<html>
<head>
<meta content="text/html; charset=UTF-8" http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
Hi Alexander,<br>
<br>
this patch (the VirtualBoxImpl.cpp part) is loosening up
synchronization a bit too much for my taste. The approved lock order
for Medium instances is "from parent to child", which is why your
MachineImpl.cpp change is a perfect catch (and solution, because the
child lock is held unnecessarily).<br>
<br>
The VirtualBoxImpl.cpp part on the other hand uses the correct lock
order, so I'm wondering if you have any evidence that this is
actually solving any deadlock. Not disputing that the code should
still work properly after the change (with a tiny bit less atomicity
which isn't strictly required), but I suspect there was nothing
wrong with it initially. If it played part of a deadlock then it's
most likely the other party where the lock order is wrong. If you're
running a debug build then the lock validator in the runtime is
active and should catch anything fishy (unless it involves event
semapshores as they have no owner).<br>
<br>
Feedback appreciated <span class="moz-smiley-s1"><span> :-) </span></span><br>
<br>
Klaus<br>
<br>
<br>
<div class="moz-cite-prefix">On 30.01.2015 15:50, <a class="moz-txt-link-abbreviated" href="mailto:a.urakov@drweb.com">a.urakov@drweb.com</a>
wrote:<br>
</div>
<blockquote cite="mid:54CB9A32.8070604@drweb.com" type="cite">
<meta content="text/html; charset=UTF-8" http-equiv="Content-Type">
Diffs attached<br>
<br>
<div class="moz-cite-prefix">30.01.2015 17:49, <a
moz-do-not-send="true" class="moz-txt-link-abbreviated"
href="mailto:a.urakov@drweb.com">a.urakov@drweb.com</a> пишет:<br>
</div>
<blockquote cite="mid:54CB99FC.2090805@drweb.com" type="cite">
<meta http-equiv="content-type" content="text/html;
charset=UTF-8">
Hello!<br>
<br>
We would like to contribute to VirtualBox under MIT license. We
send our changes to fix bug described at <a
moz-do-not-send="true" class="moz-txt-link-freetext"
href="https://www.virtualbox.org/ticket/13789">https://www.virtualbox.org/ticket/13789</a>
.<br>
<br>
The idea is that functions <i>Medium::removeRegistry()</i> and
<i>Medium::addRegistry()</i> themselves take write locks on
mediums, so we get read lock only for <i>Medium::getAnyMachineBackref()</i>
call in <i>VirtualBox::unregisterMachine()</i> function. Also
we release read lock on medium after we got its parent and
before we are going to iterate through parents in <i>Machine::detachAllMedia()</i>
function.<br>
<br>
Regards,<br>
Alexander<br>
<br>
</blockquote>
</blockquote>
</body>
</html>