diff --git a/ChangeLog b/ChangeLog index 6e2f5bd02f..cc7b911ee2 100644 --- a/ChangeLog +++ b/ChangeLog @@ -4108,3 +4108,8 @@ const char **. * sched/pthread* and include/nuttx/sched: Move pthread join data and pthread key calculation data into the "task group" structure. + * sched/atexit.c, on_exit.c, task_exithook.c and include/nuttx/sched.h: + Move atexit and on_exit data structure to task group. These + callbacks are only issued now when the final member of the task + group exits. + diff --git a/TODO b/TODO index a0f0aa1bb5..f3feb6fb0a 100644 --- a/TODO +++ b/TODO @@ -7,7 +7,7 @@ standards, things that could be improved, and ideas for enhancements. nuttx/ (11) Task/Scheduler (sched/) - (2) Memory Managment (mm/) + (1) Memory Managment (mm/) (3) Signals (sched/, arch/) (2) pthreads (sched/) (2) C++ Support @@ -186,9 +186,9 @@ o Task/Scheduler (sched/) posix_spawn(). However, posix_spawn uses the non-standard, internal NuttX interface task_reparent() to replace the childs parent task with the caller of posix_spawn(). That cannot be done with vfork() - because we don't know what vfor() is going to do. + because we don't know what vfork() is going to do. - Any solution to this is either very difficult or impossible with + Any solution to this is either very difficult or impossible without an MMU. Status: Open Priority: Low (it might as well be low since it isn't going to be fixed). @@ -200,7 +200,8 @@ o Task/Scheduler (sched/) to make this change: Just move the pterrno field from _TCB to struct task_group_s. However, I am still not sure if this should be done or not. - Status: Open + Status: Closed. The existing solution is better (although its + incompatibilities could show up in porting some code). Priority: Low o Memory Managment (mm/) @@ -269,19 +270,6 @@ o Memory Managment (mm/) Priority: Medium/Low, a good feature to prevent memory leaks but would have negative impact on memory usage and code size. - Title: CONTAINER ALLOCATOR - Description: There are several places where the logic requires allocation of - a tiny structure that just contains pointers to other things or - small amounts of data that needs to be bundled together. There - are examples net/net_poll.c and numerous other places. - - I am wondering if it would not be good create a pool of generic - containers (say void *[4]). There re-use these when we need - small containers. The code in sched/task_childstatus.c might - be generalized for this purpose. - Status: Open - Priority: Very low (I am not even sure that this is a good idea yet). - o Signals (sched/, arch/) ^^^^^^^^^^^^^^^^^^^^^^^ @@ -289,7 +277,7 @@ o Signals (sched/, arch/) Description: 'Standard' signals and signal actions are not supported. (e.g., SIGINT, SIGSEGV, etc). - Update: SIG_CHLD is support if configured. + Update: SIG_CHLD is supported if so configured. Status: Open. No changes are planned. Priority: Low, required by standards but not so critical for an embedded system. diff --git a/include/nuttx/sched.h b/include/nuttx/sched.h index 75c478220f..456135962a 100644 --- a/include/nuttx/sched.h +++ b/include/nuttx/sched.h @@ -64,7 +64,7 @@ #undef HAVE_TASK_GROUP #undef HAVE_GROUP_MEMBERS -/* We need a group an group members if we are supportint the parent/child +/* We need a group an group members if we are supporting the parent/child * relationship. */ @@ -77,13 +77,17 @@ */ #else -# if !defined(CONFIG_DISABLE_ENVIRON) +# if !defined(CONFIG_DISABLE_ENVIRON) /* Environment variales */ # define HAVE_TASK_GROUP 1 -# elif CONFIG_NFILE_DESCRIPTORS > 0 +# elif defined(CONFIG_SCHED_ATEXIT) /* Group atexit() function */ # define HAVE_TASK_GROUP 1 -# elif CONFIG_NFILE_STREAMS > 0 +# elif defined(CONFIG_SCHED_ONEXIT) /* Group on_exit() function */ # define HAVE_TASK_GROUP 1 -# elif CONFIG_NSOCKET_DESCRIPTORS > 0 +# elif CONFIG_NFILE_DESCRIPTORS > 0 /* File descriptors */ +# define HAVE_TASK_GROUP 1 +# elif CONFIG_NFILE_STREAMS > 0 /* Standard C buffered I/O */ +# define HAVE_TASK_GROUP 1 +# elif CONFIG_NSOCKET_DESCRIPTORS > 0 /* Sockets */ # define HAVE_TASK_GROUP 1 # endif #endif @@ -296,6 +300,26 @@ struct task_group_s FAR pid_t *tg_members; /* Members of the group */ #endif + /* atexit/on_exit support ****************************************************/ + +#if defined(CONFIG_SCHED_ATEXIT) && !defined(CONFIG_SCHED_ONEXIT) +# if defined(CONFIG_SCHED_ATEXIT_MAX) && CONFIG_SCHED_ATEXIT_MAX > 1 + atexitfunc_t tg_atexitfunc[CONFIG_SCHED_ATEXIT_MAX]; +# else + atexitfunc_t tg_atexitfunc; /* Called when exit is called. */ +# endif +#endif + +#ifdef CONFIG_SCHED_ONEXIT +# if defined(CONFIG_SCHED_ONEXIT_MAX) && CONFIG_SCHED_ONEXIT_MAX > 1 + onexitfunc_t tg_onexitfunc[CONFIG_SCHED_ONEXIT_MAX]; + FAR void *tg_onexitarg[CONFIG_SCHED_ONEXIT_MAX]; +# else + onexitfunc_t tg_onexitfunc; /* Called when exit is called. */ + FAR void *tg_onexitarg; /* The argument passed to the function */ +# endif +#endif + /* Child exit status **********************************************************/ #if defined(CONFIG_SCHED_HAVE_PARENT) && defined(CONFIG_SCHED_CHILD_STATUS) @@ -384,24 +408,6 @@ struct _TCB FAR void *starthookarg; /* The argument passed to the function */ #endif -#if defined(CONFIG_SCHED_ATEXIT) && !defined(CONFIG_SCHED_ONEXIT) -# if defined(CONFIG_SCHED_ATEXIT_MAX) && CONFIG_SCHED_ATEXIT_MAX > 1 - atexitfunc_t atexitfunc[CONFIG_SCHED_ATEXIT_MAX]; -# else - atexitfunc_t atexitfunc; /* Called when exit is called. */ -# endif -#endif - -#ifdef CONFIG_SCHED_ONEXIT -# if defined(CONFIG_SCHED_ONEXIT_MAX) && CONFIG_SCHED_ONEXIT_MAX > 1 - onexitfunc_t onexitfunc[CONFIG_SCHED_ONEXIT_MAX]; - FAR void *onexitarg[CONFIG_SCHED_ONEXIT_MAX]; -# else - onexitfunc_t onexitfunc; /* Called when exit is called. */ - FAR void *onexitarg; /* The argument passed to the function */ -# endif -#endif - #if defined(CONFIG_SCHED_WAITPID) && !defined(CONFIG_SCHED_HAVE_PARENT) sem_t exitsem; /* Support for waitpid */ int *stat_loc; /* Location to return exit status */ diff --git a/sched/atexit.c b/sched/atexit.c index b0559b01be..f60598884d 100644 --- a/sched/atexit.c +++ b/sched/atexit.c @@ -1,7 +1,7 @@ /**************************************************************************** * sched/atexit.c * - * Copyright (C) 2007, 2009, 2011-2012 Gregory Nutt. All rights reserved. + * Copyright (C) 2007, 2009, 2011-2013 Gregory Nutt. All rights reserved. * Author: Gregory Nutt * * Redistribution and use in source and binary forms, with or without @@ -117,12 +117,15 @@ int atexit(void (*func)(void)) * can handle a callback function that recieves more parameters than it expects). */ - return on_exit(onexitfunc_t func, NULL); + return on_exit((onexitfunc_t)func, NULL); #elif defined(CONFIG_SCHED_ATEXIT_MAX) && CONFIG_SCHED_ATEXIT_MAX > 1 - _TCB *tcb = (_TCB*)g_readytorun.head; - int index; - int ret = ERROR; + FAR _TCB *tcb = (FAR _TCB*)g_readytorun.head; + FAR struct task_group_s *group = tcb->group; + int index; + int ret = ERROR; + + DEBUGASSERT(group); /* The following must be atomic */ @@ -139,9 +142,9 @@ int atexit(void (*func)(void)) available = -1; for (index = 0; index < CONFIG_SCHED_ATEXIT_MAX; index++) { - if (!tcb->atexitfunc[index]) + if (!group->tg_atexitfunc[index]) { - tcb->atexitfunc[index] = func; + group->tg_atexitfunc[index] = func; ret = OK; break; } @@ -152,15 +155,18 @@ int atexit(void (*func)(void)) return ret; #else - _TCB *tcb = (_TCB*)g_readytorun.head; - int ret = ERROR; + FAR _TCB *tcb = (FAR _TCB*)g_readytorun.head; + FAR struct task_group_s *group = tcb->group; + int ret = ERROR; + + DEBUGASSERT(group); /* The following must be atomic */ sched_lock(); - if (func && !tcb->atexitfunc) + if (func && !group->tg_atexitfunc) { - tcb->atexitfunc = func; + group->tg_atexitfunc = func; ret = OK; } diff --git a/sched/group_leave.c b/sched/group_leave.c index 1373a25a29..e56952a7b4 100644 --- a/sched/group_leave.c +++ b/sched/group_leave.c @@ -48,8 +48,9 @@ #include #include -#include "group_internal.h" #include "env_internal.h" +#include "pthread_internal.h" +#include "group_internal.h" #ifdef HAVE_TASK_GROUP diff --git a/sched/on_exit.c b/sched/on_exit.c index 19a4f91960..89c4a2d57a 100644 --- a/sched/on_exit.c +++ b/sched/on_exit.c @@ -1,7 +1,7 @@ /**************************************************************************** * sched/on_exit.c * - * Copyright (C) 2012 Gregory Nutt. All rights reserved. + * Copyright (C) 2012-2013 Gregory Nutt. All rights reserved. * Author: Gregory Nutt * * Redistribution and use in source and binary forms, with or without @@ -115,10 +115,13 @@ int on_exit(CODE void (*func)(int, FAR void *), FAR void *arg) { #if defined(CONFIG_SCHED_ONEXIT_MAX) && CONFIG_SCHED_ONEXIT_MAX > 1 - _TCB *tcb = (_TCB*)g_readytorun.head; + FAR _TCB *tcb = (FAR _TCB*)g_readytorun.head; + FAR struct task_group_s *group = tcb->group; int index; int ret = ENOSPC; + DEBUGASSERT(group); + /* The following must be atomic */ if (func) @@ -133,10 +136,10 @@ int on_exit(CODE void (*func)(int, FAR void *), FAR void *arg) for (index = 0; index < CONFIG_SCHED_ONEXIT_MAX; index++) { - if (!tcb->onexitfunc[index]) + if (!group->tg_onexitfunc[index]) { - tcb->onexitfunc[index] = func; - tcb->onexitarg[index] = arg; + group->tg_onexitfunc[index] = func; + group->tg_onexitarg[index] = arg; ret = OK; break; } @@ -147,16 +150,19 @@ int on_exit(CODE void (*func)(int, FAR void *), FAR void *arg) return ret; #else - _TCB *tcb = (_TCB*)g_readytorun.head; + FAR _TCB *tcb = (FAR _TCB*)g_readytorun.head; + FAR struct task_group_s *group = tcb->group; int ret = ENOSPC; + DEBUGASSERT(group); + /* The following must be atomic */ sched_lock(); - if (func && !tcb->onexitfunc) + if (func && !group->tg_onexitfunc) { - tcb->onexitfunc = func; - tcb->onexitarg = arg; + group->tg_onexitfunc = func; + group->tg_onexitarg = arg; ret = OK; } diff --git a/sched/task_exithook.c b/sched/task_exithook.c index 889df25e04..20162c0d63 100644 --- a/sched/task_exithook.c +++ b/sched/task_exithook.c @@ -87,41 +87,50 @@ #if defined(CONFIG_SCHED_ATEXIT) && !defined(CONFIG_SCHED_ONEXIT) static inline void task_atexit(FAR _TCB *tcb) { -#if defined(CONFIG_SCHED_ATEXIT_MAX) && CONFIG_SCHED_ATEXIT_MAX > 1 - int index; + FAR struct task_group_s *group = tcb->group; - /* Call each atexit function in reverse order of registration atexit() - * functions are registered from lower to higher arry indices; they must - * be called in the reverse order of registration when task exists, i.e., - * from higher to lower indices. + /* Make sure that we have not already left the group. Only the final + * exitting thread in the task group should trigger the atexit() + * callbacks. */ - for (index = CONFIG_SCHED_ATEXIT_MAX-1; index >= 0; index--) + if (group && group->tg_nmembers == 1) { - if (tcb->atexitfunc[index]) +#if defined(CONFIG_SCHED_ATEXIT_MAX) && CONFIG_SCHED_ATEXIT_MAX > 1 + int index; + + /* Call each atexit function in reverse order of registration atexit() + * functions are registered from lower to higher arry indices; they + * must be called in the reverse order of registration when the task + * group exits, i.e., from higher to lower indices. + */ + + for (index = CONFIG_SCHED_ATEXIT_MAX-1; index >= 0; index--) + { + if (group->tg_atexitfunc[index]) + { + /* Call the atexit function */ + + (*group->tg_atexitfunc[index])(); + + /* Nullify the atexit function to prevent its reuse. */ + + group->tg_atexitfunc[index] = NULL; + } + } +#else + if (group->tg_atexitfunc) { /* Call the atexit function */ - (*tcb->atexitfunc[index])(); + (*group->tg_atexitfunc)(); /* Nullify the atexit function to prevent its reuse. */ - tcb->atexitfunc[index] = NULL; + group->tg_atexitfunc = NULL; } - } - -#else - if (tcb->atexitfunc) - { - /* Call the atexit function */ - - (*tcb->atexitfunc)(); - - /* Nullify the atexit function to prevent its reuse. */ - - tcb->atexitfunc = NULL; - } #endif + } } #else # define task_atexit(tcb) @@ -138,40 +147,50 @@ static inline void task_atexit(FAR _TCB *tcb) #ifdef CONFIG_SCHED_ONEXIT static inline void task_onexit(FAR _TCB *tcb, int status) { -#if defined(CONFIG_SCHED_ONEXIT_MAX) && CONFIG_SCHED_ONEXIT_MAX > 1 - int index; + FAR struct task_group_s *group = tcb->group; - /* Call each on_exit function in reverse order of registration. on_exit() - * functions are registered from lower to higher arry indices; they must - * be called in the reverse order of registration when task exists, i.e., - * from higher to lower indices. + /* Make sure that we have not already left the group. Only the final + * exitting thread in the task group should trigger the atexit() + * callbacks. */ - for (index = CONFIG_SCHED_ONEXIT_MAX-1; index >= 0; index--) + if (group && group->tg_nmembers == 1) { - if (tcb->onexitfunc[index]) +#if defined(CONFIG_SCHED_ONEXIT_MAX) && CONFIG_SCHED_ONEXIT_MAX > 1 + int index; + + /* Call each on_exit function in reverse order of registration. + * on_exit() functions are registered from lower to higher arry + * indices; they must be called in the reverse order of registration + * when the task grroup exits, i.e., from higher to lower indices. + */ + + for (index = CONFIG_SCHED_ONEXIT_MAX-1; index >= 0; index--) + { + if (group->tg_onexitfunc[index]) + { + /* Call the on_exit function */ + + (*group->tg_onexitfunc[index])(status, group->tg_onexitarg[index]); + + /* Nullify the on_exit function to prevent its reuse. */ + + group->tg_onexitfunc[index] = NULL; + } + } +#else + if (group->tg_onexitfunc) { /* Call the on_exit function */ - (*tcb->onexitfunc[index])(status, tcb->onexitarg[index]); + (*group->tg_onexitfunc)(status, group->tg_onexitarg); /* Nullify the on_exit function to prevent its reuse. */ - tcb->onexitfunc[index] = NULL; + group->tg_onexitfunc = NULL; } - } -#else - if (tcb->onexitfunc) - { - /* Call the on_exit function */ - - (*tcb->onexitfunc)(status, tcb->onexitarg); - - /* Nullify the on_exit function to prevent its reuse. */ - - tcb->onexitfunc = NULL; - } #endif + } } #else # define task_onexit(tcb,status) @@ -490,7 +509,7 @@ static inline void task_exitwakeup(FAR _TCB *tcb, int status) * Description: * This function implements some of the internal logic of exit() and * task_delete(). This function performs some cleanup and other actions - * required when a task exists: + * required when a task exits: * * - All open streams are flushed and closed. * - All functions registered with atexit() and on_exit() are called, in