嵌入式30dvd30盘2linux内核a-thread presentation

1、1 1.1 线程的定义Threads can best be described as “lightweight ses”. The IX-notion of a s has been found to be inconvenient, if not inadequate for applications in distributed systems development. The needs of these applications best served by threads,1 1.1 线程的定义Threads can best be described as “lightweigh

2、t ses”. The IX-notion of a s has been found to be inconvenient, if not inadequate for applications in distributed systems development. The needs of these applications best served by threads, which generalize the notion of a s t it can ted with multiple activities. The popularity of threads has resul

3、ted in ion on UNIX systems and thread libraries are now widely available to programmers for the development of concurrent applications.1.2 ThreadsThreadscan be implementedinone of two1.User-levelThere is no kernel support for multi-threaded ses. Hence, the kernel smulti-implemented in a library of p

4、rocedures linked pplication programs. The has no knowledge of lightweight ses (threads), and therefore cannot them independently. Athreads run-time anizes the scheduling of threads. thread would block the s and therefore all threads within it if itmade a system call, so the asynchronous I/O faciliti

5、es of UNIX are used. The disadvantage of this scheme t threads within a s cannot take advantage amulti-Theadvantages（系统消耗小）Certain thread operations are significantly less costly. le, switching n threads belonging to the same s do sarilyinvolve a system call,and avethisover-（可以修改以适应特殊的应用）User-level

6、thread implemenions can be customized or changed to suit the particular application requirements. This particularly useful for real-time multimedia sing etc. Also, it tosupport manymoreuser-level ncanby default bya2. Kernel-levelThis ion allows threads within different ses to be according to a singl

7、e scheme of relative prioritizing. This ited for exploiting concurrence of Most of the current thread library ions available today user-level threads. tosupport manymoreuser-level ncanby default bya2. Kernel-levelThis ion allows threads within different ses to be according to a single scheme of rela

8、tive prioritizing. This ited for exploiting concurrence of Most of the current thread library ions available today user-level threads. There have been several research ve some formofKernel-level threads. Notableamong these aretheMach distributedOS, which combines the advantages of user-level and ker

9、nel-level threads by user-level code to provide scheduling s to the kernel thread scheduler. providing such a two-level scheduling scheme, the kernel retains control over allocation of sor time, but also allows a s to take advantage of 1.3 LinuxXavier Leroy at INRIA (Paris, France), with input from

10、Pavel Krauz, Richard Henderson others, has developed a Pthreads t implements the One-to-One , allowing it take advantage of multiple sors. It is based on the new Linux system call, clone()2 . runs on Linux 2.0 and up, el, Alpha, SPARC, m68k, and MIPS machines. One ion itsnon-standardionofsignalLinux

11、Threadsfollowsthe so-calledone-to-one:each thread uallya separate sin the kernel. The kernel scheduler takes care of scheduling the threads, just like it regular ses. The threads are created with the Linux clone() system call, which is generalizationof fork() allowing the new signalhandlersofthe s t

12、o share the memoryspace,file descriptors, Advantagesoftheone-to-oneMinimaloverhead on ensive singbout one Asimple and robust ion (the kernel scheduler does most of the hard work The maindisadvantageis moreexpensivecontextswitches onmutexandconditionustgothrough the kernel.This is mitigatedby the tco

13、ntexthe kernelarepretty2 clone unsignedlongclone_flags=regs.ebx; newsp = regs.ecx;if2 clone unsignedlongclone_flags=regs.ebx; newsp = regs.ecx;ifnewsp= 与系统调用clone功能相似的系统调用有fork,但fork事实上只是clone的功能的一部分， 系统定义的几个 clone_flags 标志：置起 CLONE_VM 标志：*SetuptheLDTdescriptorfortheclonecopy_segments(nr, tsk, NULL)

14、; 3 Linux3.1 LinuxThreadsix1003.1c 3 Linux3.1 LinuxThreadsix1003.1c 3.2 /*Argumentspassedtothreadcreationroutine/*Thetypeofthreaddescriptors struct_pthread_descr_struct （略/*Thetypeofthreadhandles.typedefstructpthread_handle_struct* struct_pthread_fastlockh_lock;/*Fastlockforsychronizeds char */*Thre

15、addescriptororNULLifinvalid/*Lowesthestackthread/*Thetypeofmessagessenttothethreadmanagerthreadstruct pthread_request enum ngtherequest/*Requestkind req_kind; union structT, constpthread_attr_t*attr;/*threadattributesvoid*(*fn)(void*); void * arg;sigset_tstruct pthread_treq_kind; union structT, cons

16、tpthread_attr_t*attr;/*threadattributesvoid*(*fn)(void*); void * arg;sigset_tstruct pthread_t free; structvoidstartfunctionargumenttostartfunction*/ signal mask */*ForREQ_FREE:identifierofthreadtofreeexitus/*ForT:thesemaphore/*Oneendofpeforsendingrequeststothethreadmanager. pthread_manager_request=-

17、/*Otherendofpeforsendingrequeststothethreadmanager. #define(2 *1024*#define(4* /*Attributesfortypedef struct sched_param size_t stackaddr_set; void* size_t 3.2.2/*Threadcreation void*(*start_routine)(void*),size_t 3.2.2/*Threadcreation void*(*start_routine)(void*),voidpthread_descrself=thread_self()

18、; struct pthread_request request;if( pthread_manager_request0)if( pthread_initialize_manager()raise( /*Wesuspendourselfandgdbwillwakeusupwhenitreadytohandleus./*Synchronizedebuggingofthethreadmanager*/ request.req_kind = REQ_DEBUG; libc_write( pthread_manager_request,(char*)&request,returnicvoidstru

19、ctsigactionsa; sigset_t mask; struct rlimit limit;/*Ifalreadydonereadytohandleus./*Synchronizedebuggingofthethreadmanager*/ request.req_kind = REQ_DEBUG; libc_write( pthread_manager_request,(char*)&request,returnicvoidstructsigactionsa; sigset_t mask; struct rlimit limit;/*Ifalreadydone(e.g.byaconst

20、ructorcalledr!),bailoutif( !=NULL) pthread_has_cas=compare_and_swap_is_available(); /* For the initial stack, east STACK_SIZE bytes of stack below the current address,and tonaSTACK_SIZEboundary. =/*PlaywiththestacksizelimittomaketnostackeverSTACK_SIZEminustwopages(onepageforthethread,andone pagetoag

21、uardpage). max_stack=STACK_SIZE-2* getpagesize(); if (limit.rlim_cur max_stack) limit.rlim_cur = max_stack; /*Updatethedescriptorfortheinitialthread. _= sing,tfor/*Ifwehavelthread_selfhreadnow. #ifdefINIT_THREAD_SELF(& pthread_initial_thread,/*Theerrno/h_errnovariableofthehreadaretheglobal pthread_i

22、nitial_thre #ifdef SIGRTMIN_errnop=_h_errnop=/*Allocatethesignals pthread_sig_restart= /*Theerrno/h_errnovariableofthehreadaretheglobal pthread_initial_thre #ifdef SIGRTMIN_errnop=_h_errnop=/*Allocatethesignals pthread_sig_restart= libc_allocate_rtsig pthread_sig_cancel= libc_allocate_rtsig pthread_

23、sig_debug= libc_allocate_rtsig(1); if ( pthread_sig_restart 0 | pthread_sig_cancel 0 pthread_sig_debug 0) sa.sa_flags= sigaction( pthread_sig_debug,&sa,/*Initially,block pthread_sig_restart.Willbeunblockedondemand.*/ sigaddset(&mask, pthread_sig_restart); sigprocmask(SIG_BLOCK,&mask,NULL);/*Register

24、anexitfunctionto sigaction( pthread_sig_debug,&sa,/*Initially,block pthread_sig_restart.Willbeunblockedondemand.*/ sigaddset(&mask, pthread_sig_restart); sigprocmask(SIG_BLOCK,&mask,NULL);/*Registeranexitfunctiontokillallother threads.tearlytuser-registeredatexitfunctionsarebefore s.s, pthread_handl

25、e_create(pthread_t*thread,constpthread_attr_t*attr, void * (*start_routine)(void *), void *arg,sigset_t*);pthread_descrnew_thread; pthread_t new_thread_id; char *guardaddr = NULL; size_t guardsize = 0;pagesize = checkwhetherwehavetochange thepolicyandifyes,we cando this.Normally this should be done

26、by examining the returnvalueofthe sched_setschedulercallinpthread_start_thread but this is hard to implement.we cando this.Normally this should be done by examining the returnvalueofthe sched_setschedulercallinpthread_start_thread but this is hard to implement.if(attr!=NULL&attr- schedpolicy!=SCHED_

27、OTHER&geteuid()!=0) return EPERM;/*Findafreesegmentforthethread,andallocateastackifneededfor (sseg = 2; ; sseg+)if(sseg=returnif( pthread_handlessseg.h_descr!=NULL) &guardaddr, &guardsize) = 0) /* Allocate new thread identifier */分配新线程的标识符 new_thread_id = sseg + pthread_threads_counter;/*Initializet

28、hethreaddescriptor*/初始化新线程描述符 new_thread-p_nextwaiting = NULL;new_thread-p_priority = 0;new_thread-p_lock=&( pthread_handlessseg.h_lock); new_thread-p_signal = 0;new_thread-p_signal_jmp = new_thread-p_cancel_jmp=NULL; new_thread-p_terminated = 0;new_thread-p_detached=attr=NULL?0:attr- detachs new_th

29、read-p_exited = 0;new_thread-p_retval = NULL; new_thread-p_joining= e = PTHREAD_CANCEL_ENABLE; new_thread-p_canceled = 0;new_thread-p_errno = 0; new_thread-p_in_sighandler=NULL; new_thread-p_sigwaiting = 0; new_thread-p_guardaddr=guardaddr; new_thread-p_guardsize = guardsize;new_thread-p_userstack=a

30、ttr!=NULL&attr- stackaddr_set; memset (new_thread-p_specific, 0,PTHREAD_KEY_1STLEVEL_SIZE*sizeofnew_thread-p_in_sighandler=NULL; new_thread-p_sigwaiting = 0; new_thread-p_guardaddr=guardaddr; new_thread-p_guardsize = guardsize;new_thread-p_userstack=attr!=NULL&attr- stackaddr_set; memset (new_thread

31、-p_specific, 0,PTHREAD_KEY_1STLEVEL_SIZE*sizeof(new_thread-p_specific0); new_thread-p_self = new_thread;new_thread-p_nr=/*Initializethethreadhandle pthread_init_lock(& /*Determineschedulingparametersforthethreadnew_thread-p_start_args.schedpolicy=-1; if (attr != NULL) switch(attr- inheritsched)case

32、new_thread-p_start_args.schedpolicy=attr- memcpy(&new_thread-p_start_args.schedparam,&attr- schedparam, sizeof (struct sched_param);case/*schedpolicydoesntneedtobeset,onlygetpriority /*Finishsettingupargumentstopthread_start_thread new_thread-p_start_args.arg = arg; /*Dothecloning*/= clone(pthread_s

33、tart_thread,(void*) /*Checkifcloningsucceededif=-1)/*Freethestackifweallocateditif(attr=NULL|!attr- /*Freethestackifweallocateditif(attr=NULL|!attr- pthread_handles_num-; return errno;/*Insertnewthreadindoublylinkedlistofactivethreadsnew_thread-p_prevlive= new_thread-p_nextlive= pthread_main_thread-

34、 /*fieldofthenewthread,incasewegettherebeforechildstarts.*/ /*Wereallset=;return3.4 Pthread_initial_thread (CS-2*STACK_SIZE)&(STACK_SIZE-Current s pthread_descrdefault_new_thread, pthread_descr*out_new_thread, char*out_new_thread_bottom, char * out_guardaddr,size_t*pthread_descr new_thread; char*new

35、_thread_bottom; char * guardaddr;size_tstacksize,if(attr!=NULL&s pthread_descrdefault_new_thread, pthread_descr*out_new_thread, char*out_new_thread_bottom, char * out_guardaddr,size_t*pthread_descr new_thread; char*new_thread_bottom; char * guardaddr;size_tstacksize,if(attr!=NULL&attr- /*The new_thr

36、ead(pthread_descr) (long)(attr- stackaddr) & -sizeof(void *) - 1; new_thread_bottom = (char *) attr- stackaddr - attr- stacksize; guardaddr = NULL;guardsize= /*Allocatespaceforstackandthreaddescriptoratdefaultaddressnew_threaddefault_new_thread;new_thread_bottom=(char*)new_thread-INITIAL_STACK_SIZE,

37、 PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED | MAP_GROWSDOWN, -1, 0) = MAP_FAILED)/*Badluck,thissegmentisalreadymapped.*/ return -1;/*We manage togeta stack. Nowseewhetherweneedaandallocateitifsary. ttheattributes(stack_size=STACK_SIZEsizeguardsize=pagesize)donotneeda

38、guardpage,since the RLIMIT_STACK soft limit prevents stacksfromooneanother.if (attr = NULL |attr- guardsize=0(attr- guardsize=pagesize attr- stacksize=STACK_SIZE-/*Wedontneedaguardpage.*/ guardaddr = NULL;guardsize=attr- stacksize=STACK_SIZE-/*Wedontneedaguardpage.*/ guardaddr = NULL;guardsize=/*Put

39、a ageatthebottomofthestackstacksize=roundup(attr- stacksize,pagesize); if (stacksize = STACK_SIZE - pagesize)stacksize = STACK_SIZE - pagesize; guardaddr=(void*)new_thread-stacksize; guardsize = attr- guardsize; =/*Wedontmakethisanerror. */ guardaddr = NULL;guardsize=*out_new_thread= *out_guardaddr=

40、return 0;缺省保护页（1 保护页（可选3.5 4 问题及发4.1 LinuxThreads3.5 4 问题及发4.1 LinuxThreadsclone() 的 On recent kernels (late 2.1 kernels and ing 2.2 kernels), n 32 signals are he form of realtime realtime signals for When run on one of those kernels, LinuxThreads uses ernal operation, thus leaving SIGUSR1 and SIGUS

41、R2 free for user code. parentcauseproblemswith/procandsignalsgotoD 现在，实现时使用了两个信号SIGUSR1andSIGUSR2 用512。 1- Synchronous signals (generated by the thread execution, e.g. SIGFPE) are delivered to traisedthem.2- Afatal asynchronous signal terminates all he s. (OK. The thread whenathreaddiesonasignalandk

42、illsallotherthreadswiththe same3-Anasynchronoussignalwillbedeliveredtooneofthethreadsoftheprogramwhichdoesblockthesignal(itisunspecifiedwhich).(No,thesignalisdeliveredtothethreaditsbeensentwhich based on remainsd of the thread. t thread is currently blocking the signal, the 4-Thesignalwillbedelivere

43、dtoatmostonethread.(OK,exceptforsignalsgeneratedfromterminalorsenttothesgroup,whichwillbedeliveredtoall-The current ion ofwhich based on remainsd of the thread. t thread is currently blocking the signal, the 4-Thesignalwillbedeliveredtoatmostonethread.(OK,exceptforsignalsgeneratedfromterminalorsenttothesgroup,whichwillbedeliveredtoall-The current