ncsid: Import from gBMC
This is the initial code drop from gBMC.
Google-Bug-Id: 179618516
Upstream: 1e71af914bc8c54d8b91d0a1cf377e2696713c2f
Change-Id: Ic653e8271dacd205e04f2bc713071ef2ec5936a4
Signed-off-by: William A. Kennington III <wak@google.com>
diff --git a/ncsid/src/platforms/nemora/portable/ncsi_fsm.h b/ncsid/src/platforms/nemora/portable/ncsi_fsm.h
new file mode 100644
index 0000000..9ab2d82
--- /dev/null
+++ b/ncsid/src/platforms/nemora/portable/ncsi_fsm.h
@@ -0,0 +1,219 @@
+#ifndef PLATFORMS_NEMORA_PORTABLE_NCSI_FSM_H_
+#define PLATFORMS_NEMORA_PORTABLE_NCSI_FSM_H_
+
+/* Nemora NC-SI (Finite) State Machine implementation */
+
+#include <stdint.h>
+
+#include "platforms/nemora/portable/ncsi.h"
+#include "platforms/nemora/portable/net_types.h"
+
+/* TODO put this into config somewhere? */
+#define NCSI_FSM_RESTART_DELAY_COUNT 100
+#define NCSI_FSM_RETEST_DELAY_COUNT 100
+
+/* The network state is defined as a combination of the NC-SI connection state
+ * and the network configuration. However the two cannot be decoupled:
+ * - we cannot DHCP unless the NC-SI connection is up
+ * - we cannot do the OEM L3/L4 NC-SI configuration unless we have a valid
+ * network configuration
+ *
+ * For additional complexity we cannot get DHCP/ARP responses after the host
+ * has loaded the Mellanox NIC driver but we want to be able to periodically
+ * test the NC-SI connection regardless of whether we have network configuration
+ * (so that flaky cables can be troubleshooted using the host interface).
+ *
+ * For this reason there are actually 3 NC-SI finite state machines:
+ * - L2 configuration (i.e. enabling all available NC-SI channel for passthrough
+ * RX and TX, although only TX will work after the host loads the NIC driver)
+ * - L3/L4 configuration (i.e. configuring flow steering for RX traffic that
+ * matches our IP address and dedicated Nemora port so that we can receive
+ * Nemora requests even after the host loaded the NIC driver)
+ * - Connection test (i.e. periodically doing a ping test between the EC and the
+ * NIC) and also ensuring that L3/L4 configuration parameters have not been
+ * wiped out)
+ *
+ * For good karma, try to keep the state machines as linear as possible (one
+ * step after the other).
+ */
+
+typedef enum {
+ // First
+ NCSI_STATE_L2_CONFIG_BEGIN,
+ // Actual sequence
+ NCSI_STATE_RESTART = NCSI_STATE_L2_CONFIG_BEGIN,
+ NCSI_STATE_CLEAR_0,
+ NCSI_STATE_CLEAR_0_RESPONSE,
+ NCSI_STATE_GET_VERSION,
+ NCSI_STATE_GET_VERSION_RESPONSE,
+ NCSI_STATE_GET_CAPABILITIES,
+ NCSI_STATE_GET_CAPABILITIES_RESPONSE,
+ NCSI_STATE_CLEAR_1,
+ NCSI_STATE_CLEAR_1_RESPONSE,
+ NCSI_STATE_RESET_CHANNEL_0,
+ NCSI_STATE_RESET_CHANNEL_0_RESPONSE,
+ NCSI_STATE_RESET_CHANNEL_1,
+ NCSI_STATE_RESET_CHANNEL_1_RESPONSE,
+ NCSI_STATE_STOPPED,
+ NCSI_STATE_GET_MAC,
+ NCSI_STATE_GET_MAC_RESPONSE,
+ NCSI_STATE_SET_MAC_FILTER_0,
+ NCSI_STATE_SET_MAC_FILTER_0_RESPONSE,
+ NCSI_STATE_SET_MAC_FILTER_1,
+ NCSI_STATE_SET_MAC_FILTER_1_RESPONSE,
+ NCSI_STATE_ENABLE_CHANNEL_0,
+ NCSI_STATE_ENABLE_CHANNEL_0_RESPONSE,
+ NCSI_STATE_ENABLE_CHANNEL_1,
+ NCSI_STATE_ENABLE_CHANNEL_1_RESPONSE,
+ NCSI_STATE_ENABLE_TX,
+ NCSI_STATE_ENABLE_TX_RESPONSE,
+ // Last
+ NCSI_STATE_L2_CONFIG_END
+} ncsi_l2_config_state_t;
+
+typedef enum {
+ // First
+ NCSI_STATE_L3L4_CONFIG_BEGIN,
+ // Actual sequence
+ NCSI_STATE_CONFIG_FILTERS,
+ // Last
+ NCSI_STATE_L3L4_CONFIG_END
+} ncsi_l3l4_config_state_t;
+
+typedef enum {
+ // First
+ NCSI_STATE_TEST_BEGIN,
+ // Actual sequence
+ NCSI_STATE_TEST_PARAMS = NCSI_STATE_TEST_BEGIN,
+ NCSI_STATE_ECHO,
+ NCSI_STATE_ECHO_RESPONSE,
+ NCSI_STATE_CHECK_FILTERS,
+ NCSI_STATE_CHECK_FILTERS_RESPONSE,
+ NCSI_STATE_GET_PT_STATS,
+ NCSI_STATE_GET_PT_STATS_RESPONSE,
+ NCSI_STATE_GET_LINK_STATUS,
+ NCSI_STATE_GET_LINK_STATUS_RESPONSE,
+ // Last
+ NCSI_STATE_TEST_END
+} ncsi_test_state_t;
+
+typedef struct {
+ ncsi_l2_config_state_t l2_config_state;
+ ncsi_l3l4_config_state_t l3l4_config_state;
+ ncsi_test_state_t test_state;
+ // Last (OEM) command that was sent. (L3L4 SM only)
+ // Valid only if l3l4_waiting_response is true.
+ uint8_t l3l4_command;
+ // Number of the channel we are currently operating on. (L3L4 SM only)
+ uint8_t l3l4_channel;
+ // If true, means the request was sent and we are waiting for response.
+ bool l3l4_waiting_response;
+ uint8_t channel_count;
+ // The re-start and re-test delays ensures that we can flush the DMA
+ // buffers of potential out-of-sequence NC-SI packets (e.g. from
+ // packet that may have been received shortly after we timed out on
+ // them). The re-test delays also reduce the effect of NC-SI
+ // testing on more useful traffic.
+ uint8_t restart_delay_count;
+ uint8_t retest_delay_count;
+ struct {
+ uint8_t flags;
+ uint8_t regid[8];
+ } flowsteering[2];
+} ncsi_state_t;
+
+// Debug variables.
+// TODO - Change name to something more meaningful since the NC-SI test
+// is not a debug-only feature.
+typedef struct {
+ uint32_t task_count;
+ uint32_t host_ctrl_flags;
+ struct {
+ bool enabled;
+ bool pending_stop;
+ bool pending_restart;
+ bool oem_filter_disable;
+ bool loopback;
+ bool mlx_legacy;
+ uint32_t fail_count;
+ ncsi_state_t state_that_failed;
+ uint32_t tx_count;
+ uint32_t rx_count;
+ uint32_t tx_error_count;
+ struct {
+ uint32_t timeout_count;
+ uint32_t oversized_count;
+ uint32_t undersized_count;
+ uint32_t nack_count;
+ uint32_t unexpected_size_count;
+ uint32_t unexpected_type_count;
+ } rx_error;
+ struct {
+ uint32_t runs;
+ uint8_t ch_under_test;
+ uint8_t tries;
+ uint8_t max_tries; // 0 = skip test, 1 = restart on failure, > 1 = retry
+ struct {
+ uint8_t tx[NCSI_OEM_ECHO_PATTERN_SIZE];
+ uint32_t tx_count;
+ uint32_t rx_count;
+ uint32_t bad_rx_count;
+ uint8_t last_bad_rx[NCSI_OEM_ECHO_PATTERN_SIZE];
+ } ping;
+ } test;
+ ncsi_passthrough_stats_t pt_stats_be[2]; // big-endian as received from NIC
+ } ncsi;
+} network_debug_t;
+
+typedef struct {
+ uint8_t data[ETH_BUFFER_SIZE];
+ uint32_t len; // Non-zero when there's a new NC-SI response.
+} ncsi_buf_t;
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+ncsi_response_type_t ncsi_fsm_poll_l2_config(ncsi_state_t* ncsi_state,
+ network_debug_t* network_debug,
+ ncsi_buf_t* ncsi_buf,
+ mac_addr_t* mac);
+
+ncsi_response_type_t ncsi_fsm_poll_l3l4_config(ncsi_state_t* ncsi_state,
+ network_debug_t* network_debug,
+ ncsi_buf_t* ncsi_buf,
+ mac_addr_t* mac,
+ uint32_t ipv4_addr,
+ uint16_t rx_port);
+
+ncsi_response_type_t ncsi_fsm_poll_test(ncsi_state_t* ncsi_state,
+ network_debug_t* network_debug,
+ ncsi_buf_t* ncsi_buf, mac_addr_t* mac,
+ uint32_t ipv4_addr, uint16_t rx_port);
+
+/*
+ * Report a global state of the NC-SI connection as a function of the state
+ * of the 3 finite state machines.
+ * Note: Additionally for the case where the connection is down it reports
+ * whether a loopback is inferred.
+ */
+ncsi_connection_state_t ncsi_fsm_connection_state(
+ const ncsi_state_t* ncsi_state, const network_debug_t* network_debug);
+
+/*
+ * Returns true if we have executed an NC-SI Get OEM Filter command for all
+ * channels and the flags indicate that it is running in hostless mode.
+ * This means that we can DHCP/ARP if needed.
+ * Otherwise returns false.
+ *
+ * NOTE: We default to false, if we cannot complete the L2 config state
+ * machine or the test sequence.
+ */
+bool ncsi_fsm_is_nic_hostless(const ncsi_state_t* ncsi_state);
+
+#ifdef __cplusplus
+} /* extern "C" */
+#endif
+
+#endif // PLATFORMS_NEMORA_PORTABLE_NCSI_FSM_H_