Error detection and recovery of partial BIT files have unique requirements compared to loading a full BIT file. If an error is detected in a full BIT file when it is being loaded into an FPGA, the FPGA never enters user mode. The error is detected after the corrupt design has been loaded into configuration memory, and specific signals are asserted to indicate an error condition. Because the FPGA never enters user mode, the corrupt design never becomes active. You must determine the system behavior for recovering from a configuration error such as downloading a different BIT file if the error condition is detected.
When you download partial BIT files, you cannot use this methodology for error detection and recovery. The FPGA is by definition already in user mode when the partial BIT file is loaded. Because the configuration circuitry supports error detection only after a BIT file has been loaded, a corrupt partial BIT file can become active, potentially damaging the FPGA if left operating for an extended period of time.
If a CRC error is detected during a partial reconfiguration, it asserts the INIT_B pin of the FPGA (INIT_B goes Low to indicate a CRC error). In UltraScale devices, this behavior is echoed on the PRERROR output pin of the ICAP. It is important to note that if a system monitors INIT_B for CRC errors during the initial configuration, a CRC error during a partial reconfiguration might trigger the same response. To detect the presence of a CRC error from within the FPGA, the CRC status can be monitored through the ICAP block. The Status Register (STAT) indicates that the partial BIT file has a CRC error, by asserting the CRC_ERROR flag (bit 0).
There are two types of partial BIT file errors to consider: data errors and address errors (the partial BIT file is essentially address and data information). Given that static routes are free to pass through reconfigurable regions, both types of errors can corrupt the static design, although the likelihood is very small. The only method for completely safe recovery is to download a new full BIT file to ensure the state of the static logic, which requires the entire FPGA to be reset.
Many systems do not need a complex recovery mechanism because resetting the entire FPGA is not critical, or the partial BIT file is stored locally. In that case, the chance of BIT file corruption is not appreciable. Systems in which the BIT files are at risk of becoming corrupted (such as sending the partial BIT file over a radio link) should use a dedicated silicon feature that avoids the problem.
The configuration engines of all AMD devices from 7 series through Versal devices, have the ability to perform a frame-by-frame CRC check and do not load a frame into the configuration memory if that CRC check fails. A failure is reported on the INIT_B pin (it is pulled Low) and gives you the opportunity to take the next steps: retry the partial bit file, fall back to a golden partial bit file, etc. The partially loaded reconfiguration region does not have valid programming in it, but the CRC check ensures the remainder of the device (static region and any other RMs) stays operational while the system recovers from the error.
To enable this feature for these devices, set the
PerFrameCRC property prior to running
write_bitstream. The default is No, and Yes inserts the extra CRC checks. The size of an uncompressed bit file increases four to five percent with this option enabled. Note that this feature is not compatible with bitstream compression. No other specific design considerations are necessary to select this option, but your partial reconfiguration controller solution should be designed to choose the course of action should the INIT_B pin indicate a failure has occurred.
The syntax for setting the
PerFrameCRC property is:
set_property bitstream.general.perFrameCRC yes [current_design]
This property inserts per-frame CRC checks in all bitstreams created from the current checkpoint, not just partial bitstreams. Full device bitstreams for the initial configuration of the device would also contain the extra CRC checks.
After a partial bit file has been loaded (with or without the per-frame CRC checks), the overall configuration of the device has changed. If the POST_CRC feature for SEU mitigation is enabled, the SEU mitigation engine automatically recalculates the embedded SEU CRC value after the partial bitstream has been loaded and after you have de-synced the configuration interface. Upon completion of the CRC recalibration, the FRAME_ECCE2 FRAME_VALID output toggles again to indicate that SEU detection has resumed.