Experimental Physics and
Industrial Control System

Till Straumann <[email protected]> · Fri, 28 Jan 2011 12:28:42 -0600

On 01/28/2011 10:37 AM, Eric Norum wrote:

Hmm...
I see that my name is on the code in osi/os/default/osdMessageQueue.cpp
but I don't think that I actually wrote it. I would have expected that
all allocation would be done at message queue initialization -- I agree
that allocating at run-time is a poor idea.

Sure but pre-allocation of the 'event nodes' is not possible since.
There is one such node per thread that is blocking on the queue.
Unless you know how many threads may block you can't pre-allocate.

However, I'm not sure these 'event nodes' are really necessary.
They are only used to notify blocked threads in a FIFO manner.
One could have just two 'epicsEvent's (one for TX, one for RX)
and let all receivers simply block on the single RX 'epicsEvent'.

The 'epicsEvent' itself should ensure that just one thread
wakes up. However, it may be that the wake-up is not strictly
FIFO but governed by whatever policy 'epicsEvent' uses.

- Till

On Jan 28, 2011, at 8:16 AM, <[email protected]
<mailto:[email protected]>> wrote:

Hi Andrew

The PREEMPT_RT patch makes the kernel fully pre-emptive (leading to
improved maximum latency) and adds support for priority inheritance
mutexes. There are no user space API changes, the code to use
_POSIX_THREAD_PRIORITY_SCHEDULING and _POSIX_THREAD_PRIO_INHERIT is
already in libCom (if disabled by default). I'd like to use the
epicsMessageQueue from a high priority thread but the send and receive
functions may calloc an eventNode, and that might cause a page fault or
block my thread on the malloc lock. I'd like to be able to pre-allocate
the eventNodes because I know the maximum number of threads that may
wait on the queue in advance.

James

--
Eric Norum
[email protected] <mailto:[email protected]>

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System