[BUG] <- Hello world in 64 bit OBI configurations gives an error - master candidate>

[AngelaGonzalezMarino]

Code of Conduct

• I have searched the existing bug issues.
• I am a human engaging in an interpersonal interaction. During this interaction, my words are my own and are not generated. If relevant, I provide links to my sources.

CVA6 commit affected

7878f03

Bug Description

• Hello world in 64 bit OBI configurations gives an error: FIFO popped while empty

[udayaa-init]

@yanicasa The OBI agent is driving the gnt in the start of the simulation as can be seen here and also here it doesn't check for the req.
The gnt is directly provided to the pop signal of the FIFO in the amo buffer, thus failing the test with pop asserted in the empty FIFO.

[AngelaGonzalezMarino]

Thanks @udayaa-init for the explanation

I think the question is whether the core should protect against this, or whether we need to modify the testbench to not send grant when there is no req. @yanicasa what are your thoughts on this?

[yanicasa]

R-3.1.2: The manager shall not de-assert (retract) req until after the last cycle of the address phase.
Request retraction is allowed during reset activation when both manager and subordinate are being reset.
R-3.2: The subordinate shall indicate its readiness to accept the address phase transfer by asserting gnt.
R-3.2.1: The subordinate shall be allowed to assert gnt at any time (even before the corresponding req).
R-3.2.2: The subordinate shall be allowed to de-assert (retract) gnt at any time.

gnt can be tied to 1 if the subordinate is always able to accept a transaction.
The standard allows gnt to be asserted before req, and the address phase transfer only occurs when both req and gnt are high.

Similarly, for the response phase, the transfer occurs when both rvalid and rready are high. (if rready exists)

We have to modify the RTL I think

[AngelaGonzalezMarino]

Agreed, we will udpate it. Thanks for your feedback!