Register forum user name Search FAQ

Gammon Forum

Notice: Any messages purporting to come from this site telling you that your password has expired, or that you need to "verify" your details, making threats, or asking for money, are spam. We do not email users with any such messages. If you have lost your password you can obtain a new one by using the password reset link.
 Entire forum ➜ Electronics ➜ Microprocessors ➜ millis() overflow ... a bad thing?

millis() overflow ... a bad thing?

Postings by administrators only.

Refresh page


Posted by Nick Gammon   Australia  (23,057 posts)  Bio   Forum Administrator
Date Mon 26 Aug 2013 11:36 PM (UTC)

Amended on Tue 27 Aug 2013 10:19 AM (UTC) by Nick Gammon

Message
This page can be quickly reached from the link: http://www.gammon.com.au/millis


Introduction


On an almost daily basis we get a post on the Arduino forum about how "bad" it is to let the millis "timer" overflow. This thread explains why it is not a problem, if handled properly.

Background


The function millis () returns an unsigned long, which is the number of milliseconds since the processor was reset (until it overflows).

http://arduino.cc/en/Reference/Millis

Example:


unsigned long startTime = millis ();


Since there are 2^32 bits in an unsigned long it can count from 0 to 4294967295.

Computing this in terms of days we have:


2^32 / 1000 / 60 / 60 / 24 = 49.710 days


So, the number returned will overflow (go back to zero) after around 49 days (almost 50 days, as the reference page says).

Is this bad?


No. Is it bad your clock overflows (goes back to zero) at midnight? No it isn't.

Let's clear up some misconceptions:


  • The processor does not reset when the timer overflows.
  • The timer does not stop.
  • Nothing "bad" happens.


All that happens, on a timer overflow, is that it goes back to zero and starts counting up again. Just like your clock does.

How do I time an event?


There are two ways of timing something (eg. a minute from now). You can add or subtract. Because of the way that unsigned arithmetic works, subtracting is the correct way.

Example of adding - not recommended!


I want to detect a minute in the future...


unsigned long startTime = millis ();
unsigned long whenToStop = startTime + 60000;

...

if (millis () >= whenToStop)
  {
  // do something
  }


This will work most of the time, but not when the timer overflows. Let's see why:

Say the timer will overflow in 50 seconds, that is, it is currently:


2^32 - 50000 = 4294917296 (0xFFFF3CB0)


Now we add 60000, and get:


2^32 - 50000 + 60000 = 10000 (0x2710)


Our "whenToStop" variable has overflowed and is now quite small. And thus it is immediately less than the current time, and our check for when a minute is up fails, it appears to be up immediately.

Example of subtracting - recommended


I want to detect a minute in the future...


unsigned long startTime = millis ();
unsigned long interval = 60000;

...

if (millis () - startTime >= interval)
  {
  // do something
  }


This will work all the time. Let's see why:

Say the timer will overflow in 50 seconds, that is, it is currently:


2^32 - 50000 = 4294917296 (0xFFFF3CB0)


In 50 seconds the timer wraps around and goes to zero. Subtracting now gives us:


// millis () - startTime   = elapsed interval
    0        - 4294917296  = 50000 (0xC350)


The subtraction "wrapped around" and gave us a small number. If you use a scientific calculator and try the above you may get a result of 0xFF0000C350. The initial 0xFF is the "extended sign bit", in effect, however as the field is unsigned, and only four bytes long, the sign bits are dropped, giving a small positive number, not a large negative one.

Demo using the Mac Calculator app:



In 60 seconds the timer goes to 10000. Subtracting now gives us:


// millis () - startTime   = elapsed interval
    10000    - 4294917296  = 60000 (0xEA60)


So the elapsed interval (the difference) is now 60000, which is what we wanted.

Directly after we set up startTime and millis we will see a sequence like this:


millis ()   -  startTime    =  difference
                             (modulo 2^32)
4294917296     4294917296               0  <-- we start timing here
4294917297     4294917296               1
4294917298     4294917296               2
...
4294967295     4294917296           49999  <-- largest unsigned long value
         0     4294917296           50000  <-- wraps here back to zero
         1     4294917296           50001  <-- keeps counting up
...
      9999     4294917296           59999
     10000     4294917296           60000  <-- time's up!


Recommended method



  • Use unsigned long (not just long) for "time" variables.
  • Record the start time of some event (eg. a debounce, when you start feeding the fish, etc.)
  • Subtract the start time from the time now, giving a difference.
  • See if the difference exceeds the desired interval.


eg.


startedFeedingFish = millis ();

...

if (millis () - startedFeedingFish >= 20000)  // feed them for 20 seconds
  {
  // stop feeding the fish
  }


Can't I just reset millis?


You can, but you don't need to. Plus it may well throw out any library you are using that does not expect millis to be reset. Plus you have to decide exactly when to do it. It would need to be a time when you aren't using millis. For example if you happened to be feeding the fish in the above example, and your code reset millis, then the fish will get an awful lot of food!


What about micros ()?


The function micros() returns the number of microseconds since reset. This also wraps around, however since there are 1000 microseconds to a millisecond it will wrap around faster.


 2^32 / 1000 / 1000 / 60  = 71.58 minutes


http://arduino.cc/en/Reference/Micros

However you can use exactly the same technique to keep track of time to the microsecond interval. Be aware that you can't time more than 71.58 minutes this way though.

Also be aware that the resolution of micros() is four microseconds (because of the timer prescaler) so you won't ever be able to time a one or two microsecond interval.

Limitations


If you are using millis() you cannot time more than 49.71 days, and if you are using micros() you cannot time more than 71.58 minutes, without extra code to detect the wrap-around. For long intervals I suggest a real-time clock, since that would be more reliable during power failures, or when you replace batteries, etc.

[EDIT] Removed sample code for detecting the wrap-around after 49 days. I just don't recommend doing it. If you really want to, you'll be able to work it out.

- Nick Gammon

www.gammon.com.au, www.mushclient.com
Top

Posted by Nick Gammon   Australia  (23,057 posts)  Bio   Forum Administrator
Date Reply #1 on Fri 23 Jan 2015 08:25 PM (UTC)

Amended on Tue 24 Jan 2017 05:06 AM (UTC) by Nick Gammon

Message
millis() accuracy


Because of the way the library code works, the number returned by millis() will slowly drift.

Assuming a 16 MHz clock:

There is an interrupt (Timer 0 overflow interrupt) called every 1024 µs which updates the variable used by millis(). Therefore millis() will be out by 24 µs after the first interrupt, 48 µs after the second interrupt, and so on. The code eventually compensates so this inaccuracy is not cumulative over time.

In other words, millis() will run slow (it should update every 1000 µs but actually updates every 1024 µs).

However the overflow interrupt, which is called every 1024 µs, keeps track of the amount it is out, and eventually adds one to the millis() count to catch up (and reduces the overflow amount to compensate). This will happen approximately every 42 overflow interrupts. At this point, of course, the count returned by millis() will "jump" as this extra compensating amount is added in.

In detail, it adds 3 (FRACT_INC) to a variable called timer0_fract each overflow. It keeps doing that until timer0_fract is >= 125 (FRACT_MAX). When this happens it adds 1 to the millis counter (timer0_millis) and subtracts 125 from timer0_fract. Since 125 / 3 is 41.67 this happens every 42 overflows or so. (And since we are adding one every 42 overflows, that means we are adding one every 42 * 24 µs, which is every 1008 µs).

If you are timing small intervals, micros() will be much more accurate, as that reads from the hardware directly, and does not suffer from this creeping error. However it wraps after around 71 minutes. Also, micros has a resolution of 4 µs (not 1 µs) because of the way the timer is configured. (It counts to 256, using a prescaler of 64. One clock cycle is 62.5 nS, thus it "ticks" every 4 µs, and overflows every 256 * 4 µs).

- Nick Gammon

www.gammon.com.au, www.mushclient.com
Top

The dates and times for posts above are shown in Universal Co-ordinated Time (UTC).

To show them in your local time you can join the forum, and then set the 'time correction' field in your profile to the number of hours difference between your location and UTC time.


70,242 views.

Postings by administrators only.

Refresh page

Go to topic:           Search the forum


[Go to top] top

Information and images on this site are licensed under the Creative Commons Attribution 3.0 Australia License unless stated otherwise.