8
0
mirror of https://github.com/FirebirdSQL/firebird.git synced 2025-01-23 03:23:04 +01:00
firebird-mirror/extern/libcds/cds/intrusive/free_list.h
2022-10-08 20:46:39 +03:00

222 lines
8.8 KiB
C++

// Copyright (c) 2006-2018 Maxim Khizhinsky
//
// Distributed under the Boost Software License, Version 1.0. (See accompanying
// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
#ifndef CDSLIB_INTRUSIVE_FREE_LIST_H
#define CDSLIB_INTRUSIVE_FREE_LIST_H
#include <cds/algo/atomic.h>
namespace cds { namespace intrusive {
/// Lock-free free list
/** @ingroup cds_intrusive_freelist
Free list is a helper class intended for reusing objects instead of freeing them completely;
this avoids the overhead of \p malloc(), and also avoids its worst-case behavior of taking an operating system lock.
So, the free list can be considered as a specialized allocator for objects of some type.
The algorithm is taken from <a href="http://moodycamel.com/blog/2014/solving-the-aba-problem-for-lock-free-free-lists">this article</a>.
The algo does not require any SMR like Hazard Pointer to prevent ABA problem.
There is \ref TaggedFreeList "tagged pointers" variant of free list for processors with double-width CAS support.
\b How to use
\code
#include <cds/intrusive/free_list.h>
// Your struct should be derived from FreeList::node
struct Foo: public cds::intrusive::FreeList::node
{
// Foo fields
};
// Simplified Foo allocator
class FooAllocator
{
public:
// free-list clear() must be explicitly called before destroying the free-list object
~FooAllocator()
{
m_FreeList.clear( []( freelist_node * p ) { delete static_cast<Foo *>( p ); });
}
Foo * alloc()
{
freelist_node * p = m_FreeList.get();
if ( p )
return static_cast<Foo *>( p );
return new Foo;
};
void dealloc( Foo * p )
{
m_FreeList.put( static_cast<freelist_node *>( p ));
};
private:
typedef cds::intrusive::FreeList::node freelist_node;
cds::intrusive::FreeList m_FreeList;
};
\endcode
*/
class FreeList
{
public:
/// Free list node
struct node {
//@cond
atomics::atomic<uint32_t> m_freeListRefs;
atomics::atomic<node *> m_freeListNext;
node()
: m_freeListRefs( 0 )
{
m_freeListNext.store( nullptr, atomics::memory_order_release );
}
//@endcond
};
public:
/// Creates empty free list
FreeList()
: m_Head( nullptr )
{}
/// Destroys the free list. Free-list must be empty.
/**
@warning dtor does not free elements of the list.
To free elements you should manually call \p clear() with an appropriate disposer.
*/
~FreeList()
{
assert( empty());
}
/// Puts \p pNode to the free list
void put( node * pNode )
{
// We know that the should-be-on-freelist bit is 0 at this point, so it's safe to
// set it using a fetch_add
if ( pNode->m_freeListRefs.fetch_add( c_ShouldBeOnFreeList, atomics::memory_order_release ) == 0 ) {
// Oh look! We were the last ones referencing this node, and we know
// we want to add it to the free list, so let's do it!
add_knowing_refcount_is_zero( pNode );
}
}
/// Gets a node from the free list. If the list is empty, returns \p nullptr
node * get()
{
auto head = m_Head.load( atomics::memory_order_acquire );
while ( head != nullptr ) {
auto prevHead = head;
auto refs = head->m_freeListRefs.load( atomics::memory_order_relaxed );
if ( cds_unlikely( (refs & c_RefsMask) == 0 || !head->m_freeListRefs.compare_exchange_strong( refs, refs + 1,
atomics::memory_order_acquire, atomics::memory_order_relaxed )))
{
head = m_Head.load( atomics::memory_order_acquire );
continue;
}
// Good, reference count has been incremented (it wasn't at zero), which means
// we can read the next and not worry about it changing between now and the time
// we do the CAS
node * next = head->m_freeListNext.load( atomics::memory_order_relaxed );
if ( cds_likely( m_Head.compare_exchange_strong( head, next, atomics::memory_order_acquire, atomics::memory_order_relaxed ))) {
// Yay, got the node. This means it was on the list, which means
// shouldBeOnFreeList must be false no matter the refcount (because
// nobody else knows it's been taken off yet, it can't have been put back on).
assert( (head->m_freeListRefs.load( atomics::memory_order_relaxed ) & c_ShouldBeOnFreeList) == 0 );
// Decrease refcount twice, once for our ref, and once for the list's ref
head->m_freeListRefs.fetch_sub( 2, atomics::memory_order_relaxed );
return head;
}
// OK, the head must have changed on us, but we still need to decrease the refcount we
// increased
refs = prevHead->m_freeListRefs.fetch_sub( 1, atomics::memory_order_acq_rel );
if ( refs == c_ShouldBeOnFreeList + 1 )
add_knowing_refcount_is_zero( prevHead );
}
return nullptr;
}
/// Checks whether the free list is empty
bool empty() const
{
return m_Head.load( atomics::memory_order_relaxed ) == nullptr;
}
/// Clears the free list (not atomic)
/**
For each element \p disp disposer is called to free memory.
The \p Disposer interface:
\code
struct disposer
{
void operator()( FreeList::node * node );
};
\endcode
This method must be explicitly called before the free list destructor.
*/
template <typename Disposer>
void clear( Disposer disp )
{
node * head = m_Head.load( atomics::memory_order_relaxed );
m_Head.store( nullptr, atomics::memory_order_relaxed );
while ( head ) {
node * next = head->m_freeListNext.load( atomics::memory_order_relaxed );
disp( head );
head = next;
}
}
private:
//@cond
void add_knowing_refcount_is_zero( node * pNode )
{
// Since the refcount is zero, and nobody can increase it once it's zero (except us, and we
// run only one copy of this method per node at a time, i.e. the single thread case), then we
// know we can safely change the next pointer of the node; however, once the refcount is back
// above zero, then other threads could increase it (happens under heavy contention, when the
// refcount goes to zero in between a load and a refcount increment of a node in try_get, then
// back up to something non-zero, then the refcount increment is done by the other thread) --
// so, if the CAS to add the node to the actual list fails, decrease the refcount and leave
// the add operation to the next thread who puts the refcount back at zero (which could be us,
// hence the loop).
node * head = m_Head.load( atomics::memory_order_relaxed );
while ( true ) {
pNode->m_freeListNext.store( head, atomics::memory_order_relaxed );
pNode->m_freeListRefs.store( 1, atomics::memory_order_release );
if ( cds_unlikely( !m_Head.compare_exchange_strong( head, pNode, atomics::memory_order_release, atomics::memory_order_relaxed ))) {
// Hmm, the add failed, but we can only try again when the refcount goes back to zero
if ( pNode->m_freeListRefs.fetch_add( c_ShouldBeOnFreeList - 1, atomics::memory_order_release ) == 1 )
continue;
}
return;
}
}
//@endcond
private:
//@cond
static constexpr uint32_t const c_RefsMask = 0x7FFFFFFF;
static constexpr uint32_t const c_ShouldBeOnFreeList = 0x80000000;
// Implemented like a stack, but where node order doesn't matter (nodes are
// inserted out of order under contention)
atomics::atomic<node *> m_Head;
//@endcond
};
}} // namespace cds::intrusive
#endif // CDSLIB_INTRUSIVE_FREE_LIST_H