SPI Receive block reading out incorrect data on TI C2000 MCU

Hello,

I am implementing FOC on a 3 phase inverter using a rotary encoder as the position sensor on a C2000 MCU (LAUNCHXL-F28379D board). The encoder I am using communicates with BiSS-C, and I have an FPGA that translates this into SPI (it sends CS, MOSI and CLK) to send to the MCU.

My Simulink model has 3 asynchronous subsystems, and has a swithcing frequency of 20kHz. In the first one, the MCU uses SPI SYNC to get the FPGA to request encoder position. This is triggered 8us before the top of the switching cycle via PWM interrupt. SPI RX interrupt triggers a second subsystem, which just has the SPI Receive block. The third subsystem is triggered after that (at the top of the switching cycle), and is the FOC block with current sampling adcs and PWM duty cycle generation. Subsystem 2 and 3 are non-preemptable. It is using a SPI clock frequency of 6MHz, with 16bits per position measurement.

This normally works perfectly, and motor position is read out accurately. My issue is that when my motor goes to high electrical speeds relative to the base speed (constant Fsw, constant SPI clk rate, not a signal integrity issue) I start to receive incorrect SPI position readouts. When I probe the SPI pins with a logic analyser they read out the correct position (same as BiSS-C), however when I log the position in Simulink it randomly jumps up or down and is incorrect. The random position sometimes looks like a bit shifted version of what it should be, but other times it is completely random (not bit shifted, many random bit changes).

I believe this could be something due to my CPU/model getting overloaded, as when I add more stuff to my FOC block these incorrect positions start to happen more often and at lower speeds. However, the SPI SYNC subsystem always runs on time and is preemptable, so the FOC block must be finishing running well before the SPI receive subsystem. I cannot remove anything else in my FOC block, as all it has are reading current sensors and doing clark/park transforms.

My questions are:

1) Anyone heard of this issue before, or knows what could be causing this?

2) If not, how can I ensure that my subsystems are being ran on time and can I probe how long these are taking to ensure they are not running overtime and causing problems?

Massive thank you to anyone that can provide guidance.