Home | History | Annotate | Download | only in multiprocessing
      1 #
      2 # Module which supports allocation of memory from an mmap
      3 #
      4 # multiprocessing/heap.py
      5 #
      6 # Copyright (c) 2006-2008, R Oudkerk
      7 # All rights reserved.
      8 #
      9 # Redistribution and use in source and binary forms, with or without
     10 # modification, are permitted provided that the following conditions
     11 # are met:
     12 #
     13 # 1. Redistributions of source code must retain the above copyright
     14 #    notice, this list of conditions and the following disclaimer.
     15 # 2. Redistributions in binary form must reproduce the above copyright
     16 #    notice, this list of conditions and the following disclaimer in the
     17 #    documentation and/or other materials provided with the distribution.
     18 # 3. Neither the name of author nor the names of any contributors may be
     19 #    used to endorse or promote products derived from this software
     20 #    without specific prior written permission.
     21 #
     22 # THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS "AS IS" AND
     23 # ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     24 # IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     25 # ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     26 # FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     27 # DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     28 # OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     29 # HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     30 # LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     31 # OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     32 # SUCH DAMAGE.
     33 #
     34 
     35 import bisect
     36 import mmap
     37 import tempfile
     38 import os
     39 import sys
     40 import threading
     41 import itertools
     42 
     43 import _multiprocessing
     44 from multiprocessing.util import Finalize, info
     45 from multiprocessing.forking import assert_spawning
     46 
     47 __all__ = ['BufferWrapper']
     48 
     49 #
     50 # Inheirtable class which wraps an mmap, and from which blocks can be allocated
     51 #
     52 
     53 if sys.platform == 'win32':
     54 
     55     from _multiprocessing import win32
     56 
     57     class Arena(object):
     58 
     59         _counter = itertools.count()
     60 
     61         def __init__(self, size):
     62             self.size = size
     63             self.name = 'pym-%d-%d' % (os.getpid(), Arena._counter.next())
     64             self.buffer = mmap.mmap(-1, self.size, tagname=self.name)
     65             assert win32.GetLastError() == 0, 'tagname already in use'
     66             self._state = (self.size, self.name)
     67 
     68         def __getstate__(self):
     69             assert_spawning(self)
     70             return self._state
     71 
     72         def __setstate__(self, state):
     73             self.size, self.name = self._state = state
     74             self.buffer = mmap.mmap(-1, self.size, tagname=self.name)
     75             assert win32.GetLastError() == win32.ERROR_ALREADY_EXISTS
     76 
     77 else:
     78 
     79     class Arena(object):
     80 
     81         def __init__(self, size):
     82             self.buffer = mmap.mmap(-1, size)
     83             self.size = size
     84             self.name = None
     85 
     86 #
     87 # Class allowing allocation of chunks of memory from arenas
     88 #
     89 
     90 class Heap(object):
     91 
     92     _alignment = 8
     93 
     94     def __init__(self, size=mmap.PAGESIZE):
     95         self._lastpid = os.getpid()
     96         self._lock = threading.Lock()
     97         self._size = size
     98         self._lengths = []
     99         self._len_to_seq = {}
    100         self._start_to_block = {}
    101         self._stop_to_block = {}
    102         self._allocated_blocks = set()
    103         self._arenas = []
    104         # list of pending blocks to free - see free() comment below
    105         self._pending_free_blocks = []
    106 
    107     @staticmethod
    108     def _roundup(n, alignment):
    109         # alignment must be a power of 2
    110         mask = alignment - 1
    111         return (n + mask) & ~mask
    112 
    113     def _malloc(self, size):
    114         # returns a large enough block -- it might be much larger
    115         i = bisect.bisect_left(self._lengths, size)
    116         if i == len(self._lengths):
    117             length = self._roundup(max(self._size, size), mmap.PAGESIZE)
    118             self._size *= 2
    119             info('allocating a new mmap of length %d', length)
    120             arena = Arena(length)
    121             self._arenas.append(arena)
    122             return (arena, 0, length)
    123         else:
    124             length = self._lengths[i]
    125             seq = self._len_to_seq[length]
    126             block = seq.pop()
    127             if not seq:
    128                 del self._len_to_seq[length], self._lengths[i]
    129 
    130         (arena, start, stop) = block
    131         del self._start_to_block[(arena, start)]
    132         del self._stop_to_block[(arena, stop)]
    133         return block
    134 
    135     def _free(self, block):
    136         # free location and try to merge with neighbours
    137         (arena, start, stop) = block
    138 
    139         try:
    140             prev_block = self._stop_to_block[(arena, start)]
    141         except KeyError:
    142             pass
    143         else:
    144             start, _ = self._absorb(prev_block)
    145 
    146         try:
    147             next_block = self._start_to_block[(arena, stop)]
    148         except KeyError:
    149             pass
    150         else:
    151             _, stop = self._absorb(next_block)
    152 
    153         block = (arena, start, stop)
    154         length = stop - start
    155 
    156         try:
    157             self._len_to_seq[length].append(block)
    158         except KeyError:
    159             self._len_to_seq[length] = [block]
    160             bisect.insort(self._lengths, length)
    161 
    162         self._start_to_block[(arena, start)] = block
    163         self._stop_to_block[(arena, stop)] = block
    164 
    165     def _absorb(self, block):
    166         # deregister this block so it can be merged with a neighbour
    167         (arena, start, stop) = block
    168         del self._start_to_block[(arena, start)]
    169         del self._stop_to_block[(arena, stop)]
    170 
    171         length = stop - start
    172         seq = self._len_to_seq[length]
    173         seq.remove(block)
    174         if not seq:
    175             del self._len_to_seq[length]
    176             self._lengths.remove(length)
    177 
    178         return start, stop
    179 
    180     def _free_pending_blocks(self):
    181         # Free all the blocks in the pending list - called with the lock held.
    182         while True:
    183             try:
    184                 block = self._pending_free_blocks.pop()
    185             except IndexError:
    186                 break
    187             self._allocated_blocks.remove(block)
    188             self._free(block)
    189 
    190     def free(self, block):
    191         # free a block returned by malloc()
    192         # Since free() can be called asynchronously by the GC, it could happen
    193         # that it's called while self._lock is held: in that case,
    194         # self._lock.acquire() would deadlock (issue #12352). To avoid that, a
    195         # trylock is used instead, and if the lock can't be acquired
    196         # immediately, the block is added to a list of blocks to be freed
    197         # synchronously sometimes later from malloc() or free(), by calling
    198         # _free_pending_blocks() (appending and retrieving from a list is not
    199         # strictly thread-safe but under cPython it's atomic thanks to the GIL).
    200         assert os.getpid() == self._lastpid
    201         if not self._lock.acquire(False):
    202             # can't acquire the lock right now, add the block to the list of
    203             # pending blocks to free
    204             self._pending_free_blocks.append(block)
    205         else:
    206             # we hold the lock
    207             try:
    208                 self._free_pending_blocks()
    209                 self._allocated_blocks.remove(block)
    210                 self._free(block)
    211             finally:
    212                 self._lock.release()
    213 
    214     def malloc(self, size):
    215         # return a block of right size (possibly rounded up)
    216         assert 0 <= size < sys.maxint
    217         if os.getpid() != self._lastpid:
    218             self.__init__()                     # reinitialize after fork
    219         self._lock.acquire()
    220         self._free_pending_blocks()
    221         try:
    222             size = self._roundup(max(size,1), self._alignment)
    223             (arena, start, stop) = self._malloc(size)
    224             new_stop = start + size
    225             if new_stop < stop:
    226                 self._free((arena, new_stop, stop))
    227             block = (arena, start, new_stop)
    228             self._allocated_blocks.add(block)
    229             return block
    230         finally:
    231             self._lock.release()
    232 
    233 #
    234 # Class representing a chunk of an mmap -- can be inherited
    235 #
    236 
    237 class BufferWrapper(object):
    238 
    239     _heap = Heap()
    240 
    241     def __init__(self, size):
    242         assert 0 <= size < sys.maxint
    243         block = BufferWrapper._heap.malloc(size)
    244         self._state = (block, size)
    245         Finalize(self, BufferWrapper._heap.free, args=(block,))
    246 
    247     def get_address(self):
    248         (arena, start, stop), size = self._state
    249         address, length = _multiprocessing.address_of_buffer(arena.buffer)
    250         assert size <= length
    251         return address + start
    252 
    253     def get_size(self):
    254         return self._state[1]
    255