firewire: Introduce a retry mechanism for reconnects and logins.
authorKristian Høgsberg <krh@redhat.com>
Tue, 6 Feb 2007 19:49:34 +0000 (14:49 -0500)
committerStefan Richter <stefanr@s5r6.in-berlin.de>
Fri, 9 Mar 2007 21:02:52 +0000 (22:02 +0100)
Sometimes we reconnect too soon, sometimes too late.  Adding a retry
mechanism make the reconnect step much more robust.

Signed-off-by: Kristian Høgsberg <krh@redhat.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
drivers/firewire/fw-sbp2.c

index c196333e1de964e3b2e4ef3ef822e8bddc17a015..432c35aefc12b612f405183fb43e9fdeefe29a2b 100644 (file)
@@ -62,7 +62,8 @@ struct sbp2_device {
        /* Timer for flushing ORBs. */
        struct timer_list orb_timer;
 
-       struct work_struct work;
+       int retries;
+       struct delayed_work work;
        struct Scsi_Host *scsi_host;
 };
 
@@ -511,6 +512,75 @@ static int sbp2_agent_reset(struct fw_unit *unit)
 
 static int add_scsi_devices(struct fw_unit *unit);
 static void remove_scsi_devices(struct fw_unit *unit);
+static void sbp2_reconnect(struct work_struct *work);
+
+static void sbp2_login(struct work_struct *work)
+{
+       struct sbp2_device *sd =
+               container_of(work, struct sbp2_device, work.work);
+       struct fw_unit *unit = sd->unit;
+       struct fw_device *device = fw_device(unit->device.parent);
+       struct sbp2_login_response response;
+       int generation, node_id, local_node_id, lun, retval;
+
+       /* FIXME: Make this work for multi-lun devices. */
+       lun = 0;
+
+       generation    = device->card->generation;
+       node_id       = device->node->node_id;
+       local_node_id = device->card->local_node->node_id;
+
+       if (sbp2_send_management_orb(unit, node_id, generation,
+                                    SBP2_LOGIN_REQUEST, lun, &response) < 0) {
+               if (sd->retries++ < 5) {
+                       fw_error("login attempt %d for %s failed, "
+                                "rescheduling\n",
+                                sd->retries, unit->device.bus_id);
+                       schedule_delayed_work(&sd->work, DIV_ROUND_UP(HZ, 5));
+               } else {
+                       fw_error("failed to login to %s\n",
+                                unit->device.bus_id);
+                       remove_scsi_devices(unit);
+               }
+               return;
+       }
+
+       sd->generation   = generation;
+       sd->node_id      = node_id;
+       sd->address_high = local_node_id << 16;
+
+       /* Get command block agent offset and login id. */
+       sd->command_block_agent_address =
+               ((u64) response.command_block_agent.high << 32) |
+               response.command_block_agent.low;
+       sd->login_id = login_response_get_login_id(response);
+
+       fw_notify("logged in to sbp2 unit %s\n", unit->device.bus_id);
+       fw_notify(" - management_agent_address: 0x%012llx\n",
+                 (unsigned long long) sd->management_agent_address);
+       fw_notify(" - command_block_agent_address: 0x%012llx\n",
+                 (unsigned long long) sd->command_block_agent_address);
+       fw_notify(" - status write address: 0x%012llx\n",
+                 (unsigned long long) sd->address_handler.offset);
+
+#if 0
+       /* FIXME: The linux1394 sbp2 does this last step. */
+       sbp2_set_busy_timeout(scsi_id);
+#endif
+
+       INIT_DELAYED_WORK(&sd->work, sbp2_reconnect);
+       sbp2_agent_reset(unit);
+
+       retval = add_scsi_devices(unit);
+       if (retval < 0) {
+               sbp2_send_management_orb(unit, sd->node_id, sd->generation,
+                                        SBP2_LOGOUT_REQUEST, sd->login_id,
+                                        NULL);
+               /* Set this back to sbp2_login so we fall back and
+                * retry login on bus reset. */
+               INIT_DELAYED_WORK(&sd->work, sbp2_login);
+       }
+}
 
 static int sbp2_probe(struct device *dev)
 {
@@ -518,9 +588,7 @@ static int sbp2_probe(struct device *dev)
        struct fw_device *device = fw_device(unit->device.parent);
        struct sbp2_device *sd;
        struct fw_csr_iterator ci;
-       int i, key, value, lun, retval;
-       int node_id, generation, local_node_id;
-       struct sbp2_login_response response;
+       int i, key, value;
        u32 model, firmware_revision;
 
        sd = kzalloc(sizeof *sd, GFP_KERNEL);
@@ -586,58 +654,10 @@ static int sbp2_probe(struct device *dev)
                          unit->device.bus_id,
                          sd->workarounds, firmware_revision, model);
 
-       /* FIXME: Make this work for multi-lun devices. */
-       lun = 0;
-
-       generation    = device->card->generation;
-       node_id       = device->node->node_id;
-       local_node_id = device->card->local_node->node_id;
-
-       /* FIXME: We should probably do this from a keventd callback
-        * and handle retries by rescheduling the work. */
-       if (sbp2_send_management_orb(unit, node_id, generation,
-                                    SBP2_LOGIN_REQUEST, lun, &response) < 0) {
-               fw_core_remove_address_handler(&sd->address_handler);
-               del_timer_sync(&sd->orb_timer);
-               kfree(sd);
-               return -EBUSY;
-       }
-
-       sd->generation   = generation;
-       sd->node_id      = node_id;
-       sd->address_high = local_node_id << 16;
-
-       /* Get command block agent offset and login id. */
-       sd->command_block_agent_address =
-               ((u64) response.command_block_agent.high << 32) |
-               response.command_block_agent.low;
-       sd->login_id = login_response_get_login_id(response);
-
-       fw_notify("logged in to sbp2 unit %s\n", unit->device.bus_id);
-       fw_notify(" - management_agent_address: 0x%012llx\n",
-                 (unsigned long long) sd->management_agent_address);
-       fw_notify(" - command_block_agent_address: 0x%012llx\n",
-                 (unsigned long long) sd->command_block_agent_address);
-       fw_notify(" - status write address: 0x%012llx\n",
-                 (unsigned long long) sd->address_handler.offset);
-
-#if 0
-       /* FIXME: The linux1394 sbp2 does this last step. */
-       sbp2_set_busy_timeout(scsi_id);
-#endif
-
-       sbp2_agent_reset(unit);
-
-       retval = add_scsi_devices(unit);
-       if (retval < 0) {
-               sbp2_send_management_orb(unit, sd->node_id, sd->generation,
-                                        SBP2_LOGOUT_REQUEST, sd->login_id,
-                                        NULL);
-               fw_core_remove_address_handler(&sd->address_handler);
-               del_timer_sync(&sd->orb_timer);
-               kfree(sd);
-               return retval;
-       }
+       /* We schedule work to do the login so we can easily
+        * reschedule retries. */
+       INIT_DELAYED_WORK(&sd->work, sbp2_login);
+       schedule_delayed_work(&sd->work, 0);
 
        return 0;
 }
@@ -663,28 +683,41 @@ static int sbp2_remove(struct device *dev)
 
 static void sbp2_reconnect(struct work_struct *work)
 {
-       struct sbp2_device *sd = container_of(work, struct sbp2_device, work);
+       struct sbp2_device *sd =
+               container_of(work, struct sbp2_device, work.work);
        struct fw_unit *unit = sd->unit;
        struct fw_device *device = fw_device(unit->device.parent);
        int generation, node_id, local_node_id;
 
-       fw_notify("in sbp2_reconnect, reconnecting to unit %s\n",
-                 unit->device.bus_id);
-
        generation    = device->card->generation;
        node_id       = device->node->node_id;
        local_node_id = device->card->local_node->node_id;
 
-       sbp2_send_management_orb(unit, node_id, generation,
-                                SBP2_RECONNECT_REQUEST, sd->login_id, NULL);
-
-       /* FIXME: handle reconnect failures. */
-
-       sbp2_cancel_orbs(unit);
+       if (sbp2_send_management_orb(unit, node_id, generation,
+                                    SBP2_RECONNECT_REQUEST,
+                                    sd->login_id, NULL) < 0) {
+               if (sd->retries++ < 5) {
+                       fw_error("reconnect attempt %d for %s failed, "
+                                "rescheduling\n",
+                                sd->retries, unit->device.bus_id);
+               } else {
+                       fw_error("failed to reconnect to %s\n",
+                                unit->device.bus_id);
+                       /* Fall back and try to log in again. */
+                       sd->retries = 0;
+                       INIT_DELAYED_WORK(&sd->work, sbp2_login);
+               }
+               schedule_delayed_work(&sd->work, DIV_ROUND_UP(HZ, 5));
+               return;
+       }
 
        sd->generation   = generation;
        sd->node_id      = node_id;
        sd->address_high = local_node_id << 16;
+
+       fw_notify("reconnected to unit %s\n", unit->device.bus_id);
+       sbp2_agent_reset(unit);
+       sbp2_cancel_orbs(unit);
 }
 
 static void sbp2_update(struct fw_unit *unit)
@@ -692,10 +725,9 @@ static void sbp2_update(struct fw_unit *unit)
        struct fw_device *device = fw_device(unit->device.parent);
        struct sbp2_device *sd = unit->device.driver_data;
 
+       sd->retries = 0;
        fw_device_enable_phys_dma(device);
-
-       INIT_WORK(&sd->work, sbp2_reconnect);
-       schedule_work(&sd->work);
+       schedule_delayed_work(&sd->work, 0);
 }
 
 #define SBP2_UNIT_SPEC_ID_ENTRY        0x0000609e
@@ -1056,6 +1088,9 @@ static int add_scsi_devices(struct fw_unit *unit)
        struct sbp2_device *sd = unit->device.driver_data;
        int retval, lun;
 
+       if (sd->scsi_host != NULL)
+               return 0;
+
        sd->scsi_host = scsi_host_alloc(&scsi_driver_template,
                                        sizeof(unsigned long));
        if (sd->scsi_host == NULL) {
@@ -1088,8 +1123,11 @@ static void remove_scsi_devices(struct fw_unit *unit)
 {
        struct sbp2_device *sd = unit->device.driver_data;
 
-       scsi_remove_host(sd->scsi_host);
-       scsi_host_put(sd->scsi_host);
+       if (sd->scsi_host != NULL) {
+               scsi_remove_host(sd->scsi_host);
+               scsi_host_put(sd->scsi_host);
+       }
+       sd->scsi_host = NULL;
 }
 
 MODULE_AUTHOR("Kristian Hoegsberg <krh@bitplanet.net>");