Implemented custom WFE instruction #669

silabs-oysteink · 2022-09-22T12:37:15Z

Instruction behaves exactly as WFI, but can wake up to the new input wu_wfe_i in addition to interrupts and debug. WFI will never wake up due to wu_wfe_i.

misa.X is set to 1. Note that this seems to break the ISS step and compare - we might want to wait with this merge until verification can verify this behaviour and come up with a fix that avoids breaking regressions.

Silabs-ArjanB · 2022-09-26T11:09:27Z

rtl/cv32e40x_controller_bypass.sv

@@ -138,7 +138,7 @@ module cv32e40x_controller_bypass import cv32e40x_pkg::*;

  // Stall ID when WFI is active in EX.
  // Prevent load/store following a WFI in the pipeline
-  assign ctrl_byp_o.wfi_stall = (id_ex_pipe_i.sys_en && id_ex_pipe_i.sys_wfi_insn && id_ex_pipe_i.instr_valid);
+  assign ctrl_byp_o.wfi_stall = (id_ex_pipe_i.sys_en && (id_ex_pipe_i.sys_wfi_insn || id_ex_pipe_i.sys_wfe_insn) && id_ex_pipe_i.instr_valid);


wfi_stall -> wfi_wfe_stall

Update comments as well.

Silabs-ArjanB · 2022-09-26T11:16:21Z

rtl/cv32e40x_controller_fsm.sv

@@ -991,7 +998,8 @@ module cv32e40x_controller_fsm import cv32e40x_pkg::*;
  end

  // Wakeup from sleep
-  assign ctrl_fsm_o.wake_from_sleep    = irq_wu_ctrl_i || pending_debug || debug_mode_q;
+  assign ctrl_fsm_o.wake_from_sleep    = irq_wu_ctrl_i || pending_debug || debug_mode_q ||
+                                         (ex_wb_pipe_i.sys_en && ex_wb_pipe_i.sys_wfe_insn && ex_wb_pipe_i.instr_valid && wu_wfe_i); // Only WFE wakes up for wfe_wu_i


(ex_wb_pipe_i.sys_en && ex_wb_pipe_i.sys_wfe_insn && ex_wb_pipe_i.instr_valid && wu_wfe_i) ->(wfe_in_wb && wu_wfe_i)

Silabs-ArjanB · 2022-09-26T11:20:01Z

rtl/cv32e40x_controller_fsm.sv

@@ -991,7 +998,8 @@ module cv32e40x_controller_fsm import cv32e40x_pkg::*;
  end

  // Wakeup from sleep
-  assign ctrl_fsm_o.wake_from_sleep    = irq_wu_ctrl_i || pending_debug || debug_mode_q;
+  assign ctrl_fsm_o.wake_from_sleep    = irq_wu_ctrl_i || pending_debug || debug_mode_q ||
+                                         (ex_wb_pipe_i.sys_en && ex_wb_pipe_i.sys_wfe_insn && ex_wb_pipe_i.instr_valid && wu_wfe_i); // Only WFE wakes up for wfe_wu_i


debug_wfi_no_sleep -> debug_wfi_wfe_no_sleep

Silabs-ArjanB · 2022-09-26T11:23:22Z

rtl/cv32e40x_decoder.sv

@@ -43,6 +43,7 @@ module cv32e40x_decoder import cv32e40x_pkg::*;
  output logic          sys_dret_insn_o,        // Return from debug (M)
  output logic          sys_ecall_insn_o,       // Environment call (syscall) instruction encountered
  output logic          sys_wfi_insn_o,         // Pipeline flush is requested
+  output logic          sys_wfe_insn_o,


Remove above flush comment (not accurate)

Silabs-ArjanB · 2022-09-26T11:26:05Z

rtl/cv32e40x_decoder.sv

  assign sys_mret_insn_o    = decoder_i_ctrl.sys_mret_insn;                     // Only I decoder handles SYS
  assign sys_dret_insn_o    = decoder_i_ctrl.sys_dret_insn;                     // Only I decoder handles SYS
  assign sys_ebrk_insn_o    = decoder_i_ctrl.sys_ebrk_insn;                     // Only I decoder handles SYS
  assign sys_ecall_insn_o   = decoder_i_ctrl.sys_ecall_insn;                    // Only I decoder handles SYS
  assign sys_wfi_insn_o     = decoder_i_ctrl.sys_wfi_insn;                      // Only I decoder handles SYS
+  assign sys_wfe_insn_o     = decoder_x_ctrl.sys_wfe_insn;                      // Only X decoder handles WFE


Update all above comments similarly, e.g. on line 236:

// Only I decoder handles MRET

Silabs-ArjanB · 2022-09-26T11:28:33Z

rtl/cv32e40x_decoder.sv

@@ -219,12 +232,13 @@ module cv32e40x_decoder import cv32e40x_pkg::*;
  assign lsu_size_o         = decoder_ctrl_mux.lsu_size;
  assign lsu_sext_o         = decoder_ctrl_mux.lsu_sext;
  assign lsu_atop_o         = decoder_a_ctrl.lsu_atop;                          // Only A decoder handles atomics
-  assign sys_en             = decoder_i_ctrl.sys_en;                            // Only I decoder handles SYS
+  assign sys_en             = decoder_ctrl_mux.sys_en;


Can you do a synthesis and see if this change led to unwanted timing paths (as that was the reason to use decoder_i_ctrl instead of decoder_ctrl_mux). If the wfe related path shows up, then we should make WFE part if the I decoder instead (although it is very clean how you did it now)

Silabs-ArjanB · 2022-09-26T11:29:46Z

rtl/cv32e40x_x_decoder.sv

+module cv32e40x_x_decoder import cv32e40x_pkg::*;
+  (
+   // from IF/ID pipeline
+   input logic [31:0] instr_rdata_i,


Align types and signal names

Silabs-ArjanB · 2022-09-26T11:35:25Z

rtl/cv32e40x_x_decoder.sv

+    decoder_ctrl_o = DECODER_CTRL_ILLEGAL_INSN;
+    decoder_ctrl_o.illegal_insn = 1'b0;
+
+    unique case (instr_rdata_i[6:0])


Just compare all 32 bits at once against 32'h8C000073 to save some lines of code

Another PR needs to be done for the 40S to add privilege related code as for WFI.

Silabs-ArjanB · 2022-09-26T11:40:04Z

sva/cv32e40x_controller_fsm_sva.sv

  // Bubble is needed to avoid any LSU instructions to go on the bus while handling the WFI, as this
  // could cause the pipeline not to be interruptible when we wake up to an interrupt that should be taken.
  a_wfi_bubbles:
    assert property (@(posedge clk) disable iff (!rst_n)
-                      (ctrl_fsm_cs == SLEEP) |-> !(id_ex_pipe_i.instr_valid))
+                      (ctrl_fsm_cs == SLEEP)


a_wfi_bubbles -> a_wfi_wfe_bubbles

Silabs-ArjanB · 2022-09-26T11:44:23Z

sva/cv32e40x_decoder_sva.sv

@@ -138,10 +138,10 @@ module cv32e40x_decoder_sva
    else `uvm_error("decoder", "CSR related signals driven from unexpected decoder")

  // Check that SYS related signals can be used from I decoder directly (bypassing other decoders)
+  // Note that with the WFE instruction, sys_en may be driven by both I-decoder and X-decoder.


Add similar assertion for sys_wfe_insn

…input pin in addition to interrupts. Signed-off-by: Oystein Knauserud <[email protected]>

…sa.X=1. Signed-off-by: Oystein Knauserud <[email protected]> Cleanup of indentation and removed commented assertion code. Signed-off-by: Oystein Knauserud <[email protected]> Added missing file for custom decoder. Signed-off-by: Oystein Knauserud <[email protected]>

Signed-off-by: Oystein Knauserud <[email protected]>

…coder. SEC clean. Signed-off-by: Oystein Knauserud <[email protected]>

…lean change. Updated some assertions. Signed-off-by: Oystein Knauserud <[email protected]>

Signed-off-by: Oystein Knauserud <[email protected]>

silabs-oysteink added the Component:RTL For issues in the RTL (e.g. for files in the rtl directory) label Sep 22, 2022

silabs-oysteink requested review from Silabs-ArjanB and silabs-oivind September 22, 2022 12:37

Silabs-ArjanB requested changes Sep 26, 2022

View reviewed changes

silabs-oysteink added 5 commits September 27, 2022 16:27

Added WFE instruction. Behaves like WFI, but can wake up to wu_wfe_i …

0db6f25

…input pin in addition to interrupts. Signed-off-by: Oystein Knauserud <[email protected]>

Renamed signals related to wfe to include "wfe" in the signal names.

b1d374f

Signed-off-by: Oystein Knauserud <[email protected]>

Decoding custom WFE instruction in the i_decoder and removed the x_de…

2ff2f93

…coder. SEC clean. Signed-off-by: Oystein Knauserud <[email protected]>

Removed trigger_match as condition for treating WFI/WFI as nop. SEC c…

f173a06

…lean change. Updated some assertions. Signed-off-by: Oystein Knauserud <[email protected]>

silabs-oysteink force-pushed the silabs_oysteink_wfe branch from 61eb839 to f173a06 Compare September 28, 2022 07:39

Fixed typo in property name.

218f2e8

Signed-off-by: Oystein Knauserud <[email protected]>

silabs-oysteink requested a review from Silabs-ArjanB September 28, 2022 08:01

Update cv32e40x_decoder_sva.sv

4f7bd71

Silabs-ArjanB approved these changes Sep 28, 2022

View reviewed changes

Silabs-ArjanB merged commit 707899d into openhwgroup:master Sep 28, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implemented custom WFE instruction #669

Implemented custom WFE instruction #669

silabs-oysteink commented Sep 22, 2022

Silabs-ArjanB Sep 26, 2022

Silabs-ArjanB Sep 26, 2022

Silabs-ArjanB Sep 26, 2022

Silabs-ArjanB Sep 26, 2022

Silabs-ArjanB Sep 26, 2022

Silabs-ArjanB Sep 26, 2022

Silabs-ArjanB Sep 26, 2022

Silabs-ArjanB Sep 26, 2022

Silabs-ArjanB Sep 26, 2022

Silabs-ArjanB Sep 26, 2022

Silabs-ArjanB Sep 26, 2022

Silabs-ArjanB Sep 26, 2022

Implemented custom WFE instruction #669

Implemented custom WFE instruction #669

Conversation

silabs-oysteink commented Sep 22, 2022

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment