[libre-riscv-dev] [Bug 325] create POWER9 TRAP pipeline

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=325

--- Comment #71 from Cole Poirier <colepoirier at gmail.com> ---
(In reply to Luke Kenneth Casson Leighton from comment #68)
> actually... cole, can you do the same thing as i just did with
> msr_check_pr() create a function called trap(m, addr, nia)?
> 
> except this time it is a member function of the class, so trap(self, m,
> addr, nia)
> 
> and just as you see i redid comb=m.d.comb in msr_check_pr, repeat/copy the
> convenience variables that allow you to set self.o.nia etc etc
> 
> the reason for making this function is because you can see it is used twice
> already.
> 
> also i am thinking it may be a good idea to add the address (0x700, etc)
> actually to the trap_input_record and pass it in from the *decode* phase.

Sorry, I'm very shaky on the dynamic input records and how they work with the
decoder.

One difference between OP_TRAP and OP_SC is that the end of the OP_TRAP
function is:

```
# take a copy of the current PC in SRR0
comb += srr0_o.data.eq(cia_i)   # old PC
comb += srr0_o.ok.eq(1)
```
While the end of the OP_SC function is:

```
# and store the (next-after-return) PC in SRR0
comb += srr0_o.data.eq(cia_i+4) # addr to begin from on return
comb += srr0_o.ok.eq(1)
```

So should the function be:

```
def trap(self, m, addr, nia_o, srr0_o, trap_addr):
    comb = m.d.comb
    comb += nia_o.data.eq(trap_addr)
    comb += nia_o.ok.eq(1)

    comb += srr0_o.data.eq(addr) # addr to begin from on return
    comb += srr0_o.ok.eq(1)
```

The below is where my shakiness on the CompXOpSubset and the decoder starts to
confuse me. If you can briefly explain that to me here, I can use this when
doing the documentation after the tape-out deadline.

> that would allow us to do exceptions in a micro-code fashion, and also when
> it comes to illegal instructions or privileged ones, in the *decode* phase
> (PowerDecoder2) we *CHANGE* the opcode from whatever it is from the incoming
> instruction, into an OP_TRAP, set the trap address to 0x700 (or whatever)
> manually, and let the trap pipeline execute it.

(In reply to Luke Kenneth Casson Leighton from comment #69)
> see https://bugs.libre-soc.org/show_bug.cgi?id=325#c8
> 
> there, the only difference between OP_TRAP and that code is: one of the bits
> of SRR is changed (to indicate "priv execution")
> 
> we can pass that in as a flag to CompTrapOpSubset, and change the
> instruction to OP_TRAP.
> 
> illegal instruction trap happens. ta-daaa

-- 
You are receiving this mail because:
You are on the CC list for the bug.