Subject:
|
Accurate msleep and timeslicing - was Re: sys_timer
|
Newsgroups:
|
lugnet.robotics.rcx.legos
|
Date:
|
Fri, 3 May 2002 11:43:07 GMT
|
Viewed:
|
2372 times
|
| |
|
|
"Michael Obenland" <obenland@t-online.de> wrote in message
news:GvJ5vD.7MC@lugnet.com...
<snip>
> > this only seems to update by 20 ms at a time.
>
> That is right.
>
> > That doesn't really make sense though...
>
> It does, actually. The timer itself counts the millisec. But if you do a
> wait_event() call, your running program gives up its time slice, the OS
> waits for the event and then continues your task with the next task • switch.
> And the granularity for a task switch is 20ms ( or whatever you want it to
> be. Look in include/sys/tm.h and try what happens with a TM_DEFAULT_SLICE
> of 10 or 30.)
I made some modifications to the tm_sleep_wakeup that allows high priority
tasks to wake up at exactly the time specified for msleep (which uses
wait_event on tm_sleep_wakeup). The modification works well for
applications where driving or turning a specific amount of time is required.
However, I think this modification could be applied anywhere you need a high
priority task to return more quickly from a wait_event...
You would likely want to write your own 'wakeup' function, but could modify
the kernel function as well.
Here is the modified tm_sleep_wakeup - explaination follows:
-----------------------------------------
static wakeup_t tm_sleep_wakeup(wakeup_t data) {
time_t remaining = ((time_t)data) - sys_time;
if (((time_t)data)<=sys_time)
{
tm_timeslice = TM_DEFAULT_SLICE;
return -1;
}
if (remaining < tm_timeslice)
tm_timeslice = remaining;
return 0;
}
-----------------------------------------
If the condition to wake up is true, set the tm_timeslice = TM_DEFAULT_SLICE
to return to normal time slicing. Otherwise, set tm_timeslice = X; where X
is the time before the next check of this event handler.
WARNING: This method can have drastic effects on the application/OS.
Setting this to 0 would block all tasks until the event handler returns
true. Setting it to a low number (1 - 5 perhaps) could cause a lot of
overhead... slowing other tasks down.
It works well for the tm_sleep_wakeup because the amount of time remaining
is known... and the timeslice is only reduced when necessary.
// Joe
|
|
Message is in Reply To:
 | | Re: sys_timer
|
| (...) Yes. (...) That is right. (...) It does, actually. The timer itself counts the millisec. But if you do a wait_event() call, your running program gives up its time slice, the OS waits for the event and then continues your task with the next (...) (23 years ago, 3-May-02, to lugnet.robotics.rcx.legos)
|
4 Messages in This Thread:
  
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
