lhsaid Posted December 24, 2014 Report Share Posted December 24, 2014 When your front door interface is pipelined and targeting multiple slaves where read transaction responses from different slaves may not be in the same order of request transactions, the read response could be all mixed up if multiple treads are issuing read transactions. The issue is that in reg/uvm_reg_map.svh (UVM 1.2, see bellow) call to rw.parent.get_base_response(bus_rsp); used default transaction_id = -1 as opposed to the read request transaction id, which results in popping next response item from the response queue in the get_base_response task which could be wrong response as it is not matching the transaction id !! Example: - tread#1 issued read to slave#1 - then thread#2 issued read to slave#2 ---> now both threads are waiting for response and trying to pop net element available in the response_queue. - slave#2 responded first and its response get put into the response_queue. - thread#1 could be served first and get the response from slave#2 which is meant for thread#2 !!! The solution is at line# 2085 bellow, the call to get_base_response should be: rw.parent.get_base_response(bus_rsp, rw.get_transaction_id()); In file reg/uvm_reg_map.svh: ======================= 1976 task uvm_reg_map::do_bus_read ( ....... 2081 if (adapter.provides_responses) begin 2082 uvm_sequence_item bus_rsp; 2083 uvm_access_e op; 2084 // TODO: need to test for right trans type, if not put back in q 2085 rw.parent.get_base_response(bus_rsp); 2086 adapter.bus2reg(bus_rsp,rw_access); 2087 end 2088 else begin 2089 adapter.bus2reg(bus_req,rw_access); 2090 end ..... 1976 task uvm_reg_map::do_bus_read ( In file: seq/uvm_sequence_base.svh: ============================= 1157 virtual task get_base_response(output uvm_sequence_item response, input int transaction_id = -1); 1158 1159 int queue_size, i; 1160 1161 if (response_queue.size() == 0) 1162 wait (response_queue.size() != 0); 1163 1164 if (transaction_id == -1) begin 1165 response = response_queue.pop_front(); 1166 return; 1167 end 1168 1169 forever begin 1170 queue_size = response_queue.size(); 1171 for (i = 0; i < queue_size; i++) begin 1172 if (response_queue.get_transaction_id() == transaction_id) 1173 begin 1174 $cast(response,response_queue); 1175 response_queue.delete(i); 1176 return; 1177 end 1178 end 1179 wait (response_queue.size() != queue_size); 1180 end 1181 endtask Quote Link to comment Share on other sites More sharing options...
lhsaid Posted December 25, 2014 Author Report Share Posted December 25, 2014 Just realized there is already an active thread about this same issue: Pipelined RAL accesshttp://forums.accellera.org/topic/1159-pipelined-ral-access/ Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.