@CopyLeft by ICANTH , I Can do ANy THing that I CAN THink !~
Author :WenHui ,WuHan University ,2012-6-15
PDF版阅读地址 : http://www.docin.com/p1-424285718.html
普通自旋锁
自旋锁最常见的使用场景是创建一段临界区 :
static DEFINE_SPINLOCK(xxx_lock);
unsigned long flags;
spin_lock_irqsave(&xxx_lock, flags);
... critical section here ..
spin_unlock_irqrestore(&xxx_lock, flags);
自旋锁使用时值得注意的是:对于采用使用自旋锁以保证共享变量的存取安全时,仅当系统中 所有涉及 到存取该共享变量的程序部分都采用 成对的spin_lock、和spin_unlock 来进行操作才能保证其安全性。
NOTE! The spin-lock is safe only when you _also_ use the lock itself to do locking across CPU's, which implies that EVERYTHING that touches a shared variable has to agree about the spinlock they want to use.
在Linux2.6.15.5中,自旋体数据结构如下:
当配置CONFIG_SMP时,raw_spinlock_t才是一个含有slock变量的结构,该slock字段标识自旋锁是否空闲状态,用以处理多CPU处理器并发申请锁的情况;当未配置CONFIG_SMP时,对于单CPU而言,不会发生发申请自旋锁,故raw_lock为空结构体。
当配置CONFIG_SMP和CONFIG_PREEMPT时,spinlock_t才会有break_lock字段,break_lock字段用于标记自旋锁竞争状态,当break_lock = 0时表示没有多于两个的执行路径,当break_lock = 1时表示没有其它进程在忙等待该锁。当在SMP多CPU体系架构下有可能出现申请不到自旋锁、空等的情况,但LINUX内核必须保证在spin_lock的原子性,故在配置CONFIG_PREEMPT时必须禁止内核抢占。
字段 |
描述 |
spin_lock_init(lock) |
一个自旋锁时,可使用接口函数将其初始化为锁定状态 |
spin_lock(lock) |
用于锁定自旋锁,如果成功则返回;否则循环等待自旋锁变为空闲 |
spin_unlock(lock) |
释放自旋锁lock,重新设置自旋锁为锁定状态 |
spin_is_locked(lock) |
判断当前自旋锁是否处于锁定状态 |
spin_unlock_wait(lock) |
循环等待、直到自旋锁lock变为可用状态 |
spin_trylock(lock) |
尝试锁定自旋锁lock,如不成功则返回0;否则锁定,并返回1 |
spin_can_lock(lock) |
判断自旋锁lock是否处于空闲状态 |
spin_lock和spin_unlock的关系如下:
可见,在 UP 体系架构 中,由于没有必要有实际的锁以防止多CPU抢占,spin操作仅仅是禁止和开启内核抢占。
LINUX 2.6.35版本,将spin lock实现更改为 ticket lock。spin_lock数据结构除了用于内核调试之外,字段为: raw_spinlock rlock 。
ticket spinlock将rlock字段分解为如下两部分:
Next是下一个票号,而Owner是允许使用自旋锁的票号。加锁时CPU取Next,并将rlock.Next + 1。将Next与Owner相比较,若相同,则加锁成功;否则循环等待、直到Next = rlock.Owner为止。解锁则直接将Owner + 1即可。
spin_lock和spin_unlock的调用关系如下:
普通自旋锁源码分析
源程序文件目录关系图
在/include/linux/spinlock.h中通过是否配置CONFIG_SMP项判断导入哪种自旋锁定义及操作:
004 /*
005 * include/linux/spinlock.h - generic spinlock/rwlock declarations
007 * here's the role of the various spinlock/rwlock related include files:
009 * on SMP builds:
011 * asm/spinlock_types.h: contains the arch_spinlock_t/arch_rwlock_t and the
012 * initializers
014 * linux/spinlock_types.h:
015 * defines the generic type and initializers
017 * asm/spinlock.h: contains the arch_spin_*()/etc. lowlevel
018 * implementations, mostly inline assembly code
022 * linux/spinlock_api_smp.h:
023 * contains the prototypes for the _spin_*() APIs.
025 * linux/spinlock.h: builds the final spin_*() APIs.
027 * on UP builds:
029 * linux/spinlock_type_up.h:
030 * contains the generic, simplified UP spinlock type.
031 * (which is an empty structure on non-debug builds)
033 * linux/spinlock_types.h:
034 * defines the generic type and initializers
036 * linux/spinlock_up.h:
037 * contains the arch_spin_*()/etc. version of UP
038 * builds. (which are NOPs on non-debug, non-preempt
039 * builds)
041 * (included on UP-non-debug builds:)
043 * linux/spinlock_api_up.h:
044 * builds the _spin_*() APIs.
046 * linux/spinlock.h: builds the final spin_*() APIs.
047 */
082 /*
083 * Pull the arch_spin*() functions/declarations (UP-nondebug doesnt need them):
084 */
085 #ifdef CONFIG_SMP
086 # include <asm/spinlock.h>
087 #else
088 # include < linux/spinlock_up.h >
089 #endif
064 typedef struct spinlock {
065 union {
066 struct raw_spinlock rlock ;
075 };
076 } spinlock_t ;
282 static inline void spin_lock ( spinlock_t * lock )
283 {
284 raw_spin_lock (& lock -> rlock );
285 }
169 #define raw_spin_lock ( lock ) _raw_spin_lock ( lock )
322 static inline void spin_unlock ( spinlock_t * lock )
323 {
324 raw_spin_unlock (& lock -> rlock );
325 }
222 #define raw_spin_unlock ( lock ) _raw_spin_unlock ( lock )
UP 体系架构
spin_lock函数在UP体系架构中最终实现方式为:
/include/linux/spinlock_api_up.h
052 #define _raw_spin_lock ( lock ) __LOCK ( lock )
021 /*
022 * In the UP-nondebug case there's no real locking going on , so the
023 * only thing we have to do is to keep the preempt counts and irq
024 * flags straight, to suppress compiler warnings of unused lock
025 * variables, and to add the proper checker annotations:
026 */
027 #define __LOCK ( lock ) \
028 do { preempt_disable (); __acquire ( lock ); (void)( lock ); } while (0)
052 #define _raw_spin_lock ( lock ) __LOCK ( lock )
preempt_disable在未配置CONFIG_PREEMPT时为空函数,否则禁止内核抢占。而__acquire()用于内核编译过程中静态检查。(void)(lock)则是为避免编译器产生lock未被使用的警告。
spin_unlock函数在UP体系架构中最终实现方式为:
039 #define __UNLOCK ( lock ) \
040 do { preempt_enable (); __release ( lock ); (void)( lock ); } while (0)
SMP 体系架构-Tickect Spin Lock的实现方式
在Linux2.6.24中,自旋锁由一个整数表示,当为1时表示锁是空闲的,spin_lock()每次减少1,故 <=0时则表示有多个锁在忙等待,但这将导致不公平性。自linux2.6.25开始,自旋锁将整数拆为一个16位数,结构如下:
该实现机制称为“Ticket spinlocks”,Next字节表示下一次请求锁给其分配的票号,而Owner表示当前可以取得锁的票号,Next和Owner初始化为0。 当lock.Next = lock.Owner时,表示该锁处于空闲状态 。 spin_lock 执行如下过程:
1、my_ticket = slock.next
2、slock.next++
3、wait until my_ticket = slock.owner
spin_unlock 执行如下过程:
1、slock.owner++
但该锁将导致一个问题:8个bit将只能最多表示255个CPU来竞争该锁。故系统通过的方式,将实现两个tickect_spin_lock和ticket_spin_unclock的版本:
058 #if ( NR_CPUS < 256)
059 #define TICKET_SHIFT 8
106 #else
107 #define TICKET_SHIFT 16
SMP 体系架构-SPIN LOCK (ticket_shif 8)
046 #ifdef CONFIG_INLINE_SPIN_LOCK
047 #define _raw_spin_lock ( lock ) __raw_spin_lock ( lock )
048 #endif
/include/linux/spinlock_api_smp.h:
140 static inline void __raw_spin_lock ( raw_spinlock_t * lock )
141 {
142 preempt_disable ();
143 spin_acquire (& lock -> dep_map , 0, 0, _RET_IP_ );
144 LOCK_CONTENDED ( lock , do_raw_spin_trylock , do_raw_spin_lock );
145 }
在__raw_spin_lock中,首先禁止内核抢占,调用LOCK_CONTENED宏
391 #define LOCK_CONTENDED ( _lock , try, lock ) \
392 do { \
393 if (!try( _lock )) { \
394 lock_contended (&( _lock )-> dep_map , _RET_IP_ ); \
395 lock ( _lock ); \
396 } \
397 lock_acquired (&( _lock )-> dep_map , _RET_IP_ ); \
398 } while (0)
其中即在_raw_spin_lock中,即为首先调用do_raw_spin_trylock尝试加锁,若失败则继续调用do_raw_spin_lock进行加锁。而do_raw_spin_xxx具体实现与平台有关。
/include/linux/spinlock.h
136 static inline void do_raw_spin_lock ( raw_spinlock_t * lock ) __acquires ( lock )
137 {
138 __acquire ( lock );
139 arch_spin_lock (& lock -> raw_lock );
140 }
149 static inline int do_raw_spin_trylock ( raw_spinlock_t * lock )
150 {
151 return arch_spin_trylock (&( lock )-> raw_lock );
152 }
在X86平台下, do_raw_spin_lock 和 do_raw_spin_trylock 实现为两个函数:
/arch/x86/include/asm/spinlock.h
188 static __always_inline void arch_spin_lock ( arch_spinlock_t * lock )
189 {
190 __ticket_spin_lock ( lock );
191 }
192
193 static __always_inline int arch_spin_trylock ( arch_spinlock_t * lock )
194 {
195 return __ticket_spin_trylock ( lock );
196 }
058 #if ( NR_CPUS < 256)
059 #define TICKET_SHIFT 8
061 static __always_inline void __ticket_spin_lock ( arch_spinlock_t * lock )
062 {
063 short inc = 0x0100;
064
065 asm volatile (
066 LOCK_PREFIX "xaddw %w0, %1\n"
067 "1:\t"
068 "cmpb %h0, %b0\n\t"
069 "je 2f\n\t"
070 "rep ; nop\n\t"
071 "movb %1, %b0\n\t"
072 /* don't need lfence here, because loads are in-order */
073 "jmp 1b\n"
074 "2:"
075 : "+Q" ( inc ), "+m" ( lock -> slock )
076 :
077 : "memory", "cc");
078 }
066 行 :LOCK_PREFIX在UP上为空定义,而在SMP上为Lock,用以保证从 066行~074行 为原子操作,强制所有CPU缓存失效。xaddw指令用法如下:
xaddw src, dsc ==
tmp = dsc
desc = dsc + src
src = tmp
XADDW语法验证实验:
xaddw使%0和%1按1个word长度交换相加,即:%0: inc → slock, %1: slock → slock + 0x0100。%1此时高字节Next + 1。xaddw使%0和%1内容改变如下:
068 行 :比较inc中自己的Next是否与Owner中ticket相等,若相等则获取自旋锁使用权、结束循环。
070 行 ~ 073行 :如果Owner不属于自己,则执行空语句,并重新读取slock中的Owner,跳回至068行进行判断。
为什么要用LOCK_PREFIX宏来代替直接使用lock指令的方式呢?解释如下:为了避免在配置了CONFIG_SMP项编译产生的SMP内核、实际却运行在UP系统上时系统执行lock命令所带来的开销,系统创建在.smp_locks一张SMP alternatives table用以保存系统中所有lock指令的指针。当实际运行时,若从SMP→UP时,可以根据.smp_locks lock 指针表通过热补丁的方式将lock指令替换成nop指令。当然也可以实现系统运行时将锁由UP→SMP的切换。具体应用可参见参考资料《Linux 内核 LOCK_PREFIX 的含义》。
009 /*
010 * Alternative inline assembly for SMP.
011 *
012 * The LOCK_PREFIX macro defined here replaces the LOCK and
013 * LOCK_PREFIX macros used everywhere in the source tree.
014 *
015 * SMP alternatives use the same data structures as the other
016 * alternatives and the X86_FEATURE_UP flag to indicate the case of a
017 * UP system running a SMP kernel. The existing apply_alternatives()
018 * works fine for patching a SMP kernel for UP.
019 *
020 * The SMP alternative tables can be kept after boot and contain both
021 * UP and SMP versions of the instructions to allow switching back to
022 * SMP at runtime, when hotplugging in a new CPU, which is especially
023 * useful in virtualized environments.
024 *
025 * The very common lock prefix is handled as special case in a
026 * separate table which is a pure address list without replacement ptr
027 * and size information. That keeps the table sizes small.
028 */
029
030 #ifdef CONFIG_SMP
031 #define LOCK_PREFIX_HERE \
032 ".section .smp_locks,\"a\"\n" \
033 ".balign 4\n" \
034 ".long 671f - .\n" /* offset */ \
035 ".previous\n" \
036 "671:"
037
038 #define LOCK_PREFIX LOCK_PREFIX_HERE "\n\tlock; "
039
040 #else /* ! CONFIG_SMP */
041 #define LOCK_PREFIX_HERE ""
042 #define LOCK_PREFIX ""
043 #endif
032 行 “.section .smp_locks, a”,表示以下代码生成在.smp_locks段中,而“a”代表——allocatable。
033 行~034行 “.balign 4 .long 571f”,表示以4字节对齐、将671标签的地址置于.smp_locks段中,而标签671的地址即为:代码段lock指令的地址。(其实就是lock指令的指针啦~~~)
033 行~034行 “.previous”伪指令,表示恢复以前section,即代码段。故在 038行 将导致在代码段生成lock指令。
LOCK_CONTENDED 时首先尝试使用__ticket_spin_trylock对lock进行加锁,若失败则继续使用__ticket_spin_lock进行加锁。不直接调用__ticket_spin_lock而使用__ticket_spin_trylock的原因是:
trylock首先不会修改lock.slock的ticket,它只是通过再次检查,1)将slock读出,并判断slock是否处于空闲状态;2)调用LOCK执行原子操作,判断当前slock的Next是否已经被其它CPU修改,若未被修改则获得该锁,并将lock.slock.Next + 1。
spin_lock,无论如何,首先调用LOCK执行原子性操作、声明ticket;而trylock则首先进行slock.Next == slock.Owner的判断,降低第二次比较调用LOCK的概率。
080 static __always_inline int __ticket_spin_trylock ( arch_spinlock_t * lock )
081 {
082 int tmp , new;
083
084 asm volatile("movzwl %2, %0\n\t"
085 "cmpb %h0,%b0\n\t"
086 "leal 0x100(%" REG_PTR_MODE "0), %1\n\t"
087 "jne 1f\n\t"
088 LOCK_PREFIX "cmpxchgw %w1,%2\n\t"
089 "1:"
090 "sete %b1\n\t"
091 "movzbl %b1,%0\n\t"
092 : "=&a" ( tmp ), "=&q" (new), "+m" ( lock -> slock )
093 :
094 : "memory", "cc");
095
096 return tmp ;
097 }
084 行 将lock.slock的值赋给tmp。
085 行 比较tmp.next == tmp.owner,判断当前自旋锁是否空闲。
086 行 leal指令( Load effective address ),实际上是movl的变形,“leal 0x10 (%eax, %eax, 3), %edx” → “%edx = 0x10 + %eax + %eax * 3”,但leal却不像movl那样从内存取值、而直接读取寄存器。 086行 语句,根据REG_PTR_MODE不同配置,在X86平台下为:“leal 0x100(%k0), %1”,而在其它平台为:“leal 0x100(%q0), %1”,忽略占位符修饰“k”或“q”,则该行语句等价于:
“movl (%0 + 0x100),%1”,此时new = { tmp.Next + 1 , tmp.Owner }。
087 行 若tmp.next != tmp.owner,即自旋锁不空闲,则跳到089行将0赋值给tmp并返回。
088 行 原子性地执行操作cmpxchgw,用以检测当前自旋锁是否已被其它CPU修改lock.slock的Next域,若有竞争者则失败、否则获得该锁并将Next + 1,这一系列操作是原子性的!cmpxchgw操作解释如下:
the accumulator ( 8-32 bits ) with "dest". If equal the "dest" is loaded with "src", otherwise the accumulator is loaded with "dest".(在IA32下,%EAX即为累加器。)
所以,“cmpxchgw %w1, %2”等效于:
“tmp.Next == lock.slock.Next ? lock.slock = new : tmp = lock.slock”
若Next未发生变化,则将lock.slock更新为new, 实质上是将slock的Next+1 。
090 行 执行sete指令,若cmpxchgw或cmpb成功则将new的最低字节%b1赋值为1,否则赋值为0. sete的解释为:
Sets the byte in the operand to 1 if the Zero Flag is set, otherwise sets the operand to 0.
091 行 movzbl( movz from byte to long )指令将%b1赋值给tmp最低字节,且其它位补0.即将tmp置为0或1.
SMP 体系架构-SPIN UNLOCK (ticket_shif 8)
/include/linux/spinlock_api_smp.h
046 #ifdef CONFIG_INLINE_SPIN_LOCK
047 #define _raw_spin_lock ( lock ) __raw_spin_lock ( lock )
048 #endif
149 static inline void __raw_spin_unlock ( raw_spinlock_t * lock )
150 {
151 spin_release (& lock -> dep_map , 1, _RET_IP_ );
152 do_raw_spin_unlock ( lock );
153 preempt_enable ();
154 }
spin_unlock即最终调用do_raw_spin_unlock对自旋锁进行释放操作。
/include/linux/spinlock.h
136 static inline void do_raw_spin_lock ( raw_spinlock_t * lock ) __acquires ( lock )
137 {
138 __acquire ( lock );
139 arch_spin_lock (& lock -> raw_lock );
140 }
对于x86的IA32平台,arch_spin_lock实现如下:
/arch/x86/include/asm/spinlock.h
198 static __always_inline void arch_spin_unlock ( arch_spinlock_t * lock )
199 {
200 __ticket_spin_unlock ( lock );
201 }
058 #if ( NR_CPUS < 256)
059 #define TICKET_SHIFT 8
099 static __always_inline void __ticket_spin_unlock ( arch_spinlock_t * lock )
100 {
101 asm volatile( UNLOCK_LOCK_PREFIX "incb %0"
102 : "+m" ( lock -> slock )
103 :
104 : "memory", "cc");
105 }
101 行 将lock->slock的Owner + 1,表示可以让下一个拥有牌号的CPU加锁。
030 #if defined (CONFIG_X86_32) && \
031 ( defined (CONFIG_X86_OOSTORE) || defined (CONFIG_X86_PPRO_FENCE))
032 /*
033 * On PPro SMP or if we are using OOSTORE, we use a locked operation to unlock
034 * (PPro errata 66, 92)
035 */
036 # define UNLOCK_LOCK_PREFIX LOCK_PREFIX
037 #else
038 # define UNLOCK_LOCK_PREFIX
039 #endif
参考资料
自旋锁
《spinlocks.txt》,/Documentation/spinlocks.txt
《Ticket spinlocks》, http://lwn.net/Articles/267968/
《Linux x86 spinlock实现之分析》, http://blog.csdn.net/david_henry/article/details/5405093
《Linux 内核 LOCK_PREFIX 的含义》, http://blog.csdn.net/ture010love/article/details/7663008
《The Intel 8086 / 8088/ 80186 / 80286 / 80386 / 80486 Instruction Set》: http://zsmith.co/intel.html