g+
g+ Communities
Argonne National Laboratory

Experimental Physics and
Industrial Control System

2002  2003  2004  2005  <20062007  2008  2009  2010  2011  2012  2013  2014  Index 2002  2003  2004  2005  <20062007  2008  2009  2010  2011  2012  2013  2014 
<== Date ==> <== Thread ==>

Subject: Re: seq debugger
From: Andrew Johnson <anj@aps.anl.gov>
To: "Laznovsky, Michael" <lazmo@slac.stanford.edu>, EPICS core-talk <core-talk@aps.anl.gov>
Date: Wed, 21 Jun 2006 14:25:21 -0500
Hi Michael,

Laznovsky, Michael wrote:
Hi Andrew- were you the one asking about mutex's after my talk last week? Sorry if my answer was a little incoherent... I'm sometimes not at my best onstage.

Yes, that was me.

I can see a need for some kind of lock mechanism for array modification, but would scalars really need it? I guess we could maybe have an option to only modify variables at state transitions, if that sounds useful.

Some of the architectures that EPICS runs on do not have hardware floating point, thus the updating of a double precision (64 bit) floating point value can be interrupted half-way through. You wouldn't want anybody reading that floating point value in the middle of changing, so the only safe way to access it from another thread is by protecting the storage with a mutex (or otherwise guaranteeing that context switches are not occurring). Strings are another type that objiously suffer from this problem.

Note that the above isn't the only place and reason you need to protect the local variables, but its one scenario that demonstrates the issue and which is relatively easy to understand. Unfortunately if you don't do this right you may never see the problems it causes yourself, as they may only show up on certain architectures, e.g. SMP systems.

To be truly portable, if a location is written to from a different thread than the thread reading from it, there must be a mutex release after the write instruction and a take before the read.

- Andrew
--
Not everything that can be counted counts,
and not everything that counts can be counted.
  -- Albert Einstein

Replies:
RE: seq debugger Laznovsky, Michael

Navigate by Date:
Prev: Re: RE: alarm hook Andrew Johnson
Next: RE: seq debugger Laznovsky, Michael
Index: 2002  2003  2004  2005  <20062007  2008  2009  2010  2011  2012  2013  2014 
Navigate by Thread:
Prev: RE: alarm hook Liyu, Andrei
Next: RE: seq debugger Laznovsky, Michael
Index: 2002  2003  2004  2005  <20062007  2008  2009  2010  2011  2012  2013  2014 
ANJ, 02 Feb 2012 Valid HTML 4.01! · Home · News · About · Base · Modules · Extensions · Distributions · Download ·
· EPICSv4 · IRMIS · Talk · Bugs · Documents · Links · Licensing ·