Condition dcache_ctrl to MMU_PRESENT #1568
Conversation
When MMU is not present ldst does not request address translation, that's why related dcache_ctrl is not useful. Condition it to MMU_PRESENT to increase code coverage and decrease the gate count.
|
✔️ successful run, report available here. |
|
@AEzzejjari can you approve the PR ? |
|
Why is only the cache_ctrl for miss port 0 tied to |
There was a problem hiding this comment.
I think it would be better to add the condition as follows
for (genvar k = 0; k < NumPorts - 1; k++) begin : gen_rd_ports
// set these to high prio ports
if (K != 0 || MMU_PRESENT) begin
assign rd_prio[k] = 1'b1;
wt_dcache_ctrl #(
.CVA6Cfg(CVA6Cfg),
.RdTxId (RdAmoTxId)
) i_wt_dcache_ctrl (
.clk_i (clk_i),
.rst_ni (rst_ni),
.cache_en_i (cache_en),
// reqs from core
.req_port_i (req_ports_i[k]),
.req_port_o (req_ports_o[k]),
// miss interface
.miss_req_o (miss_req[k]),
.miss_ack_i (miss_ack[k]),
.miss_we_o (miss_we[k]),
.miss_wdata_o (miss_wdata[k]),
.miss_wuser_o (miss_wuser[k]),
.miss_vld_bits_o(miss_vld_bits_o[k]),
.miss_paddr_o (miss_paddr[k]),
.miss_nc_o (miss_nc[k]),
.miss_size_o (miss_size[k]),
.miss_id_o (miss_id[k]),
.miss_replay_i (miss_replay[k]),
.miss_rtrn_vld_i(miss_rtrn_vld[k]),
// used to detect readout mux collisions
.wr_cl_vld_i (wr_cl_vld),
// cache mem interface
.rd_tag_o (rd_tag[k]),
.rd_idx_o (rd_idx[k]),
.rd_off_o (rd_off[k]),
.rd_req_o (rd_req[k]),
.rd_tag_only_o (rd_tag_only[k]),
.rd_ack_i (rd_ack[k]),
.rd_data_i (rd_data),
.rd_user_i (rd_user),
.rd_vld_bits_i (rd_vld_bits),
.rd_hit_oh_i (rd_hit_oh)
);
end else begin
assign rd_prio[0] = 1'b0;
assign req_ports_o[0] = '0;
assign miss_req[0] = 1'b0;
assign miss_we[0] = 1'b0;
assign miss_wdata[0] = {{riscv::XLEN}{1'b0}};
assign miss_wuser[0] = {{DCACHE_USER_WIDTH}{1'b0}};
assign miss_vld_bits_o[0] = {{DCACHE_SET_ASSOC}{1'b0}};
assign miss_paddr[0] = {{riscv::PLEN}{1'b0}};
assign miss_nc[0] = 1'b0;
assign miss_size[0] = 3'b0;
assign miss_id[0] = {{CACHE_ID_WIDTH}{1'b0}};
assign rd_tag[0] = {{DCACHE_TAG_WIDTH}{1'b0}};
assign rd_idx[0] = {{DCACHE_CL_IDX_WIDTH}{1'b0}};
assign rd_off[0] = {{DCACHE_OFFSET_WIDTH}{1'b0}};
assign rd_req[0] = 1'b0;
assign rd_tag_only[0] = 1'b0;
end
|
@Moschn I do not understand your comment because to my mind dcache_ctrl[0] was dedicated to MMU, [1] to LOAD and [2] to STORE. If MMU is not PRESENT, we cannot disable all configurations but only [0]. Am I wrong ? |
|
Ah you probably are correct. I did not look at the details of the cache controller, I just looked at the diff in this PR, and I did not understand why only a part of it is disabled. My mistake. Maybe it would be best to add this explanation in a comment in the source code to help people like me :) |
|
Anyway, thank for your review @Moschn, I have added a comment ! |
|
❌ failed run, report available here. |
|
✔️ successful run, report available here. |
When MMU is not present ldst does not request address translation, that's why related dcache_ctrl is not useful. Condition it to MMU_PRESENT to increase code coverage and decrease the gate count.