[entropy_src] Functional coverage metrics dropped #21888
Comments
vogelpi
added
Component:DV
|
For reference: |
vogelpi
added
the
Earlgrey-PROD Candidate
|
I agree that the having the two new control fields in the cross doesn't add much value. Do we already measure non-crossed coverage and check their correct implementation in the scoreboard, though? |
Since ES tapeout, several API breaking changes had to be implemented. The DV side was mostly updated in sync but there are some reasons why we cannot meet V2S currently: - Functional and code coverage metrics are currently not met (V2 & V2S). See lowRISC#21888. - The entropy_src_err_vseq sequence should be extended to test the hardened prim_fifo_sync primitives (V2S). See lowRISC#22326. Signed-off-by: Pirmin Vogel <[email protected]>
Since ES tapeout, several API breaking changes had to be implemented. The DV side was mostly updated in sync but there are some reasons why we cannot meet V2S currently: - Functional and code coverage metrics are currently not met (V2 & V2S). See #21888. - The entropy_src_err_vseq sequence should be extended to test the hardened prim_fifo_sync primitives (V2S). See #22326. Signed-off-by: Pirmin Vogel <[email protected]>
|
Discussed moving as P1 to M5 due to RTL risk analysis performed during last issue triage. |
|
It might be more interesting to just have a separate cover group which is only for the different fips configurations and remove the fips settings from the other corver groups. Since the fips bit is now software controllable, it should not be related to the other settings. |
|
Update: @h-filali has been investigating this and it is not super straight forward. Even after removing some coverpoints from the crosses as mentioned previously, the cross coverage is lower than expected. The investigation continues as we really need this for V2S. |
|
This has been taken care of by @h-filali . All coverage metrics should now be above the V2/V2S threshold again. Closing. |
Description
After merging #21369 which made the
rng_fips_orequest bit to the RNG/noise source as well as thefips_complianceflag signaled to CSRNG software controllable, the functional coverage for ENTROPY_SRC dropped massively from above 96% to around 46%. The main reason for this massive drop is the coverage cross for the different fields of the CONF register. The PR added two new fields to the CONF register, meaning unless we substantially increase the number of seeds run in simulation, we can cover only about a 4th of the possible configuration settings.Since the two new configuration fields directly control a single output bit each and not more, I think we shouldn't spend effort/so much simulation time into getting coverage again to an acceptable level for V2/V2S (90%) and instead not consider these two settings into the crosses.
What is your view @h-filali, @andreaskurth ?
Update:
In addition, also toggle coverage dropped below 90% (currently at 88%). After all these RTL changes, we should probably regenerate the coverage exclusion files.
The text was updated successfully, but these errors were encountered: