cva6 icon indicating copy to clipboard operation
cva6 copied to clipboard

AXI UVM issue when integrating HPDCache

Open JeanRochCoulon opened this issue 1 year ago • 7 comments

Is there an existing CVA6 bug for this?

  • [X] I have searched the existing bug issues

Bug Description

When switching WT cache to HPD cache, 4 AXI assertions stop the simulations: image

When commenting the 4 assertions, the hello word test stops on timeout after 300s. When switching the AXI from active to passive, the test passes.

This failure is using vcs-uvm target. The Thales developing branch is hpdcache0

JeanRochCoulon avatar Apr 25 '24 14:04 JeanRochCoulon

When I debugged the 'hello-world' test in the smoke-tests, I found that the problem is originating from the RTL. The bug occurs when we randomize the 'aw_delay,' which indicates the delay between AWVALID and AWREADY.

AEzzejjari avatar Apr 29 '24 15:04 AEzzejjari

Hello,

Thank you for the report.

Can you please send the command line to rerun your test ?

When I debugged one of the ADD assembly tests, I saw that the problem came effectively for a write, but what I saw, is that the HPDcache sent both the address (AW) and the data (W) but it never received a response for it (B). So, in my case it looks more like a problem from the test environment.

cfuguet avatar Apr 29 '24 16:04 cfuguet

When I reviewed the log file, I didn't notice that the agent didn't sent response, but when you mentioned it, I observed the waveform and I realized that when the address is accepted by the agent after the data, the agent does not send a response. Therefore, this is a bug in the agent.

AEzzejjari avatar Apr 29 '24 19:04 AEzzejjari

Ok, yes, that was what I observed too. Let me know if I can help.

cfuguet avatar Apr 29 '24 20:04 cfuguet

Hi @AEzzejjari,

Do you have any news on this issue ? :)

Thanks and regards

cfuguet avatar May 02 '24 13:05 cfuguet

I think @AEzzejjari is oOo Today. Alae has fixed the AXI agent. I am (trying) to bump core-v-verif to CVA6... Not sure to succeed soon, because two PRs are in collision.

JeanRochCoulon avatar May 02 '24 13:05 JeanRochCoulon

Ok great ! No problem, I'm glad that you have found (and fixed) some issues.

cfuguet avatar May 02 '24 13:05 cfuguet

fixed by #2125

JeanRochCoulon avatar May 18 '24 18:05 JeanRochCoulon