// Copyright lowRISC contributors (OpenTitan project). // Licensed under the Apache License, Version 2.0, see LICENSE for details. // SPDX-License-Identifier: Apache-2.0 // // This module implements the ping mechanism. Once enabled, this module uses an // LFSR-based PRNG to // // a) determine the next peripheral index to be pinged (can be an alert receiver or an // escalation sender). If it is detected that a particular peripheral is disabled, // another index will be drawn from the PRNG. // // b) determine the amount of pause cycles to wait before pinging the peripheral selected in a). // // Once the ping timer waited for the amount of pause cycles determined in b), it asserts // the ping enable signal of the peripheral determined in a). If that peripheral does // not respond within the ping timeout window, an internal alert will be raised. // // Further, if a spurious ping_ok signal is detected (i.e., a ping ok that has not been // requested), the ping timer will also raise an internal alert. // `include "prim_assert.sv" module alert_handler_ping_timer import alert_handler_pkg::*; #( // Compile time random constants, to be overridden by topgen. parameter lfsr_seed_t RndCnstLfsrSeed = RndCnstLfsrSeedDefault, parameter lfsr_perm_t RndCnstLfsrPerm = RndCnstLfsrPermDefault, // Enable this for DV, disable this for long LFSRs in FPV parameter bit MaxLenSVA = 1'b1, // Can be disabled in cases where entropy // inputs are unused in order to not distort coverage // (the SVA will be unreachable in such cases) parameter bit LockupSVA = 1'b1 ) ( input clk_i, input rst_ni, output logic edn_req_o, // request to EDN input edn_ack_i, // ack from EDN input [LfsrWidth-1:0] edn_data_i, // from EDN input en_i, // enable ping testing input [NAlerts-1:0] alert_ping_en_i, // determines which alerts to ping input [PING_CNT_DW-1:0] ping_timeout_cyc_i, // timeout in cycles input [PING_CNT_DW-1:0] wait_cyc_mask_i, // mask to shorten the counters in DV / FPV output logic [NAlerts-1:0] alert_ping_req_o, // request to alert receivers output logic [N_ESC_SEV-1:0] esc_ping_req_o, // request to esc senders input [NAlerts-1:0] alert_ping_ok_i, // response from alert receivers input [N_ESC_SEV-1:0] esc_ping_ok_i, // response from esc senders output logic alert_ping_fail_o, // any of the alert receivers failed output logic esc_ping_fail_o // any of the esc senders failed ); localparam int unsigned IdDw = $clog2(NAlerts); // Entropy reseeding is triggered every time this counter expires. // The expected wait time between pings is 2**(PING_CNT_DW-1) on average. // We do not need to reseed the LFSR very often, and the constant below is chosen // such that on average the LFSR is reseeded every 16th ping. localparam int unsigned ReseedLfsrExtraBits = 3; localparam int unsigned ReseedLfsrWidth = PING_CNT_DW + ReseedLfsrExtraBits; // The number of bits needed for an index into the esc senders localparam int unsigned EscSenderIdxWidth = $clog2(N_ESC_SEV); // A few smoke checks for the DV mask: // 1) make sure the value is a right-aligned mask. // this can be done by checking that mask+1 is a power of 2. // 2) also make sure that the value is always >= 0x7. `ASSERT(WaitCycMaskMin_A, wait_cyc_mask_i >= 'h7) `ASSERT(WaitCycMaskIsRightAlignedMask_A, $onehot(32'(wait_cyc_mask_i) + 1)) //////////////////// // Reseed counter // //////////////////// logic reseed_en; logic [ReseedLfsrWidth-1:0] reseed_timer_d, reseed_timer_q; assign reseed_timer_d = (reseed_timer_q > '0) ? reseed_timer_q - 1'b1 : (reseed_en) ? {wait_cyc_mask_i, {ReseedLfsrExtraBits{1'b1}}} : '0; assign edn_req_o = (reseed_timer_q == '0); assign reseed_en = edn_req_o & edn_ack_i; always_ff @(posedge clk_i or negedge rst_ni) begin : p_regs if (!rst_ni) begin reseed_timer_q <= '0; end else begin reseed_timer_q <= reseed_timer_d; end end /////////////////////////// // Tandem LFSR Instances // /////////////////////////// logic cnt_set, lfsr_err; logic [LfsrWidth-1:0] entropy; logic [PING_CNT_DW + IdDw - 1:0] lfsr_state; assign entropy = (reseed_en) ? edn_data_i[LfsrWidth-1:0] : '0; // SEC_CM: PING_TIMER.LFSR.REDUN // We employ two redundant LFSRs to guard against FI attacks. // If any of the two is glitched and the two LFSR states do not agree, // the FSM below is moved into a terminal error state and all ping alerts // are permanently asserted. prim_double_lfsr #( .LfsrDw ( LfsrWidth ), .EntropyDw ( LfsrWidth ), .StateOutDw ( PING_CNT_DW + IdDw ), .DefaultSeed ( RndCnstLfsrSeed ), .StatePermEn ( 1'b1 ), .StatePerm ( RndCnstLfsrPerm ), .MaxLenSVA ( MaxLenSVA ), .LockupSVA ( LockupSVA ), .ExtSeedSVA ( 1'b0 ), // ext seed is unused .EnableAlertTriggerSVA ( 1'b0 ) ) u_prim_double_lfsr ( .clk_i, .rst_ni, .seed_en_i ( 1'b0 ), .seed_i ( '0 ), .lfsr_en_i ( reseed_en || cnt_set ), .entropy_i ( entropy ), .state_o ( lfsr_state ), .err_o ( lfsr_err ) ); logic [IdDw-1:0] id_to_ping_d, id_to_ping_q; // The subtraction below ensures that the alert ID is always in range. If // all alerts are enabled, an alert ID drawn in this way will always be // valid. This comes at the cost of a bias towards certain alert IDs that // will be pinged twice as often on average - but it ensures that we have // less alert IDs that need to be skipped since they are invalid. assign id_to_ping_d = (lfsr_state[PING_CNT_DW +: IdDw] >= NAlerts) ? lfsr_state[PING_CNT_DW +: IdDw] - NAlerts : lfsr_state[PING_CNT_DW +: IdDw]; // we need to hold the ID stable while the ping is ongoing since this will result in // spurious ping responses otherwise. always_ff @(posedge clk_i or negedge rst_ni) begin : p_id_reg if (!rst_ni) begin id_to_ping_q <= '0; end else begin if (cnt_set) begin id_to_ping_q <= id_to_ping_d; end end end // align the enable mask with powers of two for the indexing operation below. logic [2**IdDw-1:0] enable_mask; assign enable_mask = (2**IdDw)'(alert_ping_en_i); // check if the randomly drawn alert ID is actually valid and the alert is enabled logic id_vld; assign id_vld = enable_mask[id_to_ping_q]; ////////////////////////////////// // Escalation Counter Instances // ////////////////////////////////// // As opposed to the alert ID, the escalation sender ID to be pinged is not drawn at random. // Rather, we cycle through the escalation senders one by one in a deterministic fashion. // This allows us to provide guarantees needed for the ping timeout / auto escalation feature // implemented at the escalation receiver side. // // In particular, with N_ESC_SEV escalation senders in the design, we can guarantee // that each escalation channel will be pinged at least once every // // N_ESC_SEV x (NUM_WAIT_COUNT + NUM_TIMEOUT_COUNT) x 2**PING_CNT_DW // // cycles - independently of the reseeding operation. // // - N_ESC_SEV: # escalation channels to ping. // - NUM_WAIT_COUNT: # wait counts between subsequent escalation channel pings. // - NUM_TIMEOUT_COUNT: # timeout counts between subsequent escalation channel pings. // - 2**PING_CNT_DW: # maximum counter value. // // This guarantee is used inside the escalation receivers to monitor the pings sent out by the // alert handler. I.e., once the alert handler has started to send out pings, each escalation // receiver employs a timeout window within which it expects the next ping to arrive. If // escalation pings cease to arrive at an escalation receiver for any reason, this will // automatically trigger the associated escalation countermeasure. // // In order to have enough margin, the escalation receiver timeout counters use a threshold that // is 4x higher than the value calculated above. With N_ESC_SEV = 4, PING_CNT_DW = 16 and // NUM_WAIT_COUNT = NUM_TIMEOUT_COUNT = 2 this amounts to a 22bit timeout threshold. // // We employ two redundant counters to guard against FI attacks. // If any of the two is glitched and the two counter states do not agree, // the FSM below is moved into a terminal error state and all ping alerts // are permanently asserted. logic esc_cnt_en, esc_cnt_clr, esc_cnt_error; logic [EscSenderIdxWidth-1:0] esc_cnt; assign esc_cnt_clr = (esc_cnt >= EscSenderIdxWidth'(N_ESC_SEV-1)) && esc_cnt_en; // SEC_CM: PING_TIMER.CTR.REDUN prim_count #( .Width(EscSenderIdxWidth), // The alert handler behaves differently than other comportable IP. I.e., instead of sending out // an alert signal, this condition is handled internally in the alert handler. .EnableAlertTriggerSVA(0), // Pass a parameter to disable coverage for some assertions that are unreachable because set_i // and decr_en_i are tied to zero. .PossibleActions(prim_count_pkg::Clr | prim_count_pkg::Incr) ) u_prim_count_esc_cnt ( .clk_i, .rst_ni, .clr_i(esc_cnt_clr), .set_i(1'b0), .set_cnt_i('0), .incr_en_i(esc_cnt_en), .decr_en_i(1'b0), .step_i(EscSenderIdxWidth'(1)), .commit_i(1'b1), .cnt_o(esc_cnt), .cnt_after_commit_o(), .err_o(esc_cnt_error) ); ///////////////////////////// // Timer Counter Instances // ///////////////////////////// // We employ two redundant counters to guard against FI attacks. // If any of the two is glitched and the two counter states do not agree, // the FSM below is moved into a terminal error state and all ping alerts // are permanently asserted. logic [PING_CNT_DW-1:0] cnt, cnt_setval; logic wait_cnt_set, timeout_cnt_set, timer_expired, cnt_error; assign timer_expired = (cnt == '0); assign cnt_set = wait_cnt_set || timeout_cnt_set; // SEC_CM: PING_TIMER.CTR.REDUN prim_count #( .Width(PING_CNT_DW), // The alert handler behaves differently than other comportable IP. I.e., instead of sending out // an alert signal, this condition is handled internally in the alert handler. .EnableAlertTriggerSVA(0), // Pass a parameter to disable coverage for some assertions that are unreachable because clr_i // and incr_en_i are tied to zero. .PossibleActions(prim_count_pkg::Set | prim_count_pkg::Decr) ) u_prim_count_cnt ( .clk_i, .rst_ni, .clr_i(1'b0), .set_i(cnt_set), .set_cnt_i(cnt_setval), .incr_en_i(1'b0), .decr_en_i(1'b1), // we are counting down here. .step_i(PING_CNT_DW'(1'b1)), .commit_i(1'b1), .cnt_o(cnt), .cnt_after_commit_o(), .err_o(cnt_error) ); // the constant offset ensures a minimum cycle spacing between pings. logic unused_bits; logic [PING_CNT_DW-1:0] wait_cyc; assign wait_cyc = (lfsr_state[PING_CNT_DW-1:0] | PING_CNT_DW'(3'b100)); assign unused_bits = lfsr_state[2]; // note that the masks are used for DV/FPV only in order to reduce the state space. assign cnt_setval = (wait_cnt_set) ? (wait_cyc & wait_cyc_mask_i) : ping_timeout_cyc_i; //////////////////////////// // Ping and Timeout Logic // //////////////////////////// logic alert_ping_en, esc_ping_en; logic spurious_alert_ping, spurious_esc_ping; // generate ping enable vector assign alert_ping_req_o = NAlerts'(alert_ping_en) << id_to_ping_q; assign esc_ping_req_o = N_ESC_SEV'(esc_ping_en) << esc_cnt; // under normal operation, these signals should never be asserted. // we place hand instantiated buffers here such that these signals are not // optimized away during synthesis (these buffers will receive a keep or size_only // attribute in our Vivado and DC synthesis flows). prim_buf u_prim_buf_spurious_alert_ping ( .in_i(|(alert_ping_ok_i & ~alert_ping_req_o)), .out_o(spurious_alert_ping) ); prim_buf u_prim_buf_spurious_esc_ping ( .in_i(|(esc_ping_ok_i & ~esc_ping_req_o)), .out_o(spurious_esc_ping) ); // SEC_CM: PING_TIMER.FSM.SPARSE // Encoding generated with: // $ ./util/design/sparse-fsm-encode.py -d 5 -m 6 -n 9 \ // -s 728582219 --language=sv // // Hamming distance histogram: // // 0: -- // 1: -- // 2: -- // 3: -- // 4: -- // 5: |||||||||||||||||||| (60.00%) // 6: ||||||||||||| (40.00%) // 7: -- // 8: -- // 9: -- // // Minimum Hamming distance: 5 // Maximum Hamming distance: 6 // Minimum Hamming weight: 2 // Maximum Hamming weight: 6 // localparam int StateWidth = 9; typedef enum logic [StateWidth-1:0] { InitSt = 9'b011001011, AlertWaitSt = 9'b110000000, AlertPingSt = 9'b101110001, EscWaitSt = 9'b010110110, EscPingSt = 9'b000011101, FsmErrorSt = 9'b101101110 } state_e; state_e state_d, state_q; always_comb begin : p_fsm // default state_d = state_q; wait_cnt_set = 1'b0; timeout_cnt_set = 1'b0; esc_cnt_en = 1'b0; alert_ping_en = 1'b0; esc_ping_en = 1'b0; // this captures spurious ping responses alert_ping_fail_o = spurious_alert_ping; esc_ping_fail_o = spurious_esc_ping; unique case (state_q) // wait until activated // we never return to this state // once activated! InitSt: begin if (en_i) begin state_d = AlertWaitSt; wait_cnt_set = 1'b1; end end // wait for random amount of cycles AlertWaitSt: begin if (timer_expired) begin state_d = AlertPingSt; timeout_cnt_set = 1'b1; end end // SEC_CM: ALERT_RX.INTERSIG.BKGN_CHK // send out an alert ping request and wait for a ping // response or a ping timeout (whatever comes first). // if the alert ID is not valid, we drop the request and // proceed to the next ping. AlertPingSt: begin alert_ping_en = id_vld; if (timer_expired || |(alert_ping_ok_i & alert_ping_req_o) || !id_vld) begin state_d = EscWaitSt; wait_cnt_set = 1'b1; if (timer_expired) begin alert_ping_fail_o = 1'b1; end end end // wait for random amount of cycles EscWaitSt: begin if (timer_expired) begin state_d = EscPingSt; timeout_cnt_set = 1'b1; end end // SEC_CM: ESC_TX.INTERSIG.BKGN_CHK // send out an escalation ping request and wait for a ping // response or a ping timeout (whatever comes first) EscPingSt: begin esc_ping_en = 1'b1; if (timer_expired || |(esc_ping_ok_i & esc_ping_req_o)) begin state_d = AlertWaitSt; wait_cnt_set = 1'b1; esc_cnt_en = 1'b1; if (timer_expired) begin esc_ping_fail_o = 1'b1; end end end // SEC_CM: PING_TIMER.FSM.LOCAL_ESC // terminal FSM error state. // if we for some reason end up in this state (e.g. malicious glitching) // we are going to assert both ping fails continuously FsmErrorSt: begin alert_ping_fail_o = 1'b1; esc_ping_fail_o = 1'b1; end default: begin state_d = FsmErrorSt; alert_ping_fail_o = 1'b1; esc_ping_fail_o = 1'b1; end endcase // SEC_CM: PING_TIMER.FSM.LOCAL_ESC // if the two LFSR or counter states do not agree, // we move into the terminal state. if (lfsr_err || cnt_error || esc_cnt_error) begin state_d = FsmErrorSt; alert_ping_fail_o = 1'b1; esc_ping_fail_o = 1'b1; end end /////////////////// // FSM Registers // /////////////////// // The alert handler behaves differently than other comportable IP. I.e., instead of sending out // an alert signal, this condition is handled internally in the alert handler. The // EnableAlertTriggerSVA parameter is therefore set to 0. `PRIM_FLOP_SPARSE_FSM(u_state_regs, state_d, state_q, state_e, InitSt, clk_i, rst_ni, 0) //////////////// // Assertions // //////////////// // make sure the ID width is within bounds. `ASSERT_INIT(MaxIdDw_A, IdDw <= (LfsrWidth - PING_CNT_DW)) // only one module is pinged at a time. `ASSERT(PingOH0_A, $onehot0({alert_ping_req_o, esc_ping_req_o})) // we should never get into the ping state without knowing which module to ping. `ASSERT(AlertPingOH_A, alert_ping_en |-> $onehot(alert_ping_req_o)) `ASSERT(EscPingOH_A, esc_ping_en |-> $onehot(esc_ping_req_o)) endmodule : alert_handler_ping_timer