php.net |  support |  documentation |  report a bug |  advanced search |  search howto |  statistics |  random bug |  login
Bug #75368 mmap/munmap trashing on unlucky allocations
Submitted: 2017-10-12 19:53 UTC Modified: 2017-10-13 06:46 UTC
From: nikic@php.net Assigned:
Status: Closed Package: Scripting Engine problem
PHP Version: 7.0.24 OS:
Private report: No CVE-ID: None
 [2017-10-12 19:53 UTC] nikic@php.net
Description:
------------
It may happen that memory is allocated and freed in such a way that the allocation mmaps a new ZMM chunk and the deallocation immediately unmaps it again.

The attached script demonstrates this issues. It will execute quickly up to a certain point (for me on master i=94) and then hang for a while. strace shows that during this time a sequence of mmap, madvise and munmap is repeated many times:

mmap(NULL, 2097152, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f3927400000
madvise(0x7f3927400000, 2097152, MADV_HUGEPAGE) = 0
munmap(0x7f3927400000, 2097152)         = 0

This issue may be fixed by ensuring we always cache at least one chunk.


Test script:
---------------
<?php

$bigArray = range(0, 256 * 1024 - 1);

var_dump(memory_get_usage(true) / (1024*1024));
$array = [];
for ($i = 0; $i < 1024; $i++) {
    echo "$i\n";
    $array[] = range(0, 1023);
    for ($j = 0; $j < 16 * 1024; $j++) {
        $x = range(0, 1023);
        unset($x);
    }
}


Patches

Add a Patch

Pull Requests

Add a Pull Request

History

AllCommentsChangesGit/SVN commitsRelated reports
 [2017-10-13 06:46 UTC] laruence@php.net
it may still be triggered by i = 94 * 2 if you cache one chunk
 [2017-10-13 11:03 UTC] dmitry@php.net
Automatic comment on behalf of dmitry@zend.com
Revision: http://git.php.net/?p=php-src.git;a=commit;h=397f5cb687f4f31cb0dec85ba5a87237394ca12d
Log: Fixed bug #75368 (mmap/munmap trashing on unlucky allocations)
 [2017-10-13 11:03 UTC] dmitry@php.net
-Status: Open +Status: Closed
 [2017-10-13 12:10 UTC] spam2 at rhsoft dot net
thank you!

this seems to make a big difference for phpdoc on my workstation when the KDE session which seems to have small memory leaks is running a long time - phpdoc sometimes takes ages for no reason

same run, only the php-cli binary changed while currently the memleak/fragmentation is not so high after 12 hours but a noticeable change

 17/17 [============================] 100% 12.553s .. 
 17/17 [============================] 100% 6.488s ..
 
PHP Copyright © 2001-2017 The PHP Group
All rights reserved.
Last updated: Sun Nov 19 01:31:42 2017 UTC