Known/fixed concurrency issues related to memory consistency?

[Michael Tautschnig <mt...@debian.org>]

Dear Apache Devs,

We are currently looking into automatic analysis techniques to discover bugs in
concurrent software that are caused by the memory consistency models implemented
in today's multiprocessors. (Intentionally) racy program parts may thus exhibit
unexpected behaviour, caused by, e.g., delayed writes to memory. Such errors
tend to be hard to reproduce and their root causes are often very hard to
diagnose.

We thus think that automated analysis tools could be very valuable, yet we
need to test drive our tools to understand whether they are able to find these
errors on real code. As such we are asking developers of highly concurrent
software for input. Our question is thus: is anyone of you aware of current or
fixed concurrency-related bugs in Apache's httpd that might have been caused by
cache effects?

Thank you very much in advance,
Michael

Re: Known/fixed concurrency issues related to memory consistency?

[Stefan Fritsch <sf...@sfritsch.de>]

On Monday 26 September 2011, Michael Tautschnig wrote:
> > Apache HTTPD has a bugzilla database at http://issues.apache.org/
> > It's more than a little chaotic, not least because it's
> > completely public and real bugs and enhancement requests are
> > mixed with user problems that have nothing to do with us.  But
> > there are certainly some concurrency-related issues in there if
> > you look.
> >
> > 
> 
> Thank you very much for the pointer, we'll take a look at this
> database. The main hope associated with my prior email was that
> someone would know of such issues off the top of their head. If
> anybody does: please let us know.

Due to the way httpd uses memory pools (which imply the data being 
owned by the current thread), such race conditions are probably very 
rare in most parts of httpd. The only likely candidates are the MPM 
modules. But the APR library is more likely to have such issues.

Both MPMs and APR had bugs due to data races in the past (look at the 
commit logs), but I don't know if any of those are related to cache-
consistency.

Cheers,
Stefan

Re: Known/fixed concurrency issues related to memory consistency?

[Michael Tautschnig <mt...@debian.org>]

Hi,

[...]
> >    Our question is thus: is anyone of you aware of current or
> > fixed concurrency-related bugs in Apache's httpd that might have been caused by
> > cache effects?
> 
> So you're looking for test-cases for your own tools, rather
> than (directly) for bugs in ours?
> 

As we are still in the process of getting our tools in shape: we would first
like to confirm that our tools are able to diagnose known bugs; we will then
continue to look for not yet identified issues on your code. At present we have
to rely on our own examples (and kernel.org's maintenance worsens this...).

> Apache HTTPD has a bugzilla database at http://issues.apache.org/
> It's more than a little chaotic, not least because it's completely
> public and real bugs and enhancement requests are mixed with user
> problems that have nothing to do with us.  But there are certainly
> some concurrency-related issues in there if you look.
> 

Thank you very much for the pointer, we'll take a look at this database. The
main hope associated with my prior email was that someone would know of such
issues off the top of their head. If anybody does: please let us know.

Best regards,
Michael

Re: Known/fixed concurrency issues related to memory consistency?

[Nick Kew <ni...@webthing.com>]

On Mon, 26 Sep 2011 12:10:55 +0200
Michael Tautschnig <mt...@debian.org> wrote:

>    Our question is thus: is anyone of you aware of current or
> fixed concurrency-related bugs in Apache's httpd that might have been caused by
> cache effects?

So you're looking for test-cases for your own tools, rather
than (directly) for bugs in ours?

Apache HTTPD has a bugzilla database at http://issues.apache.org/
It's more than a little chaotic, not least because it's completely
public and real bugs and enhancement requests are mixed with user
problems that have nothing to do with us.  But there are certainly
some concurrency-related issues in there if you look.

-- 
Nick Kew