I do this, too, although most recently on Xilinx MicroBlaze and PPC designs. The hardware includes a free-running 32-bit counter, and I make this counter available to both arbitrary FPGA logic as well as appearing to the micro as a read-only register.

Before going into the for(;;) loop I read and save the counter value. The main loop of the micro application reads that register once per run through the loop, and compares the difference between the saved value and the current value, and if it exceeds some constant, then I do whatever I needed to do at that interval, and I save the new count, and start it all again.

This works well for millisecond counts, and at 80 MHz, the counter rolls over every 53 seconds and anyways, my compare takes rollovers into account.

I suppose if I needed microsecond-level accuracy I'd consider something else, like perhaps having the counter interrupt.

-a