Flush pipeline when PMP CSRs are cleared

[marnovandermaas]

Closes: #2193

[mndstrmr]

This is my suggested solution, but I think it's wrong: Doesn't this prevent the else case ever being entered? Indeed was the else case ever correct? It would previously not be entered e.g. on a WRITE to a debug register, which it clearly should be.

I think the if and else should be merged, and the inner if statements merged into their parents, something like the below:

assign csr_pipe_flush = 
    csr_op_en_o == 1'b1 && csr_op_o != CSR_OP_READ && (
    	csr_num_e'(instr_rdata_i[31:20]) == CSR_MSTATUS   ||
     	csr_num_e'(instr_rdata_i[31:20]) == CSR_MIE       ||
      	csr_num_e'(instr_rdata_i[31:20]) == CSR_MSECCFG   ||
      	// To catch all PMPCFG/PMPADDR registers, get the shared top most 7 bits.
      	instr_rdata_i[31:25] == 7'h1D 		  	  ||
    	csr_num_e'(instr_rdata_i[31:20]) == CSR_DCSR      ||
	csr_num_e'(instr_rdata_i[31:20]) == CSR_DPC       ||
	csr_num_e'(instr_rdata_i[31:20]) == CSR_DSCRATCH0 ||
	csr_num_e'(instr_rdata_i[31:20]) == CSR_DSCRATCH1
    );

[nbdd0121]

Very good point. Although it seems that the else case is already not being entered when doing full CSR write?

[GregAC]

I wanted to discuss different ways to approach this so please do not merge this for now.

[GregAC]

I also don't want to make any RTL changes to Ibex until I've sorted the OT earlgrey release branch

[marnovandermaas]

Closing this PR because it probably needs a different solution.