Pending NMI, but going to sleep

[silabs-robin]

Component:RTL: For issues in the RTL (e.g. for files in the rtl directory)

Steps to Reproduce

1. My branch, interrupt_assert_fix (hash 05a6d7)
2. tcf_cvverif, assert a_wfi_assert_core_not_ready
3. (See screenshot below)

There is a pending NMI when the core goes to sleep.

I don't know if this is a bug or not, because there is no outstanding transactions at the time (the rvalid has come), and NMI should not(?) be treated as a "locally enabled interrupt".
Note also that ctrl_fsm_cs == SLEEP doesn't 1-to-1 mean that the core is in "sleep mode".
The assert in question was still mid-development when the issue arose, so I don't expect the assert itself to be bug-free.

@silabs-oysteink @Silabs-ArjanB

[Silabs-ArjanB]

@silabs-robin An NMI is not supposed to be covered by the phrase 'locally enabled interrupt'. In itself the above behavior looks okay, but I wonder if we should eliminate this scenario by not entering sleep mode while there is a pending NMI. Now we wait until the LSU is no longer busy and that is one cycle too short to exclude NMIs. @silabs-oysteink let's discuss whether we change the RTL or accept above behavior (I am leaning towards accepting it).

[silabs-robin]

An NMI is not supposed to be covered by the phrase 'locally enabled interrupt'

Why not?
I can't find it in any spec or user manual.

Is it not "locally enabled"?
An action cannot be taken to enable it, so in that sense you could say it has not been "enabled".
But you cannot disable it, so in that sense you could say that it always implicitly is "enabled".

Is it not an "interrupt"?
I think it is an interrupt, because it is called "interrupt" and because of all the interrupt-like behaviors around it (e.g. mcause[31]).

[Silabs-ArjanB]

@silabs-robin Agreed that this is a bug in the RTL and that an NMI shall cause a resume from WFI (and WFE).

[Silabs-ArjanB]

Can assertions as the following be added as well:

(core_sleep_o == 1'b1) -> (instr_rvalid_i == 1'b0);

(core_sleep_o == 1'b1) -> (data_rvalid_i == 1'b0);

(core_sleep_o == 1'b1) -> 'write buffer is empty'

[silabs-robin]

Can assertions as the following be added as well:

The following asserts pass and will be added to core-v-verif's interrupt_assert:

  a_wfi_assert_sleepmode_no_ivalid: assert property (
    core_sleep_o
    |->
    !obi_iside_rvalid
  ) else `uvm_error(info_tag, "TODO");

  a_wfi_assert_sleepmode_no_dvalid: assert property (
    core_sleep_o
    |->
    !obi_dside_rvalid
  ) else `uvm_error(info_tag, "TODO");

  a_wfi_assert_sleepmode_no_wbuf: assert property (
    core_sleep_o
    |->
    (writebufstate == WBUF_EMPTY)
  ) else `uvm_error(info_tag, "TODO");

.obi_iside_rvalid (instr_rvalid_i),
.obi_dside_rvalid (data_rvalid_i),

[Silabs-ArjanB]

Fixed by openhwgroup/cv32e40x#722 (not merged yet to the 40S)

[Silabs-ArjanB]

@silabs-robin Please close this issue if you agree it has been solved.

[silabs-robin]

I will have a look at this and the other issue soon. It is not forgotten.

[Silabs-ArjanB]

@silabs-robin Can you please close this if it has been fixed?

[silabs-robin]

The relevant assertions have now turned to "passing".