Faraday (TCI6488/C6474) - SPLOOP Errata Fix - Issue

Chad Courtney

There's an SPLOOP CPU Cross Path Stall bug on Faraday Si - It's Advisory # 2.1.7 (SPRZ248)

NEC came across this using an older version of the compiler and I had the switch per the advisory. The tried version 6.0.31 (which is after the 6.0.25 version that had the fix included per our errata.) But they are still getting the issue.

I've attached the generated assembly file and will attache the trace as well on a follow up post.

Here's the interrupt information showing that it's a Cross Path Stall exception

0x00000002 (EFR)
0x008550CE (NRP)
0x0001420D (NTSR)
0x00000180 (IERR)

The NRP is right after the exit of the SPLOOP.

Can anyone confirm if the fix was in place on version 6.0.31? And if so, why we might have generated the code such as it is?

Best Regards,

Chad

cc_coding3_6.0.31.asm

over 13 years ago

0 Chad Courtney over 13 years ago

TI__Mastermind 30825 points

Added Trace File.

trace_core2.zip

0 Archaeologist over 13 years ago

TI__Guru* 84285 points

For reference, the compiler workaround for this silicon issue was tracked by the compiler team as SDSCM00035478, which was marked fixed on 6.0.25.

0 Archaeologist over 13 years ago in reply to Archaeologist

TI__Guru* 84285 points

I admit I am not very familiar with SPLOOP/SPKERNEL, but the assembly code you show does not seem to match the description of advisory 2.1.7. I cannot see "An instruction reading a register via the cross path in the first cycle after SPKERNEL instruction." If I am interpreting the code correctly, either this advisory is not the cause of the extra cycles, or there is another possible trigger condition for the advisory.

0 Chad Courtney over 13 years ago in reply to Archaeologist

TI__Mastermind 30825 points

I've went through this a few times myself and I'm not seeing it either. I figured I might be missing something.

Best Regards,

Chad

0 Todd Hahn over 13 years ago in reply to Chad Courtney

TI__Expert 3455 points

Chad,

Atsushi Yokoyama, asked me to take a look at this. Referencing the advisory 2.1.7 from the document you provided above, I can confirm the assembly source does not meet the criteria. (The first criteria is not met.) Even if the NOP after the SPKERNEL || OR wasn't there, the third criteria would not be met.

-Todd

Code Composer Studio™︎

Code Composer Studio forum

Faraday (TCI6488/C6474) - SPLOOP Errata Fix - Issue