Import last changes from 2017-03-26, including MDI hack
[revag-nm.git] / vw-nm.c
diff --git a/vw-nm.c b/vw-nm.c
index 975956c1421ee2a6adf73e2cb28bc0b0eb96d1b2..329f95e4b38fa8a52db1476faf67915afbca0aca 100644 (file)
--- a/vw-nm.c
+++ b/vw-nm.c
@@ -6,6 +6,8 @@
  * by the Free Software Foundation.
  */
 
+#include <assert.h>
+
 #include <stdio.h>
 #include <stdint.h>
 #include <unistd.h>
 #include <net/if.h>
 #include <sys/ioctl.h>
 #include <endian.h>
+#include <sys/time.h>
 
 
-enum {
-       /* OSEK/VDX NM Level 0 */
-
-       NM_MAIN_OFF      = 0x00,
-       NM_MAIN_ON       = 0x01,
-       NM_MAIN_LOGIN    = 0x02,
-       NM_MAIN_LIMPHOME = 0x04,
-       NM_MAIN_MASK     = 0x0F,
-
-       NM_SLEEP_CANCEL  = 0x00,
-       NM_SLEEP_REQUEST = 0x10,
-       NM_SLEEP_ACK     = 0x20,
-       NM_SLEEP_MASK    = 0xF0,
-};
+#include "vw-nm.h"
+#include "vw-nm-tools.h"
 
-typedef unsigned char NM_ID;
-typedef unsigned char NM_State;
-
-struct NM_Node {
-       NM_ID next;
-       NM_State state;
-};
 
-struct NM_Main {
-       unsigned max_nodes;
-       struct NM_Node *nodes;
-};
 
 
 
-
-static void can_tx(int socket, struct can_frame *frame)
+static void nm_update_my_next_id(struct NM_Main *nm)
 {
-       ssize_t ret;
+       unsigned id = nm->my_id;
 
-       ret = write(socket, frame, sizeof(*frame));
-       if (ret != sizeof(*frame)) {
-               perror("write to CAN socket");
-               exit(1);
-       }
-}
+       do {
+               NM_State state;
 
+               id++;
+               if (id >= nm->max_nodes) {
+                       id = 0;
+               }
 
+               state = nm->nodes[id].state & NM_MAIN_MASK;
 
-static char* nm_main_to_string(NM_State state)
-{
-       switch(state & NM_MAIN_MASK) {
-               case NM_MAIN_OFF:
-                       return "Off";
-               case NM_MAIN_ON:
-                       return "Ready";
-               case NM_MAIN_LOGIN:
-                       return "Login";
-               case NM_MAIN_LIMPHOME:
-                       return "Limp home";
-               default:
-                       return "Unknown?";
-       }
+               if (state == NM_MAIN_ON || state == NM_MAIN_LOGIN) {
+                       /* We skip limp home nodes */
+                       nm->nodes[nm->my_id].next = id;
+                       break;
+               }
+       } while (id != nm->my_id);
 }
 
-static char* nm_sleep_to_string(NM_State state)
-{
-       switch(state & NM_SLEEP_MASK) {
-               case NM_SLEEP_CANCEL:
-                       return "No";
-               case NM_SLEEP_REQUEST:
-                       return "Request";
-               case NM_SLEEP_ACK:
-                       return "Acknowledged";
-               default:
-                       return "Unknown?";
-       }
-}
 
 
 
-static void nm_dump_all(struct NM_Main *nm)
+static unsigned nm_num_active_nodes(struct NM_Main *nm)
 {
-       unsigned id;
-
-       printf("\n");
-       printf(" Node | next | Main      | Sleep\n");
-       printf("----------------------------------------\n");
+       unsigned id = 0;
+       unsigned active = 0;
 
        for (id = 0; id < nm->max_nodes; id++) {
-               struct NM_Node *node = &nm->nodes[id];
+               NM_State state;
 
-               if (node->state & NM_MAIN_MASK) {
-                       printf("  %02x     %02x    % 9s   %s\n",
-                               id,
-                               node->next,
-                               nm_main_to_string(node->state),
-                               nm_sleep_to_string(node->state));
+               state = nm->nodes[id].state & NM_MAIN_MASK;
 
+               if (state == NM_MAIN_ON || state == NM_MAIN_LOGIN) {
+                       /* We skip limp home nodes */
+                       active++;
                }
        }
 
-       printf("\n");
+       return active;
 }
 
 
 
 static void nm_handle_can_frame(struct NM_Main *nm, struct can_frame *frame)
 {
-       NM_ID id;
-       NM_ID next;
+       NM_ID sender, next;
        NM_State state;
 
-       //printf("Received CAN frame from CAN ID %03x\n", frame->can_id);
-
-       if (frame->can_dlc < 2) {
-               printf("Skipping short frame from CAN ID %03x\n", frame->can_id);
+       /* Is this a valid frame within our logical network? */
+       if (!nm_is_rx_frame_valid(nm, frame)) {
                return;
        }
 
+       printf("Received NM frame from CAN ID %03x\n", frame->can_id);
+
+
+       /* Parse sender, its perceived successor, and its state */
+       sender = frame->can_id & (nm->max_nodes - 1);
+       next = frame->data[0];
+       state = frame->data[1];
+
+       /* TODO: Validate state, it needs to be within the enum */
 
-       if ((frame->can_id & ~0x1f) != 0x420) {
-               printf("Skipping non-NM from CAN ID %03x\n", frame->can_id);
+       /* Skip our own frames */
+       if (sender == nm->my_id) {
                return;
        }
 
-       printf("Received NM frame from CAN ID %03x\n", frame->can_id);
+       /* If we're currently stuck in Limp Home mode, and we can see
+        * someone else's messages: reset counters, reset NM, re-login.
+        */
+       if ((nm->nodes[nm->my_id].state & NM_MAIN_MASK)
+               == NM_MAIN_LIMPHOME) {
+               nm_initreset(nm);
+               return;
+       }
 
-       id = frame->can_id & 0x1f;
-       next = frame->data[0];
-       state = frame->data[1];
+       nm->nodes[sender].next = next;
+       nm->nodes[sender].state = state;
 
-       nm->nodes[id].next = next;
-       nm->nodes[id].state = state;
+       /* Update our view of the world */
+       nm_update_my_next_id(nm);
+
+       switch (state & NM_MAIN_MASK) {
+               case NM_MAIN_ON:
+                       /* We're not alone, so let's transition to ON for now.
+                        */
+                       nm->nodes[nm->my_id].state = NM_MAIN_ON;
+
+                       /* The AWOL timeout is ONLY reset on
+                        * NM_MAIN_ON messages.
+                        */
+                       nm_set_timer_awol(nm);
+
+                       if (next == nm->nodes[nm->my_id].next
+                               && nm->nodes[nm->my_id].next != nm->my_id) {
+                               /* Sender doesn't know we exist */
+
+                               nm->nodes[nm->my_id].state = NM_MAIN_LOGIN;
+
+                               nm_set_timer_now(nm);
+
+                               /* IMPORTANT: The caller needs to check for
+                                * timeouts first, i.e. no other NM frames
+                                * are received until our correcting login
+                                * has been sent.
+                                */
+                       } else if (next == nm->nodes[nm->my_id].next) {
+                               /* where (nm->nodes[nm->my_id].next == nm->my_id) */
+
+                               /* It can happen when:
+                                *  - our sent frames don't go anywhere
+                                *  - we just logged in and immediately
+                                *    afterwards another ECU sent a regular
+                                *    NM frame.
+                                */
+
+                               /* Nothing to do. */
+                       } else if (next == nm->my_id) {
+                               /* It's our turn, do a normal timeout.
+                                * This is a period in which anyone we missed
+                                * can send its re-login frame to correct us.
+                                */
+
+                               nm_set_timer_normal(nm);
+                       } else {
+                               /* We just received a random ON message. */
+
+                               /* Nothing to do. */
+                       }
+                       break;
+               case NM_MAIN_LOGIN:
+                       /* Note: sender != nm->my_id */
+
+                       /* We're not alone anymore, so let's change state. */
+                       nm->nodes[nm->my_id].state = NM_MAIN_ON;
+
+                       /* We don't reset the timeout when somebody logs in.
+                        * Instead, we'll simply include them in the next
+                        * round.
+                        */
+
+                       /* HACK:
+                        * Special case: The Media-In's NM implementation
+                        * doesn't auto-switch to NM_ON. Let's say hello,
+                        * even if it ends up being a little late.
+                        */
+                       if (nm_num_active_nodes(nm) >= 2) {
+                               nm_set_timer_normal(nm);
+                       }
+
+                       /* Nothing else to do. */
+                       break;
+               case NM_MAIN_LIMPHOME:
+                       /* Nothing we can do. Poor guy. */
+                       break;
+       }
 
        nm_dump_all(nm);
+}
 
-       /*
-       switch (state) {
-               case 01:
-                       if (frame.data[0] == my_id) {
-                               struct can_frame txframe = {.can_id = base_id + next_id,
-                                                           .can_dlc = 8,
-                                                           .data = {next_id, 01, 00, 00, 00, 00, 00, 00},
-                                                          };
-                               can_tx(socket, &txframe);
-                       }
-               break;
-               case 02:
-                       if (ignore_counter > 0) {
-                               ignore_counter--;
-                               break;
+
+
+
+
+
+static void nm_buildframe(struct NM_Main *nm, struct can_frame *frame)
+{
+       frame->can_id = nm->can_base + nm->my_id;
+       frame->can_dlc = 2;
+       frame->data[0] = nm->nodes[nm->my_id].next;
+       frame->data[1] = nm->nodes[nm->my_id].state;
+}
+
+
+
+
+static void nm_timeout_callback(struct NM_Main *nm, struct can_frame *frame)
+{
+       switch(nm->timer_reason) {
+               case NM_TIMER_NOW:
+                       /* We're due to log in */
+                       nm_buildframe(nm, frame);
+
+                       if ((nm->nodes[nm->my_id].state & NM_MAIN_MASK)
+                               != NM_MAIN_LOGIN) {
+
+                               printf("BUG: TIMER_NOW expired in non-ON state %u\n",
+                                       nm->nodes[nm->my_id].state & NM_MAIN_MASK);
                        }
-                       if (next_id <= my_id
-                                 ? frame.can_id - base_id < next_id
-                                 : next_id == my_id || frame.can_id - base_id < next_id) {
-                               next_id = frame.can_id - base_id;
-
-                               struct can_frame txframe = {.can_id = base_id + my_id,
-                                                           .can_dlc = 8,
-                                                           .data = {my_id, 02, 01, 04, 00, 04, 00, 00},
-                                                          };
-                               can_tx(socket, &txframe);
+
+                       /* We're going to be ready, let's
+                        * change state (RCD 310 behavior)
+                        */
+                       nm->nodes[nm->my_id].state = NM_MAIN_ON;
+
+                       nm_set_timer_normal(nm);
+                       break;
+               case NM_TIMER_NORMAL:
+                       /* We're due to send our own ring message */
+                       nm_buildframe(nm, frame);
+
+                       if ((nm->nodes[nm->my_id].state & NM_MAIN_MASK)
+                               != NM_MAIN_ON) {
+
+                               printf("BUG: TIMER_NORMAL expired in non-ON state %u\n",
+                                       nm->nodes[nm->my_id].state & NM_MAIN_MASK);
                        }
-               break;
+
+                       nm_set_timer_awol(nm);
+                       break;
+               case NM_TIMER_AWOL:
+                       /* The network is silent because a node disappeared
+                        * or something bad happened.
+                        * Reset everything and start over.
+                        */
+                       nm_reset(nm);
+                       nm_buildframe(nm, frame);
+                       break;
+               case NM_TIMER_LIMPHOME:
+                       printf("Limp home timer expired again :(\n");
+
+                       nm_buildframe(nm, frame);
+                       nm_set_timer_limphome(nm);
+                       break;
        }
-       */
 }
 
 
 
 
-
 static int net_init(char *ifname)
 {
         int s;
@@ -232,45 +308,64 @@ static int net_init(char *ifname)
        return s;
 }
 
+
 int main(int argc, char **argv)
 {
-       struct NM_Node nodes[32] = {{0}};
-       struct NM_Main nm = {.max_nodes = 32, .nodes = nodes};
+       struct NM_Main *nm;
        fd_set rdfs;
        int s;
+       NM_ID my_id;
 
-       if (argc != 2) {
-               printf("syntax: %s IFNAME\n", argv[0]);
-               exit(1);
+       if (argc != 3) {
+               printf("syntax: %s IFNAME MY_ID\n", argv[0]);
+               return 1;
+       }
+
+       my_id = strtoul(argv[2], NULL, 0);
+
+       nm = nm_alloc(5, my_id, 0x420);
+       if (!nm) {
+               printf("Out of memory allocating NM struct.\n");
+               return 1;
        }
 
        s = net_init(argv[1]);
 
        while (1) {
+               int retval;
 
                FD_ZERO(&rdfs);
-
                FD_SET(s, &rdfs);
 
-               if (select(s+1, &rdfs, NULL, NULL, NULL) < 0) {
+               retval = select(s+1, &rdfs, NULL, NULL, &nm->tv);
+               /* We currently rely on Linux timeout behavior here,
+                * i.e. the timeout now reflects the remaining time */
+               if (retval < 0) {
                        perror("select");
                        return 1;
-               }
+               } else if (!retval) {
+                       /* Timeout, we NEED to check this first */
+                       struct can_frame frame;
 
-               if (FD_ISSET(s, &rdfs)) {
+                       nm_timeout_callback(nm, &frame);
+                       can_tx(s, &frame);
+               } else if (FD_ISSET(s, &rdfs)) {
                        struct can_frame frame;
                        ssize_t ret;
 
                        ret = read(s, &frame, sizeof(frame));
                        if (ret < 0) {
                                perror("recvfrom CAN socket");
-                               exit(1);
+                               return 1;
                        }
 
-                       nm_handle_can_frame(&nm, &frame);
+                       nm_handle_can_frame(nm, &frame);
                        continue;
                }
        }
 
+       nm_free(nm);
+       close(s);
+
        return 0;
 }