[LITMUS^RT] sync release on integral millisecond

[Glenn Elliott]

On Feb 17, 2013, at 2:09 AM, Björn Brandenburg <bbb at mpi-sws.org> wrote:

> 
> 
> On 17.02.2013, at 05:02, Glenn Elliott <gelliott at cs.unc.edu> wrote:
> 
>> Quick question: Why does sys_release_ts() set release times to be on integral milliseconds?  Is this for tick scheduling, or something?
> 
> A very benign reason: so that periodic release times would be aligned with integral milliseconds when visualizing the sched_trace data. Take it out if it poses a problem for something else.
> 
> - Björn

Thanks!  I have a followup question about do_wait_for_ts_release() though.  This function sets the release point of the released task to be:

54	litmus->release_at(current, wait.ts_release_time
55		+ current->rt_param.task_params.phase
56		- current->rt_param.task_params.period);

Why is the task period subtracted from wait.ts_release_time if wait.ts_release_time denotes time=0?  Why isn't the release time simply "wait.ts_release_time + current->rt_params.phase"?  Won't the "synchronous" task set not be released synchronously if all tasks have different periods?  Perhaps at some point release_at() added back the task period to the release time, but it doesn't appear to now.

-Glenn
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.litmus-rt.org/pipermail/litmus-dev/attachments/20130217/b971e53b/attachment.html>