Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / s390 / block / dasd_alias.c
diff --git a/kernel/drivers/s390/block/dasd_alias.c b/kernel/drivers/s390/block/dasd_alias.c
new file mode 100644 (file)
index 0000000..a2597e6
--- /dev/null
@@ -0,0 +1,990 @@
+/*
+ * PAV alias management for the DASD ECKD discipline
+ *
+ * Copyright IBM Corp. 2007
+ * Author(s): Stefan Weinhuber <wein@de.ibm.com>
+ */
+
+#define KMSG_COMPONENT "dasd-eckd"
+
+#include <linux/list.h>
+#include <linux/slab.h>
+#include <asm/ebcdic.h>
+#include "dasd_int.h"
+#include "dasd_eckd.h"
+
+#ifdef PRINTK_HEADER
+#undef PRINTK_HEADER
+#endif                         /* PRINTK_HEADER */
+#define PRINTK_HEADER "dasd(eckd):"
+
+
+/*
+ * General concept of alias management:
+ * - PAV and DASD alias management is specific to the eckd discipline.
+ * - A device is connected to an lcu as long as the device exists.
+ *   dasd_alias_make_device_known_to_lcu will be called wenn the
+ *   device is checked by the eckd discipline and
+ *   dasd_alias_disconnect_device_from_lcu will be called
+ *   before the device is deleted.
+ * - The dasd_alias_add_device / dasd_alias_remove_device
+ *   functions mark the point when a device is 'ready for service'.
+ * - A summary unit check is a rare occasion, but it is mandatory to
+ *   support it. It requires some complex recovery actions before the
+ *   devices can be used again (see dasd_alias_handle_summary_unit_check).
+ * - dasd_alias_get_start_dev will find an alias device that can be used
+ *   instead of the base device and does some (very simple) load balancing.
+ *   This is the function that gets called for each I/O, so when improving
+ *   something, this function should get faster or better, the rest has just
+ *   to be correct.
+ */
+
+
+static void summary_unit_check_handling_work(struct work_struct *);
+static void lcu_update_work(struct work_struct *);
+static int _schedule_lcu_update(struct alias_lcu *, struct dasd_device *);
+
+static struct alias_root aliastree = {
+       .serverlist = LIST_HEAD_INIT(aliastree.serverlist),
+       .lock = __SPIN_LOCK_UNLOCKED(aliastree.lock),
+};
+
+static struct alias_server *_find_server(struct dasd_uid *uid)
+{
+       struct alias_server *pos;
+       list_for_each_entry(pos, &aliastree.serverlist, server) {
+               if (!strncmp(pos->uid.vendor, uid->vendor,
+                            sizeof(uid->vendor))
+                   && !strncmp(pos->uid.serial, uid->serial,
+                               sizeof(uid->serial)))
+                       return pos;
+       };
+       return NULL;
+}
+
+static struct alias_lcu *_find_lcu(struct alias_server *server,
+                                  struct dasd_uid *uid)
+{
+       struct alias_lcu *pos;
+       list_for_each_entry(pos, &server->lculist, lcu) {
+               if (pos->uid.ssid == uid->ssid)
+                       return pos;
+       };
+       return NULL;
+}
+
+static struct alias_pav_group *_find_group(struct alias_lcu *lcu,
+                                          struct dasd_uid *uid)
+{
+       struct alias_pav_group *pos;
+       __u8 search_unit_addr;
+
+       /* for hyper pav there is only one group */
+       if (lcu->pav == HYPER_PAV) {
+               if (list_empty(&lcu->grouplist))
+                       return NULL;
+               else
+                       return list_first_entry(&lcu->grouplist,
+                                               struct alias_pav_group, group);
+       }
+
+       /* for base pav we have to find the group that matches the base */
+       if (uid->type == UA_BASE_DEVICE)
+               search_unit_addr = uid->real_unit_addr;
+       else
+               search_unit_addr = uid->base_unit_addr;
+       list_for_each_entry(pos, &lcu->grouplist, group) {
+               if (pos->uid.base_unit_addr == search_unit_addr &&
+                   !strncmp(pos->uid.vduit, uid->vduit, sizeof(uid->vduit)))
+                       return pos;
+       };
+       return NULL;
+}
+
+static struct alias_server *_allocate_server(struct dasd_uid *uid)
+{
+       struct alias_server *server;
+
+       server = kzalloc(sizeof(*server), GFP_KERNEL);
+       if (!server)
+               return ERR_PTR(-ENOMEM);
+       memcpy(server->uid.vendor, uid->vendor, sizeof(uid->vendor));
+       memcpy(server->uid.serial, uid->serial, sizeof(uid->serial));
+       INIT_LIST_HEAD(&server->server);
+       INIT_LIST_HEAD(&server->lculist);
+       return server;
+}
+
+static void _free_server(struct alias_server *server)
+{
+       kfree(server);
+}
+
+static struct alias_lcu *_allocate_lcu(struct dasd_uid *uid)
+{
+       struct alias_lcu *lcu;
+
+       lcu = kzalloc(sizeof(*lcu), GFP_KERNEL);
+       if (!lcu)
+               return ERR_PTR(-ENOMEM);
+       lcu->uac = kzalloc(sizeof(*(lcu->uac)), GFP_KERNEL | GFP_DMA);
+       if (!lcu->uac)
+               goto out_err1;
+       lcu->rsu_cqr = kzalloc(sizeof(*lcu->rsu_cqr), GFP_KERNEL | GFP_DMA);
+       if (!lcu->rsu_cqr)
+               goto out_err2;
+       lcu->rsu_cqr->cpaddr = kzalloc(sizeof(struct ccw1),
+                                      GFP_KERNEL | GFP_DMA);
+       if (!lcu->rsu_cqr->cpaddr)
+               goto out_err3;
+       lcu->rsu_cqr->data = kzalloc(16, GFP_KERNEL | GFP_DMA);
+       if (!lcu->rsu_cqr->data)
+               goto out_err4;
+
+       memcpy(lcu->uid.vendor, uid->vendor, sizeof(uid->vendor));
+       memcpy(lcu->uid.serial, uid->serial, sizeof(uid->serial));
+       lcu->uid.ssid = uid->ssid;
+       lcu->pav = NO_PAV;
+       lcu->flags = NEED_UAC_UPDATE | UPDATE_PENDING;
+       INIT_LIST_HEAD(&lcu->lcu);
+       INIT_LIST_HEAD(&lcu->inactive_devices);
+       INIT_LIST_HEAD(&lcu->active_devices);
+       INIT_LIST_HEAD(&lcu->grouplist);
+       INIT_WORK(&lcu->suc_data.worker, summary_unit_check_handling_work);
+       INIT_DELAYED_WORK(&lcu->ruac_data.dwork, lcu_update_work);
+       spin_lock_init(&lcu->lock);
+       init_completion(&lcu->lcu_setup);
+       return lcu;
+
+out_err4:
+       kfree(lcu->rsu_cqr->cpaddr);
+out_err3:
+       kfree(lcu->rsu_cqr);
+out_err2:
+       kfree(lcu->uac);
+out_err1:
+       kfree(lcu);
+       return ERR_PTR(-ENOMEM);
+}
+
+static void _free_lcu(struct alias_lcu *lcu)
+{
+       kfree(lcu->rsu_cqr->data);
+       kfree(lcu->rsu_cqr->cpaddr);
+       kfree(lcu->rsu_cqr);
+       kfree(lcu->uac);
+       kfree(lcu);
+}
+
+/*
+ * This is the function that will allocate all the server and lcu data,
+ * so this function must be called first for a new device.
+ * If the return value is 1, the lcu was already known before, if it
+ * is 0, this is a new lcu.
+ * Negative return code indicates that something went wrong (e.g. -ENOMEM)
+ */
+int dasd_alias_make_device_known_to_lcu(struct dasd_device *device)
+{
+       struct dasd_eckd_private *private;
+       unsigned long flags;
+       struct alias_server *server, *newserver;
+       struct alias_lcu *lcu, *newlcu;
+       struct dasd_uid uid;
+
+       private = (struct dasd_eckd_private *) device->private;
+
+       device->discipline->get_uid(device, &uid);
+       spin_lock_irqsave(&aliastree.lock, flags);
+       server = _find_server(&uid);
+       if (!server) {
+               spin_unlock_irqrestore(&aliastree.lock, flags);
+               newserver = _allocate_server(&uid);
+               if (IS_ERR(newserver))
+                       return PTR_ERR(newserver);
+               spin_lock_irqsave(&aliastree.lock, flags);
+               server = _find_server(&uid);
+               if (!server) {
+                       list_add(&newserver->server, &aliastree.serverlist);
+                       server = newserver;
+               } else {
+                       /* someone was faster */
+                       _free_server(newserver);
+               }
+       }
+
+       lcu = _find_lcu(server, &uid);
+       if (!lcu) {
+               spin_unlock_irqrestore(&aliastree.lock, flags);
+               newlcu = _allocate_lcu(&uid);
+               if (IS_ERR(newlcu))
+                       return PTR_ERR(newlcu);
+               spin_lock_irqsave(&aliastree.lock, flags);
+               lcu = _find_lcu(server, &uid);
+               if (!lcu) {
+                       list_add(&newlcu->lcu, &server->lculist);
+                       lcu = newlcu;
+               } else {
+                       /* someone was faster */
+                       _free_lcu(newlcu);
+               }
+       }
+       spin_lock(&lcu->lock);
+       list_add(&device->alias_list, &lcu->inactive_devices);
+       private->lcu = lcu;
+       spin_unlock(&lcu->lock);
+       spin_unlock_irqrestore(&aliastree.lock, flags);
+
+       return 0;
+}
+
+/*
+ * This function removes a device from the scope of alias management.
+ * The complicated part is to make sure that it is not in use by
+ * any of the workers. If necessary cancel the work.
+ */
+void dasd_alias_disconnect_device_from_lcu(struct dasd_device *device)
+{
+       struct dasd_eckd_private *private;
+       unsigned long flags;
+       struct alias_lcu *lcu;
+       struct alias_server *server;
+       int was_pending;
+       struct dasd_uid uid;
+
+       private = (struct dasd_eckd_private *) device->private;
+       lcu = private->lcu;
+       /* nothing to do if already disconnected */
+       if (!lcu)
+               return;
+       device->discipline->get_uid(device, &uid);
+       spin_lock_irqsave(&lcu->lock, flags);
+       list_del_init(&device->alias_list);
+       /* make sure that the workers don't use this device */
+       if (device == lcu->suc_data.device) {
+               spin_unlock_irqrestore(&lcu->lock, flags);
+               cancel_work_sync(&lcu->suc_data.worker);
+               spin_lock_irqsave(&lcu->lock, flags);
+               if (device == lcu->suc_data.device)
+                       lcu->suc_data.device = NULL;
+       }
+       was_pending = 0;
+       if (device == lcu->ruac_data.device) {
+               spin_unlock_irqrestore(&lcu->lock, flags);
+               was_pending = 1;
+               cancel_delayed_work_sync(&lcu->ruac_data.dwork);
+               spin_lock_irqsave(&lcu->lock, flags);
+               if (device == lcu->ruac_data.device)
+                       lcu->ruac_data.device = NULL;
+       }
+       private->lcu = NULL;
+       spin_unlock_irqrestore(&lcu->lock, flags);
+
+       spin_lock_irqsave(&aliastree.lock, flags);
+       spin_lock(&lcu->lock);
+       if (list_empty(&lcu->grouplist) &&
+           list_empty(&lcu->active_devices) &&
+           list_empty(&lcu->inactive_devices)) {
+               list_del(&lcu->lcu);
+               spin_unlock(&lcu->lock);
+               _free_lcu(lcu);
+               lcu = NULL;
+       } else {
+               if (was_pending)
+                       _schedule_lcu_update(lcu, NULL);
+               spin_unlock(&lcu->lock);
+       }
+       server = _find_server(&uid);
+       if (server && list_empty(&server->lculist)) {
+               list_del(&server->server);
+               _free_server(server);
+       }
+       spin_unlock_irqrestore(&aliastree.lock, flags);
+}
+
+/*
+ * This function assumes that the unit address configuration stored
+ * in the lcu is up to date and will update the device uid before
+ * adding it to a pav group.
+ */
+
+static int _add_device_to_lcu(struct alias_lcu *lcu,
+                             struct dasd_device *device,
+                             struct dasd_device *pos)
+{
+
+       struct dasd_eckd_private *private;
+       struct alias_pav_group *group;
+       struct dasd_uid uid;
+       unsigned long flags;
+
+       private = (struct dasd_eckd_private *) device->private;
+
+       /* only lock if not already locked */
+       if (device != pos)
+               spin_lock_irqsave_nested(get_ccwdev_lock(device->cdev), flags,
+                                        CDEV_NESTED_SECOND);
+       private->uid.type = lcu->uac->unit[private->uid.real_unit_addr].ua_type;
+       private->uid.base_unit_addr =
+               lcu->uac->unit[private->uid.real_unit_addr].base_ua;
+       uid = private->uid;
+
+       if (device != pos)
+               spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
+
+       /* if we have no PAV anyway, we don't need to bother with PAV groups */
+       if (lcu->pav == NO_PAV) {
+               list_move(&device->alias_list, &lcu->active_devices);
+               return 0;
+       }
+
+       group = _find_group(lcu, &uid);
+       if (!group) {
+               group = kzalloc(sizeof(*group), GFP_ATOMIC);
+               if (!group)
+                       return -ENOMEM;
+               memcpy(group->uid.vendor, uid.vendor, sizeof(uid.vendor));
+               memcpy(group->uid.serial, uid.serial, sizeof(uid.serial));
+               group->uid.ssid = uid.ssid;
+               if (uid.type == UA_BASE_DEVICE)
+                       group->uid.base_unit_addr = uid.real_unit_addr;
+               else
+                       group->uid.base_unit_addr = uid.base_unit_addr;
+               memcpy(group->uid.vduit, uid.vduit, sizeof(uid.vduit));
+               INIT_LIST_HEAD(&group->group);
+               INIT_LIST_HEAD(&group->baselist);
+               INIT_LIST_HEAD(&group->aliaslist);
+               list_add(&group->group, &lcu->grouplist);
+       }
+       if (uid.type == UA_BASE_DEVICE)
+               list_move(&device->alias_list, &group->baselist);
+       else
+               list_move(&device->alias_list, &group->aliaslist);
+       private->pavgroup = group;
+       return 0;
+};
+
+static void _remove_device_from_lcu(struct alias_lcu *lcu,
+                                   struct dasd_device *device)
+{
+       struct dasd_eckd_private *private;
+       struct alias_pav_group *group;
+
+       private = (struct dasd_eckd_private *) device->private;
+       list_move(&device->alias_list, &lcu->inactive_devices);
+       group = private->pavgroup;
+       if (!group)
+               return;
+       private->pavgroup = NULL;
+       if (list_empty(&group->baselist) && list_empty(&group->aliaslist)) {
+               list_del(&group->group);
+               kfree(group);
+               return;
+       }
+       if (group->next == device)
+               group->next = NULL;
+};
+
+static int
+suborder_not_supported(struct dasd_ccw_req *cqr)
+{
+       char *sense;
+       char reason;
+       char msg_format;
+       char msg_no;
+
+       sense = dasd_get_sense(&cqr->irb);
+       if (!sense)
+               return 0;
+
+       reason = sense[0];
+       msg_format = (sense[7] & 0xF0);
+       msg_no = (sense[7] & 0x0F);
+
+       /* command reject, Format 0 MSG 4 - invalid parameter */
+       if ((reason == 0x80) && (msg_format == 0x00) && (msg_no == 0x04))
+               return 1;
+
+       return 0;
+}
+
+static int read_unit_address_configuration(struct dasd_device *device,
+                                          struct alias_lcu *lcu)
+{
+       struct dasd_psf_prssd_data *prssdp;
+       struct dasd_ccw_req *cqr;
+       struct ccw1 *ccw;
+       int rc;
+       unsigned long flags;
+
+       cqr = dasd_kmalloc_request(DASD_ECKD_MAGIC, 1 /* PSF */ + 1 /* RSSD */,
+                                  (sizeof(struct dasd_psf_prssd_data)),
+                                  device);
+       if (IS_ERR(cqr))
+               return PTR_ERR(cqr);
+       cqr->startdev = device;
+       cqr->memdev = device;
+       clear_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags);
+       cqr->retries = 10;
+       cqr->expires = 20 * HZ;
+
+       /* Prepare for Read Subsystem Data */
+       prssdp = (struct dasd_psf_prssd_data *) cqr->data;
+       memset(prssdp, 0, sizeof(struct dasd_psf_prssd_data));
+       prssdp->order = PSF_ORDER_PRSSD;
+       prssdp->suborder = 0x0e;        /* Read unit address configuration */
+       /* all other bytes of prssdp must be zero */
+
+       ccw = cqr->cpaddr;
+       ccw->cmd_code = DASD_ECKD_CCW_PSF;
+       ccw->count = sizeof(struct dasd_psf_prssd_data);
+       ccw->flags |= CCW_FLAG_CC;
+       ccw->cda = (__u32)(addr_t) prssdp;
+
+       /* Read Subsystem Data - feature codes */
+       memset(lcu->uac, 0, sizeof(*(lcu->uac)));
+
+       ccw++;
+       ccw->cmd_code = DASD_ECKD_CCW_RSSD;
+       ccw->count = sizeof(*(lcu->uac));
+       ccw->cda = (__u32)(addr_t) lcu->uac;
+
+       cqr->buildclk = get_tod_clock();
+       cqr->status = DASD_CQR_FILLED;
+
+       /* need to unset flag here to detect race with summary unit check */
+       spin_lock_irqsave(&lcu->lock, flags);
+       lcu->flags &= ~NEED_UAC_UPDATE;
+       spin_unlock_irqrestore(&lcu->lock, flags);
+
+       do {
+               rc = dasd_sleep_on(cqr);
+               if (rc && suborder_not_supported(cqr))
+                       return -EOPNOTSUPP;
+       } while (rc && (cqr->retries > 0));
+       if (rc) {
+               spin_lock_irqsave(&lcu->lock, flags);
+               lcu->flags |= NEED_UAC_UPDATE;
+               spin_unlock_irqrestore(&lcu->lock, flags);
+       }
+       dasd_kfree_request(cqr, cqr->memdev);
+       return rc;
+}
+
+static int _lcu_update(struct dasd_device *refdev, struct alias_lcu *lcu)
+{
+       unsigned long flags;
+       struct alias_pav_group *pavgroup, *tempgroup;
+       struct dasd_device *device, *tempdev;
+       int i, rc;
+       struct dasd_eckd_private *private;
+
+       spin_lock_irqsave(&lcu->lock, flags);
+       list_for_each_entry_safe(pavgroup, tempgroup, &lcu->grouplist, group) {
+               list_for_each_entry_safe(device, tempdev, &pavgroup->baselist,
+                                        alias_list) {
+                       list_move(&device->alias_list, &lcu->active_devices);
+                       private = (struct dasd_eckd_private *) device->private;
+                       private->pavgroup = NULL;
+               }
+               list_for_each_entry_safe(device, tempdev, &pavgroup->aliaslist,
+                                        alias_list) {
+                       list_move(&device->alias_list, &lcu->active_devices);
+                       private = (struct dasd_eckd_private *) device->private;
+                       private->pavgroup = NULL;
+               }
+               list_del(&pavgroup->group);
+               kfree(pavgroup);
+       }
+       spin_unlock_irqrestore(&lcu->lock, flags);
+
+       rc = read_unit_address_configuration(refdev, lcu);
+       if (rc)
+               return rc;
+
+       /* need to take cdev lock before lcu lock */
+       spin_lock_irqsave_nested(get_ccwdev_lock(refdev->cdev), flags,
+                                CDEV_NESTED_FIRST);
+       spin_lock(&lcu->lock);
+       lcu->pav = NO_PAV;
+       for (i = 0; i < MAX_DEVICES_PER_LCU; ++i) {
+               switch (lcu->uac->unit[i].ua_type) {
+               case UA_BASE_PAV_ALIAS:
+                       lcu->pav = BASE_PAV;
+                       break;
+               case UA_HYPER_PAV_ALIAS:
+                       lcu->pav = HYPER_PAV;
+                       break;
+               }
+               if (lcu->pav != NO_PAV)
+                       break;
+       }
+
+       list_for_each_entry_safe(device, tempdev, &lcu->active_devices,
+                                alias_list) {
+               _add_device_to_lcu(lcu, device, refdev);
+       }
+       spin_unlock(&lcu->lock);
+       spin_unlock_irqrestore(get_ccwdev_lock(refdev->cdev), flags);
+       return 0;
+}
+
+static void lcu_update_work(struct work_struct *work)
+{
+       struct alias_lcu *lcu;
+       struct read_uac_work_data *ruac_data;
+       struct dasd_device *device;
+       unsigned long flags;
+       int rc;
+
+       ruac_data = container_of(work, struct read_uac_work_data, dwork.work);
+       lcu = container_of(ruac_data, struct alias_lcu, ruac_data);
+       device = ruac_data->device;
+       rc = _lcu_update(device, lcu);
+       /*
+        * Need to check flags again, as there could have been another
+        * prepare_update or a new device a new device while we were still
+        * processing the data
+        */
+       spin_lock_irqsave(&lcu->lock, flags);
+       if ((rc && (rc != -EOPNOTSUPP)) || (lcu->flags & NEED_UAC_UPDATE)) {
+               DBF_DEV_EVENT(DBF_WARNING, device, "could not update"
+                           " alias data in lcu (rc = %d), retry later", rc);
+               schedule_delayed_work(&lcu->ruac_data.dwork, 30*HZ);
+       } else {
+               lcu->ruac_data.device = NULL;
+               lcu->flags &= ~UPDATE_PENDING;
+       }
+       spin_unlock_irqrestore(&lcu->lock, flags);
+}
+
+static int _schedule_lcu_update(struct alias_lcu *lcu,
+                               struct dasd_device *device)
+{
+       struct dasd_device *usedev = NULL;
+       struct alias_pav_group *group;
+
+       lcu->flags |= NEED_UAC_UPDATE;
+       if (lcu->ruac_data.device) {
+               /* already scheduled or running */
+               return 0;
+       }
+       if (device && !list_empty(&device->alias_list))
+               usedev = device;
+
+       if (!usedev && !list_empty(&lcu->grouplist)) {
+               group = list_first_entry(&lcu->grouplist,
+                                        struct alias_pav_group, group);
+               if (!list_empty(&group->baselist))
+                       usedev = list_first_entry(&group->baselist,
+                                                 struct dasd_device,
+                                                 alias_list);
+               else if (!list_empty(&group->aliaslist))
+                       usedev = list_first_entry(&group->aliaslist,
+                                                 struct dasd_device,
+                                                 alias_list);
+       }
+       if (!usedev && !list_empty(&lcu->active_devices)) {
+               usedev = list_first_entry(&lcu->active_devices,
+                                         struct dasd_device, alias_list);
+       }
+       /*
+        * if we haven't found a proper device yet, give up for now, the next
+        * device that will be set active will trigger an lcu update
+        */
+       if (!usedev)
+               return -EINVAL;
+       lcu->ruac_data.device = usedev;
+       schedule_delayed_work(&lcu->ruac_data.dwork, 0);
+       return 0;
+}
+
+int dasd_alias_add_device(struct dasd_device *device)
+{
+       struct dasd_eckd_private *private;
+       struct alias_lcu *lcu;
+       unsigned long flags;
+       int rc;
+
+       private = (struct dasd_eckd_private *) device->private;
+       lcu = private->lcu;
+       rc = 0;
+
+       /* need to take cdev lock before lcu lock */
+       spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
+       spin_lock(&lcu->lock);
+       if (!(lcu->flags & UPDATE_PENDING)) {
+               rc = _add_device_to_lcu(lcu, device, device);
+               if (rc)
+                       lcu->flags |= UPDATE_PENDING;
+       }
+       if (lcu->flags & UPDATE_PENDING) {
+               list_move(&device->alias_list, &lcu->active_devices);
+               _schedule_lcu_update(lcu, device);
+       }
+       spin_unlock(&lcu->lock);
+       spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
+       return rc;
+}
+
+int dasd_alias_update_add_device(struct dasd_device *device)
+{
+       struct dasd_eckd_private *private;
+       private = (struct dasd_eckd_private *) device->private;
+       private->lcu->flags |= UPDATE_PENDING;
+       return dasd_alias_add_device(device);
+}
+
+int dasd_alias_remove_device(struct dasd_device *device)
+{
+       struct dasd_eckd_private *private;
+       struct alias_lcu *lcu;
+       unsigned long flags;
+
+       private = (struct dasd_eckd_private *) device->private;
+       lcu = private->lcu;
+       /* nothing to do if already removed */
+       if (!lcu)
+               return 0;
+       spin_lock_irqsave(&lcu->lock, flags);
+       _remove_device_from_lcu(lcu, device);
+       spin_unlock_irqrestore(&lcu->lock, flags);
+       return 0;
+}
+
+struct dasd_device *dasd_alias_get_start_dev(struct dasd_device *base_device)
+{
+
+       struct dasd_device *alias_device;
+       struct alias_pav_group *group;
+       struct alias_lcu *lcu;
+       struct dasd_eckd_private *private, *alias_priv;
+       unsigned long flags;
+
+       private = (struct dasd_eckd_private *) base_device->private;
+       group = private->pavgroup;
+       lcu = private->lcu;
+       if (!group || !lcu)
+               return NULL;
+       if (lcu->pav == NO_PAV ||
+           lcu->flags & (NEED_UAC_UPDATE | UPDATE_PENDING))
+               return NULL;
+       if (unlikely(!(private->features.feature[8] & 0x01))) {
+               /*
+                * PAV enabled but prefix not, very unlikely
+                * seems to be a lost pathgroup
+                * use base device to do IO
+                */
+               DBF_DEV_EVENT(DBF_ERR, base_device, "%s",
+                             "Prefix not enabled with PAV enabled\n");
+               return NULL;
+       }
+
+       spin_lock_irqsave(&lcu->lock, flags);
+       alias_device = group->next;
+       if (!alias_device) {
+               if (list_empty(&group->aliaslist)) {
+                       spin_unlock_irqrestore(&lcu->lock, flags);
+                       return NULL;
+               } else {
+                       alias_device = list_first_entry(&group->aliaslist,
+                                                       struct dasd_device,
+                                                       alias_list);
+               }
+       }
+       if (list_is_last(&alias_device->alias_list, &group->aliaslist))
+               group->next = list_first_entry(&group->aliaslist,
+                                              struct dasd_device, alias_list);
+       else
+               group->next = list_first_entry(&alias_device->alias_list,
+                                              struct dasd_device, alias_list);
+       spin_unlock_irqrestore(&lcu->lock, flags);
+       alias_priv = (struct dasd_eckd_private *) alias_device->private;
+       if ((alias_priv->count < private->count) && !alias_device->stopped)
+               return alias_device;
+       else
+               return NULL;
+}
+
+/*
+ * Summary unit check handling depends on the way alias devices
+ * are handled so it is done here rather then in dasd_eckd.c
+ */
+static int reset_summary_unit_check(struct alias_lcu *lcu,
+                                   struct dasd_device *device,
+                                   char reason)
+{
+       struct dasd_ccw_req *cqr;
+       int rc = 0;
+       struct ccw1 *ccw;
+
+       cqr = lcu->rsu_cqr;
+       strncpy((char *) &cqr->magic, "ECKD", 4);
+       ASCEBC((char *) &cqr->magic, 4);
+       ccw = cqr->cpaddr;
+       ccw->cmd_code = DASD_ECKD_CCW_RSCK;
+       ccw->flags = 0 ;
+       ccw->count = 16;
+       ccw->cda = (__u32)(addr_t) cqr->data;
+       ((char *)cqr->data)[0] = reason;
+
+       clear_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags);
+       cqr->retries = 255;     /* set retry counter to enable basic ERP */
+       cqr->startdev = device;
+       cqr->memdev = device;
+       cqr->block = NULL;
+       cqr->expires = 5 * HZ;
+       cqr->buildclk = get_tod_clock();
+       cqr->status = DASD_CQR_FILLED;
+
+       rc = dasd_sleep_on_immediatly(cqr);
+       return rc;
+}
+
+static void _restart_all_base_devices_on_lcu(struct alias_lcu *lcu)
+{
+       struct alias_pav_group *pavgroup;
+       struct dasd_device *device;
+       struct dasd_eckd_private *private;
+       unsigned long flags;
+
+       /* active and inactive list can contain alias as well as base devices */
+       list_for_each_entry(device, &lcu->active_devices, alias_list) {
+               private = (struct dasd_eckd_private *) device->private;
+               spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
+               if (private->uid.type != UA_BASE_DEVICE) {
+                       spin_unlock_irqrestore(get_ccwdev_lock(device->cdev),
+                                              flags);
+                       continue;
+               }
+               spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
+               dasd_schedule_block_bh(device->block);
+               dasd_schedule_device_bh(device);
+       }
+       list_for_each_entry(device, &lcu->inactive_devices, alias_list) {
+               private = (struct dasd_eckd_private *) device->private;
+               spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
+               if (private->uid.type != UA_BASE_DEVICE) {
+                       spin_unlock_irqrestore(get_ccwdev_lock(device->cdev),
+                                              flags);
+                       continue;
+               }
+               spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
+               dasd_schedule_block_bh(device->block);
+               dasd_schedule_device_bh(device);
+       }
+       list_for_each_entry(pavgroup, &lcu->grouplist, group) {
+               list_for_each_entry(device, &pavgroup->baselist, alias_list) {
+                       dasd_schedule_block_bh(device->block);
+                       dasd_schedule_device_bh(device);
+               }
+       }
+}
+
+static void flush_all_alias_devices_on_lcu(struct alias_lcu *lcu)
+{
+       struct alias_pav_group *pavgroup;
+       struct dasd_device *device, *temp;
+       struct dasd_eckd_private *private;
+       int rc;
+       unsigned long flags;
+       LIST_HEAD(active);
+
+       /*
+        * Problem here ist that dasd_flush_device_queue may wait
+        * for termination of a request to complete. We can't keep
+        * the lcu lock during that time, so we must assume that
+        * the lists may have changed.
+        * Idea: first gather all active alias devices in a separate list,
+        * then flush the first element of this list unlocked, and afterwards
+        * check if it is still on the list before moving it to the
+        * active_devices list.
+        */
+
+       spin_lock_irqsave(&lcu->lock, flags);
+       list_for_each_entry_safe(device, temp, &lcu->active_devices,
+                                alias_list) {
+               private = (struct dasd_eckd_private *) device->private;
+               if (private->uid.type == UA_BASE_DEVICE)
+                       continue;
+               list_move(&device->alias_list, &active);
+       }
+
+       list_for_each_entry(pavgroup, &lcu->grouplist, group) {
+               list_splice_init(&pavgroup->aliaslist, &active);
+       }
+       while (!list_empty(&active)) {
+               device = list_first_entry(&active, struct dasd_device,
+                                         alias_list);
+               spin_unlock_irqrestore(&lcu->lock, flags);
+               rc = dasd_flush_device_queue(device);
+               spin_lock_irqsave(&lcu->lock, flags);
+               /*
+                * only move device around if it wasn't moved away while we
+                * were waiting for the flush
+                */
+               if (device == list_first_entry(&active,
+                                              struct dasd_device, alias_list))
+                       list_move(&device->alias_list, &lcu->active_devices);
+       }
+       spin_unlock_irqrestore(&lcu->lock, flags);
+}
+
+static void __stop_device_on_lcu(struct dasd_device *device,
+                                struct dasd_device *pos)
+{
+       /* If pos == device then device is already locked! */
+       if (pos == device) {
+               dasd_device_set_stop_bits(pos, DASD_STOPPED_SU);
+               return;
+       }
+       spin_lock(get_ccwdev_lock(pos->cdev));
+       dasd_device_set_stop_bits(pos, DASD_STOPPED_SU);
+       spin_unlock(get_ccwdev_lock(pos->cdev));
+}
+
+/*
+ * This function is called in interrupt context, so the
+ * cdev lock for device is already locked!
+ */
+static void _stop_all_devices_on_lcu(struct alias_lcu *lcu,
+                                    struct dasd_device *device)
+{
+       struct alias_pav_group *pavgroup;
+       struct dasd_device *pos;
+
+       list_for_each_entry(pos, &lcu->active_devices, alias_list)
+               __stop_device_on_lcu(device, pos);
+       list_for_each_entry(pos, &lcu->inactive_devices, alias_list)
+               __stop_device_on_lcu(device, pos);
+       list_for_each_entry(pavgroup, &lcu->grouplist, group) {
+               list_for_each_entry(pos, &pavgroup->baselist, alias_list)
+                       __stop_device_on_lcu(device, pos);
+               list_for_each_entry(pos, &pavgroup->aliaslist, alias_list)
+                       __stop_device_on_lcu(device, pos);
+       }
+}
+
+static void _unstop_all_devices_on_lcu(struct alias_lcu *lcu)
+{
+       struct alias_pav_group *pavgroup;
+       struct dasd_device *device;
+       unsigned long flags;
+
+       list_for_each_entry(device, &lcu->active_devices, alias_list) {
+               spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
+               dasd_device_remove_stop_bits(device, DASD_STOPPED_SU);
+               spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
+       }
+
+       list_for_each_entry(device, &lcu->inactive_devices, alias_list) {
+               spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
+               dasd_device_remove_stop_bits(device, DASD_STOPPED_SU);
+               spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
+       }
+
+       list_for_each_entry(pavgroup, &lcu->grouplist, group) {
+               list_for_each_entry(device, &pavgroup->baselist, alias_list) {
+                       spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
+                       dasd_device_remove_stop_bits(device, DASD_STOPPED_SU);
+                       spin_unlock_irqrestore(get_ccwdev_lock(device->cdev),
+                                              flags);
+               }
+               list_for_each_entry(device, &pavgroup->aliaslist, alias_list) {
+                       spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
+                       dasd_device_remove_stop_bits(device, DASD_STOPPED_SU);
+                       spin_unlock_irqrestore(get_ccwdev_lock(device->cdev),
+                                              flags);
+               }
+       }
+}
+
+static void summary_unit_check_handling_work(struct work_struct *work)
+{
+       struct alias_lcu *lcu;
+       struct summary_unit_check_work_data *suc_data;
+       unsigned long flags;
+       struct dasd_device *device;
+
+       suc_data = container_of(work, struct summary_unit_check_work_data,
+                               worker);
+       lcu = container_of(suc_data, struct alias_lcu, suc_data);
+       device = suc_data->device;
+
+       /* 1. flush alias devices */
+       flush_all_alias_devices_on_lcu(lcu);
+
+       /* 2. reset summary unit check */
+       spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
+       dasd_device_remove_stop_bits(device,
+                                    (DASD_STOPPED_SU | DASD_STOPPED_PENDING));
+       spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
+       reset_summary_unit_check(lcu, device, suc_data->reason);
+
+       spin_lock_irqsave(&lcu->lock, flags);
+       _unstop_all_devices_on_lcu(lcu);
+       _restart_all_base_devices_on_lcu(lcu);
+       /* 3. read new alias configuration */
+       _schedule_lcu_update(lcu, device);
+       lcu->suc_data.device = NULL;
+       spin_unlock_irqrestore(&lcu->lock, flags);
+}
+
+/*
+ * note: this will be called from int handler context (cdev locked)
+ */
+void dasd_alias_handle_summary_unit_check(struct dasd_device *device,
+                                         struct irb *irb)
+{
+       struct alias_lcu *lcu;
+       char reason;
+       struct dasd_eckd_private *private;
+       char *sense;
+
+       private = (struct dasd_eckd_private *) device->private;
+
+       sense = dasd_get_sense(irb);
+       if (sense) {
+               reason = sense[8];
+               DBF_DEV_EVENT(DBF_NOTICE, device, "%s %x",
+                           "eckd handle summary unit check: reason", reason);
+       } else {
+               DBF_DEV_EVENT(DBF_WARNING, device, "%s",
+                           "eckd handle summary unit check:"
+                           " no reason code available");
+               return;
+       }
+
+       lcu = private->lcu;
+       if (!lcu) {
+               DBF_DEV_EVENT(DBF_WARNING, device, "%s",
+                           "device not ready to handle summary"
+                           " unit check (no lcu structure)");
+               return;
+       }
+       spin_lock(&lcu->lock);
+       _stop_all_devices_on_lcu(lcu, device);
+       /* prepare for lcu_update */
+       private->lcu->flags |= NEED_UAC_UPDATE | UPDATE_PENDING;
+       /* If this device is about to be removed just return and wait for
+        * the next interrupt on a different device
+        */
+       if (list_empty(&device->alias_list)) {
+               DBF_DEV_EVENT(DBF_WARNING, device, "%s",
+                           "device is in offline processing,"
+                           " don't do summary unit check handling");
+               spin_unlock(&lcu->lock);
+               return;
+       }
+       if (lcu->suc_data.device) {
+               /* already scheduled or running */
+               DBF_DEV_EVENT(DBF_WARNING, device, "%s",
+                           "previous instance of summary unit check worker"
+                           " still pending");
+               spin_unlock(&lcu->lock);
+               return ;
+       }
+       lcu->suc_data.reason = reason;
+       lcu->suc_data.device = device;
+       spin_unlock(&lcu->lock);
+       schedule_work(&lcu->suc_data.worker);
+};