FlexIO
FlexIO monitoring
HPC and Data for Lattice QCD
FlexIO monitoring
Status registers
Device |
Address |
Description |
---|---|---|
PowerXCell 8i |
0x512018 |
Retry counter |
PowerXCell 8i |
0x512020 |
CRC counter |
NWP/GBIF |
0x40c |
Retry counter |
NWP/GBIF |
0x410 |
CRC counter |
Retry errors
Known cases of retry counter on PowerXCell 8i increments:
- Sudden increment (and possibly retry error overflow checkstop) may occur after power-on. In rare cases link training is not optimal. Reboot of node will fix problem
- When speculative data credits are enabled retry errors may occur due to buffer overflow depending on traffic generated by applciation. If this happens data array overflow bit in IWC exception register will be asserted. Such overflows will have negative impact on link bandwidth.
- In other cases rare increments (<< 1/hour) due to errors on link can happen. No action required.
- Regular increments of counter at a rate of O(1) per minute or faster may indicate non-optimal setting of reference voltage. Node may require inspection.