Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix PC setup after waking from clock gating in Verilator #380

Open
wants to merge 2 commits into
base: main
Choose a base branch
from

Conversation

robertszczepanski
Copy link
Contributor

@robertszczepanski robertszczepanski commented Jan 16, 2024

This PR addresses issues reported in caliptra-sw#1220 and VeeR-EL2#88.

The TEC_RV_ICG module used in rvdffpcie operates on different logic for Verilator, enable signal is only sampled on negative clock edge. This change removes such exception and makes it possible to correctly simulate clock gating tests in Verilator. It resolves an issue with incorrect PC after waking up from clock gating and ensures proper CPU behavior. In the same time it keeps usage of the dff dedicated for PC related signals.

@bharatpillilli
Copy link
Collaborator

@robertszczepanski - Is this change an enhancement or a bug?

@Nitsirks
Copy link
Contributor

Nitsirks commented Jan 16, 2024

This path appears to work fine in simulation, are we sure this isn't just an FPGA issue?

image

@bharatpillilli
Copy link
Collaborator

@Nitsirks - can u plz paste the waveform too showing the necessary signals?

@bharatpillilli
Copy link
Collaborator

@robertszczepanski - I am inclined to reject this in the current generation of HW as we don't want to take any enhancements at this stage. Plz review the waveform that Mike pasted above and let us know if you have any questions.

@robertszczepanski
Copy link
Contributor Author

Thank you for your feedback @Nitsirks @bharatpillilli.

As pointed out as a reference in chipsalliance/Cores-VeeR-EL2#88, I was testing this using src/integration/test_suites/smoke_test_cg_wdt and I was able to reproduce a bug in the Verilator simulation.

From my understanding it's a bug caused by the usage of TEC_RV_ICG inside the rvdffpcie flip flop. This module samples an enable signal only on the falling edge of the clock. Due to such specified behavior, the first rising edge of the clock doesn't detect the asserted enable signal and din is propagated to dout 1 cycle later.

@Nitsirks can you please point me to the test that you've run in order to verify the described behavior in simulation and get the waveforms, so I can reproduce it?

@bharatpillilli
Copy link
Collaborator

It's the same test in VCS. Mike can correct me if I am wrong.

@Nitsirks
Copy link
Contributor

I ran smoke_test_clk_gating to verify the behavior.

I see why Verilator is failing now. The code in `TEC_RV_ICG is wrong for VERILATOR.

`ifdef VERILATOR
   always @(negedge CK) begin
      en_ff <= enable;
   end
`else
   always @(CK, enable) begin
      if(!CK)
        en_ff = enable;
   end
`endif

The VERILATOR code is modeling a negedge flop, while the real code is modeling a latch. In simulation the enable passes through the open latch to get captured by the first rising edge of the flop. In VERILATOR the enable can't propagate until the clock arrives, so it's 1 clock late.

@bharatpillilli
Copy link
Collaborator

Seems like Verilator is going to do best effort can than guarantee it :-)
image

@robertszczepanski robertszczepanski changed the title Fix PC setup after waking from clock gating Fix PC setup after waking from clock gating in Verilator Jan 18, 2024
Internal-tag: [#53905]
Signed-off-by: Robert Szczepanski <[email protected]>
Internal-tag: [#53905]
Signed-off-by: Robert Szczepanski <[email protected]>
@robertszczepanski
Copy link
Contributor Author

I've removed different RTL for Verilator simulation and uncommented few additional tests that now successfully pass in CI. This resolves issues in simulation, waveform analysis confirms the correct behavior of PC after waking up from clock gating.

PR description has been accordingly updated.

@bharatpillilli
Copy link
Collaborator

@Nitsirks - if you can also confirm that we need these changes and they look good, I will approve/merge the PR.

@Nitsirks
Copy link
Contributor

Nitsirks commented Jan 18, 2024

I think allowing verilator to implement the latch fixed the issue Kiran filed.

Do we know if verilator runs take longer now? The quote above mentioned disabled performance optimizations around latches.

@andreslagarcavilla
Copy link
Collaborator

Worth rebasing for conflict resolution

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants