This is the mail archive of the libstdc++@gcc.gnu.org mailing list for the libstdc++ project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[Patch] libstdc++/24469


Hi,

I'm converging to the below, which deals with the issue explained last night, without collateral damage, as far as I can see. I'm testing it on the usual targets and I mean to apply it before the end of the day...

Paolo.

//////////////////
2006-09-01  Paolo Carlini  <pcarlini@suse.de>
	    Richard Guenther  <rguenther@suse.de>

	PR libstdc++/24469
	* src/mt_allocator.cc (__pool<true>::_M_reserve_block,
	__pool<true>::_M_reclaim_block): Fix the logic to avoid
	races, exploit atomic counters stored in second part of
	the memory pointed by _M_used.
	(__pool<true>::_M_initialize): Adjust _M_used allocation.
	* include/ext/mt_allocator.h (__pool<true>::_Bin_record):
	Update comment.
Index: include/ext/mt_allocator.h
===================================================================
--- include/ext/mt_allocator.h	(revision 116606)
+++ include/ext/mt_allocator.h	(working copy)
@@ -298,8 +298,13 @@
 
 	// An "array" of counters used to keep track of the amount of
 	// blocks that are on the freelist/used for each thread id.
-	// Memory to these "arrays" is allocated in _S_initialize() for
-	// _S_max_threads + global pool 0.
+	// - Note that the second part of the allocated _M_used "array"
+	//   actually hosts (atomic) counters of reclaimed blocks:  in
+	//   _M_reserve_block and in _M_reclaim_block those numbers are
+	//   subtracted from the first ones to obtain the actual size
+	//   of the "working set" of the given thread.
+	// - Memory to these "arrays" is allocated in _S_initialize()
+	//   for _S_max_threads + global pool 0.
 	size_t*				_M_free;
 	size_t*			        _M_used;
 	
Index: src/mt_allocator.cc
===================================================================
--- src/mt_allocator.cc	(revision 116604)
+++ src/mt_allocator.cc	(working copy)
@@ -34,6 +34,7 @@
 #include <bits/c++config.h>
 #include <bits/concurrence.h>
 #include <ext/mt_allocator.h>
+#include <cstring>
 
 namespace
 {
@@ -263,13 +264,33 @@
 	// number of records is "high enough".
 	const size_t __thread_id = _M_get_thread_id();
 	const _Tune& __options = _M_get_options();	
-	const unsigned long __limit = 100 * (_M_bin_size - __which)
-		                      * __options._M_freelist_headroom;
+	const size_t __limit = (100 * (_M_bin_size - __which)
+				* __options._M_freelist_headroom);
 
-	unsigned long __remove = __bin._M_free[__thread_id];
+	size_t __remove = __bin._M_free[__thread_id];
 	__remove *= __options._M_freelist_headroom;
-	if (__remove >= __bin._M_used[__thread_id])
-	  __remove -= __bin._M_used[__thread_id];
+
+	// NB: We assume that reads of _Atomic_words are atomic.
+	const size_t __max_threads = __options._M_max_threads + 1;
+	_Atomic_word* const __reclaimed_base =
+	  reinterpret_cast<_Atomic_word*>(__bin._M_used + __max_threads);
+	const _Atomic_word __reclaimed = __reclaimed_base[__thread_id];
+	const size_t __used = __bin._M_used[__thread_id] - __reclaimed;
+
+	// NB: For performance sake we don't resync every time, in order
+	// to spare atomic ops.  Note that if __reclaimed increased by,
+	// say, 1024, since the last sync, it means that the other
+	// threads executed the atomic in the else below at least the
+	// same number of times (at least, because _M_reserve_block may
+	// have decreased the counter), therefore one more cannot hurt.
+	if (__reclaimed > 1024)
+	  {
+	    __bin._M_used[__thread_id] -= __reclaimed;
+	    __atomic_add(&__reclaimed_base[__thread_id], -__reclaimed);
+	  }
+
+	if (__remove >= __used)
+	  __remove -= __used;
 	else
 	  __remove = 0;
 	if (__remove > __limit && __remove > __bin._M_free[__thread_id])
@@ -277,7 +298,7 @@
 	    _Block_record* __first = __bin._M_first[__thread_id];
 	    _Block_record* __tmp = __first;
 	    __remove /= __options._M_freelist_headroom;
-	    const unsigned long __removed = __remove;
+	    const size_t __removed = __remove;
 	    while (--__remove > 0)
 	      __tmp = __tmp->_M_next;
 	    __bin._M_first[__thread_id] = __tmp->_M_next;
@@ -292,8 +313,11 @@
 
 	// Return this block to our list and update counters and
 	// owner id as needed.
-	--__bin._M_used[__block->_M_thread_id];
-	
+	if (__block->_M_thread_id == __thread_id)
+	  --__bin._M_used[__thread_id];
+	else
+	  __atomic_add(&__reclaimed_base[__block->_M_thread_id], 1);
+
 	__block->_M_next = __bin._M_first[__thread_id];
 	__bin._M_first[__thread_id] = __block;
 	
@@ -333,6 +357,14 @@
     _Block_record* __block = NULL;
     if (__gthread_active_p())
       {
+	// Resync the _M_used counters.
+	const size_t __max_threads = __options._M_max_threads + 1;
+	_Atomic_word* const __reclaimed_base =
+	  reinterpret_cast<_Atomic_word*>(__bin._M_used + __max_threads);
+	const _Atomic_word __reclaimed = __reclaimed_base[__thread_id];
+	__bin._M_used[__thread_id] -= __reclaimed;
+	__atomic_add(&__reclaimed_base[__thread_id], -__reclaimed);
+
 	__gthread_mutex_lock(__bin._M_mutex);
 	if (__bin._M_first[0] == NULL)
 	  {
@@ -533,14 +565,19 @@
 	  {
 	    _Bin_record& __bin = _M_bin[__n];
 	    __v = ::operator new(sizeof(_Block_record*) * __max_threads);
+	    std::memset(__v, 0, sizeof(_Block_record*) * __max_threads);    
 	    __bin._M_first = static_cast<_Block_record**>(__v);
 
 	    __bin._M_address = NULL;
 
 	    __v = ::operator new(sizeof(size_t) * __max_threads);
+	    std::memset(__v, 0, sizeof(size_t) * __max_threads);	    	    
 	    __bin._M_free = static_cast<size_t*>(__v);
-	      
-	    __v = ::operator new(sizeof(size_t) * __max_threads);
+
+	    __v = ::operator new(sizeof(size_t) * __max_threads
+				 + sizeof(_Atomic_word) * __max_threads);
+	    std::memset(__v, 0, (sizeof(size_t) * __max_threads
+				 + sizeof(_Atomic_word) * __max_threads));
 	    __bin._M_used = static_cast<size_t*>(__v);
 	      
 	    __v = ::operator new(sizeof(__gthread_mutex_t));
@@ -555,12 +592,6 @@
 #else
 	    { __GTHREAD_MUTEX_INIT_FUNCTION(__bin._M_mutex); }
 #endif
-	    for (size_t __threadn = 0; __threadn < __max_threads; ++__threadn)
-	      {
-		__bin._M_first[__threadn] = NULL;
-		__bin._M_free[__threadn] = 0;
-		__bin._M_used[__threadn] = 0;
-	      }
 	  }
       }
     else
@@ -729,16 +760,21 @@
 	  {
 	    _Bin_record& __bin = _M_bin[__n];
 	    __v = ::operator new(sizeof(_Block_record*) * __max_threads);
+	    std::memset(__v, 0, sizeof(_Block_record*) * __max_threads);
 	    __bin._M_first = static_cast<_Block_record**>(__v);
 
 	    __bin._M_address = NULL;
 
 	    __v = ::operator new(sizeof(size_t) * __max_threads);
+	    std::memset(__v, 0, sizeof(size_t) * __max_threads);
 	    __bin._M_free = static_cast<size_t*>(__v);
 	      
-	    __v = ::operator new(sizeof(size_t) * __max_threads);
+	    __v = ::operator new(sizeof(size_t) * __max_threads + 
+				 sizeof(_Atomic_word) * __max_threads);
+	    std::memset(__v, 0, (sizeof(size_t) * __max_threads
+				 + sizeof(_Atomic_word) * __max_threads));
 	    __bin._M_used = static_cast<size_t*>(__v);
-	      
+
 	    __v = ::operator new(sizeof(__gthread_mutex_t));
 	    __bin._M_mutex = static_cast<__gthread_mutex_t*>(__v);
 	      
@@ -751,12 +787,6 @@
 #else
 	    { __GTHREAD_MUTEX_INIT_FUNCTION(__bin._M_mutex); }
 #endif
-	    for (size_t __threadn = 0; __threadn < __max_threads; ++__threadn)
-	      {
-		__bin._M_first[__threadn] = NULL;
-		__bin._M_free[__threadn] = 0;
-		__bin._M_used[__threadn] = 0;
-	      }
 	  }
       }
     else

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]