 */ - - 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 */ - - /* Re-acquire interlock */ - cli /* disable interrupts */ -Lml_reget_retry: - movl %gs:CPU_ACTIVE_THREAD,%ecx - -Lml_reget_hw: - movl M_ILK,%eax /* read interlock */ - testl %eax,%eax /* unlocked? */ - jne Lml_ilk_refail /* no - slow path */ - - lock; cmpxchgl %ecx,M_ILK /* atomic compare and exchange */ - jne Lml_reget_hw /* branch on failure to retry */ - - 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 */ - -Lml_ilk_refail: +#define RW_TRY_LOCK_SHARED_MASK (LCK_RW_WANT_UPGRADE | LCK_RW_WANT_WRITE) +/* + * void lck_rw_try_lock_shared(lck_rw_t *) + * + */ +Entry(lck_rw_try_lock_shared) + LOAD_LCK_RW_REGISTER +1: + LOAD_LCK_RW_FLAGS_REGISTER /* Load state bitfield and interlock */ + testl $(LCK_RW_INTERLOCK), %eax + jne 2f + testl $(RW_TRY_LOCK_SHARED_MASK), %eax + jne 3f /* lock is busy */ + + movl %eax, %ecx /* original value in %eax for cmpxchgl */ + incl %ecx /* Increment reader refcount */ + lock + cmpxchgl %ecx, (LCK_RW_REGISTER) /* Attempt atomic exchange */ + jne 2f + +#if CONFIG_DTRACE + movl $1, %eax /* - * Slow path: call out to do the spinning. + * Dtrace lockstat event: LS_LCK_RW_TRY_LOCK_SHARED_ACQUIRE + * Implemented by swapping between return and no-op instructions. + * See bsd/dev/dtrace/lockstat.c. */ - pushl %edx /* lock address */ - call EXT(lck_mtx_interlock_spin) - popl %edx /* lock pointer */ - jmp Lml_reget_retry /* try again */ - -NONLEAF_ENTRY2(mutex_try,_mutex_try) - - movl B_ARG0,%edx /* fetch lock pointer */ + LOCKSTAT_LABEL(_lck_rw_try_lock_shared_lockstat_patch_point) + ret + /* Fall thru when patched, counting on lock pointer in LCK_RW_REGISTER */ + LOCKSTAT_RECORD(LS_LCK_RW_TRY_LOCK_SHARED_ACQUIRE, LCK_RW_REGISTER) +#endif + movl $1, %eax /* return TRUE */ + ret +2: + PAUSE + jmp 1b +3: + xorl %eax, %eax + ret - CHECK_MUTEX_TYPE() - CHECK_NO_SIMPLELOCKS() + +#define RW_LOCK_EXCLUSIVE_HELD (LCK_RW_WANT_WRITE | LCK_RW_WANT_UPGRADE) +/* + * int lck_rw_grab_shared(lck_rw_t *) + * + */ +Entry(lck_rw_grab_shared) + LOAD_LCK_RW_REGISTER +1: + LOAD_LCK_RW_FLAGS_REGISTER /* Load state bitfield and interlock */ + testl $(LCK_RW_INTERLOCK), %eax + jne 5f + testl $(RW_LOCK_EXCLUSIVE_HELD), %eax + jne 3f +2: + movl %eax, %ecx /* original value in %eax for cmpxchgl */ + incl %ecx /* Increment reader refcount */ + lock + cmpxchgl %ecx, (LCK_RW_REGISTER) /* Attempt atomic exchange */ + jne 4f - pushf /* save interrupt state */ - cli /* disable interrupts */ -Lmt_retry: - movl %gs:CPU_ACTIVE_THREAD,%ecx + movl $1, %eax /* return success */ + ret +3: + testl $(LCK_RW_SHARED_MASK), %eax + je 4f + testl $(LCK_RW_PRIV_EXCL), %eax + je 2b +4: + xorl %eax, %eax /* return failure */ + ret +5: + PAUSE + jmp 1b -Lmt_get_hw: - movl M_ILK,%eax /* read interlock */ - testl %eax,%eax /* unlocked? */ - jne Lmt_ilk_fail /* no - slow path */ - lock; cmpxchgl %ecx,M_ILK /* atomic compare and exchange */ - jne Lmt_get_hw /* branch on failure to retry */ + +#define RW_LOCK_EXCLUSIVE_MASK (LCK_RW_SHARED_MASK | LCK_RW_INTERLOCK | \ + LCK_RW_WANT_UPGRADE | LCK_RW_WANT_WRITE) +/* + * void lck_rw_lock_exclusive(lck_rw_t*) + * + */ +Entry(lck_rw_lock_exclusive) + LOAD_LCK_RW_REGISTER +1: + LOAD_LCK_RW_FLAGS_REGISTER /* Load state bitfield, interlock and shared count */ + testl $(RW_LOCK_EXCLUSIVE_MASK), %eax /* Eligible for fastpath? */ + jne 3f /* no, go slow */ - 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 + movl %eax, %ecx /* original value in %eax for cmpxchgl */ + orl $(LCK_RW_WANT_WRITE), %ecx + lock + cmpxchgl %ecx, (LCK_RW_REGISTER) /* Attempt atomic exchange */ + jne 2f -#if MACH_LDEBUG - movl %ecx,M_THREAD - movl B_PC,%ecx - movl %ecx,M_PC +#if CONFIG_DTRACE + /* + * Dtrace lockstat event: LS_LCK_RW_LOCK_EXCL_ACQUIRE + * Implemented by swapping between return and no-op instructions. + * See bsd/dev/dtrace/lockstat.c. + */ + LOCKSTAT_LABEL(_lck_rw_lock_exclusive_lockstat_patch_point) + ret + /* Fall thru when patched, counting on lock pointer in LCK_RW_REGISTER */ + LOCKSTAT_RECORD(LS_LCK_RW_LOCK_EXCL_ACQUIRE, LCK_RW_REGISTER) #endif + ret +2: + PAUSE + jmp 1b +3: + jmp EXT(lck_rw_lock_exclusive_gen) - 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 */ - - movl $1,%eax - - NONLEAF_RET -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 + +#define RW_TRY_LOCK_EXCLUSIVE_MASK (LCK_RW_SHARED_MASK | LCK_RW_WANT_UPGRADE | LCK_RW_WANT_WRITE) +/* + * void lck_rw_try_lock_exclusive(lck_rw_t *) + * + * Tries to get a write lock. + * + * Returns FALSE if the lock is not held on return. + */ +Entry(lck_rw_try_lock_exclusive) + LOAD_LCK_RW_REGISTER +1: + LOAD_LCK_RW_FLAGS_REGISTER /* Load state bitfield, interlock and shared count */ + testl $(LCK_RW_INTERLOCK), %eax + jne 2f + testl $(RW_TRY_LOCK_EXCLUSIVE_MASK), %eax + jne 3f /* can't get it */ + + movl %eax, %ecx /* original value in %eax for cmpxchgl */ + orl $(LCK_RW_WANT_WRITE), %ecx + lock + cmpxchgl %ecx, (LCK_RW_REGISTER) /* Attempt atomic exchange */ + jne 2f -Lmt_ilk_fail: +#if CONFIG_DTRACE + movl $1, %eax /* - * Slow path: call out to do the spinning. + * Dtrace lockstat event: LS_LCK_RW_TRY_LOCK_EXCL_ACQUIRE + * Implemented by swapping between return and no-op instructions. + * See bsd/dev/dtrace/lockstat.c. */ - pushl %edx /* lock address */ - call EXT(lck_mtx_interlock_spin) - popl %edx /* lock pointer */ - jmp Lmt_retry /* try again */ - -Lmt_fail: - xorl %eax,%eax - movl %eax,M_ILK - - popf /* restore interrupt state */ - - xorl %eax,%eax - - NONLEAF_RET - -NONLEAF_ENTRY(mutex_unlock) - movl B_ARG0,%edx /* fetch lock pointer */ - - CHECK_MUTEX_TYPE() - CHECK_THREAD(M_THREAD) - - pushf /* save interrupt state */ - cli /* disable interrupts */ -Lmu_retry: - movl %gs:CPU_ACTIVE_THREAD,%ecx - -Lmu_get_hw: - movl M_ILK,%eax /* read interlock */ - testl %eax,%eax /* unlocked? */ - jne Lmu_ilk_fail /* no - slow path */ - - lock; cmpxchgl %ecx,M_ILK /* atomic compare and exchange */ - jne Lmu_get_hw /* branch on failure to retry */ - - cmpw $0,M_WAITERS /* are there any waiters? */ - jne Lmu_wakeup /* yes, more work to do */ - -Lmu_doit: - -#if MACH_LDEBUG - movl $0,M_THREAD /* disown thread */ + LOCKSTAT_LABEL(_lck_rw_try_lock_exclusive_lockstat_patch_point) + ret + /* Fall thru when patched, counting on lock pointer in LCK_RW_REGISTER */ + LOCKSTAT_RECORD(LS_LCK_RW_TRY_LOCK_EXCL_ACQUIRE, LCK_RW_REGISTER) #endif + movl $1, %eax /* return TRUE */ + ret +2: + PAUSE + jmp 1b +3: + xorl %eax, %eax /* return FALSE */ + ret - xorl %ecx,%ecx - movl %ecx,M_LOCKED /* unlock the mutex */ - - movl %ecx,M_ILK - - popf /* restore interrupt state */ - - NONLEAF_RET -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(lck_mtx_unlock_wakeup)/* yes, wake a thread */ - addl $8,%esp - movl B_ARG0,%edx /* restore lock pointer */ - jmp Lmu_doit /* - * lck_mtx_lock() - * lck_mtx_try_lock() - * lck_mutex_unlock() + * void lck_rw_lock_shared_to_exclusive(lck_rw_t*) + * + * fastpath can be taken if + * the current rw_shared_count == 1 + * AND the interlock is clear + * AND RW_WANT_UPGRADE is not set + * + * note that RW_WANT_WRITE could be set, but will not + * be indicative of an exclusive hold since we have + * a read count on the lock that we have not yet released + * we can blow by that state since the lck_rw_lock_exclusive + * function will block until rw_shared_count == 0 and + * RW_WANT_UPGRADE is clear... it does this check behind + * the interlock which we are also checking for * - * These are variants of mutex_lock(), mutex_try() and mutex_unlock() without - * DEBUG checks (which require fields not present in lck_mtx_t's). + * to make the transition we must be able to atomically + * set RW_WANT_UPGRADE and get rid of the read count we hold */ -NONLEAF_ENTRY(lck_mtx_lock) - - movl B_ARG0,%edx /* fetch lock pointer */ - cmpl $(MUTEX_IND),M_ITAG /* is this indirect? */ - cmove M_PTR,%edx /* yes - take indirection */ - - CHECK_NO_SIMPLELOCKS() - CHECK_PREEMPTION_LEVEL() - - 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 - - popf /* restore interrupt state */ - - NONLEAF_RET - -Llml_waiters: - pushl %edx /* save mutex address */ - pushl %edx - call EXT(lck_mtx_lock_acquire) - addl $4,%esp - popl %edx /* restore mutex address */ - jmp Llml_return - -Llml_ilk_fail: +Entry(lck_rw_lock_shared_to_exclusive) + LOAD_LCK_RW_REGISTER +1: + LOAD_LCK_RW_FLAGS_REGISTER /* Load state bitfield, interlock and shared count */ + testl $(LCK_RW_INTERLOCK), %eax + jne 7f + testl $(LCK_RW_WANT_UPGRADE), %eax + jne 2f + + movl %eax, %ecx /* original value in %eax for cmpxchgl */ + orl $(LCK_RW_WANT_UPGRADE), %ecx /* ask for WANT_UPGRADE */ + decl %ecx /* and shed our read count */ + lock + cmpxchgl %ecx, (LCK_RW_REGISTER) /* Attempt atomic exchange */ + jne 7f + /* we now own the WANT_UPGRADE */ + testl $(LCK_RW_SHARED_MASK), %ecx /* check to see if all of the readers are drained */ + jne 8f /* if not, we need to go wait */ + +#if CONFIG_DTRACE + movl $1, %eax /* - * Slow path: call out to do the spinning. + * Dtrace lockstat event: LS_LCK_RW_LOCK_SHARED_TO_EXCL_UPGRADE + * Implemented by swapping between return and no-op instructions. + * See bsd/dev/dtrace/lockstat.c. */ - pushl %edx /* lock address */ - call EXT(lck_mtx_interlock_spin) - popl %edx /* lock pointer */ - jmp Llml_retry /* try again */ + LOCKSTAT_LABEL(_lck_rw_lock_shared_to_exclusive_lockstat_patch_point) + ret + /* Fall thru when patched, counting on lock pointer in LCK_RW_REGISTER */ + LOCKSTAT_RECORD(LS_LCK_RW_LOCK_SHARED_TO_EXCL_UPGRADE, LCK_RW_REGISTER) +#endif + movl $1, %eax /* return success */ + ret + +2: /* someone else already holds WANT_UPGRADE */ + movl %eax, %ecx /* original value in %eax for cmpxchgl */ + decl %ecx /* shed our read count */ + testl $(LCK_RW_SHARED_MASK), %ecx + jne 3f /* we were the last reader */ + andl $(~LCK_W_WAITING), %ecx /* so clear the wait indicator */ +3: + lock + cmpxchgl %ecx, (LCK_RW_REGISTER) /* Attempt atomic exchange */ + jne 7f + +#if __i386__ + pushl %eax /* go check to see if we need to */ + push %edx /* wakeup anyone */ + call EXT(lck_rw_lock_shared_to_exclusive_failure) + addl $8, %esp +#else + mov %eax, %esi /* put old flags as second arg */ + /* lock is alread in %rdi */ + call EXT(lck_rw_lock_shared_to_exclusive_failure) +#endif + ret /* and pass the failure return along */ +7: + PAUSE + jmp 1b +8: + jmp EXT(lck_rw_lock_shared_to_exclusive_success) -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 + + .cstring +rwl_release_error_str: + .asciz "Releasing non-exclusive RW lock without a reader refcount!" + .text + +/* + * lck_rw_type_t lck_rw_done(lck_rw_t *) + * + */ +Entry(lck_rw_done) + LOAD_LCK_RW_REGISTER +1: + LOAD_LCK_RW_FLAGS_REGISTER /* Load state bitfield, interlock and reader count */ + testl $(LCK_RW_INTERLOCK), %eax + jne 7f /* wait for interlock to clear */ + + movl %eax, %ecx /* keep original value in %eax for cmpxchgl */ + testl $(LCK_RW_SHARED_MASK), %ecx /* if reader count == 0, must be exclusive lock */ + je 2f + decl %ecx /* Decrement reader count */ + testl $(LCK_RW_SHARED_MASK), %ecx /* if reader count has now gone to 0, check for waiters */ + je 4f + jmp 6f +2: + testl $(LCK_RW_WANT_UPGRADE), %ecx + je 3f + andl $(~LCK_RW_WANT_UPGRADE), %ecx + jmp 4f +3: + testl $(LCK_RW_WANT_WRITE), %ecx + je 8f /* lock is not 'owned', go panic */ + andl $(~LCK_RW_WANT_WRITE), %ecx +4: /* - * 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. + * test the original values to match what + * lck_rw_done_gen is going to do to determine + * which wakeups need to happen... + * + * if !(fake_lck->lck_rw_priv_excl && fake_lck->lck_w_waiting) */ - xorl %eax,%eax - movl %eax,M_ILK /* zero interlock */ - popf - pushf /* restore interrupt state */ - - pushl %edx /* save mutex address */ - pushl %edx - call EXT(lck_mtx_lock_spin) - addl $4,%esp - popl %edx /* restore mutex address */ - - /* Re-acquire interlock */ - cli /* disable interrupts */ -Llml_reget_retry: - movl %gs:CPU_ACTIVE_THREAD,%ecx - -Llml_reget_hw: - movl M_ILK,%eax /* read interlock */ - testl %eax,%eax /* unlocked? */ - jne Llml_ilk_refail /* no - slow path */ - - 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 */ + testl $(LCK_W_WAITING), %eax + je 5f + andl $(~LCK_W_WAITING), %ecx + + testl $(LCK_RW_PRIV_EXCL), %eax + jne 6f +5: + andl $(~LCK_R_WAITING), %ecx +6: + lock + cmpxchgl %ecx, (LCK_RW_REGISTER) /* Attempt atomic exchange */ + jne 7f + +#if __i386__ + pushl %eax + push %edx + call EXT(lck_rw_done_gen) + addl $8, %esp +#else + mov %eax,%esi /* old flags in %rsi */ + /* lock is in %rdi already */ + call EXT(lck_rw_done_gen) +#endif + ret +7: + PAUSE + jmp 1b +8: + ALIGN_STACK() + LOAD_STRING_ARG0(rwl_release_error_str) + CALL_PANIC() -Llml_block: - CHECK_MYLOCK(M_THREAD) - pushl %edx /* save mutex address */ - pushl M_LOCKED - pushl %edx /* push mutex address */ - 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 */ - -Llml_ilk_refail: + + +/* + * lck_rw_type_t lck_rw_lock_exclusive_to_shared(lck_rw_t *) + * + */ +Entry(lck_rw_lock_exclusive_to_shared) + LOAD_LCK_RW_REGISTER +1: + LOAD_LCK_RW_FLAGS_REGISTER /* Load state bitfield, interlock and reader count */ + testl $(LCK_RW_INTERLOCK), %eax + jne 6f /* wait for interlock to clear */ + + movl %eax, %ecx /* keep original value in %eax for cmpxchgl */ + incl %ecx /* Increment reader count */ + + testl $(LCK_RW_WANT_UPGRADE), %ecx + je 2f + andl $(~LCK_RW_WANT_UPGRADE), %ecx + jmp 3f +2: + andl $(~LCK_RW_WANT_WRITE), %ecx +3: /* - * Slow path: call out to do the spinning. + * test the original values to match what + * lck_rw_lock_exclusive_to_shared_gen is going to do to determine + * which wakeups need to happen... + * + * if !(fake_lck->lck_rw_priv_excl && fake_lck->lck_w_waiting) */ - pushl %edx /* lock address */ - call EXT(lck_mtx_interlock_spin) - popl %edx /* lock pointer */ - jmp Llml_reget_retry /* try again */ - -NONLEAF_ENTRY(lck_mtx_try_lock) + testl $(LCK_W_WAITING), %eax + je 4f + testl $(LCK_RW_PRIV_EXCL), %eax + jne 5f +4: + andl $(~LCK_R_WAITING), %ecx +5: + lock + cmpxchgl %ecx, (LCK_RW_REGISTER) /* Attempt atomic exchange */ + jne 6f + +#if __i386__ + pushl %eax + push %edx + call EXT(lck_rw_lock_exclusive_to_shared_gen) + addl $8, %esp +#else + mov %eax,%esi + call EXT(lck_rw_lock_exclusive_to_shared_gen) +#endif + ret +6: + PAUSE + jmp 1b - movl B_ARG0,%edx /* fetch lock pointer */ - cmpl $(MUTEX_IND),M_ITAG /* is this indirect? */ - cmove M_PTR,%edx /* yes - take indirection */ - CHECK_NO_SIMPLELOCKS() - CHECK_PREEMPTION_LEVEL() - pushf /* save interrupt state */ - cli /* disable interrupts */ -Llmt_retry: - movl %gs:CPU_ACTIVE_THREAD,%ecx +/* + * int lck_rw_grab_want(lck_rw_t *) + * + */ +Entry(lck_rw_grab_want) + LOAD_LCK_RW_REGISTER +1: + LOAD_LCK_RW_FLAGS_REGISTER /* Load state bitfield, interlock and reader count */ + testl $(LCK_RW_INTERLOCK), %eax + jne 3f /* wait for interlock to clear */ + testl $(LCK_RW_WANT_WRITE), %eax /* want_write has been grabbed by someone else */ + jne 2f /* go return failure */ + + movl %eax, %ecx /* original value in %eax for cmpxchgl */ + orl $(LCK_RW_WANT_WRITE), %ecx + lock + cmpxchgl %ecx, (LCK_RW_REGISTER) /* Attempt atomic exchange */ + jne 2f + /* we now own want_write */ + movl $1, %eax /* return success */ + ret +2: + xorl %eax, %eax /* return failure */ + ret +3: + PAUSE + jmp 1b -Llmt_get_hw: - movl M_ILK,%eax /* read interlock */ - testl %eax,%eax /* unlocked? */ - jne Llmt_ilk_fail /* no - slow path */ + +#define RW_LOCK_SHARED_OR_UPGRADE_MASK (LCK_RW_SHARED_MASK | LCK_RW_INTERLOCK | LCK_RW_WANT_UPGRADE) +/* + * int lck_rw_held_read_or_upgrade(lck_rw_t *) + * + */ +Entry(lck_rw_held_read_or_upgrade) + LOAD_LCK_RW_REGISTER + LOAD_LCK_RW_FLAGS_REGISTER /* Load state bitfield, interlock and reader count */ + andl $(RW_LOCK_SHARED_OR_UPGRADE_MASK), %eax + ret - lock; cmpxchgl %ecx,M_ILK /* atomic compare and exchange */ - jne Llmt_get_hw /* branch on failure to retry */ - 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 + +/* + * N.B.: On x86, statistics are currently recorded for all indirect mutexes. + * Also, only the acquire attempt count (GRP_MTX_STAT_UTIL) is maintained + * as a 64-bit quantity (this matches the existing PowerPC implementation, + * and the new x86 specific statistics are also maintained as 32-bit + * quantities). + * + * + * Enable this preprocessor define to record the first miss alone + * By default, we count every miss, hence multiple misses may be + * recorded for a single lock acquire attempt via lck_mtx_lock + */ +#undef LOG_FIRST_MISS_ALONE - 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 +/* + * This preprocessor define controls whether the R-M-W update of the + * per-group statistics elements are atomic (LOCK-prefixed) + * Enabled by default. + */ +#define ATOMIC_STAT_UPDATES 1 + +#if defined(ATOMIC_STAT_UPDATES) +#define LOCK_IF_ATOMIC_STAT_UPDATES lock +#else +#define LOCK_IF_ATOMIC_STAT_UPDATES +#endif /* ATOMIC_STAT_UPDATES */ + + +/* + * For most routines, the lck_mtx_t pointer is loaded into a + * register initially, and the owner field checked for indirection. + * Eventually the lock owner is loaded into a register and examined. + */ + +#define M_OWNER MUTEX_OWNER +#define M_PTR MUTEX_PTR +#define M_STATE MUTEX_STATE + +#if defined(__i386__) + +#define LMTX_ARG0 B_ARG0 +#define LMTX_ARG1 B_ARG1 +#define LMTX_REG %edx +#define LMTX_A_REG %eax +#define LMTX_A_REG32 %eax +#define LMTX_C_REG %ecx +#define LMTX_C_REG32 %ecx +#define LMTX_RET_REG %eax +#define LMTX_RET_REG32 %eax +#define LMTX_LGROUP_REG %esi +#define LMTX_SSTATE_REG %edi +#define LOAD_LMTX_REG(arg) mov arg, LMTX_REG +#define LMTX_CHK_EXTENDED cmp LMTX_REG, LMTX_ARG0 +#define LMTX_ASSERT_OWNED cmpl $(MUTEX_ASSERT_OWNED), LMTX_ARG1 + +#define LMTX_ENTER_EXTENDED \ + mov M_PTR(LMTX_REG), LMTX_REG ; \ + push LMTX_LGROUP_REG ; \ + push LMTX_SSTATE_REG ; \ + xor LMTX_SSTATE_REG, LMTX_SSTATE_REG ; \ + mov MUTEX_GRP(LMTX_REG), LMTX_LGROUP_REG ; \ + LOCK_IF_ATOMIC_STAT_UPDATES ; \ + addl $1, GRP_MTX_STAT_UTIL(LMTX_LGROUP_REG) ; \ + jnc 11f ; \ + incl GRP_MTX_STAT_UTIL+4(LMTX_LGROUP_REG) ; \ +11: + +#define LMTX_EXIT_EXTENDED \ + pop LMTX_SSTATE_REG ; \ + pop LMTX_LGROUP_REG + + +#define LMTX_CHK_EXTENDED_EXIT \ + cmp LMTX_REG, LMTX_ARG0 ; \ + je 12f ; \ + pop LMTX_SSTATE_REG ; \ + pop LMTX_LGROUP_REG ; \ +12: + + +#if LOG_FIRST_MISS_ALONE +#define LMTX_UPDATE_MISS \ + test $1, LMTX_SSTATE_REG ; \ + jnz 11f ; \ + LOCK_IF_ATOMIC_STAT_UPDATES ; \ + incl GRP_MTX_STAT_MISS(LMTX_LGROUP_REG) ; \ + or $1, LMTX_SSTATE_REG ; \ +11: +#else +#define LMTX_UPDATE_MISS \ + LOCK_IF_ATOMIC_STAT_UPDATES ; \ + incl GRP_MTX_STAT_MISS(LMTX_LGROUP_REG) +#endif + + +#if LOG_FIRST_MISS_ALONE +#define LMTX_UPDATE_WAIT \ + test $2, LMTX_SSTATE_REG ; \ + jnz 11f ; \ + LOCK_IF_ATOMIC_STAT_UPDATES ; \ + incl GRP_MTX_STAT_WAIT(LMTX_LGROUP_REG) ; \ + or $2, LMTX_SSTATE_REG ; \ +11: +#else +#define LMTX_UPDATE_WAIT \ + LOCK_IF_ATOMIC_STAT_UPDATES ; \ + incl GRP_MTX_STAT_WAIT(LMTX_LGROUP_REG) +#endif - popf /* restore interrupt state */ + +/* + * Record the "direct wait" statistic, which indicates if a + * miss proceeded to block directly without spinning--occurs + * if the owner of the mutex isn't running on another processor + * at the time of the check. + */ +#define LMTX_UPDATE_DIRECT_WAIT \ + LOCK_IF_ATOMIC_STAT_UPDATES ; \ + incl GRP_MTX_STAT_DIRECT_WAIT(LMTX_LGROUP_REG) + + +#define LMTX_CALLEXT1(func_name) \ + push LMTX_REG ; \ + push LMTX_REG ; \ + call EXT(func_name) ; \ + add $4, %esp ; \ + pop LMTX_REG + +#define LMTX_CALLEXT2(func_name, reg) \ + push LMTX_REG ; \ + push reg ; \ + push LMTX_REG ; \ + call EXT(func_name) ; \ + add $8, %esp ; \ + pop LMTX_REG + +#elif defined(__x86_64__) + +#define LMTX_ARG0 %rdi +#define LMTX_ARG1 %rsi +#define LMTX_REG_ORIG %rdi +#define LMTX_REG %rdx +#define LMTX_A_REG %rax +#define LMTX_A_REG32 %eax +#define LMTX_C_REG %rcx +#define LMTX_C_REG32 %ecx +#define LMTX_RET_REG %rax +#define LMTX_RET_REG32 %eax +#define LMTX_LGROUP_REG %r10 +#define LMTX_SSTATE_REG %r11 +#define LOAD_LMTX_REG(arg) mov %rdi, %rdx +#define LMTX_CHK_EXTENDED cmp LMTX_REG, LMTX_REG_ORIG +#define LMTX_ASSERT_OWNED cmp $(MUTEX_ASSERT_OWNED), LMTX_ARG1 + +#define LMTX_ENTER_EXTENDED \ + mov M_PTR(LMTX_REG), LMTX_REG ; \ + xor LMTX_SSTATE_REG, LMTX_SSTATE_REG ; \ + mov MUTEX_GRP(LMTX_REG), LMTX_LGROUP_REG ; \ + LOCK_IF_ATOMIC_STAT_UPDATES ; \ + incq GRP_MTX_STAT_UTIL(LMTX_LGROUP_REG) + +#define LMTX_EXIT_EXTENDED + +#define LMTX_CHK_EXTENDED_EXIT + + +#if LOG_FIRST_MISS_ALONE +#define LMTX_UPDATE_MISS \ + test $1, LMTX_SSTATE_REG ; \ + jnz 11f ; \ + LOCK_IF_ATOMIC_STAT_UPDATES ; \ + incl GRP_MTX_STAT_MISS(LMTX_LGROUP_REG) ; \ + or $1, LMTX_SSTATE_REG ; \ +11: +#else +#define LMTX_UPDATE_MISS \ + LOCK_IF_ATOMIC_STAT_UPDATES ; \ + incl GRP_MTX_STAT_MISS(LMTX_LGROUP_REG) +#endif + + +#if LOG_FIRST_MISS_ALONE +#define LMTX_UPDATE_WAIT \ + test $2, LMTX_SSTATE_REG ; \ + jnz 11f ; \ + LOCK_IF_ATOMIC_STAT_UPDATES ; \ + incl GRP_MTX_STAT_WAIT(LMTX_LGROUP_REG) ; \ + or $2, LMTX_SSTATE_REG ; \ +11: +#else +#define LMTX_UPDATE_WAIT \ + LOCK_IF_ATOMIC_STAT_UPDATES ; \ + incl GRP_MTX_STAT_WAIT(LMTX_LGROUP_REG) +#endif + + +/* + * Record the "direct wait" statistic, which indicates if a + * miss proceeded to block directly without spinning--occurs + * if the owner of the mutex isn't running on another processor + * at the time of the check. + */ +#define LMTX_UPDATE_DIRECT_WAIT \ + LOCK_IF_ATOMIC_STAT_UPDATES ; \ + incl GRP_MTX_STAT_DIRECT_WAIT(LMTX_LGROUP_REG) + + +#define LMTX_CALLEXT1(func_name) \ + LMTX_CHK_EXTENDED ; \ + je 12f ; \ + push LMTX_LGROUP_REG ; \ + push LMTX_SSTATE_REG ; \ +12: push LMTX_REG_ORIG ; \ + push LMTX_REG ; \ + mov LMTX_REG, LMTX_ARG0 ; \ + call EXT(func_name) ; \ + pop LMTX_REG ; \ + pop LMTX_REG_ORIG ; \ + LMTX_CHK_EXTENDED ; \ + je 12f ; \ + pop LMTX_SSTATE_REG ; \ + pop LMTX_LGROUP_REG ; \ +12: + +#define LMTX_CALLEXT2(func_name, reg) \ + LMTX_CHK_EXTENDED ; \ + je 12f ; \ + push LMTX_LGROUP_REG ; \ + push LMTX_SSTATE_REG ; \ +12: push LMTX_REG_ORIG ; \ + push LMTX_REG ; \ + mov reg, LMTX_ARG1 ; \ + mov LMTX_REG, LMTX_ARG0 ; \ + call EXT(func_name) ; \ + pop LMTX_REG ; \ + pop LMTX_REG_ORIG ; \ + LMTX_CHK_EXTENDED ; \ + je 12f ; \ + pop LMTX_SSTATE_REG ; \ + pop LMTX_LGROUP_REG ; \ +12: + +#else +#error Unsupported architecture +#endif + + +#define M_WAITERS_MSK 0x0000ffff +#define M_PRIORITY_MSK 0x00ff0000 +#define M_ILOCKED_MSK 0x01000000 +#define M_MLOCKED_MSK 0x02000000 +#define M_PROMOTED_MSK 0x04000000 +#define M_SPIN_MSK 0x08000000 - movl $1,%eax /* return success */ +/* + * void lck_mtx_assert(lck_mtx_t* l, unsigned int) + * Takes the address of a lock, and an assertion type as parameters. + * The assertion can take one of two forms determine by the type + * parameter: either the lock is held by the current thread, and the + * type is LCK_MTX_ASSERT_OWNED, or it isn't and the type is + * LCK_MTX_ASSERT_NOTOWNED. Calls panic on assertion failure. + * + */ + +NONLEAF_ENTRY(lck_mtx_assert) + LOAD_LMTX_REG(B_ARG0) /* Load lock address */ + mov %gs:CPU_ACTIVE_THREAD, LMTX_A_REG /* Load current thread */ + + mov M_STATE(LMTX_REG), LMTX_C_REG32 + cmp $(MUTEX_IND), LMTX_C_REG32 /* Is this an indirect mutex? */ + jne 0f + mov M_PTR(LMTX_REG), LMTX_REG /* If so, take indirection */ +0: + mov M_OWNER(LMTX_REG), LMTX_C_REG /* Load owner */ + LMTX_ASSERT_OWNED + jne 2f /* Assert ownership? */ + cmp LMTX_A_REG, LMTX_C_REG /* Current thread match? */ + jne 3f /* no, go panic */ + testl $(M_ILOCKED_MSK | M_MLOCKED_MSK), M_STATE(LMTX_REG) + je 3f +1: /* yes, we own it */ NONLEAF_RET +2: + cmp LMTX_A_REG, LMTX_C_REG /* Current thread match? */ + jne 1b /* No, return */ + ALIGN_STACK() + LOAD_PTR_ARG1(LMTX_REG) + LOAD_STRING_ARG0(mutex_assert_owned_str) + jmp 4f +3: + ALIGN_STACK() + LOAD_PTR_ARG1(LMTX_REG) + LOAD_STRING_ARG0(mutex_assert_not_owned_str) +4: + CALL_PANIC() -Llmt_waiters: - pushl %edx /* save mutex address */ - pushl %edx - call EXT(lck_mtx_lock_acquire) - addl $4,%esp - popl %edx /* restore mutex address */ - 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 */ +lck_mtx_destroyed: + ALIGN_STACK() + LOAD_PTR_ARG1(LMTX_REG) + LOAD_STRING_ARG0(mutex_interlock_destroyed_str) + CALL_PANIC() + + +.data +mutex_assert_not_owned_str: + .asciz "mutex (%p) not owned\n" +mutex_assert_owned_str: + .asciz "mutex (%p) owned\n" +mutex_interlock_destroyed_str: + .asciz "trying to interlock destroyed mutex (%p)" +.text + + + +/* + * lck_mtx_lock() + * lck_mtx_try_lock() + * lck_mtx_unlock() + * lck_mtx_lock_spin() + * lck_mtx_lock_spin_always() + * lck_mtx_convert_spin() + */ +NONLEAF_ENTRY(lck_mtx_lock_spin_always) + LOAD_LMTX_REG(B_ARG0) /* fetch lock pointer */ + jmp Llmls_avoid_check + +NONLEAF_ENTRY(lck_mtx_lock_spin) + LOAD_LMTX_REG(B_ARG0) /* fetch lock pointer */ + + CHECK_PREEMPTION_LEVEL() +Llmls_avoid_check: + mov M_STATE(LMTX_REG), LMTX_C_REG32 + test $(M_ILOCKED_MSK | M_MLOCKED_MSK), LMTX_C_REG32 /* is the interlock or mutex held */ + jnz Llmls_slow +Llmls_try: /* no - can't be INDIRECT, DESTROYED or locked */ + mov LMTX_C_REG, LMTX_A_REG /* eax contains snapshot for cmpxchgl */ + or $(M_ILOCKED_MSK | M_SPIN_MSK), LMTX_C_REG32 + + PREEMPTION_DISABLE + lock + cmpxchg LMTX_C_REG32, M_STATE(LMTX_REG) /* atomic compare and exchange */ + jne Llmls_busy_disabled + + mov %gs:CPU_ACTIVE_THREAD, LMTX_A_REG + mov LMTX_A_REG, M_OWNER(LMTX_REG) /* record owner of interlock */ +#if MACH_LDEBUG + test LMTX_A_REG, LMTX_A_REG + jz 1f + incl TH_MUTEX_COUNT(LMTX_A_REG) /* lock statistic */ +1: +#endif /* MACH_LDEBUG */ + + LMTX_CHK_EXTENDED_EXIT + /* return with the interlock held and preemption disabled */ + leave +#if CONFIG_DTRACE + LOCKSTAT_LABEL(_lck_mtx_lock_spin_lockstat_patch_point) + ret + /* inherit lock pointer in LMTX_REG above */ + LOCKSTAT_RECORD(LS_LCK_MTX_LOCK_SPIN_ACQUIRE, LMTX_REG) +#endif + ret + +Llmls_slow: + test $M_ILOCKED_MSK, LMTX_C_REG32 /* is the interlock held */ + jz Llml_contended /* no, must have been the mutex */ + + cmp $(MUTEX_DESTROYED), LMTX_C_REG32 /* check to see if its marked destroyed */ + je lck_mtx_destroyed + cmp $(MUTEX_IND), LMTX_C_REG32 /* Is this an indirect mutex */ + jne Llmls_loop /* no... must be interlocked */ + + LMTX_ENTER_EXTENDED + + mov M_STATE(LMTX_REG), LMTX_C_REG32 + test $(M_SPIN_MSK), LMTX_C_REG32 + jz Llmls_loop1 + + LMTX_UPDATE_MISS /* M_SPIN_MSK was set, so M_ILOCKED_MSK must also be present */ +Llmls_loop: + PAUSE + mov M_STATE(LMTX_REG), LMTX_C_REG32 +Llmls_loop1: + test $(M_ILOCKED_MSK | M_MLOCKED_MSK), LMTX_C_REG32 + jz Llmls_try + test $(M_MLOCKED_MSK), LMTX_C_REG32 + jnz Llml_contended /* mutex owned by someone else, go contend for it */ + jmp Llmls_loop + +Llmls_busy_disabled: + PREEMPTION_ENABLE + jmp Llmls_loop + + + +NONLEAF_ENTRY(lck_mtx_lock) + LOAD_LMTX_REG(B_ARG0) /* fetch lock pointer */ + + CHECK_PREEMPTION_LEVEL() + + mov M_STATE(LMTX_REG), LMTX_C_REG32 + test $(M_ILOCKED_MSK | M_MLOCKED_MSK), LMTX_C_REG32 /* is the interlock or mutex held */ + jnz Llml_slow +Llml_try: /* no - can't be INDIRECT, DESTROYED or locked */ + mov LMTX_C_REG, LMTX_A_REG /* eax contains snapshot for cmpxchgl */ + or $(M_ILOCKED_MSK | M_MLOCKED_MSK), LMTX_C_REG32 + + PREEMPTION_DISABLE + lock + cmpxchg LMTX_C_REG32, M_STATE(LMTX_REG) /* atomic compare and exchange */ + jne Llml_busy_disabled + + mov %gs:CPU_ACTIVE_THREAD, LMTX_A_REG + mov LMTX_A_REG, M_OWNER(LMTX_REG) /* record owner of mutex */ +#if MACH_LDEBUG + test LMTX_A_REG, LMTX_A_REG + jz 1f + incl TH_MUTEX_COUNT(LMTX_A_REG) /* lock statistic */ +1: +#endif /* MACH_LDEBUG */ + + testl $(M_WAITERS_MSK), M_STATE(LMTX_REG) + jz Llml_finish + + LMTX_CALLEXT1(lck_mtx_lock_acquire_x86) + +Llml_finish: + andl $(~M_ILOCKED_MSK), M_STATE(LMTX_REG) + PREEMPTION_ENABLE + + LMTX_CHK_EXTENDED /* is this an extended mutex */ + jne 2f + + leave +#if CONFIG_DTRACE + LOCKSTAT_LABEL(_lck_mtx_lock_lockstat_patch_point) + ret + /* inherit lock pointer in LMTX_REG above */ + LOCKSTAT_RECORD(LS_LCK_MTX_LOCK_ACQUIRE, LMTX_REG) +#endif + ret +2: + LMTX_EXIT_EXTENDED + leave +#if CONFIG_DTRACE + LOCKSTAT_LABEL(_lck_mtx_lock_ext_lockstat_patch_point) + ret + /* inherit lock pointer in LMTX_REG above */ + LOCKSTAT_RECORD(LS_LCK_MTX_EXT_LOCK_ACQUIRE, LMTX_REG) +#endif + ret + + +Llml_slow: + test $M_ILOCKED_MSK, LMTX_C_REG32 /* is the interlock held */ + jz Llml_contended /* no, must have been the mutex */ + + cmp $(MUTEX_DESTROYED), LMTX_C_REG32 /* check to see if its marked destroyed */ + je lck_mtx_destroyed + cmp $(MUTEX_IND), LMTX_C_REG32 /* Is this an indirect mutex? */ + jne Llml_loop /* no... must be interlocked */ + + LMTX_ENTER_EXTENDED + + mov M_STATE(LMTX_REG), LMTX_C_REG32 + test $(M_SPIN_MSK), LMTX_C_REG32 + jz Llml_loop1 + + LMTX_UPDATE_MISS /* M_SPIN_MSK was set, so M_ILOCKED_MSK must also be present */ +Llml_loop: + PAUSE + mov M_STATE(LMTX_REG), LMTX_C_REG32 +Llml_loop1: + test $(M_ILOCKED_MSK | M_MLOCKED_MSK), LMTX_C_REG32 + jz Llml_try + test $(M_MLOCKED_MSK), LMTX_C_REG32 + jnz Llml_contended /* mutex owned by someone else, go contend for it */ + jmp Llml_loop + +Llml_busy_disabled: + PREEMPTION_ENABLE + jmp Llml_loop + + +Llml_contended: + LMTX_CHK_EXTENDED /* is this an extended mutex */ + je 0f + LMTX_UPDATE_MISS +0: + LMTX_CALLEXT1(lck_mtx_lock_spinwait_x86) + + test LMTX_RET_REG, LMTX_RET_REG + jz Llml_acquired /* acquired mutex, interlock held and preemption disabled */ + + cmp $1, LMTX_RET_REG /* check for direct wait status */ + je 2f + LMTX_CHK_EXTENDED /* is this an extended mutex */ + je 2f + LMTX_UPDATE_DIRECT_WAIT +2: + mov M_STATE(LMTX_REG), LMTX_C_REG32 + test $(M_ILOCKED_MSK), LMTX_C_REG32 + jnz 6f + + mov LMTX_C_REG, LMTX_A_REG /* eax contains snapshot for cmpxchgl */ + or $(M_ILOCKED_MSK), LMTX_C_REG32 /* try to take the interlock */ + + PREEMPTION_DISABLE + lock + cmpxchg LMTX_C_REG32, M_STATE(LMTX_REG) /* atomic compare and exchange */ + jne 5f + + test $(M_MLOCKED_MSK), LMTX_C_REG32 /* we've got the interlock and */ + jnz 3f + or $(M_MLOCKED_MSK), LMTX_C_REG32 /* the mutex is free... grab it directly */ + mov LMTX_C_REG32, M_STATE(LMTX_REG) + + mov %gs:CPU_ACTIVE_THREAD, LMTX_A_REG + mov LMTX_A_REG, M_OWNER(LMTX_REG) /* record owner of mutex */ +#if MACH_LDEBUG + test LMTX_A_REG, LMTX_A_REG + jz 1f + incl TH_MUTEX_COUNT(LMTX_A_REG) /* lock statistic */ +1: +#endif /* MACH_LDEBUG */ + +Llml_acquired: + testl $(M_WAITERS_MSK), M_STATE(LMTX_REG) + jnz 1f + mov M_OWNER(LMTX_REG), LMTX_A_REG + mov TH_WAS_PROMOTED_ON_WAKEUP(LMTX_A_REG), LMTX_A_REG32 + test LMTX_A_REG32, LMTX_A_REG32 + jz Llml_finish +1: + LMTX_CALLEXT1(lck_mtx_lock_acquire_x86) + jmp Llml_finish + +3: /* interlock held, mutex busy */ + LMTX_CHK_EXTENDED /* is this an extended mutex */ + je 4f + LMTX_UPDATE_WAIT +4: + LMTX_CALLEXT1(lck_mtx_lock_wait_x86) + jmp Llml_contended +5: + PREEMPTION_ENABLE +6: + PAUSE + jmp 2b + + + +NONLEAF_ENTRY(lck_mtx_try_lock_spin) + LOAD_LMTX_REG(B_ARG0) /* fetch lock pointer */ + + mov M_STATE(LMTX_REG), LMTX_C_REG32 + test $(M_ILOCKED_MSK | M_MLOCKED_MSK), LMTX_C_REG32 /* is the interlock or mutex held */ + jnz Llmts_slow +Llmts_try: /* no - can't be INDIRECT, DESTROYED or locked */ + mov LMTX_C_REG, LMTX_A_REG /* eax contains snapshot for cmpxchgl */ + or $(M_ILOCKED_MSK | M_SPIN_MSK), LMTX_C_REG + + PREEMPTION_DISABLE + lock + cmpxchg LMTX_C_REG32, M_STATE(LMTX_REG) /* atomic compare and exchange */ + jne Llmts_busy_disabled + + mov %gs:CPU_ACTIVE_THREAD, LMTX_A_REG + mov LMTX_A_REG, M_OWNER(LMTX_REG) /* record owner of mutex */ +#if MACH_LDEBUG + test LMTX_A_REG, LMTX_A_REG + jz 1f + incl TH_MUTEX_COUNT(LMTX_A_REG) /* lock statistic */ +1: +#endif /* MACH_LDEBUG */ + + LMTX_CHK_EXTENDED_EXIT + leave + +#if CONFIG_DTRACE + mov $1, LMTX_RET_REG /* return success */ + LOCKSTAT_LABEL(_lck_mtx_try_lock_spin_lockstat_patch_point) + ret + /* inherit lock pointer in LMTX_REG above */ + LOCKSTAT_RECORD(LS_LCK_MTX_TRY_SPIN_LOCK_ACQUIRE, LMTX_REG) +#endif + mov $1, LMTX_RET_REG /* return success */ + ret + +Llmts_slow: + test $(M_ILOCKED_MSK), LMTX_C_REG32 /* is the interlock held */ + jz Llmts_fail /* no, must be held as a mutex */ + + cmp $(MUTEX_DESTROYED), LMTX_C_REG32 /* check to see if its marked destroyed */ + je lck_mtx_destroyed + cmp $(MUTEX_IND), LMTX_C_REG32 /* Is this an indirect mutex? */ + jne Llmts_loop1 + + LMTX_ENTER_EXTENDED +Llmts_loop: + PAUSE + mov M_STATE(LMTX_REG), LMTX_C_REG32 +Llmts_loop1: + test $(M_MLOCKED_MSK | M_SPIN_MSK), LMTX_C_REG32 + jnz Llmts_fail + test $(M_ILOCKED_MSK), LMTX_C_REG32 + jz Llmts_try + jmp Llmts_loop + +Llmts_busy_disabled: + PREEMPTION_ENABLE + jmp Llmts_loop + + + +NONLEAF_ENTRY(lck_mtx_try_lock) + LOAD_LMTX_REG(B_ARG0) /* fetch lock pointer */ + + mov M_STATE(LMTX_REG), LMTX_C_REG32 + test $(M_ILOCKED_MSK | M_MLOCKED_MSK), LMTX_C_REG32 /* is the interlock or mutex held */ + jnz Llmt_slow +Llmt_try: /* no - can't be INDIRECT, DESTROYED or locked */ + mov LMTX_C_REG, LMTX_A_REG /* eax contains snapshot for cmpxchgl */ + or $(M_ILOCKED_MSK | M_MLOCKED_MSK), LMTX_C_REG32 + + PREEMPTION_DISABLE + lock + cmpxchg LMTX_C_REG32, M_STATE(LMTX_REG) /* atomic compare and exchange */ + jne Llmt_busy_disabled + + mov %gs:CPU_ACTIVE_THREAD, LMTX_A_REG + mov LMTX_A_REG, M_OWNER(LMTX_REG) /* record owner of mutex */ +#if MACH_LDEBUG + test LMTX_A_REG, LMTX_A_REG + jz 1f + incl TH_MUTEX_COUNT(LMTX_A_REG) /* lock statistic */ +1: +#endif /* MACH_LDEBUG */ + + LMTX_CHK_EXTENDED_EXIT + + test $(M_WAITERS_MSK), LMTX_C_REG32 + jz 0f + + LMTX_CALLEXT1(lck_mtx_lock_acquire_x86) +0: + andl $(~M_ILOCKED_MSK), M_STATE(LMTX_REG) + PREEMPTION_ENABLE + + leave +#if CONFIG_DTRACE + mov $1, LMTX_RET_REG /* return success */ + /* Dtrace probe: LS_LCK_MTX_TRY_LOCK_ACQUIRE */ + LOCKSTAT_LABEL(_lck_mtx_try_lock_lockstat_patch_point) + ret + /* inherit lock pointer in LMTX_REG from above */ + LOCKSTAT_RECORD(LS_LCK_MTX_TRY_LOCK_ACQUIRE, LMTX_REG) +#endif + mov $1, LMTX_RET_REG /* return success */ + ret + +Llmt_slow: + test $(M_ILOCKED_MSK), LMTX_C_REG32 /* is the interlock held */ + jz Llmt_fail /* no, must be held as a mutex */ + + cmp $(MUTEX_DESTROYED), LMTX_C_REG32 /* check to see if its marked destroyed */ + je lck_mtx_destroyed + cmp $(MUTEX_IND), LMTX_C_REG32 /* Is this an indirect mutex? */ + jne Llmt_loop + + LMTX_ENTER_EXTENDED +Llmt_loop: + PAUSE + mov M_STATE(LMTX_REG), LMTX_C_REG32 +Llmt_loop1: + test $(M_MLOCKED_MSK | M_SPIN_MSK), LMTX_C_REG32 + jnz Llmt_fail + test $(M_ILOCKED_MSK), LMTX_C_REG32 + jz Llmt_try + jmp Llmt_loop + +Llmt_busy_disabled: + PREEMPTION_ENABLE + jmp Llmt_loop + Llmt_fail: - xorl %eax,%eax - movl %eax,M_ILK +Llmts_fail: + LMTX_CHK_EXTENDED /* is this an extended mutex */ + je 0f + LMTX_UPDATE_MISS + LMTX_EXIT_EXTENDED +0: + xor LMTX_RET_REG, LMTX_RET_REG + NONLEAF_RET + - popf /* restore interrupt state */ - xorl %eax,%eax /* return failure */ +NONLEAF_ENTRY(lck_mtx_convert_spin) + LOAD_LMTX_REG(B_ARG0) /* fetch lock pointer */ + + mov M_STATE(LMTX_REG), LMTX_C_REG32 + cmp $(MUTEX_IND), LMTX_C_REG32 /* Is this an indirect mutex? */ + jne 0f + mov M_PTR(LMTX_REG), LMTX_REG /* If so, take indirection */ + mov M_STATE(LMTX_REG), LMTX_C_REG32 +0: + test $(M_MLOCKED_MSK), LMTX_C_REG32 /* already owned as a mutex, just return */ + jnz 2f + test $(M_WAITERS_MSK), LMTX_C_REG32 /* are there any waiters? */ + jz 1f + + LMTX_CALLEXT1(lck_mtx_lock_acquire_x86) + mov M_STATE(LMTX_REG), LMTX_C_REG32 +1: + and $(~(M_ILOCKED_MSK | M_SPIN_MSK)), LMTX_C_REG32 /* convert from spin version to mutex */ + or $(M_MLOCKED_MSK), LMTX_C_REG32 + mov LMTX_C_REG32, M_STATE(LMTX_REG) /* since I own the interlock, I don't need an atomic update */ + + PREEMPTION_ENABLE +2: NONLEAF_RET + + +#if defined(__i386__) NONLEAF_ENTRY(lck_mtx_unlock) + LOAD_LMTX_REG(B_ARG0) /* fetch lock pointer */ + mov M_OWNER(LMTX_REG), LMTX_A_REG + test LMTX_A_REG, LMTX_A_REG + jnz Llmu_entry + leave + ret +NONLEAF_ENTRY(lck_mtx_unlock_darwin10) +#else +NONLEAF_ENTRY(lck_mtx_unlock) +#endif + LOAD_LMTX_REG(B_ARG0) /* fetch lock pointer */ +Llmu_entry: + mov M_STATE(LMTX_REG), LMTX_C_REG32 +Llmu_prim: + cmp $(MUTEX_IND), LMTX_C_REG32 /* Is this an indirect mutex? */ + je Llmu_ext + +Llmu_chktype: + test $(M_MLOCKED_MSK), LMTX_C_REG32 /* check for full mutex */ + jz Llmu_unlock +Llmu_mutex: + test $(M_ILOCKED_MSK), LMTX_C_REG /* have to wait for interlock to clear */ + jnz Llmu_busy + + mov LMTX_C_REG, LMTX_A_REG /* eax contains snapshot for cmpxchgl */ + and $(~M_MLOCKED_MSK), LMTX_C_REG32 /* drop mutex */ + or $(M_ILOCKED_MSK), LMTX_C_REG32 /* pick up interlock */ + + PREEMPTION_DISABLE + lock + cmpxchg LMTX_C_REG32, M_STATE(LMTX_REG) /* atomic compare and exchange */ + jne Llmu_busy_disabled /* branch on failure to spin loop */ - movl B_ARG0,%edx /* fetch lock pointer */ - cmpl $(MUTEX_IND),M_ITAG /* is this indirect? */ - cmove M_PTR,%edx /* yes - take indirection */ +Llmu_unlock: + xor LMTX_A_REG, LMTX_A_REG + mov LMTX_A_REG, M_OWNER(LMTX_REG) + mov LMTX_C_REG, LMTX_A_REG /* keep original state in %ecx for later evaluation */ + and $(~(M_ILOCKED_MSK | M_SPIN_MSK | M_PROMOTED_MSK)), LMTX_A_REG - pushf /* save interrupt state */ - cli /* disable interrupts */ -Llmu_retry: - movl %gs:CPU_ACTIVE_THREAD,%ecx + test $(M_WAITERS_MSK), LMTX_A_REG32 + jz 2f + dec LMTX_A_REG32 /* decrement waiter count */ +2: + mov LMTX_A_REG32, M_STATE(LMTX_REG) /* since I own the interlock, I don't need an atomic update */ -Llmu_get_hw: - movl M_ILK,%eax /* read interlock */ - testl %eax,%eax /* unlocked? */ - jne Llmu_ilk_fail /* no - slow path */ +#if MACH_LDEBUG + /* perform lock statistics after drop to prevent delay */ + mov %gs:CPU_ACTIVE_THREAD, LMTX_A_REG + test LMTX_A_REG, LMTX_A_REG + jz 1f + decl TH_MUTEX_COUNT(LMTX_A_REG) /* lock statistic */ +1: +#endif /* MACH_LDEBUG */ - lock; cmpxchgl %ecx,M_ILK /* atomic compare and exchange */ - jne Llmu_get_hw /* branch on failure to retry */ + test $(M_PROMOTED_MSK | M_WAITERS_MSK), LMTX_C_REG32 + jz 3f - cmpw $0,M_WAITERS /* are there any waiters? */ - jne Llmu_wakeup /* yes, more work to do */ + LMTX_CALLEXT2(lck_mtx_unlock_wakeup_x86, LMTX_C_REG) +3: + PREEMPTION_ENABLE -Llmu_doit: - xorl %ecx,%ecx - movl %ecx,M_LOCKED /* unlock the mutex */ + LMTX_CHK_EXTENDED + jne 4f - movl %ecx,M_ILK + leave +#if CONFIG_DTRACE + /* Dtrace: LS_LCK_MTX_UNLOCK_RELEASE */ + LOCKSTAT_LABEL(_lck_mtx_unlock_lockstat_patch_point) + ret + /* inherit lock pointer in LMTX_REG from above */ + LOCKSTAT_RECORD(LS_LCK_MTX_UNLOCK_RELEASE, LMTX_REG) +#endif + ret +4: + leave +#if CONFIG_DTRACE + /* Dtrace: LS_LCK_MTX_EXT_UNLOCK_RELEASE */ + LOCKSTAT_LABEL(_lck_mtx_ext_unlock_lockstat_patch_point) + ret + /* inherit lock pointer in LMTX_REG from above */ + LOCKSTAT_RECORD(LS_LCK_MTX_EXT_UNLOCK_RELEASE, LMTX_REG) +#endif + ret - popf /* restore interrupt state */ - NONLEAF_RET +Llmu_busy_disabled: + PREEMPTION_ENABLE +Llmu_busy: + PAUSE + mov M_STATE(LMTX_REG), LMTX_C_REG32 + jmp Llmu_mutex + +Llmu_ext: + mov M_PTR(LMTX_REG), LMTX_REG + mov M_OWNER(LMTX_REG), LMTX_A_REG + mov %gs:CPU_ACTIVE_THREAD, LMTX_C_REG + CHECK_UNLOCK(LMTX_C_REG, LMTX_A_REG) + mov M_STATE(LMTX_REG), LMTX_C_REG32 + jmp Llmu_chktype -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 */ - -Llmu_wakeup: - pushl %edx /* save mutex address */ - pushl M_LOCKED - pushl %edx /* push mutex address */ - call EXT(lck_mtx_unlock_wakeup)/* yes, wake a thread */ - addl $8,%esp - popl %edx /* restore mutex pointer */ - jmp Llmu_doit + LEAF_ENTRY(lck_mtx_ilk_unlock) - movl L_ARG0,%edx /* no indirection here */ + LOAD_LMTX_REG(L_ARG0) /* fetch lock pointer - no indirection here */ - xorl %eax,%eax - movl %eax,M_ILK + andl $(~M_ILOCKED_MSK), M_STATE(LMTX_REG) + + PREEMPTION_ENABLE /* need to re-enable preemption */ + + LEAF_RET + + + +LEAF_ENTRY(lck_mtx_lock_grab_mutex) + LOAD_LMTX_REG(L_ARG0) /* fetch lock pointer - no indirection here */ + mov M_STATE(LMTX_REG), LMTX_C_REG32 + + test $(M_ILOCKED_MSK | M_MLOCKED_MSK), LMTX_C_REG32 /* can't have the mutex yet */ + jnz 3f + + mov LMTX_C_REG, LMTX_A_REG /* eax contains snapshot for cmpxchgl */ + or $(M_ILOCKED_MSK | M_MLOCKED_MSK), LMTX_C_REG32 + + PREEMPTION_DISABLE + lock + cmpxchg LMTX_C_REG32, M_STATE(LMTX_REG) /* atomic compare and exchange */ + jne 2f /* branch on failure to spin loop */ + + mov %gs:CPU_ACTIVE_THREAD, LMTX_A_REG + mov LMTX_A_REG, M_OWNER(LMTX_REG) /* record owner of mutex */ +#if MACH_LDEBUG + test LMTX_A_REG, LMTX_A_REG + jz 1f + incl TH_MUTEX_COUNT(LMTX_A_REG) /* lock statistic */ +1: +#endif /* MACH_LDEBUG */ + + mov $1, LMTX_RET_REG /* return success */ + LEAF_RET +2: + PREEMPTION_ENABLE +3: + xor LMTX_RET_REG, LMTX_RET_REG /* return failure */ LEAF_RET + + +LEAF_ENTRY(lck_mtx_lock_mark_destroyed) + LOAD_LMTX_REG(L_ARG0) +1: + mov M_STATE(LMTX_REG), LMTX_C_REG32 + cmp $(MUTEX_IND), LMTX_C_REG32 /* Is this an indirect mutex? */ + jne 2f + + movl $(MUTEX_DESTROYED), M_STATE(LMTX_REG) /* convert to destroyed state */ + jmp 3f +2: + test $(M_ILOCKED_MSK), LMTX_C_REG /* have to wait for interlock to clear */ + jnz 5f + + PREEMPTION_DISABLE + mov LMTX_C_REG, LMTX_A_REG /* eax contains snapshot for cmpxchgl */ + or $(M_ILOCKED_MSK), LMTX_C_REG32 + lock + cmpxchg LMTX_C_REG32, M_STATE(LMTX_REG) /* atomic compare and exchange */ + jne 4f /* branch on failure to spin loop */ + movl $(MUTEX_DESTROYED), M_STATE(LMTX_REG) /* convert to destroyed state */ + PREEMPTION_ENABLE +3: + LEAF_RET /* return with M_ILOCKED set */ +4: + PREEMPTION_ENABLE +5: + PAUSE + jmp 1b + +LEAF_ENTRY(preemption_underflow_panic) + FRAME + incl %gs:CPU_PREEMPTION_LEVEL + ALIGN_STACK() + LOAD_STRING_ARG0(16f) + CALL_PANIC() + hlt + .data +16: String "Preemption level underflow, possible cause unlocking an unlocked mutex or spinlock" + .text + + LEAF_ENTRY(_disable_preemption) #if MACH_RT - _DISABLE_PREEMPTION + PREEMPTION_DISABLE #endif /* MACH_RT */ LEAF_RET @@ -920,16 +2018,22 @@ LEAF_ENTRY(_enable_preemption) #if MACH_ASSERT cmpl $0,%gs:CPU_PREEMPTION_LEVEL jg 1f +#if __i386__ pushl %gs:CPU_PREEMPTION_LEVEL - pushl $2f - call EXT(panic) +#else + movl %gs:CPU_PREEMPTION_LEVEL,%esi +#endif + ALIGN_STACK() + LOAD_STRING_ARG0(_enable_preemption_less_than_zero) + CALL_PANIC() hlt - .data -2: String "_enable_preemption: preemption_level(%d) < 0!" + .cstring +_enable_preemption_less_than_zero: + .asciz "_enable_preemption: preemption_level(%d) < 0!" .text 1: #endif /* MACH_ASSERT */ - _ENABLE_PREEMPTION + PREEMPTION_ENABLE #endif /* MACH_RT */ LEAF_RET @@ -938,11 +2042,13 @@ LEAF_ENTRY(_enable_preemption_no_check) #if MACH_ASSERT cmpl $0,%gs:CPU_PREEMPTION_LEVEL jg 1f - pushl $2f - call EXT(panic) + ALIGN_STACK() + LOAD_STRING_ARG0(_enable_preemption_no_check_less_than_zero) + CALL_PANIC() hlt - .data -2: String "_enable_preemption_no_check: preemption_level <= 0!" + .cstring +_enable_preemption_no_check_less_than_zero: + .asciz "_enable_preemption_no_check: preemption_level <= 0!" .text 1: #endif /* MACH_ASSERT */ @@ -953,7 +2059,7 @@ LEAF_ENTRY(_enable_preemption_no_check) LEAF_ENTRY(_mp_disable_preemption) #if MACH_RT - _DISABLE_PREEMPTION + PREEMPTION_DISABLE #endif /* MACH_RT */ LEAF_RET @@ -962,16 +2068,22 @@ LEAF_ENTRY(_mp_enable_preemption) #if MACH_ASSERT cmpl $0,%gs:CPU_PREEMPTION_LEVEL jg 1f +#if __i386__ pushl %gs:CPU_PREEMPTION_LEVEL - pushl $2f - call EXT(panic) +#else + movl %gs:CPU_PREEMPTION_LEVEL,%esi +#endif + ALIGN_PANIC() + LOAD_STRING_ARG0(_mp_enable_preemption_less_than_zero) + CALL_PANIC() hlt - .data -2: String "_mp_enable_preemption: preemption_level (%d) <= 0!" + .cstring +_mp_enable_preemption_less_than_zero: + .asciz "_mp_enable_preemption: preemption_level (%d) <= 0!" .text 1: #endif /* MACH_ASSERT */ - _ENABLE_PREEMPTION + PREEMPTION_ENABLE #endif /* MACH_RT */ LEAF_RET @@ -980,11 +2092,13 @@ LEAF_ENTRY(_mp_enable_preemption_no_check) #if MACH_ASSERT cmpl $0,%gs:CPU_PREEMPTION_LEVEL jg 1f - pushl $2f - call EXT(panic) + ALIGN_STACK() + LOAD_STRING_ARG0(_mp_enable_preemption_no_check_less_than_zero) + CALL_PANIC() hlt - .data -2: String "_mp_enable_preemption_no_check: preemption_level <= 0!" + .cstring +_mp_enable_preemption_no_check_less_than_zero: + .asciz "_mp_enable_preemption_no_check: preemption_level <= 0!" .text 1: #endif /* MACH_ASSERT */ @@ -992,6 +2106,7 @@ LEAF_ENTRY(_mp_enable_preemption_no_check) #endif /* MACH_RT */ LEAF_RET +#if __i386__ LEAF_ENTRY(i_bit_set) movl L_ARG0,%edx @@ -1007,6 +2122,7 @@ LEAF_ENTRY(i_bit_clear) btr %edx,(%eax) LEAF_RET + LEAF_ENTRY(bit_lock) movl L_ARG0,%ecx movl L_ARG1,%eax @@ -1016,6 +2132,7 @@ LEAF_ENTRY(bit_lock) jb 1b LEAF_RET + LEAF_ENTRY(bit_lock_try) movl L_ARG0,%ecx movl L_ARG1,%eax @@ -1033,3 +2150,200 @@ LEAF_ENTRY(bit_unlock) lock btr %ecx,(%eax) LEAF_RET + +/* + * Atomic primitives, prototyped in kern/simple_lock.h + */ +LEAF_ENTRY(hw_atomic_add) + movl L_ARG0, %ecx /* Load address of operand */ + movl L_ARG1, %eax /* Load addend */ + movl %eax, %edx + lock + xaddl %eax, (%ecx) /* Atomic exchange and add */ + addl %edx, %eax /* Calculate result */ + LEAF_RET + +LEAF_ENTRY(hw_atomic_sub) + movl L_ARG0, %ecx /* Load address of operand */ + movl L_ARG1, %eax /* Load subtrahend */ + negl %eax + movl %eax, %edx + lock + xaddl %eax, (%ecx) /* Atomic exchange and add */ + addl %edx, %eax /* Calculate result */ + LEAF_RET + +LEAF_ENTRY(hw_atomic_or) + movl L_ARG0, %ecx /* Load address of operand */ + movl (%ecx), %eax +1: + movl L_ARG1, %edx /* Load mask */ + orl %eax, %edx + lock + cmpxchgl %edx, (%ecx) /* Atomic CAS */ + jne 1b + movl %edx, %eax /* Result */ + LEAF_RET +/* + * A variant of hw_atomic_or which doesn't return a value. + * The implementation is thus comparatively more efficient. + */ + +LEAF_ENTRY(hw_atomic_or_noret) + movl L_ARG0, %ecx /* Load address of operand */ + movl L_ARG1, %edx /* Load mask */ + lock + orl %edx, (%ecx) /* Atomic OR */ + LEAF_RET + +LEAF_ENTRY(hw_atomic_and) + movl L_ARG0, %ecx /* Load address of operand */ + movl (%ecx), %eax +1: + movl L_ARG1, %edx /* Load mask */ + andl %eax, %edx + lock + cmpxchgl %edx, (%ecx) /* Atomic CAS */ + jne 1b + movl %edx, %eax /* Result */ + LEAF_RET +/* + * A variant of hw_atomic_and which doesn't return a value. + * The implementation is thus comparatively more efficient. + */ + +LEAF_ENTRY(hw_atomic_and_noret) + movl L_ARG0, %ecx /* Load address of operand */ + movl L_ARG1, %edx /* Load mask */ + lock + andl %edx, (%ecx) /* Atomic AND */ + LEAF_RET + +#else /* !__i386__ */ + +LEAF_ENTRY(i_bit_set) + lock + bts %edi,(%rsi) + LEAF_RET + +LEAF_ENTRY(i_bit_clear) + lock + btr %edi,(%rsi) + LEAF_RET + + +LEAF_ENTRY(bit_lock) +1: + lock + bts %edi,(%rsi) + jb 1b + LEAF_RET + + +LEAF_ENTRY(bit_lock_try) + lock + bts %edi,(%rsi) + jb bit_lock_failed + movl $1, %eax + LEAF_RET +bit_lock_failed: + xorl %eax,%eax + LEAF_RET + +LEAF_ENTRY(bit_unlock) + lock + btr %edi,(%rsi) + LEAF_RET + + +/* + * Atomic primitives, prototyped in kern/simple_lock.h + */ +LEAF_ENTRY(hw_atomic_add) +#if MACH_LDEBUG + test $3, %rdi + jz 1f + ud2 +1: +#endif + movl %esi, %eax /* Load addend */ + lock xaddl %eax, (%rdi) /* Atomic exchange and add */ + addl %esi, %eax /* Calculate result */ + LEAF_RET + +LEAF_ENTRY(hw_atomic_sub) +#if MACH_LDEBUG + test $3, %rdi + jz 1f + ud2 +1: +#endif + negl %esi + movl %esi, %eax + lock xaddl %eax, (%rdi) /* Atomic exchange and add */ + addl %esi, %eax /* Calculate result */ + LEAF_RET + +LEAF_ENTRY(hw_atomic_or) +#if MACH_LDEBUG + test $3, %rdi + jz 1f + ud2 +1: +#endif + movl (%rdi), %eax +1: + movl %esi, %edx /* Load mask */ + orl %eax, %edx + lock cmpxchgl %edx, (%rdi) /* Atomic CAS */ + jne 1b + movl %edx, %eax /* Result */ + LEAF_RET +/* + * A variant of hw_atomic_or which doesn't return a value. + * The implementation is thus comparatively more efficient. + */ + +LEAF_ENTRY(hw_atomic_or_noret) +#if MACH_LDEBUG + test $3, %rdi + jz 1f + ud2 +1: +#endif + lock + orl %esi, (%rdi) /* Atomic OR */ + LEAF_RET + + +LEAF_ENTRY(hw_atomic_and) +#if MACH_LDEBUG + test $3, %rdi + jz 1f + ud2 +1: +#endif + movl (%rdi), %eax +1: + movl %esi, %edx /* Load mask */ + andl %eax, %edx + lock cmpxchgl %edx, (%rdi) /* Atomic CAS */ + jne 1b + movl %edx, %eax /* Result */ + LEAF_RET +/* + * A variant of hw_atomic_and which doesn't return a value. + * The implementation is thus comparatively more efficient. + */ + +LEAF_ENTRY(hw_atomic_and_noret) +#if MACH_LDEBUG + test $3, %rdi + jz 1f + ud2 +1: +#endif + lock andl %esi, (%rdi) /* Atomic OR */ + LEAF_RET + +#endif /* !__i386 __ */