OpenCores
Issue List
bug in the implementation of instructions RTI and RTS #4
Closed abi01shek opened this issue almost 11 years ago
abi01shek commented almost 11 years ago

Desc: when an interrupt occurs, the control saves the return address onto the stack and jumps to the ISR. after executing the ISR, the RTI instruction is executed to return to normal mode of execution, where the stack is popped off to load the return address into the PC

bug: the PC is not loaded with the return address but loaded with 0x0000 when RTI or RTS instructions are used.

Solution: make the following changes into the code

when RTS_C1 => CPU_Next_State <= RTS_C2; AS_Ctrl.Src <= ADDR_SP; SP_Ctrl.Oper <= SP_POP;

when RTS_C2 =>
    CPU_Next_State       <= RTS_C3;
    AS_Ctrl.Src          <= ADDR_SP;
    --Abishek start                                                                                                                                      
    Cache_Ctrl           <= CACHE_OPER1; --line added
    --Abishek end                                                                                                                                        
    -- if this is an RTI, then we need to POP the flags                                                                                                  
    if( SubOp = SOP_RTI )then
      SP_Ctrl.Oper       <= SP_POP;
    end if;
  when RTS_C3 =>
    CPU_Next_State       <= RTS_C4;
    --Abishek start                                                                                                                                      
    Cache_Ctrl           <= CACHE_OPER2;   --line added
    --Cache_Ctrl           <= CACHE_OPER1;  --original line                                                                                                             
    --Abishek end                                                                                                                                        
    -- It doesn't really matter what is on the address bus for RTS, while                                                                                
    --  it does for RTI, so we make this the default                                                                                                     
    AS_Ctrl.Src          <= ADDR_SP;
  when RTS_C4 =>
    CPU_Next_State       <= RTS_C5;
    --Abishek start                                                                                                                                      
    PC_Ctrl.Oper         <= PC_LOAD;   --these lines are added
    PC_Ctrl.Addr         <= Operand2 & Operand1;
    Cache_Ctrl           <= CACHE_OPER2;
    --Abishek end                                                                                                                                        
  when RTS_C5 =>
    CPU_Next_State       <= PIPE_FILL_0;
    --Abishek start                        
    --PC_Ctrl.Oper         <= PC_LOAD; --these lines are commented
    --PC_Ctrl.Addr         <= Operand2 & Operand1;                                                                                                       
    --Abishek end                                                                                                                                        
    if( SubOp = SOP_RTI )then
      CPU_Next_State     <= RTI_C6;
      Cache_Ctrl         <= CACHE_OPER1;
    end if;
  when RTI_C6 =>
    CPU_Next_State       <= PIPE_FILL_1;
    PC_Ctrl.Oper         <= PC_INCR;
    ALU_Ctrl.Oper        <= ALU_RFLG;
    ALU_Ctrl.Data        <= Operand1;
    PC_Ctrl.Oper         <= PC_INCR;
    Int_RTI_D            <= '1';

I have mailed you a detailed document reporting the bug and solution along with the waveforms for your convenience. Thanks Abishek Ramdas

jshamlet was assigned almost 11 years ago
khays commented almost 11 years ago

The Open8 implementation assumes that RAM and register files take two cycles to return valid data. The RAM model used returns data in a single cycle.

khays closed this almost 11 years ago

Assignee
jshamlet
Labels
Bug