Software Watchdog Timer resets MPC875 CPU too fast

Question

I have a problem with using Software Watchdog Timer on a MPC875 cpu:

The timer triggers reset signal VERY fast: I get less than about half of a second betwen turning the timer on and receiving the reset signal, although I set Countdown value (SWTC) to maximum with prescaling enabled.

Here is how I do it (the JTAG script):

CF   TAR      875
CF   GRP      400F

INN

SR  PLPRCRK 0x55CCAA33
SR  PLPRCR  0x1A4D5000  ; Configure Phase-Lock Loop

SR  SCCRK   0x55CCAA33
SR  SCCR    0xF47F0002  ; Configure System Clock

SR  SYPCR   0xFFFFFF87  ; Enable the Software Watchdog Timer


SR SWSR 0x556c  ; Reset the timer
SR SWSR 0xaa39

SR SWSR 0x556c
SR SWSR 0xaa39

SR SWSR 0x556c
SR SWSR 0xaa39

SR SWSR 0x556c
SR SWSR 0xaa39

SR SWSR 0x556c
SR SWSR 0xaa39

SR SWSR 0x556c
SR SWSR 0xaa39

SR SWSR 0x556c
SR SWSR 0xaa39  ; The HRESET issued after few moments from this point

With prescaling enabled (SWP bit in SYPCR) it asserts HRESET signal in about a half of a second after the last counter reset (the latest SR SWSR command). And if I disable the prescaler the HRESET signal asserted even before even betwen first modification of the SWSR register (very short time). Thus, it seems the Timer in some way responds to the setting changes, but it is something wrong with the timing.

The System Clock and Phase-Lock Loop supposed to be configured correctly because we get same configuration on already established and running application (on top of VxWorks), but with the Watchdog Timer turned off.

Also I tried to clear SWF bit in SYPCR register to prevent the Timer counting while JTAG stopped the cpu (by asserting FRZ signal): then it works only until I exit from break/stepping mode, when I run an application the reset occurs in a very short time.

According to the Reference Manual the Timer works in the following way:

There is a down-counter, when it reaches zero the circuit asserts HRESET signal or raises a System Reset Interrupt. The counter is two bytes long optionally prescaled by a factor of 2048. It is decremented at rate of System Clock divided by 2048. So the expected timeout with the prescaler enabled and maximum counter value is 1 / (80MHz / 2048) * (65535 * 2048) which is about of 3435 seconds. With the prescaler disabled it should be about 1.7 seconds. Actual values are far smaller: about 0.5 seconds with the prescaled counter and much smaller (can't even measure) with the prescaler turned off.

According to the Diagram the SWT depends only on Core Clock and the SYPCR register, here is the register description:

I set the value 0xFFFFFF87 (actually tried different options), which means:

• SWTC: 0xFFFF (Timer Count, the maximum value is loaded into the internal down-counter (see diagram) upon writing magic sequence to the SWSR register).
• BMT: 0XFF (Bus Monitor Timer Count, the maximum value)
• BME: 1 (Bus Monitor enabled, and there is no sence to turn the bit off because it is always enabled with JTAG regardless of the configured value).
• SWF: 0 (the timer counts even when JTAG asserts FRZ signal).
• SWE: 1 (Watchdog is Enabled).
• SWRI: 1 (configured to assert HRESET, switching to NMI didn't helped).
• SWP: 1 (SWTC is prescaled by a factor of 2048).

Any suggestions?

Answer 1

Are you sure of your formula ? It looks weird. I would expect something like 1/(80MHz/2048/65536)=1.5 sec to be the correct value for the maximum period of the watchdog.