[PATCH 2/2] posix: Reimplement POSIX Key manager to use a red-black tree.
Ashi
ashi08104 at gmail.com
Sat Feb 23 05:01:57 UTC 2013
On Fri, Feb 22, 2013 at 10:38 PM, Gedare Bloom <gedare at rtems.org> wrote:
> I pulled this from github using the latest pull request you sent.
> Maybe you posted an updated patch somewhere? Check to make sure there
> are no other major issues to deal with and let me know if there are
> any updates for me to pull. I will be looking to commit this work
> soon. I will figure out the confdefs.h issue myself.
>
I've checked there no other major issues to deal with. And the updated
patches(without fixing confdefs.h issue) are attached.
>
> On Thu, Feb 21, 2013 at 10:19 PM, Ashi <ashi08104 at gmail.com> wrote:
> > Gedare, I feel this patch is a little old than my actually work, because
> > some line I have removed before(like the "TODO" line), I haven't figured
> out
> > which commit this patch is based.
> > Anyway, I've updated according your comments.
> >
> >
> > On Sat, Feb 9, 2013 at 9:15 PM, Gedare Bloom <gedare at rtems.org> wrote:
> >>
> >> On Thu, Feb 7, 2013 at 9:20 PM, Gedare Bloom <gedare at rtems.org> wrote:
> >> > From: Zhongwei Yao <ashi08104 at gmail.com>
> >> >
> >> > The POSIX Key manager is reimplemented with a red-black tree to store
> >> > the keys and values. This code was contributed as part of GSOC 2012.
> >> > ---
> >> > cpukit/posix/include/rtems/posix/key.h | 56
> >> > ++++++++++++++++-----
> >> > cpukit/posix/include/rtems/posix/threadsup.h | 4 ++
> >> > cpukit/posix/inline/rtems/posix/key.inl | 2 +-
> >> > cpukit/posix/src/key.c | 52
> >> > +++++++++++++++++++-
> >> > cpukit/posix/src/keycreate.c | 53
> >> > +-------------------
> >> > cpukit/posix/src/keydelete.c | 4 +-
> >> > cpukit/posix/src/keyfreememory.c | 35 ++++++++++++-
> >> > cpukit/posix/src/keygetspecific.c | 21 +++++---
> >> > cpukit/posix/src/keyrundestructors.c | 68
> >> > +++++++++++++++-----------
> >> > cpukit/posix/src/keysetspecific.c | 32 +++++++++---
> >> > cpukit/posix/src/pthread.c | 3 +
> >> > cpukit/sapi/src/posixapi.c | 2 +-
> >> > 12 files changed, 216 insertions(+), 116 deletions(-)
> >> >
> >> > diff --git a/cpukit/posix/include/rtems/posix/key.h
> >> > b/cpukit/posix/include/rtems/posix/key.h
> >> > index 0bb1dbe..80e534b 100644
> >> > --- a/cpukit/posix/include/rtems/posix/key.h
> >> > +++ b/cpukit/posix/include/rtems/posix/key.h
> >> > @@ -8,6 +8,7 @@
> >> > */
> >> >
> >> > /*
> >> > + * Copyright (c) 2012 Zhongwei Yao.
> >> > * COPYRIGHT (c) 1989-2011.
> >> > * On-Line Applications Research Corporation (OAR).
> >> > *
> >> > @@ -20,6 +21,8 @@
> >> > #define _RTEMS_POSIX_KEY_H
> >> >
> >> > #include <rtems/score/object.h>
> >> > +#include <rtems/score/rbtree.h>
> >> > +#include <rtems/score/chain.h>
> >> >
> >> > /**
> >> > * @defgroup POSIX_KEY POSIX Key
> >> > @@ -34,32 +37,57 @@ extern "C" {
> >> > #endif
> >> >
> >> > /**
> >> > - * This is the data Structure used to manage a POSIX key.
> >> > - *
> >> > - * NOTE: The Values is a table indexed by the index portion of the
> >> > - * ID of the currently executing thread.
> >> > + * @brief The rbtree node used to manage a POSIX key and value.
> >> > */
> >> > typedef struct {
> >> > - /** This field is the Object control structure. */
> >> > - Objects_Control Object;
> >> > - /** This field points to the optional destructor method. */
> >> > - void (*destructor)( void * );
> >> > - /** This field points to the values per thread. */
> >> > - void **Values[ OBJECTS_APIS_LAST + 1 ];
> >> > -} POSIX_Keys_Control;
> >> > + /** This field is the chain node structure. */
> >> > + Chain_Node ch_node;
> >> > + /** This field is the rbtree node structure. */
> >> > + RBTree_Node rb_node;
> >> > + /** This field is the POSIX key used as an rbtree key */
> >> > + pthread_key_t key;
> >> > + /** This field is the Thread id also used as an rbtree key */
> >> > + Objects_Id thread_id;
> >> > + /** This field points to the POSIX key value of specific thread */
> >> > + void *value;
> >> > + } POSIX_Keys_Rbtree_node;
> >> >
> >> > /**
> >> > - * The following defines the information control block used to manage
> >> > - * this class of objects.
> >> > + * @brief The data structure used to manage a POSIX key.
> >> > + */
> >> > +typedef struct {
> >> > + /** This field is the Object control structure. */
> >> > + Objects_Control object;
> >> > + /** This field is the data destructor. */
> >> > + void (*destructor) (void *);
> >> > + } POSIX_Keys_Control;
> >> > +
> >> > +/**
> >> > + * @brief The information control block used to manage this class of
> >> > objects.
> >> > */
> >> > POSIX_EXTERN Objects_Information _POSIX_Keys_Information;
> >> >
> >> > /**
> >> > + * @brief The rbtree control block used to manage all key values
> >> > + */
> >> > +POSIX_EXTERN RBTree_Control _POSIX_Keys_Rbtree;
> >> > +
> >> > +/**
> >> > * @brief POSIX keys manager initialization.
> >> > *
> >> > * This routine performs the initialization necessary for this
> manager.
> >> > */
> >> > -void _POSIX_Key_Manager_initialization(void);
> >> > +void _POSIX_Keys_Manager_initialization(void);
> >> > +
> >> > +/**
> >> > + * @brief POSIX keys Red-Black tree node comparison.
> >> > + *
> >> > + * This routine compares the rbtree node
> >> > + */
> >> > +int _POSIX_Keys_Rbtree_compare_function(
> >> > + const RBTree_Node *node1,
> >> > + const RBTree_Node *node2
> >> > +);
> >> >
> >> > /**
> >> > * @brief Create thread-specific data POSIX key.
> >> > diff --git a/cpukit/posix/include/rtems/posix/threadsup.h
> >> > b/cpukit/posix/include/rtems/posix/threadsup.h
> >> > index 80f64dc..f47a7d4 100644
> >> > --- a/cpukit/posix/include/rtems/posix/threadsup.h
> >> > +++ b/cpukit/posix/include/rtems/posix/threadsup.h
> >> > @@ -21,6 +21,7 @@
> >> > #include <sys/signal.h>
> >> > #include <rtems/score/coresem.h>
> >> > #include <rtems/score/tqdata.h>
> >> > +#include <rtems/score/chain.h>
> >> >
> >> > /**
> >> > * @defgroup POSIX_THREAD POSIX Thread API Extension
> >> > @@ -78,6 +79,9 @@ typedef struct {
> >> > int cancelation_requested;
> >> > /** This is the set of cancelation handlers. */
> >> > Chain_Control Cancellation_Handlers;
> >> > +
> >> > + /** This is the thread key value chain's control */
> >> > + Chain_Control the_chain;
> >> >
> >> Perhaps a more descriptive variable name? Also the name the_chain
> >> should be in initial capitals because it is an aggregate structure.
> >> Finally, we should make a note of why this chain exists. it is
> >>
> >> needed to iterate through the keys of a given thread when destroying
> >> the thread.
> >>
> > Yeah, that's the point. I've add a little more detail in comment.
> >>
> >> > } POSIX_API_Control;
> >> >
> >> > diff --git a/cpukit/posix/inline/rtems/posix/key.inl
> >> > b/cpukit/posix/inline/rtems/posix/key.inl
> >> > index ce5601b..9469580 100644
> >> > --- a/cpukit/posix/inline/rtems/posix/key.inl
> >> > +++ b/cpukit/posix/inline/rtems/posix/key.inl
> >> > @@ -45,7 +45,7 @@ RTEMS_INLINE_ROUTINE void _POSIX_Keys_Free (
> >> > POSIX_Keys_Control *the_key
> >> > )
> >> > {
> >> > - _Objects_Free( &_POSIX_Keys_Information, &the_key->Object );
> >> > + _Objects_Free( &_POSIX_Keys_Information, &the_key->object );
> >> > }
> >> >
> >> > /**
> >> > diff --git a/cpukit/posix/src/key.c b/cpukit/posix/src/key.c
> >> > index 6eace26..b0a1f86 100644
> >> > --- a/cpukit/posix/src/key.c
> >> > +++ b/cpukit/posix/src/key.c
> >> > @@ -29,6 +29,49 @@
> >> > #include <rtems/score/thread.h>
> >> > #include <rtems/score/wkspace.h>
> >> > #include <rtems/posix/key.h>
> >> > +#include <rtems/score/rbtree.h>
> >> > +
> >> > +/*
> >> > + * _POSIX_Keys_Rbtree_compare_function
> >> > + *
> >> > + * DESCRIPTION:
> >> > + * This routine compares the rbtree node
> >> > + * by comparing POSIX key first and comparing thread id second.
> >> > + * And if either of the input nodes's thread_id member is 0, then
> >> > + * it will only compare the pthread_key_t member. That is when we
> >> > + * pass thread_id = 0 node as a search node, the search is done only
> >> > + * by pthread_key_t.
> >> > + *
> >> > + * Input parameters: two rbtree node
> >> > + *
> >> > + * Output parameters: return positive if first node
> >> > + * has higher key than second, negative if lower, 0 if equal,
> >> > + * and for all the thread id is unique, then return 0 is impossible
> >> > + */
> >> > +
> >> > +int _POSIX_Keys_Rbtree_compare_function(
> >> > + const RBTree_Node *node1,
> >> > + const RBTree_Node *node2
> >> > +)
> >> > +{
> >> > + pthread_key_t key1 = _RBTree_Container_of(node1,
> >> > POSIX_Keys_Rbtree_node, rb_node)->key;
> >> > + pthread_key_t key2 = _RBTree_Container_of(node2,
> >> > POSIX_Keys_Rbtree_node, rb_node)->key;
> >> > +
> >> > + Objects_Id thread_id1 = _RBTree_Container_of(node1,
> >> > POSIX_Keys_Rbtree_node, rb_node)->thread_id;
> >> > + Objects_Id thread_id2 = _RBTree_Container_of(node2,
> >> > POSIX_Keys_Rbtree_node, rb_node)->thread_id;
> >> > +
> >> These lines are too long (>80 char) and need to be broken. I think
> >> there are a few others in this patch too.
> >
> > I remember there is a format check script, which I also have run it my
> > code...
> >>
> >>
> >> > + int diff = key1 - key2;
> >> > + if ( diff )
> >> > + return diff;
> >> > + /**
> >> > + * if thread_id1 or thread_id2 equals to 0, only key1 and key2 is
> >> > valued.
> >> > + * it enables us search node only by pthread_key_t type key.
> >> > + */
> >> > + if ( thread_id1 && thread_id2 )
> >> > + return thread_id1 - thread_id2;
> >> > + return 0;
> >> > +}
> >> > +
> >> >
> >> > /*
> >> > * _POSIX_Key_Manager_initialization
> >> > @@ -42,7 +85,7 @@
> >> > * Output parameters: NONE
> >> > */
> >> >
> >> > -void _POSIX_Key_Manager_initialization(void)
> >> > +void _POSIX_Keys_Manager_initialization(void)
> >> > {
> >> > _Objects_Initialize_information(
> >> > &_POSIX_Keys_Information, /* object information table */
> >> > @@ -60,4 +103,11 @@ void _POSIX_Key_Manager_initialization(void)
> >> > NULL /* Proxy extraction support callout
> */
> >> > #endif
> >> > );
> >> > +
> >> > + _RBTree_Initialize_empty(
> >> > + &_POSIX_Keys_Rbtree, /* the rbtree control block */
> >> > + _POSIX_Keys_Rbtree_compare_function,
> >> > + /* the rbtree compare function */
> >> > + true /* true if each rbtree node is unique */
> >> > + );
> >> I don't think the parameter comments here are necessary.
> >>
> >> > }
> >> > diff --git a/cpukit/posix/src/keycreate.c
> b/cpukit/posix/src/keycreate.c
> >> > index b41b590..84a4d9c 100644
> >> > --- a/cpukit/posix/src/keycreate.c
> >> > +++ b/cpukit/posix/src/keycreate.c
> >> > @@ -37,10 +37,6 @@ int pthread_key_create(
> >> > )
> >> > {
> >> > POSIX_Keys_Control *the_key;
> >> > - void *table;
> >> > - uint32_t the_api;
> >> > - uint32_t bytes_to_allocate;
> >> > -
> >> >
> >> > _Thread_Disable_dispatch();
> >> >
> >> > @@ -52,52 +48,9 @@ int pthread_key_create(
> >> > }
> >> >
> >> > the_key->destructor = destructor;
> >> > -
> >> > - /*
> >> > - * This is a bit more complex than one might initially expect
> >> > because
> >> > - * APIs are optional.
> >> > - *
> >> > - * NOTE: Currently RTEMS Classic API tasks are always enabled.
> >> > - */
> >> > - for ( the_api = 1; the_api <= OBJECTS_APIS_LAST; the_api++ ) {
> >> > - the_key->Values[ the_api ] = NULL;
> >> > -
> >> > - #if defined(RTEMS_DEBUG)
> >> > - /*
> >> > - * Since the removal of ITRON, this cannot occur.
> >> > - */
> >> > - if ( !_Objects_Information_table[ the_api ] )
> >> > - continue;
> >> > -
> >> > - /*
> >> > - * Currently all managers are installed if the API is
> installed.
> >> > - * This would be a horrible implementation error.
> >> > - */
> >> > - if (_Objects_Information_table[ the_api ][ 1 ] == NULL )
> >> > - _Internal_error_Occurred(
> >> > - INTERNAL_ERROR_CORE,
> >> > - true,
> >> > - INTERNAL_ERROR_IMPLEMENTATION_KEY_CREATE_INCONSISTENCY
> >> > - );
> >> > - #endif
> >> > -
> >> > - bytes_to_allocate = sizeof( void * ) *
> >> > - (_Objects_Information_table[ the_api ][ 1 ]->maximum + 1);
> >> > - table = _Workspace_Allocate( bytes_to_allocate );
> >> > - if ( !table ) {
> >> > - _POSIX_Keys_Free_memory( the_key );
> >> > -
> >> > - _POSIX_Keys_Free( the_key );
> >> > - _Thread_Enable_dispatch();
> >> > - return ENOMEM;
> >> > - }
> >> > -
> >> > - the_key->Values[ the_api ] = table;
> >> > - memset( table, '\0', bytes_to_allocate );
> >> > - }
> >> > -
> >> > - _Objects_Open_u32( &_POSIX_Keys_Information, &the_key->Object, 0 );
> >> > - *key = the_key->Object.id;
> >> > + /*problem: not clear about _Objects_Open_u32() */
> >> > + _Objects_Open_u32( &_POSIX_Keys_Information, &the_key->object, 0 );
> >> I believe you need to "open" the object for RTEMS object
> >> tracking..like reference counting. Not sure though.
> >
> > OK.
> >>
> >>
> >> > + *key = the_key->object.id;
> >> > _Thread_Enable_dispatch();
> >> > return 0;
> >> > }
> >> > diff --git a/cpukit/posix/src/keydelete.c
> b/cpukit/posix/src/keydelete.c
> >> > index 5ef6261..f45e264 100644
> >> > --- a/cpukit/posix/src/keydelete.c
> >> > +++ b/cpukit/posix/src/keydelete.c
> >> > @@ -42,9 +42,9 @@ int pthread_key_delete(
> >> > switch ( location ) {
> >> >
> >> > case OBJECTS_LOCAL:
> >> > - _Objects_Close( &_POSIX_Keys_Information, &the_key->Object );
> >> > -
> >> > _POSIX_Keys_Free_memory( the_key );
> >> > + /* problem: should it done before _POSIX_Keys_Free_memory? */
> >> > + _Objects_Close( &_POSIX_Keys_Information, &the_key->object );
> >> I think you need to be finished with the_key before closing it.
> >>
> > I see.
> >>
> >> >
> >> > /*
> >> > * NOTE: The destructor is not called and it is the
> >> > responsibility
> >> > diff --git a/cpukit/posix/src/keyfreememory.c
> >> > b/cpukit/posix/src/keyfreememory.c
> >> > index f71af4f..50f560e 100644
> >> > --- a/cpukit/posix/src/keyfreememory.c
> >> > +++ b/cpukit/posix/src/keyfreememory.c
> >> > @@ -21,14 +21,43 @@
> >> > #include <rtems/system.h>
> >> > #include <rtems/score/thread.h>
> >> > #include <rtems/score/wkspace.h>
> >> > +#include <rtems/score/rbtree.h>
> >> > #include <rtems/posix/key.h>
> >> >
> >> > void _POSIX_Keys_Free_memory(
> >> > POSIX_Keys_Control *the_key
> >> > )
> >> > {
> >> > - uint32_t the_api;
> >> > + POSIX_Keys_Rbtree_node search_node;
> >> > + POSIX_Keys_Rbtree_node *p;
> >> > + RBTree_Node *iter, *next;
> >> >
> >> > - for ( the_api = 1; the_api <= OBJECTS_APIS_LAST; the_api++ )
> >> > - _Workspace_Free( the_key->Values[ the_api ] );
> >> > + search_node.key = the_key->object.id;
> >> > + search_node.thread_id = 0;
> >> > + iter = _RBTree_Find_unprotected( &_POSIX_Keys_Rbtree,
> >> > &search_node.rb_node);
> >> > + if ( !iter )
> >> > + return;
> >> > + /**
> >> > + * find the smallest thread_id node in the rbtree.
> >> > + */
> >> > + next = _RBTree_Next_unprotected( iter, RBT_LEFT );
> >> > + p = _RBTree_Container_of( next, POSIX_Keys_Rbtree_node, rb_node );
> >> > + while ( p->key == the_key->object.id) {
> >> > + iter = next;
> >> > + next = _RBTree_Next_unprotected( iter, RBT_LEFT );
> >> > + p = _RBTree_Container_of( next, POSIX_Keys_Rbtree_node, rb_node
> );
> >> > + }
> >> > +
> >> > + /**
> >> > + * delete all nodes belongs to the_key from the rbtree and chain.
> >> > + */
> >> > + p = _RBTree_Container_of( iter, POSIX_Keys_Rbtree_node, rb_node );
> >> > + while ( p->key == the_key->object.id ) {
> >> > + next = _RBTree_Next_unprotected( iter, RBT_RIGHT );
> >> > + _RBTree_Extract_unprotected( &_POSIX_Keys_Rbtree, iter );
> >> > + _Chain_Extract_unprotected( &p->ch_node );
> >> > + _Workspace_Free( p );
> >> > + iter = next;
> >> > + p = _RBTree_Container_of( iter, POSIX_Keys_Rbtree_node, rb_node
> );
> >> > + }
> >> > }
> >> > diff --git a/cpukit/posix/src/keygetspecific.c
> >> > b/cpukit/posix/src/keygetspecific.c
> >> > index debad83..61adea6 100644
> >> > --- a/cpukit/posix/src/keygetspecific.c
> >> > +++ b/cpukit/posix/src/keygetspecific.c
> >> > @@ -26,6 +26,7 @@
> >> > #include <rtems/system.h>
> >> > #include <rtems/score/thread.h>
> >> > #include <rtems/score/wkspace.h>
> >> > +#include <rtems/score/rbtree.h>
> >> > #include <rtems/posix/key.h>
> >> >
> >> > /*
> >> > @@ -36,19 +37,23 @@ void *pthread_getspecific(
> >> > pthread_key_t key
> >> > )
> >> > {
> >> > - register POSIX_Keys_Control *the_key;
> >> > - uint32_t api;
> >> > - uint32_t index;
> >> > Objects_Locations location;
> >> > + POSIX_Keys_Rbtree_node search_node;
> >> > + RBTree_Node *p;
> >> > void *key_data;
> >> >
> >> > - the_key = _POSIX_Keys_Get( key, &location );
> >> > + _POSIX_Keys_Get( key, &location );
> >> > switch ( location ) {
> >> > -
> >> > +
> >> > case OBJECTS_LOCAL:
> >> > - api = _Objects_Get_API( _Thread_Executing->Object.id );
> >> > - index = _Objects_Get_index( _Thread_Executing->Object.id );
> >> > - key_data = (void *) the_key->Values[ api ][ index ];
> >> > + /** TODO: search the node in TCB's chain(maybe the rbtree) to
> >> > speed up the search */
> >> Is this TODO still relevant? I don't quite understand it.
> >>
> > I thought search a node in chain's runtime is O(n1) and in rbtree is
> > O(log(n2), where n2 is bigger than n1,
> > so I think there is no big difference between these two and this TODO
> can be
> > removed.
> >>
> >> > + search_node.key = key;
> >> > + search_node.thread_id = _Thread_Executing->Object.id;
> >> > + p = _RBTree_Find_unprotected( &_POSIX_Keys_Rbtree,
> >> > &search_node.rb_node);
> >> > + key_data = NULL;
> >> > + if ( p ) {
> >> > + key_data = _RBTree_Container_of( p, POSIX_Keys_Rbtree_node,
> >> > rb_node )->value;
> >> > + }
> >> > _Thread_Enable_dispatch();
> >> > return key_data;
> >> >
> >> > diff --git a/cpukit/posix/src/keyrundestructors.c
> >> > b/cpukit/posix/src/keyrundestructors.c
> >> > index 9f48888..687217c 100644
> >> > --- a/cpukit/posix/src/keyrundestructors.c
> >> > +++ b/cpukit/posix/src/keyrundestructors.c
> >> > @@ -23,7 +23,10 @@
> >> > #include <rtems/system.h>
> >> > #include <rtems/score/object.h>
> >> > #include <rtems/score/thread.h>
> >> > +#include <rtems/score/wkspace.h>
> >> > +#include <rtems/score/chain.h>
> >> > #include <rtems/posix/key.h>
> >> > +#include <rtems/posix/threadsup.h>
> >> >
> >> > /*
> >> > * _POSIX_Keys_Run_destructors
> >> > @@ -38,36 +41,43 @@ void _POSIX_Keys_Run_destructors(
> >> > Thread_Control *thread
> >> > )
> >> > {
> >> > - Objects_Maximum thread_index = _Objects_Get_index(
> thread->Object.id
> >> > );
> >> > - Objects_APIs thread_api = _Objects_Get_API( thread->Object.id );
> >> > - bool done = false;
> >> > + Chain_Control *chain;
> >> > + Chain_Node *iter, *next;
> >> > + void *value;
> >> > + void (*destructor) (void *);
> >> > + POSIX_Keys_Control *the_key;
> >> > + Objects_Locations location;
> >> >
> >> > - /*
> >> > - * The standard allows one to avoid a potential infinite loop and
> >> > limit the
> >> > - * number of iterations. An infinite loop may happen if
> destructors
> >> > set
> >> > - * thread specific data. This can be considered dubious.
> >> > - *
> >> > - * Reference: 17.1.1.2 P1003.1c/Draft 10, p. 163, line 99.
> >> > - */
> >> > - while ( !done ) {
> >> > - Objects_Maximum index = 0;
> >> > - Objects_Maximum max = _POSIX_Keys_Information.maximum;
> >> > + _Thread_Disable_dispatch();
> >> > +
> >> > + chain = &((POSIX_API_Control *)thread->API_Extensions[
> >> > THREAD_API_POSIX ])->the_chain;
> >> > + iter = _Chain_First( chain );
> >> > + while ( !_Chain_Is_tail( chain, iter ) ) {
> >> > + next = _Chain_Next( iter );
> >> > +
> >> > + /**
> >> > + * remove key from rbtree and chain.
> >> > + * here Chain_Node *iter can be convert to POSIX_Keys_Rbtree_node
> >> > *,
> >> > + * because Chain_Node is the first member of
> POSIX_Keys_Rbtree_node
> >> > structure.
> >> > + */
> >> > + _RBTree_Extract_unprotected( &_POSIX_Keys_Rbtree,
> >> > &((POSIX_Keys_Rbtree_node *)iter)->rb_node );
> >> > + _Chain_Extract_unprotected( iter );
> >> >
> >> > - done = true;
> >> > -
> >> > - for ( index = 1 ; index <= max ; ++index ) {
> >> > - POSIX_Keys_Control *key = (POSIX_Keys_Control *)
> >> > - _POSIX_Keys_Information.local_table [ index ];
> >> > -
> >> > - if ( key != NULL && key->destructor != NULL ) {
> >> > - void *value = key->Values [ thread_api ][ thread_index ];
> >> > -
> >> > - if ( value != NULL ) {
> >> > - key->Values [ thread_api ][ thread_index ] = NULL;
> >> > - (*key->destructor)( value );
> >> > - done = false;
> >> > - }
> >> > - }
> >> > - }
> >> > + /**
> >> > + * run key value's destructor if destructor and value are both
> >> > non-null.
> >> > + */
> >> > + the_key = _POSIX_Keys_Get( ((POSIX_Keys_Rbtree_node *)iter)->key,
> >> > &location);
> >> > + destructor = the_key->destructor;
> >> > + value = ((POSIX_Keys_Rbtree_node *)iter)->value;
> >> > + if ( destructor != NULL && value != NULL )
> >> > + (*destructor)( value );
> >> > + /**
> >> > + * disable dispatch is nested here
> >> > + */
> >> > + _Thread_Enable_dispatch();
> >> > +
> >> > + _Workspace_Free( (POSIX_Keys_Rbtree_node *)iter );
> >> > + iter = next;
> >> > }
> >> > + _Thread_Enable_dispatch();
> >> > }
> >> > diff --git a/cpukit/posix/src/keysetspecific.c
> >> > b/cpukit/posix/src/keysetspecific.c
> >> > index b25e44c..eeb4d84 100644
> >> > --- a/cpukit/posix/src/keysetspecific.c
> >> > +++ b/cpukit/posix/src/keysetspecific.c
> >> > @@ -26,7 +26,9 @@
> >> > #include <rtems/system.h>
> >> > #include <rtems/score/thread.h>
> >> > #include <rtems/score/wkspace.h>
> >> > +#include <rtems/score/rbtree.h>
> >> > #include <rtems/posix/key.h>
> >> > +#include <rtems/posix/threadsup.h>
> >> >
> >> > /*
> >> > * 17.1.2 Thread-Specific Data Management, P1003.1c/Draft 10, p. 165
> >> > @@ -37,18 +39,34 @@ int pthread_setspecific(
> >> > const void *value
> >> > )
> >> > {
> >> > - register POSIX_Keys_Control *the_key;
> >> > - uint32_t api;
> >> > - uint32_t index;
> >> > Objects_Locations location;
> >> > + POSIX_Keys_Rbtree_node *rb_node;
> >> > + POSIX_API_Control *api;
> >> >
> >> > - the_key = _POSIX_Keys_Get( key, &location );
> >> > + /** _POSIX_Keys_Get() would call _Thread_Disable_dispatch()
> >> > implicitly*/
> >> > + _POSIX_Keys_Get( key, &location );
> >> > switch ( location ) {
> >> >
> >> > case OBJECTS_LOCAL:
> >> > - api = _Objects_Get_API( _Thread_Executing->Object.id );
> >> > - index = _Objects_Get_index( _Thread_Executing->Object.id );
> >> > - the_key->Values[ api ][ index ] = (void *) value;
> >> > + rb_node = _Workspace_Allocate( sizeof( POSIX_Keys_Rbtree_node )
> >> > );
> >> This allocation needs to be accounted for in confdefs.h
> >>
> > I remember we have discussed this before, the exactly post
> > is:http://www.rtems.org/pipermail/rtems-devel/2012-August/001712.html,
> > but seem there is no conclusion yet.
> >>
> >> > + if ( !rb_node ) {
> >> > + _Thread_Enable_dispatch();
> >> > + return ENOMEM;
> >> > + }
> >> > +
> >> > + rb_node->key = key;
> >> > + rb_node->thread_id = _Thread_Executing->Object.id;
> >> > + rb_node->value = value;
> >> > + if (_RBTree_Insert_unprotected( &_POSIX_Keys_Rbtree,
> >> > &(rb_node->rb_node) ) ) {
> >> > + _Workspace_Free( rb_node );
> >> > + _Thread_Enable_dispatch();
> >> > + return EAGAIN;
> >> > + }
> >> > +
> >> > + /** append rb_node to the thread API extension's chain */
> >> > + api = (POSIX_API_Control
> >> > *)(_Thread_Executing->API_Extensions[THREAD_API_POSIX]);
> >> > + _Chain_Append_unprotected( &api->the_chain, &rb_node->ch_node
> );
> >> > +
> >> > _Thread_Enable_dispatch();
> >> > return 0;
> >> >
> >> > diff --git a/cpukit/posix/src/pthread.c b/cpukit/posix/src/pthread.c
> >> > index 873449a..22c835f 100644
> >> > --- a/cpukit/posix/src/pthread.c
> >> > +++ b/cpukit/posix/src/pthread.c
> >> > @@ -234,6 +234,9 @@ static bool _POSIX_Threads_Create_extension(
> >> > created
> >> > );
> >> >
> >> > + /** initialize thread's key vaule node chain */
> >> > + _Chain_Initialize_empty( &api->the_chain );
> >> > +
> >> > return true;
> >> > }
> >> >
> >> > diff --git a/cpukit/sapi/src/posixapi.c b/cpukit/sapi/src/posixapi.c
> >> > index 3f65442..b5e62c2 100644
> >> > --- a/cpukit/sapi/src/posixapi.c
> >> > +++ b/cpukit/sapi/src/posixapi.c
> >> > @@ -64,7 +64,7 @@ void _POSIX_API_Initialize(void)
> >> > _POSIX_signals_Manager_Initialization();
> >> > _POSIX_Threads_Manager_initialization();
> >> > _POSIX_Condition_variables_Manager_initialization();
> >> > - _POSIX_Key_Manager_initialization();
> >> > + _POSIX_Keys_Manager_initialization();
> >> > _POSIX_Mutex_Manager_initialization();
> >> > _POSIX_Message_queue_Manager_initialization();
> >> > _POSIX_Semaphore_Manager_initialization();
> >> > --
> >> > 1.7.1
> >> >
> >
> >
>
Thanks,
Zhongwei Yao
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/devel/attachments/20130223/fe425c77/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: POSIX_Key1.patch
Type: application/octet-stream
Size: 18473 bytes
Desc: not available
URL: <http://lists.rtems.org/pipermail/devel/attachments/20130223/fe425c77/attachment-0002.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: POSIX_Key2.patch
Type: application/octet-stream
Size: 45167 bytes
Desc: not available
URL: <http://lists.rtems.org/pipermail/devel/attachments/20130223/fe425c77/attachment-0003.obj>
More information about the devel
mailing list