Brandon Kim | dab96f1 | 2021-02-18 11:21:37 -0800 | [diff] [blame^] | 1 | /* |
| 2 | * Copyright 2021 Google LLC |
| 3 | * |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | * |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | * |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
| 16 | |
William A. Kennington III | 7d6fa42 | 2021-02-08 17:04:02 -0800 | [diff] [blame] | 17 | #ifndef PLATFORMS_NEMORA_PORTABLE_NCSI_FSM_H_ |
| 18 | #define PLATFORMS_NEMORA_PORTABLE_NCSI_FSM_H_ |
| 19 | |
| 20 | /* Nemora NC-SI (Finite) State Machine implementation */ |
| 21 | |
| 22 | #include <stdint.h> |
| 23 | |
| 24 | #include "platforms/nemora/portable/ncsi.h" |
| 25 | #include "platforms/nemora/portable/net_types.h" |
| 26 | |
| 27 | /* TODO put this into config somewhere? */ |
| 28 | #define NCSI_FSM_RESTART_DELAY_COUNT 100 |
| 29 | #define NCSI_FSM_RETEST_DELAY_COUNT 100 |
| 30 | |
| 31 | /* The network state is defined as a combination of the NC-SI connection state |
| 32 | * and the network configuration. However the two cannot be decoupled: |
| 33 | * - we cannot DHCP unless the NC-SI connection is up |
| 34 | * - we cannot do the OEM L3/L4 NC-SI configuration unless we have a valid |
| 35 | * network configuration |
| 36 | * |
| 37 | * For additional complexity we cannot get DHCP/ARP responses after the host |
| 38 | * has loaded the Mellanox NIC driver but we want to be able to periodically |
| 39 | * test the NC-SI connection regardless of whether we have network configuration |
| 40 | * (so that flaky cables can be troubleshooted using the host interface). |
| 41 | * |
| 42 | * For this reason there are actually 3 NC-SI finite state machines: |
| 43 | * - L2 configuration (i.e. enabling all available NC-SI channel for passthrough |
| 44 | * RX and TX, although only TX will work after the host loads the NIC driver) |
| 45 | * - L3/L4 configuration (i.e. configuring flow steering for RX traffic that |
| 46 | * matches our IP address and dedicated Nemora port so that we can receive |
| 47 | * Nemora requests even after the host loaded the NIC driver) |
| 48 | * - Connection test (i.e. periodically doing a ping test between the EC and the |
| 49 | * NIC) and also ensuring that L3/L4 configuration parameters have not been |
| 50 | * wiped out) |
| 51 | * |
| 52 | * For good karma, try to keep the state machines as linear as possible (one |
| 53 | * step after the other). |
| 54 | */ |
| 55 | |
| 56 | typedef enum { |
| 57 | // First |
| 58 | NCSI_STATE_L2_CONFIG_BEGIN, |
| 59 | // Actual sequence |
| 60 | NCSI_STATE_RESTART = NCSI_STATE_L2_CONFIG_BEGIN, |
| 61 | NCSI_STATE_CLEAR_0, |
| 62 | NCSI_STATE_CLEAR_0_RESPONSE, |
| 63 | NCSI_STATE_GET_VERSION, |
| 64 | NCSI_STATE_GET_VERSION_RESPONSE, |
| 65 | NCSI_STATE_GET_CAPABILITIES, |
| 66 | NCSI_STATE_GET_CAPABILITIES_RESPONSE, |
| 67 | NCSI_STATE_CLEAR_1, |
| 68 | NCSI_STATE_CLEAR_1_RESPONSE, |
| 69 | NCSI_STATE_RESET_CHANNEL_0, |
| 70 | NCSI_STATE_RESET_CHANNEL_0_RESPONSE, |
| 71 | NCSI_STATE_RESET_CHANNEL_1, |
| 72 | NCSI_STATE_RESET_CHANNEL_1_RESPONSE, |
| 73 | NCSI_STATE_STOPPED, |
| 74 | NCSI_STATE_GET_MAC, |
| 75 | NCSI_STATE_GET_MAC_RESPONSE, |
| 76 | NCSI_STATE_SET_MAC_FILTER_0, |
| 77 | NCSI_STATE_SET_MAC_FILTER_0_RESPONSE, |
| 78 | NCSI_STATE_SET_MAC_FILTER_1, |
| 79 | NCSI_STATE_SET_MAC_FILTER_1_RESPONSE, |
| 80 | NCSI_STATE_ENABLE_CHANNEL_0, |
| 81 | NCSI_STATE_ENABLE_CHANNEL_0_RESPONSE, |
| 82 | NCSI_STATE_ENABLE_CHANNEL_1, |
| 83 | NCSI_STATE_ENABLE_CHANNEL_1_RESPONSE, |
| 84 | NCSI_STATE_ENABLE_TX, |
| 85 | NCSI_STATE_ENABLE_TX_RESPONSE, |
| 86 | // Last |
| 87 | NCSI_STATE_L2_CONFIG_END |
| 88 | } ncsi_l2_config_state_t; |
| 89 | |
| 90 | typedef enum { |
| 91 | // First |
| 92 | NCSI_STATE_L3L4_CONFIG_BEGIN, |
| 93 | // Actual sequence |
| 94 | NCSI_STATE_CONFIG_FILTERS, |
| 95 | // Last |
| 96 | NCSI_STATE_L3L4_CONFIG_END |
| 97 | } ncsi_l3l4_config_state_t; |
| 98 | |
| 99 | typedef enum { |
| 100 | // First |
| 101 | NCSI_STATE_TEST_BEGIN, |
| 102 | // Actual sequence |
| 103 | NCSI_STATE_TEST_PARAMS = NCSI_STATE_TEST_BEGIN, |
| 104 | NCSI_STATE_ECHO, |
| 105 | NCSI_STATE_ECHO_RESPONSE, |
| 106 | NCSI_STATE_CHECK_FILTERS, |
| 107 | NCSI_STATE_CHECK_FILTERS_RESPONSE, |
| 108 | NCSI_STATE_GET_PT_STATS, |
| 109 | NCSI_STATE_GET_PT_STATS_RESPONSE, |
| 110 | NCSI_STATE_GET_LINK_STATUS, |
| 111 | NCSI_STATE_GET_LINK_STATUS_RESPONSE, |
| 112 | // Last |
| 113 | NCSI_STATE_TEST_END |
| 114 | } ncsi_test_state_t; |
| 115 | |
| 116 | typedef struct { |
| 117 | ncsi_l2_config_state_t l2_config_state; |
| 118 | ncsi_l3l4_config_state_t l3l4_config_state; |
| 119 | ncsi_test_state_t test_state; |
| 120 | // Last (OEM) command that was sent. (L3L4 SM only) |
| 121 | // Valid only if l3l4_waiting_response is true. |
| 122 | uint8_t l3l4_command; |
| 123 | // Number of the channel we are currently operating on. (L3L4 SM only) |
| 124 | uint8_t l3l4_channel; |
| 125 | // If true, means the request was sent and we are waiting for response. |
| 126 | bool l3l4_waiting_response; |
| 127 | uint8_t channel_count; |
| 128 | // The re-start and re-test delays ensures that we can flush the DMA |
| 129 | // buffers of potential out-of-sequence NC-SI packets (e.g. from |
| 130 | // packet that may have been received shortly after we timed out on |
| 131 | // them). The re-test delays also reduce the effect of NC-SI |
| 132 | // testing on more useful traffic. |
| 133 | uint8_t restart_delay_count; |
| 134 | uint8_t retest_delay_count; |
| 135 | struct { |
| 136 | uint8_t flags; |
| 137 | uint8_t regid[8]; |
| 138 | } flowsteering[2]; |
| 139 | } ncsi_state_t; |
| 140 | |
| 141 | // Debug variables. |
| 142 | // TODO - Change name to something more meaningful since the NC-SI test |
| 143 | // is not a debug-only feature. |
| 144 | typedef struct { |
| 145 | uint32_t task_count; |
| 146 | uint32_t host_ctrl_flags; |
| 147 | struct { |
| 148 | bool enabled; |
| 149 | bool pending_stop; |
| 150 | bool pending_restart; |
| 151 | bool oem_filter_disable; |
| 152 | bool loopback; |
| 153 | bool mlx_legacy; |
| 154 | uint32_t fail_count; |
| 155 | ncsi_state_t state_that_failed; |
| 156 | uint32_t tx_count; |
| 157 | uint32_t rx_count; |
| 158 | uint32_t tx_error_count; |
| 159 | struct { |
| 160 | uint32_t timeout_count; |
| 161 | uint32_t oversized_count; |
| 162 | uint32_t undersized_count; |
| 163 | uint32_t nack_count; |
| 164 | uint32_t unexpected_size_count; |
| 165 | uint32_t unexpected_type_count; |
| 166 | } rx_error; |
| 167 | struct { |
| 168 | uint32_t runs; |
| 169 | uint8_t ch_under_test; |
| 170 | uint8_t tries; |
| 171 | uint8_t max_tries; // 0 = skip test, 1 = restart on failure, > 1 = retry |
| 172 | struct { |
| 173 | uint8_t tx[NCSI_OEM_ECHO_PATTERN_SIZE]; |
| 174 | uint32_t tx_count; |
| 175 | uint32_t rx_count; |
| 176 | uint32_t bad_rx_count; |
| 177 | uint8_t last_bad_rx[NCSI_OEM_ECHO_PATTERN_SIZE]; |
| 178 | } ping; |
| 179 | } test; |
| 180 | ncsi_passthrough_stats_t pt_stats_be[2]; // big-endian as received from NIC |
| 181 | } ncsi; |
| 182 | } network_debug_t; |
| 183 | |
| 184 | typedef struct { |
| 185 | uint8_t data[ETH_BUFFER_SIZE]; |
| 186 | uint32_t len; // Non-zero when there's a new NC-SI response. |
| 187 | } ncsi_buf_t; |
| 188 | |
| 189 | |
| 190 | #ifdef __cplusplus |
| 191 | extern "C" { |
| 192 | #endif |
| 193 | |
| 194 | ncsi_response_type_t ncsi_fsm_poll_l2_config(ncsi_state_t* ncsi_state, |
| 195 | network_debug_t* network_debug, |
| 196 | ncsi_buf_t* ncsi_buf, |
| 197 | mac_addr_t* mac); |
| 198 | |
| 199 | ncsi_response_type_t ncsi_fsm_poll_l3l4_config(ncsi_state_t* ncsi_state, |
| 200 | network_debug_t* network_debug, |
| 201 | ncsi_buf_t* ncsi_buf, |
| 202 | mac_addr_t* mac, |
| 203 | uint32_t ipv4_addr, |
| 204 | uint16_t rx_port); |
| 205 | |
| 206 | ncsi_response_type_t ncsi_fsm_poll_test(ncsi_state_t* ncsi_state, |
| 207 | network_debug_t* network_debug, |
| 208 | ncsi_buf_t* ncsi_buf, mac_addr_t* mac, |
| 209 | uint32_t ipv4_addr, uint16_t rx_port); |
| 210 | |
| 211 | /* |
| 212 | * Report a global state of the NC-SI connection as a function of the state |
| 213 | * of the 3 finite state machines. |
| 214 | * Note: Additionally for the case where the connection is down it reports |
| 215 | * whether a loopback is inferred. |
| 216 | */ |
| 217 | ncsi_connection_state_t ncsi_fsm_connection_state( |
| 218 | const ncsi_state_t* ncsi_state, const network_debug_t* network_debug); |
| 219 | |
| 220 | /* |
| 221 | * Returns true if we have executed an NC-SI Get OEM Filter command for all |
| 222 | * channels and the flags indicate that it is running in hostless mode. |
| 223 | * This means that we can DHCP/ARP if needed. |
| 224 | * Otherwise returns false. |
| 225 | * |
| 226 | * NOTE: We default to false, if we cannot complete the L2 config state |
| 227 | * machine or the test sequence. |
| 228 | */ |
| 229 | bool ncsi_fsm_is_nic_hostless(const ncsi_state_t* ncsi_state); |
| 230 | |
| 231 | #ifdef __cplusplus |
| 232 | } /* extern "C" */ |
| 233 | #endif |
| 234 | |
| 235 | #endif // PLATFORMS_NEMORA_PORTABLE_NCSI_FSM_H_ |