Fix hang caused by la_interval == 0

[tombrazier]

Description

#25474 Introduced a change to calc_timer_interval() that exposed a pre-existing linear advance bug by causing a hang.

Under linear advance, the stepper ISR can be called at a much higher rate (but have far less work to do) than the rate normal steps generate. This is to do with how LA uses the bresenham algorithm as a way to drive both advance and ordinary E steps.

In extreme edge cases, calc_timer_interval() may even be called with a parameter greater than 65535, i.e. above 16 bits. The case I encountered was for very short segments with backlash compensation enabled where the added backlash confuses the LA acceleration limiting code in planner.cpp. The segment was so short that it really didn't matter, except that it now causes a hang.

With the old calc_timer_interval(), a spurious (and much too large) value for la_interval is calculated. With the new one, it is just possible that an overflow in the lookup into speed_lookuptable_fast could hit the dummy line which is all zeros. This results in la_interval == 0 and, consequently, an infinite loop in the stepper ISR.

This PR responds with two fixes. First, the effect of current_block->la_scaling is now applied before calling calc_timer_interval() which reduces the bitcount of the parameter. Second, calc_timer_interval() now handles bitcounts above 16 bits by returning a fixed result (which is in line with the AVR's maximum call rate of the ISR). This new code costs an extra two CPU cycles per call to calc_timer_interval().

Requirements

LIN_ADVANCE and BACKLASH_COMPENSATION and a higher steps/mm for the extruder (on mine it was 551 steps/mm).

Benefits

Print doesn't hang.

Configurations

stepperzerointervalfix.zip

Related Issues

#25474